diff --git a/config.json b/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..51d1535433b1953b69c9c84331af943303a34171
--- /dev/null
+++ b/config.json
@@ -0,0 +1,40 @@
+{
+  "apply_residual_connection_post_layernorm": false,
+  "architectures": [
+    "TelechatForCausalLM"
+  ],
+  "auto_map": {
+	"AutoConfig": "configuration_telechat.TelechatConfig",
+	"AutoModelForCausalLM": "modeling_telechat.TelechatForCausalLM"
+  },
+  "attention_dropout": 0.0,
+  "attention_softmax_in_fp32": true,
+  "bias_dropout_fusion": true,
+  "bos_token_id": 1,
+  "eos_token_id": 2,
+  "hidden_dropout": 0.0,
+  "hidden_size": 8192,
+  "initializer_range": 0.02,
+  "layer_norm_epsilon": 1e-08,
+  "masked_softmax_fusion": true,
+  "model_type": "telechat",
+  "n_head": 64,
+  "n_inner": null,
+  "num_key_value_heads": 8,
+  "n_layer": 96,
+  "pad_token_id": 3,
+  "pretraining_tp": 2,
+  "skip_bias_add": false,
+  "skip_bias_add_qkv": false,
+  "slow_but_exact": false,
+  "unk_token_id": 0,
+  "use_cache": true,
+  "vocab_size": 131072,
+  "ffn_hidden_size": 40960,
+  "flash_attn":true,
+  "tie_word_embeddings":false,
+  "training_seqlen":8192,
+  "base_seqlen":8192,
+  "seq_length": 8192
+}
+
diff --git a/configuration.json b/configuration.json
new file mode 100644
index 0000000000000000000000000000000000000000..9fbcf957f0338085ab0d7245f17f9fd1a1b64a65
--- /dev/null
+++ b/configuration.json
@@ -0,0 +1 @@
+{"task":"text-generation"}
\ No newline at end of file
diff --git a/configuration_telechat.py b/configuration_telechat.py
new file mode 100644
index 0000000000000000000000000000000000000000..6c6169db242f100ed18215302d25dc375e7e5033
--- /dev/null
+++ b/configuration_telechat.py
@@ -0,0 +1,94 @@
+# coding=utf-8
+# Copyright 2022 the Big Science Workshop and HuggingFace Inc. team.  All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+""" Telechat configuration"""
+
+from packaging import version
+from collections import OrderedDict
+from transformers.utils import is_torch_available, logging
+from transformers.configuration_utils import PretrainedConfig
+from typing import TYPE_CHECKING, Any, List, Mapping, Optional
+
+logger = logging.get_logger(__name__)
+
+class TelechatConfig(PretrainedConfig):
+    """
+    Args:
+        vocab_size (`int`, *optional*, defaults to 160256): Vocabulary size of the Telechat model.
+        hidden_size (`int`, *optional*, defaults to 4096): Dimensionality of the embeddings and hidden states.
+        ffn_hidden_size (`int`, *optional*, defaults to 12288): Dimensionality of the feed-forward hidden states.
+        n_layer (`int`, *optional*, defaults to 30): Number of hidden layers in the Transformer
+        n_head (`int`, *optional*, defaults to 32): Number of attention heads for each attention layer.
+        layer_norm_epsilon (`float`, *optional*, defaults to 1e-5): The epsilon to use in the layer normalization layers.
+        initializer_range (`float`, *optional*, defaults to 0.02): The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
+        apply_residual_connection_post_layernorm (`bool`, *optional*, defaults to `False`): If enabled, use the layer norm of the hidden states as the residual in the transformer blocks
+        hidden_dropout (`float`, *optional*, defaults to 0.0): Dropout rate of the dropout function on the bias dropout.
+        attention_dropout (`float`, *optional*, defaults to 0.0): Dropout rate applied to the attention probs
+        use_cache (`bool`, *optional*, defaults to `True`): Whether or not the model should return the last key/values attentions.
+        training_seqlen (`int`, *optional*, defaults to 8192): Sequence length during last finetuning.
+        logn (`bool`, *optional*, defaults to `True`): Whether or not to use logN during extrapolation.
+        embed_layernorm (`bool`, *optional*, defaults to `True`): Whether or not to use embedding layernorm.
+
+    """
+
+    model_type = "telechat"
+    keys_to_ignore_at_inference = ["past_key_values"]
+    attribute_map = {
+        "num_hidden_layers": "n_layer",
+        "num_attention_heads": "n_head",
+    }
+
+    def __init__(
+        self,
+        vocab_size=160256,
+        hidden_size=4096,
+        n_layer=30,
+        n_head=32,
+        layer_norm_epsilon=1e-5,
+        initializer_range=0.02,
+        use_cache=True,
+        bos_token_id=1,
+        eos_token_id=2,
+        apply_residual_connection_post_layernorm=False,
+        hidden_dropout=0.0,
+        attention_dropout=0.0,
+        ffn_hidden_size=12288,
+        training_seqlen = 8192,
+        logn = True,
+        embed_layernorm = False,
+        **kwargs,
+    ):
+        self.vocab_size = vocab_size
+        n_embed = kwargs.pop("n_embed", None)
+        self.hidden_size = hidden_size if n_embed is None else n_embed
+        self.n_layer = n_layer
+        self.n_head = n_head
+        self.layer_norm_epsilon = layer_norm_epsilon
+        self.initializer_range = initializer_range
+        self.use_cache = use_cache
+        self.apply_residual_connection_post_layernorm = apply_residual_connection_post_layernorm
+        self.hidden_dropout = hidden_dropout
+        self.attention_dropout = attention_dropout
+        self.bos_token_id = bos_token_id
+        self.eos_token_id = eos_token_id
+        self.logn = logn
+        self.ffn_hidden_size = ffn_hidden_size
+        self.training_seqlen = training_seqlen
+        self.embed_layernorm = embed_layernorm
+        self.num_key_value_heads= kwargs.pop("num_key_value_heads", None)
+
+
+        super().__init__(bos_token_id=bos_token_id, eos_token_id=eos_token_id, **kwargs)
+
diff --git a/generation_config.json b/generation_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..67fd08a69f94fd250bd5c50c8905f64e9297ecb8
--- /dev/null
+++ b/generation_config.json
@@ -0,0 +1,14 @@
+{
+  "max_length": 8192,
+  "do_sample": false,
+  "use_cache": true,
+  "temperature": 0.3,
+  "top_k": 5,
+  "top_p": 0.85,
+  "repetition_penalty": 1.03,
+  "pad_token_id": 3,
+  "bos_token_id": 1,
+  "eos_token_id": 2,
+  "user_token_id": 4,
+  "bot_token_id": 5
+}
diff --git a/generation_utils.py b/generation_utils.py
new file mode 100644
index 0000000000000000000000000000000000000000..82410f2eeb3e8ef64f995d7786f2da4419c0f0e7
--- /dev/null
+++ b/generation_utils.py
@@ -0,0 +1,162 @@
+from typing import Optional
+from collections import deque
+from queue import Queue
+import copy
+
+
+class History:
+
+    def __init__(self, tokenizer, history):
+        '''
+        init from a list of dict
+        '''
+        # use deque to meet some special situation
+        self.input_history = deque()
+        self.tokenizer = tokenizer
+        if history:
+            self._transfer_from_list(history)
+
+    def _transfer_from_list(self, history):
+        for message in history:
+            content = message.get("content")
+            # the token result may not be equal to the result model gen
+            message.update(self.tokenizer(content))
+            self.input_history.append(message)
+
+    def append(self, message):
+        content = message.get("content")
+        if "input_ids" not in message or "attention_mask" not in message:
+            message.update(self.tokenizer(content))
+        self.input_history.append(message)
+
+    def append_left(self, message):
+        content = message.get("content")
+        if "input_ids" not in message or "attention_mask" not in message:
+            message.update(self.tokenizer(content))
+        self.input_history.appendleft(message)
+
+    def pop(self):
+        x = self.input_history.pop()
+        return x
+
+    def pop_left(self):
+        x = self.input_history.pop_left()
+        return x
+
+    def update(self, message):
+        self.input_history.pop()
+        self.append(message)
+
+    def __len__(self):
+        return self.input_history.__len__()
+
+    def __str__(self):
+        return self.input_history.__str__()
+
+    def __copy__(self):
+        new_instance = type(self)(self.tokenizer, [])
+        new_instance.input_history = copy.copy(self.input_history)
+        return new_instance
+
+    def __deepcopy__(self, memodict={}):
+        new_instance = type(self)(self.tokenizer, [])
+        new_instance.input_history = copy.deepcopy(self.input_history)
+        return new_instance
+
+
+class TelechatIterTextStreamer:
+    """
+    With reference to the TextIterStreamers in transformers, we have rewritten this class
+    """
+
+    def __init__(
+            self, tokenizer, history: History = None, skip_prompt: bool = False, timeout: Optional[float] = None,
+            **decode_kwargs
+    ):
+
+        self.tokenizer = tokenizer
+        self.history = history
+        self.skip_prompt = skip_prompt
+        self.timeout = timeout
+        self.decode_kwargs = decode_kwargs
+
+        self.text_queue = Queue()
+        self.cache_time = 0
+        self.text_until = ""
+        self.token_until = []
+        self.stop_signal = None
+        self.next_tokens_are_prompt = True
+
+        self.history.append({"role": "bot", "content": self.text_until})
+
+    def put(self, value):
+        """
+        put printable text into queue
+        """
+        if len(value.shape) > 1 and value.shape[0] > 1:
+            raise ValueError("TextStreamer only supports batch size 1")
+        elif len(value.shape) > 1:
+            value = value[0]
+
+        if self.skip_prompt and self.next_tokens_are_prompt:
+            self.next_tokens_are_prompt = False
+            return
+
+        if value[-1] == self.tokenizer.eos_token_id:
+            return
+
+        # there may be some smart way to decode.
+        self.token_until.extend(value.tolist())
+        text = self.tokenizer.decode(self.token_until, **self.decode_kwargs)
+
+
+        if self._is_printable(text) or self.cache_time >= 6:
+            output_text = text[len(self.text_until):]
+            self.text_until = text
+
+        else:
+            self.cache_time+=1
+            return
+
+        self.on_finalized_text(output_text)
+
+    def end(self):
+        """Flushes any remaining cache and prints a newline to stdout."""
+        # Flush the cache, if it exists
+        text = self.tokenizer.decode(self.token_until, **self.decode_kwargs)
+        output_text = text[len(self.text_until):]
+        self.text_until = text
+        self.on_finalized_text(output_text, stream_end=True)
+        self.clear_cache()
+
+    def clear_cache(self):
+        self.cache_time = 0
+        self.token_until = []
+        self.text_until = ""
+        self.history = None
+        self.next_tokens_are_prompt = True
+
+    def on_finalized_text(self, text: str, stream_end: bool = False):
+        """Put the text tuple in the queue."""
+        self.history.update({"role": "bot", "content": self.text_until, "input_ids": self.token_until,
+                             "attention_mask": [1] * len(self.token_until)})
+        self.text_queue.put((text, self.history), timeout=self.timeout)
+        if stream_end:
+            self.text_queue.put((self.stop_signal, self.history), timeout=self.timeout)
+
+    @staticmethod
+    def _is_printable(cp):
+        """Checks whether tokens can be decoded or not"""
+        if "�" in cp:
+            return False
+        return True
+
+    def __iter__(self):
+        return self
+
+    def __next__(self):
+        value_now, history_until = self.text_queue.get(timeout=self.timeout)
+        if value_now == self.stop_signal:
+            raise StopIteration()
+        else:
+            return value_now, history_until
\ No newline at end of file
diff --git a/model-00001-of-00050.safetensors b/model-00001-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..426d53ab14b3b78d70fd6986e3708f0b9b51388d
--- /dev/null
+++ b/model-00001-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:50a4549e91bf98dcb213f6455f92440454fdd012cd0369e0fdb2f019c83fe4c7
+size 8925611272
diff --git a/model-00002-of-00050.safetensors b/model-00002-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..a7eda89723d05ca4657a914a798299392cf46f20
--- /dev/null
+++ b/model-00002-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:548f67873b1f538bf325590f30dbe267600356ccd1bef73c1f1cc7a2902c2aa7
+size 9261287712
diff --git a/model-00003-of-00050.safetensors b/model-00003-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..9da07177d5e2d3d3bf9862fb05de0e0d78370ad3
--- /dev/null
+++ b/model-00003-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e855d3528914e522dcad5348c2229dc2e88fa55e87416f7d5106437d3679a8d3
+size 9261287712
diff --git a/model-00004-of-00050.safetensors b/model-00004-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..0464ad79419900ee9bcf6af5db3290d2a546feba
--- /dev/null
+++ b/model-00004-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:509b7084b8dcb308c7c7d22fa988e805b6f6947e356526f60cccdd40ceb2a11d
+size 9261287712
diff --git a/model-00005-of-00050.safetensors b/model-00005-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..8b7c32a147d32e433942d3f2a714b791b44459ea
--- /dev/null
+++ b/model-00005-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bb044a2e88e6e7a0817459f7f409eeb5abd00836346f69f8ebfd9d0c47028ebf
+size 9261287712
diff --git a/model-00006-of-00050.safetensors b/model-00006-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..011d2601c9ed9767d5723c0213147e7fa9499a08
--- /dev/null
+++ b/model-00006-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a3e410d45493b53ba09cdb8d368ecad07df8c1db96a58d928cd7045c8ed7a48f
+size 9261287728
diff --git a/model-00007-of-00050.safetensors b/model-00007-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..5a5b58baa52f37d9c9be321568bc79faeacce13a
--- /dev/null
+++ b/model-00007-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b9cdcab0486ef69e5eedbb99b08f0b55c918a83f227194cb8b8937f010c950a7
+size 9261287736
diff --git a/model-00008-of-00050.safetensors b/model-00008-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..e4982fc575729421b355442381ea2880be3bc04d
--- /dev/null
+++ b/model-00008-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f2a7cf122475960568ea086f9b98b2283dbf9344dda2720726a52bff7f31edbc
+size 9261287736
diff --git a/model-00009-of-00050.safetensors b/model-00009-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..e9517e6ccc3ca3e12848e779f565fc39a4f920be
--- /dev/null
+++ b/model-00009-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cc5d19f0add344ac71fb3855ca8f04e588fd4d5263e05709ec4dd668ebcf8ee4
+size 9261287736
diff --git a/model-00010-of-00050.safetensors b/model-00010-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..f04889e5eb226789f9e74c0df9161d4251a1b6b2
--- /dev/null
+++ b/model-00010-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5d9763dc2004ea25ca751d271f92a1cce636f92c6991077ea391788428cfc6c1
+size 9261287736
diff --git a/model-00011-of-00050.safetensors b/model-00011-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..1e1e2df0ad9ffaec520448de27d3ad7376eb5044
--- /dev/null
+++ b/model-00011-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ca7e26c5ed4c0fedec09a4cceef1ac7e03d037c921c18d1df0b782ca4fe7b71b
+size 9261287736
diff --git a/model-00012-of-00050.safetensors b/model-00012-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..d51d04ce2d8564fada56337ec00df57e6953487d
--- /dev/null
+++ b/model-00012-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e0390b0e55acc47cebac2bc81e57e1e7d3f02f72a0cad2b01b5165caf976ab0b
+size 9261287736
diff --git a/model-00013-of-00050.safetensors b/model-00013-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..3ae3f60416daf48143e336694ea18a9fed769c22
--- /dev/null
+++ b/model-00013-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bbb3ccd22fc048c59e67dab8df54264acf53d7d62fc98fb49ffeafe7ea181739
+size 9261287736
diff --git a/model-00014-of-00050.safetensors b/model-00014-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..9892baad14cfb948ec60e2f3b547ad273de6dac5
--- /dev/null
+++ b/model-00014-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1e2c53dae773a55d42d3265d0d85ae02c8f6f70543a90d923823cab68d44d5ce
+size 9261287736
diff --git a/model-00015-of-00050.safetensors b/model-00015-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..28f47ab7f8e76554e4295337e8b372e7be705908
--- /dev/null
+++ b/model-00015-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b813a2adab5ff776a06986e60d22e917c596122c417a7d6ae1930b8ed5d61a4f
+size 9261287736
diff --git a/model-00016-of-00050.safetensors b/model-00016-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..1533631fbb8cfb1c1bf074a2af340150acc69a97
--- /dev/null
+++ b/model-00016-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:33976aac9f40e0ff430435fe0295ba1f4443dc4b78233c077de95453eff51d01
+size 9261287736
diff --git a/model-00017-of-00050.safetensors b/model-00017-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..4d003d91f97e4c6b1703c4b015ca9c8665e159d8
--- /dev/null
+++ b/model-00017-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:096850a1c187e804c370956830ef94d20731f79ee1078fa0613604c93015d305
+size 9261287736
diff --git a/model-00018-of-00050.safetensors b/model-00018-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..87b81a764935743e9e0fb9fb2d34a543af72ef87
--- /dev/null
+++ b/model-00018-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b7ed47b73f1171ee30de651924ab58aaa7088037b9f96203be23b4b2514dfb27
+size 9261287736
diff --git a/model-00019-of-00050.safetensors b/model-00019-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..d4db36b8ff81c6314f40c883e0059f70864c1065
--- /dev/null
+++ b/model-00019-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fe1e87f96e41804e06c966605e6b00e0a9c6b51c504ee5e22dfdb28e90d1308f
+size 9261287736
diff --git a/model-00020-of-00050.safetensors b/model-00020-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..18711a652f15f28872d05e599c10bc6e53fe0b6b
--- /dev/null
+++ b/model-00020-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:93bc200e651f67d4910170de85235af15b3228b23192afb5c32d5126daaaf996
+size 9261287736
diff --git a/model-00021-of-00050.safetensors b/model-00021-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..5a1112a2baba2f36f507b60ede9c8a05ddd5fd37
--- /dev/null
+++ b/model-00021-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:645df22333bb21b8cef574bdadcaecaf0efcf43d3e747a0b0c0d462740d0b7c4
+size 9261287736
diff --git a/model-00022-of-00050.safetensors b/model-00022-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..5ff162d27c9d68bbbe2399757e925e52907a3582
--- /dev/null
+++ b/model-00022-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3a288d0dafad8ca11e21c869a31d62057f6bc13036934e306cb68cd7fdd5861f
+size 9261287736
diff --git a/model-00023-of-00050.safetensors b/model-00023-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..0c367f7dbe88ec4f307e559496cca9d517825059
--- /dev/null
+++ b/model-00023-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:94e110e13a08711bfd02cbb3d20e32baecd675f2593a8ae66ba1fcf039e34073
+size 9261287736
diff --git a/model-00024-of-00050.safetensors b/model-00024-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..72909e4735f5e9b59848cec7d4bfa2920275e8c1
--- /dev/null
+++ b/model-00024-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:56894143f2c2a3ff2af49e7f3ec2af9929ff8d95fe7e8e63b709280ba4babe52
+size 9261287736
diff --git a/model-00025-of-00050.safetensors b/model-00025-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..75fb1501f0c277e5b899d9d90a32c5f23a98c881
--- /dev/null
+++ b/model-00025-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:72d4aee125d0a41ceb2afc164c78d25b9edc89cf845da5f91e9c60b743b2f3a7
+size 9261287736
diff --git a/model-00026-of-00050.safetensors b/model-00026-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..51d9835113b524c45ee2fafa75800b4f8627551f
--- /dev/null
+++ b/model-00026-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:72fdd61ca5c065b386092a0701f6fe8ed7ae8d99b0eaba87e50e60acff058876
+size 9261287736
diff --git a/model-00027-of-00050.safetensors b/model-00027-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..4392013dc21137991c5a8a5deab5000abadf7db4
--- /dev/null
+++ b/model-00027-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:764f995de7385ab9c14e7393ee6adc73b557c1a182efdcbbf143c3fe22f353c3
+size 9261287736
diff --git a/model-00028-of-00050.safetensors b/model-00028-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..19149dc27dc0cc7bb6e261e9684389015aed22ac
--- /dev/null
+++ b/model-00028-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ba9bd63f6e97a17e03b467629d9a4bafd581da1b9445b134423c75d5aa409c08
+size 9261287736
diff --git a/model-00029-of-00050.safetensors b/model-00029-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..59260a818ad863f170b8ca7e7742f71b46a1fc52
--- /dev/null
+++ b/model-00029-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:21c3f8b212fef1678419b95693121290430930dd2930bd531aa7a3faa40fac41
+size 9261287736
diff --git a/model-00030-of-00050.safetensors b/model-00030-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..a2e8070dff10c5e27fa805247542da801756b9c9
--- /dev/null
+++ b/model-00030-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b861e871ac1d1ae14b6fcadaefe876c45d694f715eb91d68bacf114c2c426adf
+size 9261287736
diff --git a/model-00031-of-00050.safetensors b/model-00031-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..8e8ee1d2cd088bf29429505bc34949f8c1751be0
--- /dev/null
+++ b/model-00031-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:495ecf892bce401191ce699032b3d95b957d935987ee462df76de2b5928c1127
+size 9261287736
diff --git a/model-00032-of-00050.safetensors b/model-00032-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..f6c1ed3f2979c4d960f167af198cded2596d9020
--- /dev/null
+++ b/model-00032-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2613dc2c31320d4c300e260aac5a313dc183b8fdd140f6b3aea338667126ffae
+size 9261287736
diff --git a/model-00033-of-00050.safetensors b/model-00033-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..db920738a5a9dcdb89c0b025e89363aa48f96ecf
--- /dev/null
+++ b/model-00033-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:67348068c452d63d6ea33f468f810157993b1431c823fc4fc68d1e4c7cf1b653
+size 9261287736
diff --git a/model-00034-of-00050.safetensors b/model-00034-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..f5dd0b2830c789a0af78e05c7b0bff71d3b963ed
--- /dev/null
+++ b/model-00034-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e28056f9391f029ddea9743445dbf763c265feb31a184ba73a7e8de91280d981
+size 9261287736
diff --git a/model-00035-of-00050.safetensors b/model-00035-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..76839e63506bc6c1209ebc20111182ca8944174f
--- /dev/null
+++ b/model-00035-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7c966308e58af945da780ddf143e6f436f21a8019ddb1824200822ac8ecf72c8
+size 9261287736
diff --git a/model-00036-of-00050.safetensors b/model-00036-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..bc7027d7c1398ca59cf58c9724c8dc0a43536196
--- /dev/null
+++ b/model-00036-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:998076e5d150bea2630b24af7bf1da793a951f5c5b9ddfeb7b9400eb1188eea6
+size 9261287736
diff --git a/model-00037-of-00050.safetensors b/model-00037-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..5ee15196b1d3982741e117b6c8a57bebfca0fb6f
--- /dev/null
+++ b/model-00037-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1facd7663e7db14ec43876cc727f3d8e75b40bb025cb7a25fb7d2f9a60de68f9
+size 9261287736
diff --git a/model-00038-of-00050.safetensors b/model-00038-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..3582e0a48dc68a40fd9b79163495df571f7031fa
--- /dev/null
+++ b/model-00038-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c06aafaf597e562ebce7672650df8ae3fcabf22977c4d47004f817ca4f8fe536
+size 9261287736
diff --git a/model-00039-of-00050.safetensors b/model-00039-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..1d299b60693e3ee1cc1745162624fcef38d8981a
--- /dev/null
+++ b/model-00039-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d3bda7f14ffa81266bd9ec07e05e9c3bc16726f8fcd75281d4af4002508bfceb
+size 9261287736
diff --git a/model-00040-of-00050.safetensors b/model-00040-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..4aa7642b76fcea370ed65104c83e33d6ac03aedd
--- /dev/null
+++ b/model-00040-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bf19d0fbf517593390ccc10b1c00c1410b54286ca5c651c1fe3ee8de7b02c4d1
+size 9261287736
diff --git a/model-00041-of-00050.safetensors b/model-00041-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..701adb43db72cc034547664107939802c16c7e7d
--- /dev/null
+++ b/model-00041-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f98f5ea063046592be25906b7dc84f0096e67f104f1bcd847858ee778ba4e1eb
+size 9261287736
diff --git a/model-00042-of-00050.safetensors b/model-00042-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..dc00db6eb6f59f07a92b232674153abaf18d4f29
--- /dev/null
+++ b/model-00042-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:033ba66aecb88ed950b1c49bd2c6f6d5e5173ed4882b22ff189f4caef542ec9f
+size 9261287736
diff --git a/model-00043-of-00050.safetensors b/model-00043-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..621342bbdd38b2a3c23bf85c703e3654994aac63
--- /dev/null
+++ b/model-00043-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5a46195067f9f791eee221c2195d30136f0f8eb723dd0c1674d68c6c2409fa68
+size 9261287736
diff --git a/model-00044-of-00050.safetensors b/model-00044-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..a7601a75167bc353c3914ef3f277d135995ab62e
--- /dev/null
+++ b/model-00044-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d5becd94d0a4e65b2b0496f21ca679ba4b171ffb32851eabd15af17004bda901
+size 9261287736
diff --git a/model-00045-of-00050.safetensors b/model-00045-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..2e27e51423a18e1ce3a30dee2b8fe446255e1056
--- /dev/null
+++ b/model-00045-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:460bd2d228da729c1ef006c65582a8f8d709c748a731f56ff6129ec8beb7cd57
+size 9261287736
diff --git a/model-00046-of-00050.safetensors b/model-00046-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..aa2e89240ef87d36b2e42111b0d2bb517462fd95
--- /dev/null
+++ b/model-00046-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:008d1a031198ab127c27761d605accc0fd48f1d872afb3e0711613b1722cc14f
+size 9261287736
diff --git a/model-00047-of-00050.safetensors b/model-00047-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..77897c6cebb6b7dae751ee810dbeb1058f67da96
--- /dev/null
+++ b/model-00047-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d7f43eceff211a0f79ac9315091a2afc585a10b0aaee46c1b2caa0ca57aaffc4
+size 9261287736
diff --git a/model-00048-of-00050.safetensors b/model-00048-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..adcf5beed7177079df6f99c86dc8b2b294337f39
--- /dev/null
+++ b/model-00048-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c15f954ccbe5a57682184600a4f55fd6c5eb0fe98d18d0ad96ae48c48a2f4e3c
+size 9261287736
diff --git a/model-00049-of-00050.safetensors b/model-00049-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..313d23061828e08fe3682020e1063fd46df66d0c
--- /dev/null
+++ b/model-00049-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9ec2e464fb26c94e5862f52d3b5bcf239e01e8903e587fb1bd416ef2051a225f
+size 4630643872
diff --git a/model-00050-of-00050.safetensors b/model-00050-of-00050.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..cf4495d91a3fc94f49e8a4091294b5728f15e1ab
--- /dev/null
+++ b/model-00050-of-00050.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0dcbfe5b08c5190c2a12bd635d9a23e53aca6d1f1dabc42bae7566d1b121db78
+size 4295000288
diff --git a/model.safetensors.index.json b/model.safetensors.index.json
new file mode 100644
index 0000000000000000000000000000000000000000..6bc2a4c59294fc86fac4dccaab02b0f4d50196f3
--- /dev/null
+++ b/model.safetensors.index.json
@@ -0,0 +1 @@
+{"metadata": {"total_size": 453131665408}, "weight_map": {"transformer.word_embeddings.weight": "model-00001-of-00050.safetensors", "transformer.h.0.input_layernorm.weight": "model-00001-of-00050.safetensors", "transformer.h.0.post_attention_layernorm.weight": "model-00001-of-00050.safetensors", "transformer.h.0.self_attention.dense.weight": "model-00001-of-00050.safetensors", "transformer.h.0.self_attention.dense.bias": "model-00001-of-00050.safetensors", "transformer.h.0.self_attention.query.weight": "model-00001-of-00050.safetensors", "transformer.h.0.self_attention.key_value.weight": "model-00001-of-00050.safetensors", "transformer.h.0.mlp.gate_proj.weight": "model-00001-of-00050.safetensors", "transformer.h.0.mlp.down_proj.weight": "model-00001-of-00050.safetensors", "transformer.h.0.mlp.down_proj.bias": "model-00001-of-00050.safetensors", "transformer.h.0.mlp.up_proj.weight": "model-00001-of-00050.safetensors", "transformer.h.1.input_layernorm.weight": "model-00002-of-00050.safetensors", "transformer.h.1.post_attention_layernorm.weight": "model-00002-of-00050.safetensors", "transformer.h.1.self_attention.dense.weight": "model-00002-of-00050.safetensors", "transformer.h.1.self_attention.dense.bias": "model-00002-of-00050.safetensors", "transformer.h.1.self_attention.query.weight": "model-00002-of-00050.safetensors", "transformer.h.1.self_attention.key_value.weight": "model-00002-of-00050.safetensors", "transformer.h.1.mlp.gate_proj.weight": "model-00002-of-00050.safetensors", "transformer.h.1.mlp.down_proj.weight": "model-00002-of-00050.safetensors", "transformer.h.1.mlp.down_proj.bias": "model-00002-of-00050.safetensors", "transformer.h.1.mlp.up_proj.weight": "model-00002-of-00050.safetensors", "transformer.h.2.input_layernorm.weight": "model-00002-of-00050.safetensors", "transformer.h.2.post_attention_layernorm.weight": "model-00002-of-00050.safetensors", "transformer.h.2.self_attention.dense.weight": "model-00002-of-00050.safetensors", "transformer.h.2.self_attention.dense.bias": "model-00002-of-00050.safetensors", "transformer.h.2.self_attention.query.weight": "model-00002-of-00050.safetensors", "transformer.h.2.self_attention.key_value.weight": "model-00002-of-00050.safetensors", "transformer.h.2.mlp.gate_proj.weight": "model-00002-of-00050.safetensors", "transformer.h.2.mlp.down_proj.weight": "model-00002-of-00050.safetensors", "transformer.h.2.mlp.down_proj.bias": "model-00002-of-00050.safetensors", "transformer.h.2.mlp.up_proj.weight": "model-00002-of-00050.safetensors", "transformer.h.3.input_layernorm.weight": "model-00003-of-00050.safetensors", "transformer.h.3.post_attention_layernorm.weight": "model-00003-of-00050.safetensors", "transformer.h.3.self_attention.dense.weight": "model-00003-of-00050.safetensors", "transformer.h.3.self_attention.dense.bias": "model-00003-of-00050.safetensors", "transformer.h.3.self_attention.query.weight": "model-00003-of-00050.safetensors", "transformer.h.3.self_attention.key_value.weight": "model-00003-of-00050.safetensors", "transformer.h.3.mlp.gate_proj.weight": "model-00003-of-00050.safetensors", "transformer.h.3.mlp.down_proj.weight": "model-00003-of-00050.safetensors", "transformer.h.3.mlp.down_proj.bias": "model-00003-of-00050.safetensors", "transformer.h.3.mlp.up_proj.weight": "model-00003-of-00050.safetensors", "transformer.h.4.input_layernorm.weight": "model-00003-of-00050.safetensors", "transformer.h.4.post_attention_layernorm.weight": "model-00003-of-00050.safetensors", "transformer.h.4.self_attention.dense.weight": "model-00003-of-00050.safetensors", "transformer.h.4.self_attention.dense.bias": "model-00003-of-00050.safetensors", "transformer.h.4.self_attention.query.weight": "model-00003-of-00050.safetensors", "transformer.h.4.self_attention.key_value.weight": "model-00003-of-00050.safetensors", "transformer.h.4.mlp.gate_proj.weight": "model-00003-of-00050.safetensors", "transformer.h.4.mlp.down_proj.weight": "model-00003-of-00050.safetensors", "transformer.h.4.mlp.down_proj.bias": "model-00003-of-00050.safetensors", "transformer.h.4.mlp.up_proj.weight": "model-00003-of-00050.safetensors", "transformer.h.5.input_layernorm.weight": "model-00004-of-00050.safetensors", "transformer.h.5.post_attention_layernorm.weight": "model-00004-of-00050.safetensors", "transformer.h.5.self_attention.dense.weight": "model-00004-of-00050.safetensors", "transformer.h.5.self_attention.dense.bias": "model-00004-of-00050.safetensors", "transformer.h.5.self_attention.query.weight": "model-00004-of-00050.safetensors", "transformer.h.5.self_attention.key_value.weight": "model-00004-of-00050.safetensors", "transformer.h.5.mlp.gate_proj.weight": "model-00004-of-00050.safetensors", "transformer.h.5.mlp.down_proj.weight": "model-00004-of-00050.safetensors", "transformer.h.5.mlp.down_proj.bias": "model-00004-of-00050.safetensors", "transformer.h.5.mlp.up_proj.weight": "model-00004-of-00050.safetensors", "transformer.h.6.input_layernorm.weight": "model-00004-of-00050.safetensors", "transformer.h.6.post_attention_layernorm.weight": "model-00004-of-00050.safetensors", "transformer.h.6.self_attention.dense.weight": "model-00004-of-00050.safetensors", "transformer.h.6.self_attention.dense.bias": "model-00004-of-00050.safetensors", "transformer.h.6.self_attention.query.weight": "model-00004-of-00050.safetensors", "transformer.h.6.self_attention.key_value.weight": "model-00004-of-00050.safetensors", "transformer.h.6.mlp.gate_proj.weight": "model-00004-of-00050.safetensors", "transformer.h.6.mlp.down_proj.weight": "model-00004-of-00050.safetensors", "transformer.h.6.mlp.down_proj.bias": "model-00004-of-00050.safetensors", "transformer.h.6.mlp.up_proj.weight": "model-00004-of-00050.safetensors", "transformer.h.7.input_layernorm.weight": "model-00005-of-00050.safetensors", "transformer.h.7.post_attention_layernorm.weight": "model-00005-of-00050.safetensors", "transformer.h.7.self_attention.dense.weight": "model-00005-of-00050.safetensors", "transformer.h.7.self_attention.dense.bias": "model-00005-of-00050.safetensors", "transformer.h.7.self_attention.query.weight": "model-00005-of-00050.safetensors", "transformer.h.7.self_attention.key_value.weight": "model-00005-of-00050.safetensors", "transformer.h.7.mlp.gate_proj.weight": "model-00005-of-00050.safetensors", "transformer.h.7.mlp.down_proj.weight": "model-00005-of-00050.safetensors", "transformer.h.7.mlp.down_proj.bias": "model-00005-of-00050.safetensors", "transformer.h.7.mlp.up_proj.weight": "model-00005-of-00050.safetensors", "transformer.h.8.input_layernorm.weight": "model-00005-of-00050.safetensors", "transformer.h.8.post_attention_layernorm.weight": "model-00005-of-00050.safetensors", "transformer.h.8.self_attention.dense.weight": "model-00005-of-00050.safetensors", "transformer.h.8.self_attention.dense.bias": "model-00005-of-00050.safetensors", "transformer.h.8.self_attention.query.weight": "model-00005-of-00050.safetensors", "transformer.h.8.self_attention.key_value.weight": "model-00005-of-00050.safetensors", "transformer.h.8.mlp.gate_proj.weight": "model-00005-of-00050.safetensors", "transformer.h.8.mlp.down_proj.weight": "model-00005-of-00050.safetensors", "transformer.h.8.mlp.down_proj.bias": "model-00005-of-00050.safetensors", "transformer.h.8.mlp.up_proj.weight": "model-00005-of-00050.safetensors", "transformer.h.9.input_layernorm.weight": "model-00006-of-00050.safetensors", "transformer.h.9.post_attention_layernorm.weight": "model-00006-of-00050.safetensors", "transformer.h.9.self_attention.dense.weight": "model-00006-of-00050.safetensors", "transformer.h.9.self_attention.dense.bias": "model-00006-of-00050.safetensors", "transformer.h.9.self_attention.query.weight": "model-00006-of-00050.safetensors", "transformer.h.9.self_attention.key_value.weight": "model-00006-of-00050.safetensors", "transformer.h.9.mlp.gate_proj.weight": "model-00006-of-00050.safetensors", "transformer.h.9.mlp.down_proj.weight": "model-00006-of-00050.safetensors", "transformer.h.9.mlp.down_proj.bias": "model-00006-of-00050.safetensors", "transformer.h.9.mlp.up_proj.weight": "model-00006-of-00050.safetensors", "transformer.h.10.input_layernorm.weight": "model-00006-of-00050.safetensors", "transformer.h.10.post_attention_layernorm.weight": "model-00006-of-00050.safetensors", "transformer.h.10.self_attention.dense.weight": "model-00006-of-00050.safetensors", "transformer.h.10.self_attention.dense.bias": "model-00006-of-00050.safetensors", "transformer.h.10.self_attention.query.weight": "model-00006-of-00050.safetensors", "transformer.h.10.self_attention.key_value.weight": "model-00006-of-00050.safetensors", "transformer.h.10.mlp.gate_proj.weight": "model-00006-of-00050.safetensors", "transformer.h.10.mlp.down_proj.weight": "model-00006-of-00050.safetensors", "transformer.h.10.mlp.down_proj.bias": "model-00006-of-00050.safetensors", "transformer.h.10.mlp.up_proj.weight": "model-00006-of-00050.safetensors", "transformer.h.11.input_layernorm.weight": "model-00007-of-00050.safetensors", "transformer.h.11.post_attention_layernorm.weight": "model-00007-of-00050.safetensors", "transformer.h.11.self_attention.dense.weight": "model-00007-of-00050.safetensors", "transformer.h.11.self_attention.dense.bias": "model-00007-of-00050.safetensors", "transformer.h.11.self_attention.query.weight": "model-00007-of-00050.safetensors", "transformer.h.11.self_attention.key_value.weight": "model-00007-of-00050.safetensors", "transformer.h.11.mlp.gate_proj.weight": "model-00007-of-00050.safetensors", "transformer.h.11.mlp.down_proj.weight": "model-00007-of-00050.safetensors", "transformer.h.11.mlp.down_proj.bias": "model-00007-of-00050.safetensors", "transformer.h.11.mlp.up_proj.weight": "model-00007-of-00050.safetensors", "transformer.h.12.input_layernorm.weight": "model-00007-of-00050.safetensors", "transformer.h.12.post_attention_layernorm.weight": "model-00007-of-00050.safetensors", "transformer.h.12.self_attention.dense.weight": "model-00007-of-00050.safetensors", "transformer.h.12.self_attention.dense.bias": "model-00007-of-00050.safetensors", "transformer.h.12.self_attention.query.weight": "model-00007-of-00050.safetensors", "transformer.h.12.self_attention.key_value.weight": "model-00007-of-00050.safetensors", "transformer.h.12.mlp.gate_proj.weight": "model-00007-of-00050.safetensors", "transformer.h.12.mlp.down_proj.weight": "model-00007-of-00050.safetensors", "transformer.h.12.mlp.down_proj.bias": "model-00007-of-00050.safetensors", "transformer.h.12.mlp.up_proj.weight": "model-00007-of-00050.safetensors", "transformer.h.13.input_layernorm.weight": "model-00008-of-00050.safetensors", "transformer.h.13.post_attention_layernorm.weight": "model-00008-of-00050.safetensors", "transformer.h.13.self_attention.dense.weight": "model-00008-of-00050.safetensors", "transformer.h.13.self_attention.dense.bias": "model-00008-of-00050.safetensors", "transformer.h.13.self_attention.query.weight": "model-00008-of-00050.safetensors", "transformer.h.13.self_attention.key_value.weight": "model-00008-of-00050.safetensors", "transformer.h.13.mlp.gate_proj.weight": "model-00008-of-00050.safetensors", "transformer.h.13.mlp.down_proj.weight": "model-00008-of-00050.safetensors", "transformer.h.13.mlp.down_proj.bias": "model-00008-of-00050.safetensors", "transformer.h.13.mlp.up_proj.weight": "model-00008-of-00050.safetensors", "transformer.h.14.input_layernorm.weight": "model-00008-of-00050.safetensors", "transformer.h.14.post_attention_layernorm.weight": "model-00008-of-00050.safetensors", "transformer.h.14.self_attention.dense.weight": "model-00008-of-00050.safetensors", "transformer.h.14.self_attention.dense.bias": "model-00008-of-00050.safetensors", "transformer.h.14.self_attention.query.weight": "model-00008-of-00050.safetensors", "transformer.h.14.self_attention.key_value.weight": "model-00008-of-00050.safetensors", "transformer.h.14.mlp.gate_proj.weight": "model-00008-of-00050.safetensors", "transformer.h.14.mlp.down_proj.weight": "model-00008-of-00050.safetensors", "transformer.h.14.mlp.down_proj.bias": "model-00008-of-00050.safetensors", "transformer.h.14.mlp.up_proj.weight": "model-00008-of-00050.safetensors", "transformer.h.15.input_layernorm.weight": "model-00009-of-00050.safetensors", "transformer.h.15.post_attention_layernorm.weight": "model-00009-of-00050.safetensors", "transformer.h.15.self_attention.dense.weight": "model-00009-of-00050.safetensors", "transformer.h.15.self_attention.dense.bias": "model-00009-of-00050.safetensors", "transformer.h.15.self_attention.query.weight": "model-00009-of-00050.safetensors", "transformer.h.15.self_attention.key_value.weight": "model-00009-of-00050.safetensors", "transformer.h.15.mlp.gate_proj.weight": "model-00009-of-00050.safetensors", "transformer.h.15.mlp.down_proj.weight": "model-00009-of-00050.safetensors", "transformer.h.15.mlp.down_proj.bias": "model-00009-of-00050.safetensors", "transformer.h.15.mlp.up_proj.weight": "model-00009-of-00050.safetensors", "transformer.h.16.input_layernorm.weight": "model-00009-of-00050.safetensors", "transformer.h.16.post_attention_layernorm.weight": "model-00009-of-00050.safetensors", "transformer.h.16.self_attention.dense.weight": "model-00009-of-00050.safetensors", "transformer.h.16.self_attention.dense.bias": "model-00009-of-00050.safetensors", "transformer.h.16.self_attention.query.weight": "model-00009-of-00050.safetensors", "transformer.h.16.self_attention.key_value.weight": "model-00009-of-00050.safetensors", "transformer.h.16.mlp.gate_proj.weight": "model-00009-of-00050.safetensors", "transformer.h.16.mlp.down_proj.weight": "model-00009-of-00050.safetensors", "transformer.h.16.mlp.down_proj.bias": "model-00009-of-00050.safetensors", "transformer.h.16.mlp.up_proj.weight": "model-00009-of-00050.safetensors", "transformer.h.17.input_layernorm.weight": "model-00010-of-00050.safetensors", "transformer.h.17.post_attention_layernorm.weight": "model-00010-of-00050.safetensors", "transformer.h.17.self_attention.dense.weight": "model-00010-of-00050.safetensors", "transformer.h.17.self_attention.dense.bias": "model-00010-of-00050.safetensors", "transformer.h.17.self_attention.query.weight": "model-00010-of-00050.safetensors", "transformer.h.17.self_attention.key_value.weight": "model-00010-of-00050.safetensors", "transformer.h.17.mlp.gate_proj.weight": "model-00010-of-00050.safetensors", "transformer.h.17.mlp.down_proj.weight": "model-00010-of-00050.safetensors", "transformer.h.17.mlp.down_proj.bias": "model-00010-of-00050.safetensors", "transformer.h.17.mlp.up_proj.weight": "model-00010-of-00050.safetensors", "transformer.h.18.input_layernorm.weight": "model-00010-of-00050.safetensors", "transformer.h.18.post_attention_layernorm.weight": "model-00010-of-00050.safetensors", "transformer.h.18.self_attention.dense.weight": "model-00010-of-00050.safetensors", "transformer.h.18.self_attention.dense.bias": "model-00010-of-00050.safetensors", "transformer.h.18.self_attention.query.weight": "model-00010-of-00050.safetensors", "transformer.h.18.self_attention.key_value.weight": "model-00010-of-00050.safetensors", "transformer.h.18.mlp.gate_proj.weight": "model-00010-of-00050.safetensors", "transformer.h.18.mlp.down_proj.weight": "model-00010-of-00050.safetensors", "transformer.h.18.mlp.down_proj.bias": "model-00010-of-00050.safetensors", "transformer.h.18.mlp.up_proj.weight": "model-00010-of-00050.safetensors", "transformer.h.19.input_layernorm.weight": "model-00011-of-00050.safetensors", "transformer.h.19.post_attention_layernorm.weight": "model-00011-of-00050.safetensors", "transformer.h.19.self_attention.dense.weight": "model-00011-of-00050.safetensors", "transformer.h.19.self_attention.dense.bias": "model-00011-of-00050.safetensors", "transformer.h.19.self_attention.query.weight": "model-00011-of-00050.safetensors", "transformer.h.19.self_attention.key_value.weight": "model-00011-of-00050.safetensors", "transformer.h.19.mlp.gate_proj.weight": "model-00011-of-00050.safetensors", "transformer.h.19.mlp.down_proj.weight": "model-00011-of-00050.safetensors", "transformer.h.19.mlp.down_proj.bias": "model-00011-of-00050.safetensors", "transformer.h.19.mlp.up_proj.weight": "model-00011-of-00050.safetensors", "transformer.h.20.input_layernorm.weight": "model-00011-of-00050.safetensors", "transformer.h.20.post_attention_layernorm.weight": "model-00011-of-00050.safetensors", "transformer.h.20.self_attention.dense.weight": "model-00011-of-00050.safetensors", "transformer.h.20.self_attention.dense.bias": "model-00011-of-00050.safetensors", "transformer.h.20.self_attention.query.weight": "model-00011-of-00050.safetensors", "transformer.h.20.self_attention.key_value.weight": "model-00011-of-00050.safetensors", "transformer.h.20.mlp.gate_proj.weight": "model-00011-of-00050.safetensors", "transformer.h.20.mlp.down_proj.weight": "model-00011-of-00050.safetensors", "transformer.h.20.mlp.down_proj.bias": "model-00011-of-00050.safetensors", "transformer.h.20.mlp.up_proj.weight": "model-00011-of-00050.safetensors", "transformer.h.21.input_layernorm.weight": "model-00012-of-00050.safetensors", "transformer.h.21.post_attention_layernorm.weight": "model-00012-of-00050.safetensors", "transformer.h.21.self_attention.dense.weight": "model-00012-of-00050.safetensors", "transformer.h.21.self_attention.dense.bias": "model-00012-of-00050.safetensors", "transformer.h.21.self_attention.query.weight": "model-00012-of-00050.safetensors", "transformer.h.21.self_attention.key_value.weight": "model-00012-of-00050.safetensors", "transformer.h.21.mlp.gate_proj.weight": "model-00012-of-00050.safetensors", "transformer.h.21.mlp.down_proj.weight": "model-00012-of-00050.safetensors", "transformer.h.21.mlp.down_proj.bias": "model-00012-of-00050.safetensors", "transformer.h.21.mlp.up_proj.weight": "model-00012-of-00050.safetensors", "transformer.h.22.input_layernorm.weight": "model-00012-of-00050.safetensors", "transformer.h.22.post_attention_layernorm.weight": "model-00012-of-00050.safetensors", "transformer.h.22.self_attention.dense.weight": "model-00012-of-00050.safetensors", "transformer.h.22.self_attention.dense.bias": "model-00012-of-00050.safetensors", "transformer.h.22.self_attention.query.weight": "model-00012-of-00050.safetensors", "transformer.h.22.self_attention.key_value.weight": "model-00012-of-00050.safetensors", "transformer.h.22.mlp.gate_proj.weight": "model-00012-of-00050.safetensors", "transformer.h.22.mlp.down_proj.weight": "model-00012-of-00050.safetensors", "transformer.h.22.mlp.down_proj.bias": "model-00012-of-00050.safetensors", "transformer.h.22.mlp.up_proj.weight": "model-00012-of-00050.safetensors", "transformer.h.23.input_layernorm.weight": "model-00013-of-00050.safetensors", "transformer.h.23.post_attention_layernorm.weight": "model-00013-of-00050.safetensors", "transformer.h.23.self_attention.dense.weight": "model-00013-of-00050.safetensors", "transformer.h.23.self_attention.dense.bias": "model-00013-of-00050.safetensors", "transformer.h.23.self_attention.query.weight": "model-00013-of-00050.safetensors", "transformer.h.23.self_attention.key_value.weight": "model-00013-of-00050.safetensors", "transformer.h.23.mlp.gate_proj.weight": "model-00013-of-00050.safetensors", "transformer.h.23.mlp.down_proj.weight": "model-00013-of-00050.safetensors", "transformer.h.23.mlp.down_proj.bias": "model-00013-of-00050.safetensors", "transformer.h.23.mlp.up_proj.weight": "model-00013-of-00050.safetensors", "transformer.h.24.input_layernorm.weight": "model-00013-of-00050.safetensors", "transformer.h.24.post_attention_layernorm.weight": "model-00013-of-00050.safetensors", "transformer.h.24.self_attention.dense.weight": "model-00013-of-00050.safetensors", "transformer.h.24.self_attention.dense.bias": "model-00013-of-00050.safetensors", "transformer.h.24.self_attention.query.weight": "model-00013-of-00050.safetensors", "transformer.h.24.self_attention.key_value.weight": "model-00013-of-00050.safetensors", "transformer.h.24.mlp.gate_proj.weight": "model-00013-of-00050.safetensors", "transformer.h.24.mlp.down_proj.weight": "model-00013-of-00050.safetensors", "transformer.h.24.mlp.down_proj.bias": "model-00013-of-00050.safetensors", "transformer.h.24.mlp.up_proj.weight": "model-00013-of-00050.safetensors", "transformer.h.25.input_layernorm.weight": "model-00014-of-00050.safetensors", "transformer.h.25.post_attention_layernorm.weight": "model-00014-of-00050.safetensors", "transformer.h.25.self_attention.dense.weight": "model-00014-of-00050.safetensors", "transformer.h.25.self_attention.dense.bias": "model-00014-of-00050.safetensors", "transformer.h.25.self_attention.query.weight": "model-00014-of-00050.safetensors", "transformer.h.25.self_attention.key_value.weight": "model-00014-of-00050.safetensors", "transformer.h.25.mlp.gate_proj.weight": "model-00014-of-00050.safetensors", "transformer.h.25.mlp.down_proj.weight": "model-00014-of-00050.safetensors", "transformer.h.25.mlp.down_proj.bias": "model-00014-of-00050.safetensors", "transformer.h.25.mlp.up_proj.weight": "model-00014-of-00050.safetensors", "transformer.h.26.input_layernorm.weight": "model-00014-of-00050.safetensors", "transformer.h.26.post_attention_layernorm.weight": "model-00014-of-00050.safetensors", "transformer.h.26.self_attention.dense.weight": "model-00014-of-00050.safetensors", "transformer.h.26.self_attention.dense.bias": "model-00014-of-00050.safetensors", "transformer.h.26.self_attention.query.weight": "model-00014-of-00050.safetensors", "transformer.h.26.self_attention.key_value.weight": "model-00014-of-00050.safetensors", "transformer.h.26.mlp.gate_proj.weight": "model-00014-of-00050.safetensors", "transformer.h.26.mlp.down_proj.weight": "model-00014-of-00050.safetensors", "transformer.h.26.mlp.down_proj.bias": "model-00014-of-00050.safetensors", "transformer.h.26.mlp.up_proj.weight": "model-00014-of-00050.safetensors", "transformer.h.27.input_layernorm.weight": "model-00015-of-00050.safetensors", "transformer.h.27.post_attention_layernorm.weight": "model-00015-of-00050.safetensors", "transformer.h.27.self_attention.dense.weight": "model-00015-of-00050.safetensors", "transformer.h.27.self_attention.dense.bias": "model-00015-of-00050.safetensors", "transformer.h.27.self_attention.query.weight": "model-00015-of-00050.safetensors", "transformer.h.27.self_attention.key_value.weight": "model-00015-of-00050.safetensors", "transformer.h.27.mlp.gate_proj.weight": "model-00015-of-00050.safetensors", "transformer.h.27.mlp.down_proj.weight": "model-00015-of-00050.safetensors", "transformer.h.27.mlp.down_proj.bias": "model-00015-of-00050.safetensors", "transformer.h.27.mlp.up_proj.weight": "model-00015-of-00050.safetensors", "transformer.h.28.input_layernorm.weight": "model-00015-of-00050.safetensors", "transformer.h.28.post_attention_layernorm.weight": "model-00015-of-00050.safetensors", "transformer.h.28.self_attention.dense.weight": "model-00015-of-00050.safetensors", "transformer.h.28.self_attention.dense.bias": "model-00015-of-00050.safetensors", "transformer.h.28.self_attention.query.weight": "model-00015-of-00050.safetensors", "transformer.h.28.self_attention.key_value.weight": "model-00015-of-00050.safetensors", "transformer.h.28.mlp.gate_proj.weight": "model-00015-of-00050.safetensors", "transformer.h.28.mlp.down_proj.weight": "model-00015-of-00050.safetensors", "transformer.h.28.mlp.down_proj.bias": "model-00015-of-00050.safetensors", "transformer.h.28.mlp.up_proj.weight": "model-00015-of-00050.safetensors", "transformer.h.29.input_layernorm.weight": "model-00016-of-00050.safetensors", "transformer.h.29.post_attention_layernorm.weight": "model-00016-of-00050.safetensors", "transformer.h.29.self_attention.dense.weight": "model-00016-of-00050.safetensors", "transformer.h.29.self_attention.dense.bias": "model-00016-of-00050.safetensors", "transformer.h.29.self_attention.query.weight": "model-00016-of-00050.safetensors", "transformer.h.29.self_attention.key_value.weight": "model-00016-of-00050.safetensors", "transformer.h.29.mlp.gate_proj.weight": "model-00016-of-00050.safetensors", "transformer.h.29.mlp.down_proj.weight": "model-00016-of-00050.safetensors", "transformer.h.29.mlp.down_proj.bias": "model-00016-of-00050.safetensors", "transformer.h.29.mlp.up_proj.weight": "model-00016-of-00050.safetensors", "transformer.h.30.input_layernorm.weight": "model-00016-of-00050.safetensors", "transformer.h.30.post_attention_layernorm.weight": "model-00016-of-00050.safetensors", "transformer.h.30.self_attention.dense.weight": "model-00016-of-00050.safetensors", "transformer.h.30.self_attention.dense.bias": "model-00016-of-00050.safetensors", "transformer.h.30.self_attention.query.weight": "model-00016-of-00050.safetensors", "transformer.h.30.self_attention.key_value.weight": "model-00016-of-00050.safetensors", "transformer.h.30.mlp.gate_proj.weight": "model-00016-of-00050.safetensors", "transformer.h.30.mlp.down_proj.weight": "model-00016-of-00050.safetensors", "transformer.h.30.mlp.down_proj.bias": "model-00016-of-00050.safetensors", "transformer.h.30.mlp.up_proj.weight": "model-00016-of-00050.safetensors", "transformer.h.31.input_layernorm.weight": "model-00017-of-00050.safetensors", "transformer.h.31.post_attention_layernorm.weight": "model-00017-of-00050.safetensors", "transformer.h.31.self_attention.dense.weight": "model-00017-of-00050.safetensors", "transformer.h.31.self_attention.dense.bias": "model-00017-of-00050.safetensors", "transformer.h.31.self_attention.query.weight": "model-00017-of-00050.safetensors", "transformer.h.31.self_attention.key_value.weight": "model-00017-of-00050.safetensors", "transformer.h.31.mlp.gate_proj.weight": "model-00017-of-00050.safetensors", "transformer.h.31.mlp.down_proj.weight": "model-00017-of-00050.safetensors", "transformer.h.31.mlp.down_proj.bias": "model-00017-of-00050.safetensors", "transformer.h.31.mlp.up_proj.weight": "model-00017-of-00050.safetensors", "transformer.h.32.input_layernorm.weight": "model-00017-of-00050.safetensors", "transformer.h.32.post_attention_layernorm.weight": "model-00017-of-00050.safetensors", "transformer.h.32.self_attention.dense.weight": "model-00017-of-00050.safetensors", "transformer.h.32.self_attention.dense.bias": "model-00017-of-00050.safetensors", "transformer.h.32.self_attention.query.weight": "model-00017-of-00050.safetensors", "transformer.h.32.self_attention.key_value.weight": "model-00017-of-00050.safetensors", "transformer.h.32.mlp.gate_proj.weight": "model-00017-of-00050.safetensors", "transformer.h.32.mlp.down_proj.weight": "model-00017-of-00050.safetensors", "transformer.h.32.mlp.down_proj.bias": "model-00017-of-00050.safetensors", "transformer.h.32.mlp.up_proj.weight": "model-00017-of-00050.safetensors", "transformer.h.33.input_layernorm.weight": "model-00018-of-00050.safetensors", "transformer.h.33.post_attention_layernorm.weight": "model-00018-of-00050.safetensors", "transformer.h.33.self_attention.dense.weight": "model-00018-of-00050.safetensors", "transformer.h.33.self_attention.dense.bias": "model-00018-of-00050.safetensors", "transformer.h.33.self_attention.query.weight": "model-00018-of-00050.safetensors", "transformer.h.33.self_attention.key_value.weight": "model-00018-of-00050.safetensors", "transformer.h.33.mlp.gate_proj.weight": "model-00018-of-00050.safetensors", "transformer.h.33.mlp.down_proj.weight": "model-00018-of-00050.safetensors", "transformer.h.33.mlp.down_proj.bias": "model-00018-of-00050.safetensors", "transformer.h.33.mlp.up_proj.weight": "model-00018-of-00050.safetensors", "transformer.h.34.input_layernorm.weight": "model-00018-of-00050.safetensors", "transformer.h.34.post_attention_layernorm.weight": "model-00018-of-00050.safetensors", "transformer.h.34.self_attention.dense.weight": "model-00018-of-00050.safetensors", "transformer.h.34.self_attention.dense.bias": "model-00018-of-00050.safetensors", "transformer.h.34.self_attention.query.weight": "model-00018-of-00050.safetensors", "transformer.h.34.self_attention.key_value.weight": "model-00018-of-00050.safetensors", "transformer.h.34.mlp.gate_proj.weight": "model-00018-of-00050.safetensors", "transformer.h.34.mlp.down_proj.weight": "model-00018-of-00050.safetensors", "transformer.h.34.mlp.down_proj.bias": "model-00018-of-00050.safetensors", "transformer.h.34.mlp.up_proj.weight": "model-00018-of-00050.safetensors", "transformer.h.35.input_layernorm.weight": "model-00019-of-00050.safetensors", "transformer.h.35.post_attention_layernorm.weight": "model-00019-of-00050.safetensors", "transformer.h.35.self_attention.dense.weight": "model-00019-of-00050.safetensors", "transformer.h.35.self_attention.dense.bias": "model-00019-of-00050.safetensors", "transformer.h.35.self_attention.query.weight": "model-00019-of-00050.safetensors", "transformer.h.35.self_attention.key_value.weight": "model-00019-of-00050.safetensors", "transformer.h.35.mlp.gate_proj.weight": "model-00019-of-00050.safetensors", "transformer.h.35.mlp.down_proj.weight": "model-00019-of-00050.safetensors", "transformer.h.35.mlp.down_proj.bias": "model-00019-of-00050.safetensors", "transformer.h.35.mlp.up_proj.weight": "model-00019-of-00050.safetensors", "transformer.h.36.input_layernorm.weight": "model-00019-of-00050.safetensors", "transformer.h.36.post_attention_layernorm.weight": "model-00019-of-00050.safetensors", "transformer.h.36.self_attention.dense.weight": "model-00019-of-00050.safetensors", "transformer.h.36.self_attention.dense.bias": "model-00019-of-00050.safetensors", "transformer.h.36.self_attention.query.weight": "model-00019-of-00050.safetensors", "transformer.h.36.self_attention.key_value.weight": "model-00019-of-00050.safetensors", "transformer.h.36.mlp.gate_proj.weight": "model-00019-of-00050.safetensors", "transformer.h.36.mlp.down_proj.weight": "model-00019-of-00050.safetensors", "transformer.h.36.mlp.down_proj.bias": "model-00019-of-00050.safetensors", "transformer.h.36.mlp.up_proj.weight": "model-00019-of-00050.safetensors", "transformer.h.37.input_layernorm.weight": "model-00020-of-00050.safetensors", "transformer.h.37.post_attention_layernorm.weight": "model-00020-of-00050.safetensors", "transformer.h.37.self_attention.dense.weight": "model-00020-of-00050.safetensors", "transformer.h.37.self_attention.dense.bias": "model-00020-of-00050.safetensors", "transformer.h.37.self_attention.query.weight": "model-00020-of-00050.safetensors", "transformer.h.37.self_attention.key_value.weight": "model-00020-of-00050.safetensors", "transformer.h.37.mlp.gate_proj.weight": "model-00020-of-00050.safetensors", "transformer.h.37.mlp.down_proj.weight": "model-00020-of-00050.safetensors", "transformer.h.37.mlp.down_proj.bias": "model-00020-of-00050.safetensors", "transformer.h.37.mlp.up_proj.weight": "model-00020-of-00050.safetensors", "transformer.h.38.input_layernorm.weight": "model-00020-of-00050.safetensors", "transformer.h.38.post_attention_layernorm.weight": "model-00020-of-00050.safetensors", "transformer.h.38.self_attention.dense.weight": "model-00020-of-00050.safetensors", "transformer.h.38.self_attention.dense.bias": "model-00020-of-00050.safetensors", "transformer.h.38.self_attention.query.weight": "model-00020-of-00050.safetensors", "transformer.h.38.self_attention.key_value.weight": "model-00020-of-00050.safetensors", "transformer.h.38.mlp.gate_proj.weight": "model-00020-of-00050.safetensors", "transformer.h.38.mlp.down_proj.weight": "model-00020-of-00050.safetensors", "transformer.h.38.mlp.down_proj.bias": "model-00020-of-00050.safetensors", "transformer.h.38.mlp.up_proj.weight": "model-00020-of-00050.safetensors", "transformer.h.39.input_layernorm.weight": "model-00021-of-00050.safetensors", "transformer.h.39.post_attention_layernorm.weight": "model-00021-of-00050.safetensors", "transformer.h.39.self_attention.dense.weight": "model-00021-of-00050.safetensors", "transformer.h.39.self_attention.dense.bias": "model-00021-of-00050.safetensors", "transformer.h.39.self_attention.query.weight": "model-00021-of-00050.safetensors", "transformer.h.39.self_attention.key_value.weight": "model-00021-of-00050.safetensors", "transformer.h.39.mlp.gate_proj.weight": "model-00021-of-00050.safetensors", "transformer.h.39.mlp.down_proj.weight": "model-00021-of-00050.safetensors", "transformer.h.39.mlp.down_proj.bias": "model-00021-of-00050.safetensors", "transformer.h.39.mlp.up_proj.weight": "model-00021-of-00050.safetensors", "transformer.h.40.input_layernorm.weight": "model-00021-of-00050.safetensors", "transformer.h.40.post_attention_layernorm.weight": "model-00021-of-00050.safetensors", "transformer.h.40.self_attention.dense.weight": "model-00021-of-00050.safetensors", "transformer.h.40.self_attention.dense.bias": "model-00021-of-00050.safetensors", "transformer.h.40.self_attention.query.weight": "model-00021-of-00050.safetensors", "transformer.h.40.self_attention.key_value.weight": "model-00021-of-00050.safetensors", "transformer.h.40.mlp.gate_proj.weight": "model-00021-of-00050.safetensors", "transformer.h.40.mlp.down_proj.weight": "model-00021-of-00050.safetensors", "transformer.h.40.mlp.down_proj.bias": "model-00021-of-00050.safetensors", "transformer.h.40.mlp.up_proj.weight": "model-00021-of-00050.safetensors", "transformer.h.41.input_layernorm.weight": "model-00022-of-00050.safetensors", "transformer.h.41.post_attention_layernorm.weight": "model-00022-of-00050.safetensors", "transformer.h.41.self_attention.dense.weight": "model-00022-of-00050.safetensors", "transformer.h.41.self_attention.dense.bias": "model-00022-of-00050.safetensors", "transformer.h.41.self_attention.query.weight": "model-00022-of-00050.safetensors", "transformer.h.41.self_attention.key_value.weight": "model-00022-of-00050.safetensors", "transformer.h.41.mlp.gate_proj.weight": "model-00022-of-00050.safetensors", "transformer.h.41.mlp.down_proj.weight": "model-00022-of-00050.safetensors", "transformer.h.41.mlp.down_proj.bias": "model-00022-of-00050.safetensors", "transformer.h.41.mlp.up_proj.weight": "model-00022-of-00050.safetensors", "transformer.h.42.input_layernorm.weight": "model-00022-of-00050.safetensors", "transformer.h.42.post_attention_layernorm.weight": "model-00022-of-00050.safetensors", "transformer.h.42.self_attention.dense.weight": "model-00022-of-00050.safetensors", "transformer.h.42.self_attention.dense.bias": "model-00022-of-00050.safetensors", "transformer.h.42.self_attention.query.weight": "model-00022-of-00050.safetensors", "transformer.h.42.self_attention.key_value.weight": "model-00022-of-00050.safetensors", "transformer.h.42.mlp.gate_proj.weight": "model-00022-of-00050.safetensors", "transformer.h.42.mlp.down_proj.weight": "model-00022-of-00050.safetensors", "transformer.h.42.mlp.down_proj.bias": "model-00022-of-00050.safetensors", "transformer.h.42.mlp.up_proj.weight": "model-00022-of-00050.safetensors", "transformer.h.43.input_layernorm.weight": "model-00023-of-00050.safetensors", "transformer.h.43.post_attention_layernorm.weight": "model-00023-of-00050.safetensors", "transformer.h.43.self_attention.dense.weight": "model-00023-of-00050.safetensors", "transformer.h.43.self_attention.dense.bias": "model-00023-of-00050.safetensors", "transformer.h.43.self_attention.query.weight": "model-00023-of-00050.safetensors", "transformer.h.43.self_attention.key_value.weight": "model-00023-of-00050.safetensors", "transformer.h.43.mlp.gate_proj.weight": "model-00023-of-00050.safetensors", "transformer.h.43.mlp.down_proj.weight": "model-00023-of-00050.safetensors", "transformer.h.43.mlp.down_proj.bias": "model-00023-of-00050.safetensors", "transformer.h.43.mlp.up_proj.weight": "model-00023-of-00050.safetensors", "transformer.h.44.input_layernorm.weight": "model-00023-of-00050.safetensors", "transformer.h.44.post_attention_layernorm.weight": "model-00023-of-00050.safetensors", "transformer.h.44.self_attention.dense.weight": "model-00023-of-00050.safetensors", "transformer.h.44.self_attention.dense.bias": "model-00023-of-00050.safetensors", "transformer.h.44.self_attention.query.weight": "model-00023-of-00050.safetensors", "transformer.h.44.self_attention.key_value.weight": "model-00023-of-00050.safetensors", "transformer.h.44.mlp.gate_proj.weight": "model-00023-of-00050.safetensors", "transformer.h.44.mlp.down_proj.weight": "model-00023-of-00050.safetensors", "transformer.h.44.mlp.down_proj.bias": "model-00023-of-00050.safetensors", "transformer.h.44.mlp.up_proj.weight": "model-00023-of-00050.safetensors", "transformer.h.45.input_layernorm.weight": "model-00024-of-00050.safetensors", "transformer.h.45.post_attention_layernorm.weight": "model-00024-of-00050.safetensors", "transformer.h.45.self_attention.dense.weight": "model-00024-of-00050.safetensors", "transformer.h.45.self_attention.dense.bias": "model-00024-of-00050.safetensors", "transformer.h.45.self_attention.query.weight": "model-00024-of-00050.safetensors", "transformer.h.45.self_attention.key_value.weight": "model-00024-of-00050.safetensors", "transformer.h.45.mlp.gate_proj.weight": "model-00024-of-00050.safetensors", "transformer.h.45.mlp.down_proj.weight": "model-00024-of-00050.safetensors", "transformer.h.45.mlp.down_proj.bias": "model-00024-of-00050.safetensors", "transformer.h.45.mlp.up_proj.weight": "model-00024-of-00050.safetensors", "transformer.h.46.input_layernorm.weight": "model-00024-of-00050.safetensors", "transformer.h.46.post_attention_layernorm.weight": "model-00024-of-00050.safetensors", "transformer.h.46.self_attention.dense.weight": "model-00024-of-00050.safetensors", "transformer.h.46.self_attention.dense.bias": "model-00024-of-00050.safetensors", "transformer.h.46.self_attention.query.weight": "model-00024-of-00050.safetensors", "transformer.h.46.self_attention.key_value.weight": "model-00024-of-00050.safetensors", "transformer.h.46.mlp.gate_proj.weight": "model-00024-of-00050.safetensors", "transformer.h.46.mlp.down_proj.weight": "model-00024-of-00050.safetensors", "transformer.h.46.mlp.down_proj.bias": "model-00024-of-00050.safetensors", "transformer.h.46.mlp.up_proj.weight": "model-00024-of-00050.safetensors", "transformer.h.47.input_layernorm.weight": "model-00025-of-00050.safetensors", "transformer.h.47.post_attention_layernorm.weight": "model-00025-of-00050.safetensors", "transformer.h.47.self_attention.dense.weight": "model-00025-of-00050.safetensors", "transformer.h.47.self_attention.dense.bias": "model-00025-of-00050.safetensors", "transformer.h.47.self_attention.query.weight": "model-00025-of-00050.safetensors", "transformer.h.47.self_attention.key_value.weight": "model-00025-of-00050.safetensors", "transformer.h.47.mlp.gate_proj.weight": "model-00025-of-00050.safetensors", "transformer.h.47.mlp.down_proj.weight": "model-00025-of-00050.safetensors", "transformer.h.47.mlp.down_proj.bias": "model-00025-of-00050.safetensors", "transformer.h.47.mlp.up_proj.weight": "model-00025-of-00050.safetensors", "transformer.h.48.input_layernorm.weight": "model-00025-of-00050.safetensors", "transformer.h.48.post_attention_layernorm.weight": "model-00025-of-00050.safetensors", "transformer.h.48.self_attention.dense.weight": "model-00025-of-00050.safetensors", "transformer.h.48.self_attention.dense.bias": "model-00025-of-00050.safetensors", "transformer.h.48.self_attention.query.weight": "model-00025-of-00050.safetensors", "transformer.h.48.self_attention.key_value.weight": "model-00025-of-00050.safetensors", "transformer.h.48.mlp.gate_proj.weight": "model-00025-of-00050.safetensors", "transformer.h.48.mlp.down_proj.weight": "model-00025-of-00050.safetensors", "transformer.h.48.mlp.down_proj.bias": "model-00025-of-00050.safetensors", "transformer.h.48.mlp.up_proj.weight": "model-00025-of-00050.safetensors", "transformer.h.49.input_layernorm.weight": "model-00026-of-00050.safetensors", "transformer.h.49.post_attention_layernorm.weight": "model-00026-of-00050.safetensors", "transformer.h.49.self_attention.dense.weight": "model-00026-of-00050.safetensors", "transformer.h.49.self_attention.dense.bias": "model-00026-of-00050.safetensors", "transformer.h.49.self_attention.query.weight": "model-00026-of-00050.safetensors", "transformer.h.49.self_attention.key_value.weight": "model-00026-of-00050.safetensors", "transformer.h.49.mlp.gate_proj.weight": "model-00026-of-00050.safetensors", "transformer.h.49.mlp.down_proj.weight": "model-00026-of-00050.safetensors", "transformer.h.49.mlp.down_proj.bias": "model-00026-of-00050.safetensors", "transformer.h.49.mlp.up_proj.weight": "model-00026-of-00050.safetensors", "transformer.h.50.input_layernorm.weight": "model-00026-of-00050.safetensors", "transformer.h.50.post_attention_layernorm.weight": "model-00026-of-00050.safetensors", "transformer.h.50.self_attention.dense.weight": "model-00026-of-00050.safetensors", "transformer.h.50.self_attention.dense.bias": "model-00026-of-00050.safetensors", "transformer.h.50.self_attention.query.weight": "model-00026-of-00050.safetensors", "transformer.h.50.self_attention.key_value.weight": "model-00026-of-00050.safetensors", "transformer.h.50.mlp.gate_proj.weight": "model-00026-of-00050.safetensors", "transformer.h.50.mlp.down_proj.weight": "model-00026-of-00050.safetensors", "transformer.h.50.mlp.down_proj.bias": "model-00026-of-00050.safetensors", "transformer.h.50.mlp.up_proj.weight": "model-00026-of-00050.safetensors", "transformer.h.51.input_layernorm.weight": "model-00027-of-00050.safetensors", "transformer.h.51.post_attention_layernorm.weight": "model-00027-of-00050.safetensors", "transformer.h.51.self_attention.dense.weight": "model-00027-of-00050.safetensors", "transformer.h.51.self_attention.dense.bias": "model-00027-of-00050.safetensors", "transformer.h.51.self_attention.query.weight": "model-00027-of-00050.safetensors", "transformer.h.51.self_attention.key_value.weight": "model-00027-of-00050.safetensors", "transformer.h.51.mlp.gate_proj.weight": "model-00027-of-00050.safetensors", "transformer.h.51.mlp.down_proj.weight": "model-00027-of-00050.safetensors", "transformer.h.51.mlp.down_proj.bias": "model-00027-of-00050.safetensors", "transformer.h.51.mlp.up_proj.weight": "model-00027-of-00050.safetensors", "transformer.h.52.input_layernorm.weight": "model-00027-of-00050.safetensors", "transformer.h.52.post_attention_layernorm.weight": "model-00027-of-00050.safetensors", "transformer.h.52.self_attention.dense.weight": "model-00027-of-00050.safetensors", "transformer.h.52.self_attention.dense.bias": "model-00027-of-00050.safetensors", "transformer.h.52.self_attention.query.weight": "model-00027-of-00050.safetensors", "transformer.h.52.self_attention.key_value.weight": "model-00027-of-00050.safetensors", "transformer.h.52.mlp.gate_proj.weight": "model-00027-of-00050.safetensors", "transformer.h.52.mlp.down_proj.weight": "model-00027-of-00050.safetensors", "transformer.h.52.mlp.down_proj.bias": "model-00027-of-00050.safetensors", "transformer.h.52.mlp.up_proj.weight": "model-00027-of-00050.safetensors", "transformer.h.53.input_layernorm.weight": "model-00028-of-00050.safetensors", "transformer.h.53.post_attention_layernorm.weight": "model-00028-of-00050.safetensors", "transformer.h.53.self_attention.dense.weight": "model-00028-of-00050.safetensors", "transformer.h.53.self_attention.dense.bias": "model-00028-of-00050.safetensors", "transformer.h.53.self_attention.query.weight": "model-00028-of-00050.safetensors", "transformer.h.53.self_attention.key_value.weight": "model-00028-of-00050.safetensors", "transformer.h.53.mlp.gate_proj.weight": "model-00028-of-00050.safetensors", "transformer.h.53.mlp.down_proj.weight": "model-00028-of-00050.safetensors", "transformer.h.53.mlp.down_proj.bias": "model-00028-of-00050.safetensors", "transformer.h.53.mlp.up_proj.weight": "model-00028-of-00050.safetensors", "transformer.h.54.input_layernorm.weight": "model-00028-of-00050.safetensors", "transformer.h.54.post_attention_layernorm.weight": "model-00028-of-00050.safetensors", "transformer.h.54.self_attention.dense.weight": "model-00028-of-00050.safetensors", "transformer.h.54.self_attention.dense.bias": "model-00028-of-00050.safetensors", "transformer.h.54.self_attention.query.weight": "model-00028-of-00050.safetensors", "transformer.h.54.self_attention.key_value.weight": "model-00028-of-00050.safetensors", "transformer.h.54.mlp.gate_proj.weight": "model-00028-of-00050.safetensors", "transformer.h.54.mlp.down_proj.weight": "model-00028-of-00050.safetensors", "transformer.h.54.mlp.down_proj.bias": "model-00028-of-00050.safetensors", "transformer.h.54.mlp.up_proj.weight": "model-00028-of-00050.safetensors", "transformer.h.55.input_layernorm.weight": "model-00029-of-00050.safetensors", "transformer.h.55.post_attention_layernorm.weight": "model-00029-of-00050.safetensors", "transformer.h.55.self_attention.dense.weight": "model-00029-of-00050.safetensors", "transformer.h.55.self_attention.dense.bias": "model-00029-of-00050.safetensors", "transformer.h.55.self_attention.query.weight": "model-00029-of-00050.safetensors", "transformer.h.55.self_attention.key_value.weight": "model-00029-of-00050.safetensors", "transformer.h.55.mlp.gate_proj.weight": "model-00029-of-00050.safetensors", "transformer.h.55.mlp.down_proj.weight": "model-00029-of-00050.safetensors", "transformer.h.55.mlp.down_proj.bias": "model-00029-of-00050.safetensors", "transformer.h.55.mlp.up_proj.weight": "model-00029-of-00050.safetensors", "transformer.h.56.input_layernorm.weight": "model-00029-of-00050.safetensors", "transformer.h.56.post_attention_layernorm.weight": "model-00029-of-00050.safetensors", "transformer.h.56.self_attention.dense.weight": "model-00029-of-00050.safetensors", "transformer.h.56.self_attention.dense.bias": "model-00029-of-00050.safetensors", "transformer.h.56.self_attention.query.weight": "model-00029-of-00050.safetensors", "transformer.h.56.self_attention.key_value.weight": "model-00029-of-00050.safetensors", "transformer.h.56.mlp.gate_proj.weight": "model-00029-of-00050.safetensors", "transformer.h.56.mlp.down_proj.weight": "model-00029-of-00050.safetensors", "transformer.h.56.mlp.down_proj.bias": "model-00029-of-00050.safetensors", "transformer.h.56.mlp.up_proj.weight": "model-00029-of-00050.safetensors", "transformer.h.57.input_layernorm.weight": "model-00030-of-00050.safetensors", "transformer.h.57.post_attention_layernorm.weight": "model-00030-of-00050.safetensors", "transformer.h.57.self_attention.dense.weight": "model-00030-of-00050.safetensors", "transformer.h.57.self_attention.dense.bias": "model-00030-of-00050.safetensors", "transformer.h.57.self_attention.query.weight": "model-00030-of-00050.safetensors", "transformer.h.57.self_attention.key_value.weight": "model-00030-of-00050.safetensors", "transformer.h.57.mlp.gate_proj.weight": "model-00030-of-00050.safetensors", "transformer.h.57.mlp.down_proj.weight": "model-00030-of-00050.safetensors", "transformer.h.57.mlp.down_proj.bias": "model-00030-of-00050.safetensors", "transformer.h.57.mlp.up_proj.weight": "model-00030-of-00050.safetensors", "transformer.h.58.input_layernorm.weight": "model-00030-of-00050.safetensors", "transformer.h.58.post_attention_layernorm.weight": "model-00030-of-00050.safetensors", "transformer.h.58.self_attention.dense.weight": "model-00030-of-00050.safetensors", "transformer.h.58.self_attention.dense.bias": "model-00030-of-00050.safetensors", "transformer.h.58.self_attention.query.weight": "model-00030-of-00050.safetensors", "transformer.h.58.self_attention.key_value.weight": "model-00030-of-00050.safetensors", "transformer.h.58.mlp.gate_proj.weight": "model-00030-of-00050.safetensors", "transformer.h.58.mlp.down_proj.weight": "model-00030-of-00050.safetensors", "transformer.h.58.mlp.down_proj.bias": "model-00030-of-00050.safetensors", "transformer.h.58.mlp.up_proj.weight": "model-00030-of-00050.safetensors", "transformer.h.59.input_layernorm.weight": "model-00031-of-00050.safetensors", "transformer.h.59.post_attention_layernorm.weight": "model-00031-of-00050.safetensors", "transformer.h.59.self_attention.dense.weight": "model-00031-of-00050.safetensors", "transformer.h.59.self_attention.dense.bias": "model-00031-of-00050.safetensors", "transformer.h.59.self_attention.query.weight": "model-00031-of-00050.safetensors", "transformer.h.59.self_attention.key_value.weight": "model-00031-of-00050.safetensors", "transformer.h.59.mlp.gate_proj.weight": "model-00031-of-00050.safetensors", "transformer.h.59.mlp.down_proj.weight": "model-00031-of-00050.safetensors", "transformer.h.59.mlp.down_proj.bias": "model-00031-of-00050.safetensors", "transformer.h.59.mlp.up_proj.weight": "model-00031-of-00050.safetensors", "transformer.h.60.input_layernorm.weight": "model-00031-of-00050.safetensors", "transformer.h.60.post_attention_layernorm.weight": "model-00031-of-00050.safetensors", "transformer.h.60.self_attention.dense.weight": "model-00031-of-00050.safetensors", "transformer.h.60.self_attention.dense.bias": "model-00031-of-00050.safetensors", "transformer.h.60.self_attention.query.weight": "model-00031-of-00050.safetensors", "transformer.h.60.self_attention.key_value.weight": "model-00031-of-00050.safetensors", "transformer.h.60.mlp.gate_proj.weight": "model-00031-of-00050.safetensors", "transformer.h.60.mlp.down_proj.weight": "model-00031-of-00050.safetensors", "transformer.h.60.mlp.down_proj.bias": "model-00031-of-00050.safetensors", "transformer.h.60.mlp.up_proj.weight": "model-00031-of-00050.safetensors", "transformer.h.61.input_layernorm.weight": "model-00032-of-00050.safetensors", "transformer.h.61.post_attention_layernorm.weight": "model-00032-of-00050.safetensors", "transformer.h.61.self_attention.dense.weight": "model-00032-of-00050.safetensors", "transformer.h.61.self_attention.dense.bias": "model-00032-of-00050.safetensors", "transformer.h.61.self_attention.query.weight": "model-00032-of-00050.safetensors", "transformer.h.61.self_attention.key_value.weight": "model-00032-of-00050.safetensors", "transformer.h.61.mlp.gate_proj.weight": "model-00032-of-00050.safetensors", "transformer.h.61.mlp.down_proj.weight": "model-00032-of-00050.safetensors", "transformer.h.61.mlp.down_proj.bias": "model-00032-of-00050.safetensors", "transformer.h.61.mlp.up_proj.weight": "model-00032-of-00050.safetensors", "transformer.h.62.input_layernorm.weight": "model-00032-of-00050.safetensors", "transformer.h.62.post_attention_layernorm.weight": "model-00032-of-00050.safetensors", "transformer.h.62.self_attention.dense.weight": "model-00032-of-00050.safetensors", "transformer.h.62.self_attention.dense.bias": "model-00032-of-00050.safetensors", "transformer.h.62.self_attention.query.weight": "model-00032-of-00050.safetensors", "transformer.h.62.self_attention.key_value.weight": "model-00032-of-00050.safetensors", "transformer.h.62.mlp.gate_proj.weight": "model-00032-of-00050.safetensors", "transformer.h.62.mlp.down_proj.weight": "model-00032-of-00050.safetensors", "transformer.h.62.mlp.down_proj.bias": "model-00032-of-00050.safetensors", "transformer.h.62.mlp.up_proj.weight": "model-00032-of-00050.safetensors", "transformer.h.63.input_layernorm.weight": "model-00033-of-00050.safetensors", "transformer.h.63.post_attention_layernorm.weight": "model-00033-of-00050.safetensors", "transformer.h.63.self_attention.dense.weight": "model-00033-of-00050.safetensors", "transformer.h.63.self_attention.dense.bias": "model-00033-of-00050.safetensors", "transformer.h.63.self_attention.query.weight": "model-00033-of-00050.safetensors", "transformer.h.63.self_attention.key_value.weight": "model-00033-of-00050.safetensors", "transformer.h.63.mlp.gate_proj.weight": "model-00033-of-00050.safetensors", "transformer.h.63.mlp.down_proj.weight": "model-00033-of-00050.safetensors", "transformer.h.63.mlp.down_proj.bias": "model-00033-of-00050.safetensors", "transformer.h.63.mlp.up_proj.weight": "model-00033-of-00050.safetensors", "transformer.h.64.input_layernorm.weight": "model-00033-of-00050.safetensors", "transformer.h.64.post_attention_layernorm.weight": "model-00033-of-00050.safetensors", "transformer.h.64.self_attention.dense.weight": "model-00033-of-00050.safetensors", "transformer.h.64.self_attention.dense.bias": "model-00033-of-00050.safetensors", "transformer.h.64.self_attention.query.weight": "model-00033-of-00050.safetensors", "transformer.h.64.self_attention.key_value.weight": "model-00033-of-00050.safetensors", "transformer.h.64.mlp.gate_proj.weight": "model-00033-of-00050.safetensors", "transformer.h.64.mlp.down_proj.weight": "model-00033-of-00050.safetensors", "transformer.h.64.mlp.down_proj.bias": "model-00033-of-00050.safetensors", "transformer.h.64.mlp.up_proj.weight": "model-00033-of-00050.safetensors", "transformer.h.65.input_layernorm.weight": "model-00034-of-00050.safetensors", "transformer.h.65.post_attention_layernorm.weight": "model-00034-of-00050.safetensors", "transformer.h.65.self_attention.dense.weight": "model-00034-of-00050.safetensors", "transformer.h.65.self_attention.dense.bias": "model-00034-of-00050.safetensors", "transformer.h.65.self_attention.query.weight": "model-00034-of-00050.safetensors", "transformer.h.65.self_attention.key_value.weight": "model-00034-of-00050.safetensors", "transformer.h.65.mlp.gate_proj.weight": "model-00034-of-00050.safetensors", "transformer.h.65.mlp.down_proj.weight": "model-00034-of-00050.safetensors", "transformer.h.65.mlp.down_proj.bias": "model-00034-of-00050.safetensors", "transformer.h.65.mlp.up_proj.weight": "model-00034-of-00050.safetensors", "transformer.h.66.input_layernorm.weight": "model-00034-of-00050.safetensors", "transformer.h.66.post_attention_layernorm.weight": "model-00034-of-00050.safetensors", "transformer.h.66.self_attention.dense.weight": "model-00034-of-00050.safetensors", "transformer.h.66.self_attention.dense.bias": "model-00034-of-00050.safetensors", "transformer.h.66.self_attention.query.weight": "model-00034-of-00050.safetensors", "transformer.h.66.self_attention.key_value.weight": "model-00034-of-00050.safetensors", "transformer.h.66.mlp.gate_proj.weight": "model-00034-of-00050.safetensors", "transformer.h.66.mlp.down_proj.weight": "model-00034-of-00050.safetensors", "transformer.h.66.mlp.down_proj.bias": "model-00034-of-00050.safetensors", "transformer.h.66.mlp.up_proj.weight": "model-00034-of-00050.safetensors", "transformer.h.67.input_layernorm.weight": "model-00035-of-00050.safetensors", "transformer.h.67.post_attention_layernorm.weight": "model-00035-of-00050.safetensors", "transformer.h.67.self_attention.dense.weight": "model-00035-of-00050.safetensors", "transformer.h.67.self_attention.dense.bias": "model-00035-of-00050.safetensors", "transformer.h.67.self_attention.query.weight": "model-00035-of-00050.safetensors", "transformer.h.67.self_attention.key_value.weight": "model-00035-of-00050.safetensors", "transformer.h.67.mlp.gate_proj.weight": "model-00035-of-00050.safetensors", "transformer.h.67.mlp.down_proj.weight": "model-00035-of-00050.safetensors", "transformer.h.67.mlp.down_proj.bias": "model-00035-of-00050.safetensors", "transformer.h.67.mlp.up_proj.weight": "model-00035-of-00050.safetensors", "transformer.h.68.input_layernorm.weight": "model-00035-of-00050.safetensors", "transformer.h.68.post_attention_layernorm.weight": "model-00035-of-00050.safetensors", "transformer.h.68.self_attention.dense.weight": "model-00035-of-00050.safetensors", "transformer.h.68.self_attention.dense.bias": "model-00035-of-00050.safetensors", "transformer.h.68.self_attention.query.weight": "model-00035-of-00050.safetensors", "transformer.h.68.self_attention.key_value.weight": "model-00035-of-00050.safetensors", "transformer.h.68.mlp.gate_proj.weight": "model-00035-of-00050.safetensors", "transformer.h.68.mlp.down_proj.weight": "model-00035-of-00050.safetensors", "transformer.h.68.mlp.down_proj.bias": "model-00035-of-00050.safetensors", "transformer.h.68.mlp.up_proj.weight": "model-00035-of-00050.safetensors", "transformer.h.69.input_layernorm.weight": "model-00036-of-00050.safetensors", "transformer.h.69.post_attention_layernorm.weight": "model-00036-of-00050.safetensors", "transformer.h.69.self_attention.dense.weight": "model-00036-of-00050.safetensors", "transformer.h.69.self_attention.dense.bias": "model-00036-of-00050.safetensors", "transformer.h.69.self_attention.query.weight": "model-00036-of-00050.safetensors", "transformer.h.69.self_attention.key_value.weight": "model-00036-of-00050.safetensors", "transformer.h.69.mlp.gate_proj.weight": "model-00036-of-00050.safetensors", "transformer.h.69.mlp.down_proj.weight": "model-00036-of-00050.safetensors", "transformer.h.69.mlp.down_proj.bias": "model-00036-of-00050.safetensors", "transformer.h.69.mlp.up_proj.weight": "model-00036-of-00050.safetensors", "transformer.h.70.input_layernorm.weight": "model-00036-of-00050.safetensors", "transformer.h.70.post_attention_layernorm.weight": "model-00036-of-00050.safetensors", "transformer.h.70.self_attention.dense.weight": "model-00036-of-00050.safetensors", "transformer.h.70.self_attention.dense.bias": "model-00036-of-00050.safetensors", "transformer.h.70.self_attention.query.weight": "model-00036-of-00050.safetensors", "transformer.h.70.self_attention.key_value.weight": "model-00036-of-00050.safetensors", "transformer.h.70.mlp.gate_proj.weight": "model-00036-of-00050.safetensors", "transformer.h.70.mlp.down_proj.weight": "model-00036-of-00050.safetensors", "transformer.h.70.mlp.down_proj.bias": "model-00036-of-00050.safetensors", "transformer.h.70.mlp.up_proj.weight": "model-00036-of-00050.safetensors", "transformer.h.71.input_layernorm.weight": "model-00037-of-00050.safetensors", "transformer.h.71.post_attention_layernorm.weight": "model-00037-of-00050.safetensors", "transformer.h.71.self_attention.dense.weight": "model-00037-of-00050.safetensors", "transformer.h.71.self_attention.dense.bias": "model-00037-of-00050.safetensors", "transformer.h.71.self_attention.query.weight": "model-00037-of-00050.safetensors", "transformer.h.71.self_attention.key_value.weight": "model-00037-of-00050.safetensors", "transformer.h.71.mlp.gate_proj.weight": "model-00037-of-00050.safetensors", "transformer.h.71.mlp.down_proj.weight": "model-00037-of-00050.safetensors", "transformer.h.71.mlp.down_proj.bias": "model-00037-of-00050.safetensors", "transformer.h.71.mlp.up_proj.weight": "model-00037-of-00050.safetensors", "transformer.h.72.input_layernorm.weight": "model-00037-of-00050.safetensors", "transformer.h.72.post_attention_layernorm.weight": "model-00037-of-00050.safetensors", "transformer.h.72.self_attention.dense.weight": "model-00037-of-00050.safetensors", "transformer.h.72.self_attention.dense.bias": "model-00037-of-00050.safetensors", "transformer.h.72.self_attention.query.weight": "model-00037-of-00050.safetensors", "transformer.h.72.self_attention.key_value.weight": "model-00037-of-00050.safetensors", "transformer.h.72.mlp.gate_proj.weight": "model-00037-of-00050.safetensors", "transformer.h.72.mlp.down_proj.weight": "model-00037-of-00050.safetensors", "transformer.h.72.mlp.down_proj.bias": "model-00037-of-00050.safetensors", "transformer.h.72.mlp.up_proj.weight": "model-00037-of-00050.safetensors", "transformer.h.73.input_layernorm.weight": "model-00038-of-00050.safetensors", "transformer.h.73.post_attention_layernorm.weight": "model-00038-of-00050.safetensors", "transformer.h.73.self_attention.dense.weight": "model-00038-of-00050.safetensors", "transformer.h.73.self_attention.dense.bias": "model-00038-of-00050.safetensors", "transformer.h.73.self_attention.query.weight": "model-00038-of-00050.safetensors", "transformer.h.73.self_attention.key_value.weight": "model-00038-of-00050.safetensors", "transformer.h.73.mlp.gate_proj.weight": "model-00038-of-00050.safetensors", "transformer.h.73.mlp.down_proj.weight": "model-00038-of-00050.safetensors", "transformer.h.73.mlp.down_proj.bias": "model-00038-of-00050.safetensors", "transformer.h.73.mlp.up_proj.weight": "model-00038-of-00050.safetensors", "transformer.h.74.input_layernorm.weight": "model-00038-of-00050.safetensors", "transformer.h.74.post_attention_layernorm.weight": "model-00038-of-00050.safetensors", "transformer.h.74.self_attention.dense.weight": "model-00038-of-00050.safetensors", "transformer.h.74.self_attention.dense.bias": "model-00038-of-00050.safetensors", "transformer.h.74.self_attention.query.weight": "model-00038-of-00050.safetensors", "transformer.h.74.self_attention.key_value.weight": "model-00038-of-00050.safetensors", "transformer.h.74.mlp.gate_proj.weight": "model-00038-of-00050.safetensors", "transformer.h.74.mlp.down_proj.weight": "model-00038-of-00050.safetensors", "transformer.h.74.mlp.down_proj.bias": "model-00038-of-00050.safetensors", "transformer.h.74.mlp.up_proj.weight": "model-00038-of-00050.safetensors", "transformer.h.75.input_layernorm.weight": "model-00039-of-00050.safetensors", "transformer.h.75.post_attention_layernorm.weight": "model-00039-of-00050.safetensors", "transformer.h.75.self_attention.dense.weight": "model-00039-of-00050.safetensors", "transformer.h.75.self_attention.dense.bias": "model-00039-of-00050.safetensors", "transformer.h.75.self_attention.query.weight": "model-00039-of-00050.safetensors", "transformer.h.75.self_attention.key_value.weight": "model-00039-of-00050.safetensors", "transformer.h.75.mlp.gate_proj.weight": "model-00039-of-00050.safetensors", "transformer.h.75.mlp.down_proj.weight": "model-00039-of-00050.safetensors", "transformer.h.75.mlp.down_proj.bias": "model-00039-of-00050.safetensors", "transformer.h.75.mlp.up_proj.weight": "model-00039-of-00050.safetensors", "transformer.h.76.input_layernorm.weight": "model-00039-of-00050.safetensors", "transformer.h.76.post_attention_layernorm.weight": "model-00039-of-00050.safetensors", "transformer.h.76.self_attention.dense.weight": "model-00039-of-00050.safetensors", "transformer.h.76.self_attention.dense.bias": "model-00039-of-00050.safetensors", "transformer.h.76.self_attention.query.weight": "model-00039-of-00050.safetensors", "transformer.h.76.self_attention.key_value.weight": "model-00039-of-00050.safetensors", "transformer.h.76.mlp.gate_proj.weight": "model-00039-of-00050.safetensors", "transformer.h.76.mlp.down_proj.weight": "model-00039-of-00050.safetensors", "transformer.h.76.mlp.down_proj.bias": "model-00039-of-00050.safetensors", "transformer.h.76.mlp.up_proj.weight": "model-00039-of-00050.safetensors", "transformer.h.77.input_layernorm.weight": "model-00040-of-00050.safetensors", "transformer.h.77.post_attention_layernorm.weight": "model-00040-of-00050.safetensors", "transformer.h.77.self_attention.dense.weight": "model-00040-of-00050.safetensors", "transformer.h.77.self_attention.dense.bias": "model-00040-of-00050.safetensors", "transformer.h.77.self_attention.query.weight": "model-00040-of-00050.safetensors", "transformer.h.77.self_attention.key_value.weight": "model-00040-of-00050.safetensors", "transformer.h.77.mlp.gate_proj.weight": "model-00040-of-00050.safetensors", "transformer.h.77.mlp.down_proj.weight": "model-00040-of-00050.safetensors", "transformer.h.77.mlp.down_proj.bias": "model-00040-of-00050.safetensors", "transformer.h.77.mlp.up_proj.weight": "model-00040-of-00050.safetensors", "transformer.h.78.input_layernorm.weight": "model-00040-of-00050.safetensors", "transformer.h.78.post_attention_layernorm.weight": "model-00040-of-00050.safetensors", "transformer.h.78.self_attention.dense.weight": "model-00040-of-00050.safetensors", "transformer.h.78.self_attention.dense.bias": "model-00040-of-00050.safetensors", "transformer.h.78.self_attention.query.weight": "model-00040-of-00050.safetensors", "transformer.h.78.self_attention.key_value.weight": "model-00040-of-00050.safetensors", "transformer.h.78.mlp.gate_proj.weight": "model-00040-of-00050.safetensors", "transformer.h.78.mlp.down_proj.weight": "model-00040-of-00050.safetensors", "transformer.h.78.mlp.down_proj.bias": "model-00040-of-00050.safetensors", "transformer.h.78.mlp.up_proj.weight": "model-00040-of-00050.safetensors", "transformer.h.79.input_layernorm.weight": "model-00041-of-00050.safetensors", "transformer.h.79.post_attention_layernorm.weight": "model-00041-of-00050.safetensors", "transformer.h.79.self_attention.dense.weight": "model-00041-of-00050.safetensors", "transformer.h.79.self_attention.dense.bias": "model-00041-of-00050.safetensors", "transformer.h.79.self_attention.query.weight": "model-00041-of-00050.safetensors", "transformer.h.79.self_attention.key_value.weight": "model-00041-of-00050.safetensors", "transformer.h.79.mlp.gate_proj.weight": "model-00041-of-00050.safetensors", "transformer.h.79.mlp.down_proj.weight": "model-00041-of-00050.safetensors", "transformer.h.79.mlp.down_proj.bias": "model-00041-of-00050.safetensors", "transformer.h.79.mlp.up_proj.weight": "model-00041-of-00050.safetensors", "transformer.h.80.input_layernorm.weight": "model-00041-of-00050.safetensors", "transformer.h.80.post_attention_layernorm.weight": "model-00041-of-00050.safetensors", "transformer.h.80.self_attention.dense.weight": "model-00041-of-00050.safetensors", "transformer.h.80.self_attention.dense.bias": "model-00041-of-00050.safetensors", "transformer.h.80.self_attention.query.weight": "model-00041-of-00050.safetensors", "transformer.h.80.self_attention.key_value.weight": "model-00041-of-00050.safetensors", "transformer.h.80.mlp.gate_proj.weight": "model-00041-of-00050.safetensors", "transformer.h.80.mlp.down_proj.weight": "model-00041-of-00050.safetensors", "transformer.h.80.mlp.down_proj.bias": "model-00041-of-00050.safetensors", "transformer.h.80.mlp.up_proj.weight": "model-00041-of-00050.safetensors", "transformer.h.81.input_layernorm.weight": "model-00042-of-00050.safetensors", "transformer.h.81.post_attention_layernorm.weight": "model-00042-of-00050.safetensors", "transformer.h.81.self_attention.dense.weight": "model-00042-of-00050.safetensors", "transformer.h.81.self_attention.dense.bias": "model-00042-of-00050.safetensors", "transformer.h.81.self_attention.query.weight": "model-00042-of-00050.safetensors", "transformer.h.81.self_attention.key_value.weight": "model-00042-of-00050.safetensors", "transformer.h.81.mlp.gate_proj.weight": "model-00042-of-00050.safetensors", "transformer.h.81.mlp.down_proj.weight": "model-00042-of-00050.safetensors", "transformer.h.81.mlp.down_proj.bias": "model-00042-of-00050.safetensors", "transformer.h.81.mlp.up_proj.weight": "model-00042-of-00050.safetensors", "transformer.h.82.input_layernorm.weight": "model-00042-of-00050.safetensors", "transformer.h.82.post_attention_layernorm.weight": "model-00042-of-00050.safetensors", "transformer.h.82.self_attention.dense.weight": "model-00042-of-00050.safetensors", "transformer.h.82.self_attention.dense.bias": "model-00042-of-00050.safetensors", "transformer.h.82.self_attention.query.weight": "model-00042-of-00050.safetensors", "transformer.h.82.self_attention.key_value.weight": "model-00042-of-00050.safetensors", "transformer.h.82.mlp.gate_proj.weight": "model-00042-of-00050.safetensors", "transformer.h.82.mlp.down_proj.weight": "model-00042-of-00050.safetensors", "transformer.h.82.mlp.down_proj.bias": "model-00042-of-00050.safetensors", "transformer.h.82.mlp.up_proj.weight": "model-00042-of-00050.safetensors", "transformer.h.83.input_layernorm.weight": "model-00043-of-00050.safetensors", "transformer.h.83.post_attention_layernorm.weight": "model-00043-of-00050.safetensors", "transformer.h.83.self_attention.dense.weight": "model-00043-of-00050.safetensors", "transformer.h.83.self_attention.dense.bias": "model-00043-of-00050.safetensors", "transformer.h.83.self_attention.query.weight": "model-00043-of-00050.safetensors", "transformer.h.83.self_attention.key_value.weight": "model-00043-of-00050.safetensors", "transformer.h.83.mlp.gate_proj.weight": "model-00043-of-00050.safetensors", "transformer.h.83.mlp.down_proj.weight": "model-00043-of-00050.safetensors", "transformer.h.83.mlp.down_proj.bias": "model-00043-of-00050.safetensors", "transformer.h.83.mlp.up_proj.weight": "model-00043-of-00050.safetensors", "transformer.h.84.input_layernorm.weight": "model-00043-of-00050.safetensors", "transformer.h.84.post_attention_layernorm.weight": "model-00043-of-00050.safetensors", "transformer.h.84.self_attention.dense.weight": "model-00043-of-00050.safetensors", "transformer.h.84.self_attention.dense.bias": "model-00043-of-00050.safetensors", "transformer.h.84.self_attention.query.weight": "model-00043-of-00050.safetensors", "transformer.h.84.self_attention.key_value.weight": "model-00043-of-00050.safetensors", "transformer.h.84.mlp.gate_proj.weight": "model-00043-of-00050.safetensors", "transformer.h.84.mlp.down_proj.weight": "model-00043-of-00050.safetensors", "transformer.h.84.mlp.down_proj.bias": "model-00043-of-00050.safetensors", "transformer.h.84.mlp.up_proj.weight": "model-00043-of-00050.safetensors", "transformer.h.85.input_layernorm.weight": "model-00044-of-00050.safetensors", "transformer.h.85.post_attention_layernorm.weight": "model-00044-of-00050.safetensors", "transformer.h.85.self_attention.dense.weight": "model-00044-of-00050.safetensors", "transformer.h.85.self_attention.dense.bias": "model-00044-of-00050.safetensors", "transformer.h.85.self_attention.query.weight": "model-00044-of-00050.safetensors", "transformer.h.85.self_attention.key_value.weight": "model-00044-of-00050.safetensors", "transformer.h.85.mlp.gate_proj.weight": "model-00044-of-00050.safetensors", "transformer.h.85.mlp.down_proj.weight": "model-00044-of-00050.safetensors", "transformer.h.85.mlp.down_proj.bias": "model-00044-of-00050.safetensors", "transformer.h.85.mlp.up_proj.weight": "model-00044-of-00050.safetensors", "transformer.h.86.input_layernorm.weight": "model-00044-of-00050.safetensors", "transformer.h.86.post_attention_layernorm.weight": "model-00044-of-00050.safetensors", "transformer.h.86.self_attention.dense.weight": "model-00044-of-00050.safetensors", "transformer.h.86.self_attention.dense.bias": "model-00044-of-00050.safetensors", "transformer.h.86.self_attention.query.weight": "model-00044-of-00050.safetensors", "transformer.h.86.self_attention.key_value.weight": "model-00044-of-00050.safetensors", "transformer.h.86.mlp.gate_proj.weight": "model-00044-of-00050.safetensors", "transformer.h.86.mlp.down_proj.weight": "model-00044-of-00050.safetensors", "transformer.h.86.mlp.down_proj.bias": "model-00044-of-00050.safetensors", "transformer.h.86.mlp.up_proj.weight": "model-00044-of-00050.safetensors", "transformer.h.87.input_layernorm.weight": "model-00045-of-00050.safetensors", "transformer.h.87.post_attention_layernorm.weight": "model-00045-of-00050.safetensors", "transformer.h.87.self_attention.dense.weight": "model-00045-of-00050.safetensors", "transformer.h.87.self_attention.dense.bias": "model-00045-of-00050.safetensors", "transformer.h.87.self_attention.query.weight": "model-00045-of-00050.safetensors", "transformer.h.87.self_attention.key_value.weight": "model-00045-of-00050.safetensors", "transformer.h.87.mlp.gate_proj.weight": "model-00045-of-00050.safetensors", "transformer.h.87.mlp.down_proj.weight": "model-00045-of-00050.safetensors", "transformer.h.87.mlp.down_proj.bias": "model-00045-of-00050.safetensors", "transformer.h.87.mlp.up_proj.weight": "model-00045-of-00050.safetensors", "transformer.h.88.input_layernorm.weight": "model-00045-of-00050.safetensors", "transformer.h.88.post_attention_layernorm.weight": "model-00045-of-00050.safetensors", "transformer.h.88.self_attention.dense.weight": "model-00045-of-00050.safetensors", "transformer.h.88.self_attention.dense.bias": "model-00045-of-00050.safetensors", "transformer.h.88.self_attention.query.weight": "model-00045-of-00050.safetensors", "transformer.h.88.self_attention.key_value.weight": "model-00045-of-00050.safetensors", "transformer.h.88.mlp.gate_proj.weight": "model-00045-of-00050.safetensors", "transformer.h.88.mlp.down_proj.weight": "model-00045-of-00050.safetensors", "transformer.h.88.mlp.down_proj.bias": "model-00045-of-00050.safetensors", "transformer.h.88.mlp.up_proj.weight": "model-00045-of-00050.safetensors", "transformer.h.89.input_layernorm.weight": "model-00046-of-00050.safetensors", "transformer.h.89.post_attention_layernorm.weight": "model-00046-of-00050.safetensors", "transformer.h.89.self_attention.dense.weight": "model-00046-of-00050.safetensors", "transformer.h.89.self_attention.dense.bias": "model-00046-of-00050.safetensors", "transformer.h.89.self_attention.query.weight": "model-00046-of-00050.safetensors", "transformer.h.89.self_attention.key_value.weight": "model-00046-of-00050.safetensors", "transformer.h.89.mlp.gate_proj.weight": "model-00046-of-00050.safetensors", "transformer.h.89.mlp.down_proj.weight": "model-00046-of-00050.safetensors", "transformer.h.89.mlp.down_proj.bias": "model-00046-of-00050.safetensors", "transformer.h.89.mlp.up_proj.weight": "model-00046-of-00050.safetensors", "transformer.h.90.input_layernorm.weight": "model-00046-of-00050.safetensors", "transformer.h.90.post_attention_layernorm.weight": "model-00046-of-00050.safetensors", "transformer.h.90.self_attention.dense.weight": "model-00046-of-00050.safetensors", "transformer.h.90.self_attention.dense.bias": "model-00046-of-00050.safetensors", "transformer.h.90.self_attention.query.weight": "model-00046-of-00050.safetensors", "transformer.h.90.self_attention.key_value.weight": "model-00046-of-00050.safetensors", "transformer.h.90.mlp.gate_proj.weight": "model-00046-of-00050.safetensors", "transformer.h.90.mlp.down_proj.weight": "model-00046-of-00050.safetensors", "transformer.h.90.mlp.down_proj.bias": "model-00046-of-00050.safetensors", "transformer.h.90.mlp.up_proj.weight": "model-00046-of-00050.safetensors", "transformer.h.91.input_layernorm.weight": "model-00047-of-00050.safetensors", "transformer.h.91.post_attention_layernorm.weight": "model-00047-of-00050.safetensors", "transformer.h.91.self_attention.dense.weight": "model-00047-of-00050.safetensors", "transformer.h.91.self_attention.dense.bias": "model-00047-of-00050.safetensors", "transformer.h.91.self_attention.query.weight": "model-00047-of-00050.safetensors", "transformer.h.91.self_attention.key_value.weight": "model-00047-of-00050.safetensors", "transformer.h.91.mlp.gate_proj.weight": "model-00047-of-00050.safetensors", "transformer.h.91.mlp.down_proj.weight": "model-00047-of-00050.safetensors", "transformer.h.91.mlp.down_proj.bias": "model-00047-of-00050.safetensors", "transformer.h.91.mlp.up_proj.weight": "model-00047-of-00050.safetensors", "transformer.h.92.input_layernorm.weight": "model-00047-of-00050.safetensors", "transformer.h.92.post_attention_layernorm.weight": "model-00047-of-00050.safetensors", "transformer.h.92.self_attention.dense.weight": "model-00047-of-00050.safetensors", "transformer.h.92.self_attention.dense.bias": "model-00047-of-00050.safetensors", "transformer.h.92.self_attention.query.weight": "model-00047-of-00050.safetensors", "transformer.h.92.self_attention.key_value.weight": "model-00047-of-00050.safetensors", "transformer.h.92.mlp.gate_proj.weight": "model-00047-of-00050.safetensors", "transformer.h.92.mlp.down_proj.weight": "model-00047-of-00050.safetensors", "transformer.h.92.mlp.down_proj.bias": "model-00047-of-00050.safetensors", "transformer.h.92.mlp.up_proj.weight": "model-00047-of-00050.safetensors", "transformer.h.93.input_layernorm.weight": "model-00048-of-00050.safetensors", "transformer.h.93.post_attention_layernorm.weight": "model-00048-of-00050.safetensors", "transformer.h.93.self_attention.dense.weight": "model-00048-of-00050.safetensors", "transformer.h.93.self_attention.dense.bias": "model-00048-of-00050.safetensors", "transformer.h.93.self_attention.query.weight": "model-00048-of-00050.safetensors", "transformer.h.93.self_attention.key_value.weight": "model-00048-of-00050.safetensors", "transformer.h.93.mlp.gate_proj.weight": "model-00048-of-00050.safetensors", "transformer.h.93.mlp.down_proj.weight": "model-00048-of-00050.safetensors", "transformer.h.93.mlp.down_proj.bias": "model-00048-of-00050.safetensors", "transformer.h.93.mlp.up_proj.weight": "model-00048-of-00050.safetensors", "transformer.h.94.input_layernorm.weight": "model-00048-of-00050.safetensors", "transformer.h.94.post_attention_layernorm.weight": "model-00048-of-00050.safetensors", "transformer.h.94.self_attention.dense.weight": "model-00048-of-00050.safetensors", "transformer.h.94.self_attention.dense.bias": "model-00048-of-00050.safetensors", "transformer.h.94.self_attention.query.weight": "model-00048-of-00050.safetensors", "transformer.h.94.self_attention.key_value.weight": "model-00048-of-00050.safetensors", "transformer.h.94.mlp.gate_proj.weight": "model-00048-of-00050.safetensors", "transformer.h.94.mlp.down_proj.weight": "model-00048-of-00050.safetensors", "transformer.h.94.mlp.down_proj.bias": "model-00048-of-00050.safetensors", "transformer.h.94.mlp.up_proj.weight": "model-00048-of-00050.safetensors", "transformer.h.95.input_layernorm.weight": "model-00049-of-00050.safetensors", "transformer.h.95.post_attention_layernorm.weight": "model-00049-of-00050.safetensors", "transformer.h.95.self_attention.dense.weight": "model-00049-of-00050.safetensors", "transformer.h.95.self_attention.dense.bias": "model-00049-of-00050.safetensors", "transformer.h.95.self_attention.query.weight": "model-00049-of-00050.safetensors", "transformer.h.95.self_attention.key_value.weight": "model-00049-of-00050.safetensors", "transformer.h.95.mlp.gate_proj.weight": "model-00049-of-00050.safetensors", "transformer.h.95.mlp.down_proj.weight": "model-00049-of-00050.safetensors", "transformer.h.95.mlp.down_proj.bias": "model-00049-of-00050.safetensors", "transformer.h.95.mlp.up_proj.weight": "model-00049-of-00050.safetensors", "transformer.ln_f.weight": "model-00050-of-00050.safetensors", "lm_head.weight": "model-00050-of-00050.safetensors"}}
\ No newline at end of file
diff --git a/modeling_telechat.py b/modeling_telechat.py
new file mode 100644
index 0000000000000000000000000000000000000000..a7e1c8890288de2ea68b5fb6d3f32248d34a39aa
--- /dev/null
+++ b/modeling_telechat.py
@@ -0,0 +1,939 @@
+# coding=utf-8
+# Copyright 2022 HuggingFace Inc. team and BigScience workshop.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+# Copyright (c) 2023, NVIDIA CORPORATION.  All rights reserved.
+
+# Copyright (c) 2021 EleutherAI
+# This file is based on code by the authors denoted below and has been modified from its original version.
+#
+# Copyright (c) 2020, NVIDIA CORPORATION.  All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+
+"""PyTorch TELECHAT model."""
+
+import warnings
+from typing import Optional, Tuple, Union, List, Dict
+from threading import Thread
+
+import torch
+import math
+import copy
+from torch import nn
+import torch.utils.checkpoint
+from torch.nn import functional as F
+from torch.nn import BCEWithLogitsLoss, CrossEntropyLoss, LayerNorm, MSELoss
+from transformers.modeling_outputs import (
+    BaseModelOutputWithPastAndCrossAttentions,
+    CausalLMOutputWithCrossAttentions
+)
+from transformers.modeling_utils import PreTrainedModel
+from transformers.utils import logging
+from transformers import GenerationConfig
+
+from .configuration_telechat import TelechatConfig
+from .generation_utils import History, TelechatIterTextStreamer
+
+logger = logging.get_logger(__name__)
+
+_CHECKPOINT_FOR_DOC = "telechat"
+_CONFIG_FOR_DOC = "TelechatConfig"
+
+TELECHAT_PRETRAINED_MODEL_ARCHIVE_LIST = []
+
+try:
+    from einops import rearrange
+except ImportError:
+    rearrange = None
+
+use_flash_attn = True
+try:
+    from flash_attn.flash_attn_interface import flash_attn_unpadded_func
+except ImportError:
+    try:
+        from flash_attn.flash_attn_interface import flash_attn_varlen_func as flash_attn_unpadded_func
+    except ImportError:
+        flash_attn_unpadded_func = None
+
+
+class RotaryEmbedding(torch.nn.Module):
+    # Extracted from: https://github.com/EleutherAI/gpt-neox
+    def __init__(self, dim, config, base=10000, precision=torch.half):
+        super().__init__()
+        self.config = config
+        self.dim = dim
+        self.base = base
+        self.inv_freq = 1. / (base ** (torch.arange(0, dim, 2).float().half() / dim)).cuda()
+        self.max_seq_len_cached = None
+        self.cos_cached = None
+        self.sin_cached = None
+        self.precision = precision
+
+    def get_mscale(self, scale=1):
+        if scale <= 1:
+            return 1.0
+        return 0.1 * math.log(scale) + 1.0
+
+    def get_ntk_alpha(self, true_seq_len):
+        context_value = math.log(true_seq_len / self.config.base_seqlen, 2) + 1
+        # ntk_alpha = 2 ** context_value - 1
+        ntk_alpha = 2 ** math.ceil(context_value) - 1
+        ntk_alpha = max(ntk_alpha, 1)
+        return ntk_alpha
+
+    def forward(self, x, seq_dim=0, seq_len=None):
+        if seq_len is None:
+            seq_len = x.shape[seq_dim]
+        seq_len = max(seq_len, self.config.training_seqlen)
+        ntk_alpha = self.get_ntk_alpha(seq_len)
+        self.mscale = float(self.get_mscale(seq_len / self.config.training_seqlen))
+        if True:
+            base = self.base * ntk_alpha ** (self.dim / (self.dim - 2))
+            self.inv_freq = 1.0 / (base ** (torch.arange(0, self.dim, 2, device=x.device).float() / self.dim))
+            self.max_seq_len_cached = seq_len
+            t = torch.arange(self.max_seq_len_cached, device=x.device, dtype=self.inv_freq.dtype)
+            freqs = torch.einsum('i,j->ij', t, self.inv_freq)
+            # Different from paper, but it uses a different permutation in order to obtain the same calculation
+            emb = torch.cat((freqs, freqs), dim=-1).to(x.device)
+            if self.precision == torch.bfloat16:
+                emb = emb.float()
+            # [sx, 1 (b * np), hn]
+            self.cos_cached = self.mscale * emb.cos()[:, None, :].half()
+            self.sin_cached = self.mscale * emb.sin()[:, None, :].half()
+            if self.precision == torch.bfloat16:
+                self.cos_cached = self.cos_cached.bfloat16()
+                self.sin_cached = self.sin_cached.bfloat16()
+        return self.cos_cached[:seq_len, ...], self.sin_cached[:seq_len, ...]
+
+
+# rotary pos emb helpers:
+def rotate_half(x):
+    x1, x2 = x[..., :x.shape[-1] // 2], x[..., x.shape[-1] // 2:]
+    return torch.cat((-x2, x1), dim=x1.ndim - 1)  # dim=-1 triggers a bug in earlier torch versions
+
+
+def apply_rotary_pos_emb_torch(q, k, cos, sin, offset: int = 0):  # jitting fails with bf16
+    cos, sin = cos[offset:q.shape[0] + offset, ...], sin[offset:q.shape[0] + offset, ...]
+    return (q * cos) + (rotate_half(q) * sin), (k * cos) + (rotate_half(k) * sin)
+
+
+class MixedFusedRMSNorm(nn.Module):
+    # Extracted from https://github.com/huggingface/transformers/blob/main/src/transformers/models/llama/modeling_llama.py
+    def __init__(self, hidden_size, eps=1e-6):
+        super().__init__()
+        self.weight = nn.Parameter(torch.ones(hidden_size))
+        self.variance_epsilon = eps
+
+    def forward(self, hidden_states):
+        input_dtype = hidden_states.dtype
+        hidden_states = hidden_states.to(torch.float32)
+        variance = hidden_states.pow(2).mean(-1, keepdim=True)
+        hidden_states = hidden_states * torch.rsqrt(variance + self.variance_epsilon)
+        return self.weight * hidden_states.to(input_dtype)
+
+
+class FlashSelfAttention(torch.nn.Module):
+    # Extracted from https://github.com/microsoft/Megatron-DeepSpeed/blob/main/megatron/model/transformer.py
+    """Implement the scaled dot product attention with softmax.
+    Arguments
+    ---------
+        softmax_scale: The temperature to use for the softmax attention.
+                      (default: 1/sqrt(d_keys) where d_keys is computed at
+                      runtime)
+        attention_dropout: The dropout rate to apply to the attention
+                           (default: 0.0)
+    """
+
+    def __init__(self, causal=False, softmax_scale=None, attention_dropout=0.0,
+                 device=None, dtype=None):
+        super().__init__()
+        assert flash_attn_unpadded_func is not None, ('Please install FlashAttention first, '
+                                                      'e.g., with pip install flash-attn')
+        assert rearrange is not None, 'Please install einops first, e.g., with pip install einops'
+        self.causal = causal
+        self.softmax_scale = softmax_scale
+        self.dropout_p = attention_dropout
+
+    def forward(self, q, k, v):
+        """Implements the multihead softmax attention.
+        Arguments
+        ---------
+            q, k, v: The tensor containing the query, key, and value. (B, S, H, D)
+        """
+        assert all((i.dtype in [torch.float16, torch.bfloat16] for i in (q, k, v)))
+        assert all((i.is_cuda for i in (q, k, v)))
+
+        batch_size, seqlen_q = q.shape[0], q.shape[1]
+        seqlen_k = k.shape[1]
+
+        q, k, v = [rearrange(x, 'b s ... -> (b s) ...') for x in [q, k, v]]
+        cu_seqlens_q = torch.arange(0, (batch_size + 1) * seqlen_q, step=seqlen_q, dtype=torch.int32,
+                                    device=q.device)
+        # self.training = False
+        if self.training:
+            # during training q,k,v always have same seqlen
+            assert seqlen_k == seqlen_q
+
+            is_causal = self.causal
+            cu_seqlens_k = cu_seqlens_q
+            dropout_p = self.dropout_p
+        else:
+            # turn off FA causal mask after first inference autoregressive iteration
+            # only on first autoregressive step q,k,v have same seqlen
+            is_causal = seqlen_q == seqlen_k
+            cu_seqlens_k = torch.arange(0, (batch_size + 1) * seqlen_k, step=seqlen_k, dtype=torch.int32,
+                                        device=q.device)
+            dropout_p = 0
+
+        output = flash_attn_unpadded_func(
+            q, k, v, cu_seqlens_q, cu_seqlens_k, seqlen_q, seqlen_k,
+            dropout_p=dropout_p,
+            softmax_scale=self.softmax_scale, causal=is_causal
+        )
+
+        output = rearrange(output, '(b s) ... -> b s ...', b=batch_size)
+        return output
+
+
+def _make_causal_mask(
+        input_ids_shape: torch.Size, device: torch.device, past_key_values_length: int
+) -> torch.BoolTensor:
+    """
+    Make causal mask used for self-attention.
+    """
+    batch_size, target_length = input_ids_shape
+    mask = torch.empty((target_length, target_length + past_key_values_length), dtype=torch.bool, device=device)
+    # ONNX doesn't support `torch.Tensor.triu` properly, thus we use this workaround
+    seq_ids = torch.arange(target_length, device=device)
+    mask[:, past_key_values_length:] = seq_ids[:, None] < seq_ids[None, :]
+
+    if past_key_values_length > 0:
+        mask[:, :past_key_values_length] = False
+
+    expanded_mask = mask[None, None, :, :].expand(batch_size, 1, target_length, target_length + past_key_values_length)
+    return expanded_mask
+
+
+def _expand_mask(mask: torch.Tensor, tgt_length: int) -> torch.BoolTensor:
+    """
+    Expands attention_mask from `[batch_size, src_length]` to `[batch_size, 1, tgt_length, src_length]`.
+    """
+    batch_size, src_length = mask.shape
+    tgt_length = tgt_length if tgt_length is not None else src_length
+
+    expanded_mask = ~(mask[:, None, None, :].to(torch.bool))
+    return expanded_mask.expand(batch_size, 1, tgt_length, src_length)
+
+
+def dropout_add(x: torch.Tensor, residual: torch.Tensor, prob: float, training: bool) -> torch.Tensor:
+    """
+    Dropout add function
+
+    Args:
+        x (`torch.tensor`, *required*):
+            input tensor
+        residual (`torch.tensor`, *required*):
+            residual tensor
+        prob (`float`, *required*):
+            dropout probability
+        training (`bool`, *required*):
+            training mode
+    """
+    out = F.dropout(x, p=prob, training=training)
+    out = residual + out
+    return out
+
+
+def telechat_gelu_forward(x: torch.Tensor) -> torch.Tensor:
+    """
+    Custom bias GELU function. Adapted from Megatron-DeepSpeed code. Here we use a simple implementation (inference) to
+    make the model jitable.
+
+    Args:
+        x (`torch.tensor`, *required*):
+            input hidden states
+    """
+    return x * 0.5 * (1.0 + torch.tanh(0.79788456 * x * (1 + 0.044715 * x * x)))
+
+
+def telechat_gelu_back(g: torch.Tensor, x: torch.Tensor) -> torch.Tensor:
+    """
+    gradient of tanh approximation of gelu gradient of actual gelu is: 0.5 * (1. + torch.erf(x * 0.70710678)) +
+    0.3989423 * x * torch.exp(-0.5 * x * x)
+
+    Args:
+        g (`torch.tensor`, *required*):
+            gradient output tensor
+        x (`torch.tensor`, *required*):
+            input tensor
+    """
+    x = x[0]  # x is a tuple of 1 element, needs to unpack it first
+    tanh_out = torch.tanh(0.79788456 * x * (1 + 0.044715 * x * x))
+    # sqrt(2/pi) * 3 * 0.044715 -> 0.1070322243
+    ff = 0.5 * x * ((1 - tanh_out * tanh_out) * (0.79788456 + 0.1070322243 * x * x)) + 0.5 * (1 + tanh_out)
+    return ff * g
+
+
+class GeLUFunction(torch.autograd.Function):
+    @staticmethod
+    def forward(ctx, input: torch.Tensor) -> torch.Tensor:
+        ctx.save_for_backward(input)
+        return telechat_gelu_forward(input)
+
+    @staticmethod
+    def backward(ctx, grad_output: torch.Tensor) -> torch.Tensor:
+        input = ctx.saved_tensors
+        tmp = telechat_gelu_back(grad_output, input)
+        return tmp
+
+
+class TelechatGelu(nn.Module):
+    """
+    TelechatBiasGelu wrapper function that make use of the simple function on inference mode to make the model
+    torchscriptable and use the autograd function in training mode to get the accurate results of the gradients Partly
+    copied from Megatron-DeepSpeed code and adapted for our needs
+
+    See here why autograd functions are not torchscriptable: https://github.com/pytorch/pytorch/issues/22329
+    """
+
+    def __init__(self):
+        super().__init__()
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        if self.training:
+            return GeLUFunction.apply(x)
+        else:
+            return telechat_gelu_forward(x)
+
+
+class TelechatAttention(nn.Module):
+    def __init__(self, config: TelechatConfig, layer_idx):
+        super().__init__()
+        self.kv_cache = None
+        self.layer_idx = layer_idx
+
+        self.hidden_size = config.hidden_size
+        self.num_heads = config.n_head
+        self.head_dim = self.hidden_size // self.num_heads
+        self.split_size = self.hidden_size
+        self.hidden_dropout = config.hidden_dropout
+        self.config = config
+
+        if self.head_dim * self.num_heads != self.hidden_size:
+            raise ValueError(
+                f"`hidden_size` must be divisible by num_heads (got `hidden_size`: {self.hidden_size} and `num_heads`:"
+                f" {self.num_heads})."
+            )
+
+        # Layer-wise attention scaling
+        self.inv_norm_factor = 1.0 / math.sqrt(self.head_dim)
+        self.beta = 1.0
+
+        self.num_key_value_heads = config.num_key_value_heads if config.num_key_value_heads else self.num_heads
+        self.kv_projection_size = self.head_dim * self.num_key_value_heads
+        self.num_key_value_groups = self.num_heads // self.num_key_value_heads
+        self.query = nn.Linear(self.hidden_size, self.hidden_size, bias=False)
+        self.key_value = nn.Linear(self.hidden_size, self.kv_projection_size * 2, bias=False)
+        self.dense = nn.Linear(self.hidden_size, self.hidden_size)
+        self.attention_dropout = nn.Dropout(config.attention_dropout)
+        self.rotary_emb = RotaryEmbedding(self.head_dim, config=config)
+
+        self.core_attention_flash = FlashSelfAttention(
+            causal=True, attention_dropout=config.attention_dropout
+        )
+
+        self.last_key_layer = None
+        # logn_list = [math.log(i, 4096) if i > 4096 else 1 for i in range(1, 32768)]
+        # self.logn_tensor = torch.tensor(logn_list)[None, :, None, None].half().cuda()
+
+    def repeat_kv(self, hidden_states, n_rep):
+        slen, batch, num_key_value_heads_per_partition, head_dim = hidden_states.shape
+        if n_rep == 1:
+            return hidden_states
+        hidden_states = hidden_states[:, :, :, None, :].expand(slen, batch, num_key_value_heads_per_partition, n_rep,
+                                                               head_dim)
+        return hidden_states.reshape(slen, batch, num_key_value_heads_per_partition * n_rep, head_dim)
+
+    def split_tensor_along_last_dim(self,
+                                    tensor: torch.Tensor,
+                                    num_partitions: int,
+                                    contiguous_split_chunks: bool = False,
+                                    ):
+
+        # Get the size and dimension.
+        last_dim = tensor.dim() - 1
+        last_dim_size = tensor.size()[last_dim] // num_partitions
+        # Split.
+        tensor_list = torch.split(tensor, last_dim_size, dim=last_dim)
+        # Note: torch.split does not create contiguous tensors by default.
+        if contiguous_split_chunks:
+            return tuple(chunk.contiguous() for chunk in tensor_list)
+
+        return tensor_list
+
+    def _merge_heads(self, x: torch.Tensor) -> torch.Tensor:
+        batch_size_and_num_heads, seq_length, _ = x.shape
+        batch_size = batch_size_and_num_heads // self.num_heads
+        x = x.view(batch_size, self.num_heads, seq_length, self.head_dim)
+        x = x.permute(0, 2, 1, 3)
+        return x.reshape(batch_size, seq_length, self.num_heads * self.head_dim)
+
+    def forward(
+            self,
+            hidden_states: torch.Tensor,
+            residual: torch.Tensor,
+            attention_mask: torch.Tensor,
+            layer_past: Optional[Tuple[torch.Tensor, torch.Tensor]] = None,
+            use_cache: bool = False,
+            output_attentions: bool = False,
+    ):
+        hidden_states = hidden_states.transpose(1, 0)
+        query_layer = self.query(hidden_states)
+        new_tensor_shape = query_layer.size()[:-1] + \
+                           (self.num_heads,
+                            self.head_dim)
+        query_layer = query_layer.view(*new_tensor_shape)
+
+        mixed_kv_layer = self.key_value(hidden_states)
+        new_tensor_shape = mixed_kv_layer.size()[:-1] + \
+                           (self.num_key_value_heads,
+                            2 * self.head_dim)
+        mixed_kv_layer = mixed_kv_layer.view(*new_tensor_shape)
+        (key_layer, value_layer) = self.split_tensor_along_last_dim(mixed_kv_layer, 2)
+
+        output_size = (query_layer.size(1),
+                       query_layer.size(2),
+                       query_layer.size(0),
+                       key_layer.size(0),
+                       key_layer.size(2)
+                       )
+
+        query_layer = query_layer.view(output_size[2], output_size[0] * output_size[1], -1)
+        key_layer = key_layer.view(output_size[3], output_size[0] * output_size[4], -1)
+
+        apply_rotary_fn = apply_rotary_pos_emb_torch
+
+        seq_len = key_layer.shape[0]
+        offset = 0
+
+        if use_cache and layer_past != None:
+            past_key, past_value = layer_past
+            offset = past_key.shape[0]
+            seq_len += offset
+
+        cos, sin = self.rotary_emb(value_layer, seq_len=seq_len)
+
+        query_layer, key_layer = apply_rotary_fn(query_layer, key_layer, cos, sin, offset=offset)
+        if use_cache:
+            if layer_past != None:
+                past_key, past_value = layer_past
+                key_layer = torch.cat((past_key, key_layer[-1, ...].unsqueeze(0)), dim=0)
+                value_layer = torch.cat((past_value, value_layer[-1, ...].unsqueeze(0)), dim=0)
+            layer_past = key_layer, value_layer
+
+        s_value, bz, kv_head, dim = value_layer.shape
+        s_key = key_layer.shape[0]
+        s_query = query_layer.shape[0]
+        q_head = output_size[1]
+
+        query_layer = query_layer.reshape((s_query, bz, q_head, dim))
+        key_layer = key_layer.reshape((s_key, bz, kv_head, dim))
+
+        key_layer = self.repeat_kv(key_layer, self.num_key_value_groups)
+        value_layer = self.repeat_kv(value_layer, self.num_key_value_groups)
+
+        if self.config.flash_attn:
+            q, k, v = [rearrange(x, 's b ... -> b s ...').contiguous() for x in
+                       (query_layer, key_layer, value_layer)]
+            context_layer = self.core_attention_flash(q, k, v)
+            context_layer = rearrange(context_layer, 'b s h d -> b s (h d)').contiguous()
+        else:
+            ##[sq, b, np, hn] -> [sq, b * np, hn]
+            query_layer = query_layer.reshape(s_query, bz * self.num_heads, dim)
+            # [sk, b, np, hn] -> [sk, b * np, hn]
+            key_layer = key_layer.reshape(s_key, bz * self.num_heads, dim)
+            matmul_result = self.inv_norm_factor * torch.einsum('bik,bkj->bij', query_layer.transpose(0, 1),
+                                                                key_layer.transpose(0, 1).transpose(1, 2))
+
+            attention_scores = matmul_result.view(bz, self.num_heads, s_query, s_key)
+
+            input_dtype = attention_scores.dtype
+            if input_dtype == torch.float16:
+                attention_scores = attention_scores.to(torch.float)
+            attn_weights = torch.masked_fill(attention_scores, attention_mask, torch.finfo(attention_scores.dtype).min)
+            attention_probs = F.softmax(attn_weights, dim=-1).to(input_dtype)  ##dtype = torch.float32
+            attention_probs = self.attention_dropout(attention_probs)
+            attention_probs_reshaped = attention_probs.view(bz * self.num_heads, s_query, s_key)
+
+            value_layer = value_layer.reshape(s_key, bz * self.num_heads, dim)
+            context_layer = torch.bmm(attention_probs_reshaped, value_layer.transpose(0, 1))
+            context_layer = self._merge_heads(context_layer)
+        output_tensor = self.dense(context_layer)
+
+        output_tensor = dropout_add(output_tensor, residual, self.hidden_dropout, self.training)
+        present = None
+        outputs = (output_tensor, present)
+        if output_attentions:
+            outputs += (attention_probs,)
+
+        return output_tensor, layer_past
+
+
+class TelechatMLP(nn.Module):
+    def __init__(self, config: TelechatConfig):
+        super().__init__()
+        hidden_size = config.hidden_size
+        self.gate_proj = nn.Linear(hidden_size, config.ffn_hidden_size, bias=False)
+        self.up_proj = nn.Linear(hidden_size, config.ffn_hidden_size, bias=False)
+        self.down_proj = nn.Linear(config.ffn_hidden_size, hidden_size, bias=True)
+        self.hidden_dropout = config.hidden_dropout
+
+    def forward(self, hidden_states: torch.Tensor, residual: torch.Tensor) -> torch.Tensor:
+        intermediate_output = self.down_proj(F.silu(self.gate_proj(hidden_states)) * self.up_proj(hidden_states))
+        output = dropout_add(intermediate_output, residual, self.hidden_dropout, self.training)
+        return output
+
+
+class TelechatBlock(nn.Module):
+    def __init__(self, config: TelechatConfig, layer_idx):
+        super().__init__()
+        hidden_size = config.hidden_size
+
+        self.input_layernorm = MixedFusedRMSNorm(hidden_size, eps=config.layer_norm_epsilon)
+        self.num_heads = config.n_head
+        self.layer_idx = layer_idx
+        self.self_attention = TelechatAttention(config, layer_idx)
+        self.post_attention_layernorm = MixedFusedRMSNorm(hidden_size, eps=config.layer_norm_epsilon)
+
+        self.mlp = TelechatMLP(config)
+
+        self.apply_residual_connection_post_layernorm = config.apply_residual_connection_post_layernorm
+        self.hidden_dropout = config.hidden_dropout
+
+    def forward(
+            self,
+            hidden_states: torch.Tensor,
+            attention_mask: torch.Tensor,
+            layer_past: Optional[Tuple[torch.Tensor, torch.Tensor]] = None,
+            use_cache: bool = False,
+            output_attentions: bool = False,
+    ):
+        layernorm_output = self.input_layernorm(hidden_states)
+        if self.apply_residual_connection_post_layernorm:
+            residual = layernorm_output
+        else:
+            residual = hidden_states
+
+        attn_outputs = self.self_attention(
+            layernorm_output,
+            residual,
+            layer_past=layer_past,
+            attention_mask=attention_mask,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+        )
+
+        attention_output = attn_outputs[0]
+        outputs = attn_outputs[1:]
+        layernorm_output = self.post_attention_layernorm(attention_output)
+
+        if self.apply_residual_connection_post_layernorm:
+            residual = layernorm_output
+        else:
+            residual = attention_output
+        output = self.mlp(layernorm_output, residual)
+
+        if use_cache:
+            outputs = (output,) + outputs
+        else:
+            outputs = (output,) + outputs[1:]
+
+        return outputs
+
+
+class TelechatPreTrainedModel(PreTrainedModel):
+    config_class = TelechatConfig
+    base_model_prefix = "transformer"
+    supports_gradient_checkpointing = True
+    _no_split_modules = ["TelechatBlock"]
+    _skip_keys_device_placement = "past_key_values"
+
+    def __init__(self, *inputs, **kwargs):
+        super().__init__(*inputs, **kwargs)
+
+    def _init_weights(self, module: nn.Module):
+        """Initialize the weights."""
+        if isinstance(module, nn.Linear):
+            module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
+            if module.bias is not None:
+                module.bias.data.zero_()
+
+        elif isinstance(module, nn.Embedding):
+            module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
+            if module.padding_idx is not None:
+                module.weight.data[module.padding_idx].zero_()
+
+        elif isinstance(module, LayerNorm):
+            module.bias.data.zero_()
+            module.weight.data.fill_(1.0)
+
+    def _set_gradient_checkpointing(self, module: nn.Module, value: bool = False):
+        if isinstance(module, TelechatModel):
+            module.gradient_checkpointing = value
+
+
+class TelechatModel(TelechatPreTrainedModel):
+    def __init__(self, config: TelechatConfig):
+        super().__init__(config)
+
+        self.embed_dim = config.hidden_size
+        self.num_heads = config.n_head
+        self.config = config
+        self.word_embeddings = nn.Embedding(config.vocab_size, self.embed_dim)
+        if self.config.embed_layernorm:
+            self.word_embeddings_layernorm = MixedFusedRMSNorm(self.embed_dim, eps=config.layer_norm_epsilon)
+
+        self.h = nn.ModuleList([TelechatBlock(config, _) for _ in range(config.num_hidden_layers)])
+        self.ln_f = MixedFusedRMSNorm(self.embed_dim, eps=config.layer_norm_epsilon)
+        self.gradient_checkpointing = False
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.word_embeddings
+
+    def _prepare_attn_mask(
+            self, attention_mask: torch.Tensor, input_shape: Tuple[int, int], past_key_values_length: int
+    ) -> torch.BoolTensor:
+        combined_attention_mask = None
+        device = attention_mask.device
+        _, src_length = input_shape
+
+        if src_length > 1:
+            combined_attention_mask = _make_causal_mask(
+                input_shape, device=device, past_key_values_length=past_key_values_length
+            )
+        expanded_attn_mask = _expand_mask(attention_mask, tgt_length=src_length)
+        combined_attention_mask = (
+            expanded_attn_mask if combined_attention_mask is None else expanded_attn_mask | combined_attention_mask
+        )
+
+        return combined_attention_mask
+
+    def set_input_embeddings(self, new_embeddings: torch.Tensor):
+        self.word_embeddings = new_embeddings
+
+    def forward(
+            self,
+            input_ids: Optional[torch.LongTensor] = None,
+            past_key_values: Optional[Tuple[Tuple[torch.Tensor, torch.Tensor], ...]] = None,
+            attention_mask: Optional[torch.Tensor] = None,
+            inputs_embeds: Optional[torch.LongTensor] = None,
+            use_cache: Optional[bool] = None,
+            output_attentions: Optional[bool] = None,
+            output_hidden_states: Optional[bool] = None,
+            return_dict: Optional[bool] = None,
+            **deprecated_arguments,
+    ) -> Union[Tuple[torch.Tensor, ...], BaseModelOutputWithPastAndCrossAttentions]:
+
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        use_cache = use_cache if use_cache is not None else self.config.use_cache
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        if input_ids is not None:
+            batch_size, seq_length = input_ids.shape
+        elif inputs_embeds is not None:
+            batch_size, seq_length, _ = inputs_embeds.shape
+
+        if past_key_values is None:
+            past_key_values = tuple([None] * len(self.h))
+        # input_ids = torch.load("Megatron-LM-0624-3B/tensors/input_ids.pt").to(input_ids.device)
+        if inputs_embeds is None:
+            inputs_embeds = self.word_embeddings(input_ids)
+        hidden_states = inputs_embeds
+        # print(f"[INFO_Telechat]: inputs_embeds={inputs_embeds}")
+        if self.config.embed_layernorm:
+            hidden_states = self.word_embeddings_layernorm(inputs_embeds)
+
+        presents = () if use_cache else None
+        all_self_attentions = () if output_attentions else None
+        all_hidden_states = () if output_hidden_states else None
+
+        if self.gradient_checkpointing and self.training:
+            if use_cache:
+                use_cache = False
+
+        seq_length_with_past = seq_length
+        past_key_values_length = 0
+        if past_key_values[0] is not None:
+            past_key_values_length = past_key_values[0][0].shape[2]
+            seq_length_with_past = seq_length_with_past + past_key_values_length
+        if attention_mask is None:
+            attention_mask = torch.ones((batch_size, seq_length_with_past), device=hidden_states.device)
+        else:
+            attention_mask = attention_mask.to(hidden_states.device)
+        causal_mask = self._prepare_attn_mask(
+            attention_mask,
+            input_shape=(batch_size, seq_length),
+            past_key_values_length=past_key_values_length,
+        )
+
+        # print(f"[INFO_Telechat]: word_embeddings_layernorm={hidden_states}")
+        for i, (block, layer_past) in enumerate(zip(self.h, past_key_values)):
+            if output_hidden_states:
+                all_hidden_states = all_hidden_states + (hidden_states,)
+
+            if self.gradient_checkpointing and self.training:
+
+                def create_custom_forward(module):
+                    def custom_forward(*inputs):
+                        # None for past_key_value
+                        return module(*inputs, use_cache=use_cache, output_attentions=output_attentions)
+
+                    return custom_forward
+
+                outputs = torch.utils.checkpoint.checkpoint(
+                    create_custom_forward(block),
+                    hidden_states,
+                    causal_mask,
+                    layer_past,
+                )
+            else:
+                outputs = block(
+                    hidden_states,
+                    layer_past=layer_past,
+                    attention_mask=causal_mask,
+                    use_cache=use_cache,
+                    output_attentions=output_attentions,
+                )
+
+            # print(f"[INFO_Telechat]: outputs{i}={outputs}")
+            hidden_states = outputs[0]
+            if use_cache is True:
+                presents = presents + (outputs[1],)
+
+            if output_attentions:
+                all_self_attentions = all_self_attentions + (outputs[2 if use_cache else 1],)
+        hidden_states = self.ln_f(hidden_states)
+        # print(f"[INFO_Telechat]: hidden_states={hidden_states}")
+        # ref = torch.load("Megatron-LM-0624-3B/tensors/final_layernorm.pt")
+        # print(hidden_states.squeeze()[2048:])
+        # print(ref.squeeze())
+        # print(torch.max(hidden_states.squeeze()[2048:] - ref.squeeze().to(hidden_states.device)))
+        # exit()
+        # print(ref.shape,hidden_states.shape)
+        # print(hidden_states)
+        # exit()
+        if output_hidden_states:
+            all_hidden_states = all_hidden_states + (hidden_states,)
+        if not return_dict:
+            return tuple(v for v in [hidden_states, presents, all_hidden_states, all_self_attentions] if v is not None)
+        return BaseModelOutputWithPastAndCrossAttentions(
+            last_hidden_state=hidden_states,
+            past_key_values=presents,
+            hidden_states=all_hidden_states,
+            attentions=all_self_attentions,
+        )
+
+
+class TelechatForCausalLM(TelechatPreTrainedModel):
+    # _tied_weights_keys = ["lm_head.weight"]
+    _keys_to_ignore_on_load_missing = [r"lm_head.weight"]
+
+    def __init__(self, config: TelechatConfig):
+        super().__init__(config)
+        self.transformer = TelechatModel(config)
+        self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
+        self.post_init()
+
+    def get_output_embeddings(self):
+        return self.lm_head
+
+    def set_output_embeddings(self, new_embeddings: torch.Tensor):
+        self.lm_head = new_embeddings
+
+    def prepare_inputs_for_generation(
+            self,
+            input_ids: torch.LongTensor,
+            past_key_values: Optional[torch.Tensor] = None,
+            attention_mask: Optional[torch.Tensor] = None,
+            inputs_embeds: Optional[torch.Tensor] = None,
+            **kwargs,
+    ) -> dict:
+        if past_key_values:
+            input_ids = input_ids[:, -1].unsqueeze(-1)
+        if inputs_embeds is not None and past_key_values is None:
+            model_inputs = {"inputs_embeds": inputs_embeds}
+        else:
+            model_inputs = {"input_ids": input_ids}
+
+        model_inputs.update(
+            {
+                "past_key_values": past_key_values,
+                "use_cache": kwargs.get("use_cache"),
+                "attention_mask": attention_mask,
+            }
+        )
+        return model_inputs
+
+    def forward(
+            self,
+            input_ids: Optional[torch.LongTensor] = None,
+            past_key_values: Optional[Tuple[Tuple[torch.Tensor, torch.Tensor], ...]] = None,
+            attention_mask: Optional[torch.Tensor] = None,
+            inputs_embeds: Optional[torch.Tensor] = None,
+            labels: Optional[torch.Tensor] = None,
+            use_cache: Optional[bool] = None,
+            output_attentions: Optional[bool] = None,
+            output_hidden_states: Optional[bool] = None,
+            return_dict: Optional[bool] = None,
+            **deprecated_arguments,
+    ) -> Union[Tuple[torch.Tensor], CausalLMOutputWithCrossAttentions]:
+
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        transformer_outputs = self.transformer(
+            input_ids,
+            past_key_values=past_key_values,
+            attention_mask=attention_mask,
+            inputs_embeds=inputs_embeds,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+        hidden_states = transformer_outputs[0]
+        lm_logits = self.lm_head(hidden_states)
+
+        loss = None
+        if labels is not None:
+            labels = labels.to(lm_logits.device)
+            shift_logits = lm_logits[..., :-1, :].contiguous()
+            shift_labels = labels[..., 1:].contiguous()
+            batch_size, seq_length, vocab_size = shift_logits.shape
+            loss_fct = CrossEntropyLoss()
+            loss = loss_fct(
+                shift_logits.view(batch_size * seq_length, vocab_size), shift_labels.view(batch_size * seq_length)
+            )
+
+        if not return_dict:
+            output = (lm_logits,) + transformer_outputs[1:]
+            return ((loss,) + output) if loss is not None else output
+
+        return CausalLMOutputWithCrossAttentions(
+            loss=loss,
+            logits=lm_logits,
+            past_key_values=transformer_outputs.past_key_values,
+            hidden_states=transformer_outputs.hidden_states,
+            attentions=transformer_outputs.attentions,
+        )
+
+    def chat(self, tokenizer, question: str = '', history: Union[List[Dict], History] = None, stream: bool = False,
+             generation_config: Optional[GenerationConfig] = None, **kwargs):
+        """
+        Args:
+            tokenizer:  the tokenizer of  telechat
+            question: question which the model reply in this turn
+            history: history which will format the input for telechat
+            stream: if return the full text at last or yield the text in token
+            generation_config:  configuration for generation
+            **kwargs: args which will update the generation config or pass to model forward
+        """
+        generation_config = generation_config or self.generation_config
+        if not generation_config:
+            logger.error("generation_config is None")
+            raise ValueError("generation_config must not be None")
+        if not question:
+            logger.error("question is empty")
+            raise ValueError("question must not be empty")
+        if history is None:
+            history = []
+
+        # we update and check generate_config here for building inputs.
+
+        generation_config = copy.deepcopy(generation_config)
+        user_id = generation_config.user_token_id
+        bot_id = generation_config.bot_token_id
+        model_kwargs = generation_config.update(**kwargs)
+        generation_config.validate()
+
+        # transfer to History
+        if not isinstance(history, History):
+            history = History(tokenizer, history)
+
+        inputs = self.build_inputs_for_chat(tokenizer, question, history, generation_config, user_id, bot_id)
+        history.append({"role": "user", "content": question})
+        if stream:
+            streamer = TelechatIterTextStreamer(tokenizer, history, skip_prompt=True)
+            Thread(target=self.generate, kwargs=dict(
+                inputs=inputs.to(self.device), streamer=streamer,
+                generation_config=generation_config, **model_kwargs
+            )).start()
+            return streamer
+        else:
+            outputs = self.generate(inputs.to(self.device), generation_config=generation_config, **model_kwargs)
+            response = tokenizer.decode(outputs[0][len(inputs[0]):-1])
+            history.append({"role": "bot", "content": response})
+            return response, history
+
+    def build_inputs_for_chat(self, tokenizer, question, history, generation_config, usr_id, bot_id):
+        """
+        check history and  build inputs here
+        """
+        # first tokenize question
+        q_token = tokenizer(question)
+        qa_history = copy.deepcopy(history)
+
+        # get the max length we should build our inputs in
+        model_max_length = self.config.seq_length
+        build_max_length = max(0, model_max_length - generation_config.max_new_tokens - 1) \
+            if generation_config.max_new_tokens else max(0, generation_config.max_length)
+        if build_max_length < 3:
+            logger.warning("the model can not meet the  requirements of input length,Please check config")
+            raise ValueError("")
+
+        # trunc left
+        input_tokens = [usr_id] + q_token["input_ids"][-build_max_length + 1:] + [bot_id]
+        length = len(input_tokens)
+
+        while len(qa_history) != 0:
+            message = qa_history.pop()
+            if message["role"] == "user":
+                tokens = [usr_id] + message["input_ids"]
+            elif message["role"] == "bot":
+                tokens = [bot_id] + message["input_ids"] + [generation_config.eos_token_id]
+            else:
+                tokens = []
+            if len(tokens) + length >= build_max_length:
+                break
+            else:
+                input_tokens = tokens + input_tokens
+
+        input_tokens = [generation_config.bos_token_id] + input_tokens
+
+        return torch.tensor([input_tokens], dtype=torch.int64)
diff --git a/tokenization_telechat.py b/tokenization_telechat.py
new file mode 100644
index 0000000000000000000000000000000000000000..6ac4fb87adaa33ad7850e7964157b9f5b335b435
--- /dev/null
+++ b/tokenization_telechat.py
@@ -0,0 +1,220 @@
+import os
+from shutil import copyfile
+from typing import Any, Dict, List, Optional, Tuple
+import sentencepiece as spm
+from transformers.tokenization_utils import AddedToken, PreTrainedTokenizer
+from transformers.utils import logging
+
+logger = logging.get_logger(__name__)
+
+VOCAB_FILES_NAMES = {"vocab_file": "tokenizer.model"}
+
+# TODO: when we get download url from huggingface, refresh the map
+PRETRAINED_VOCAB_FILES_MAP = {
+    "vocab_file": {},
+    "tokenizer_file": {},
+}
+
+
+class TelechatTokenizer(PreTrainedTokenizer):
+
+    vocab_files_names = VOCAB_FILES_NAMES
+    pretrained_vocab_files_map = PRETRAINED_VOCAB_FILES_MAP
+    model_input_names = ["input_ids", "attention_mask"]
+
+    def __init__(
+        self,
+        vocab_file,
+        unk_token="<unk>",
+        bos_token="<_start>",
+        eos_token="<_end>",
+        pad_token="<_pad>",
+        sp_model_kwargs: Optional[Dict[str, Any]] = None,
+        add_bos_token=True,
+        add_eos_token=False,
+        clean_up_tokenization_spaces=False,
+        **kwargs,
+    ):
+        self.sp_model_kwargs = {} if sp_model_kwargs is None else sp_model_kwargs
+        bos_token = AddedToken(bos_token, lstrip=False, rstrip=False) if isinstance(bos_token, str) else bos_token
+        eos_token = AddedToken(eos_token, lstrip=False, rstrip=False) if isinstance(eos_token, str) else eos_token
+        unk_token = AddedToken(unk_token, lstrip=False, rstrip=False) if isinstance(unk_token, str) else unk_token
+        pad_token = AddedToken(pad_token, lstrip=False, rstrip=False) if isinstance(pad_token, str) else pad_token
+        self.sp_model = spm.SentencePieceProcessor(**self.sp_model_kwargs)
+        self.sp_model.Load(vocab_file)
+        super().__init__(
+            bos_token=bos_token,
+            eos_token=eos_token,
+            unk_token=unk_token,
+            pad_token=pad_token,
+            add_bos_token=add_bos_token,
+            add_eos_token=add_eos_token,
+            sp_model_kwargs=self.sp_model_kwargs,
+            clean_up_tokenization_spaces=clean_up_tokenization_spaces,
+            **kwargs,
+        )
+        self.vocab_file = vocab_file
+        self.add_bos_token = add_bos_token
+        self.add_eos_token = add_eos_token
+
+
+    def __getstate__(self):
+        state = self.__dict__.copy()
+        state["sp_model"] = None
+        return state
+
+    def __setstate__(self, d):
+        self.__dict__ = d
+        self.sp_model = spm.SentencePieceProcessor(**self.sp_model_kwargs)
+        self.sp_model.Load(self.vocab_file)
+
+    @property
+    def vocab_size(self):
+        """Returns vocab size"""
+        return self.sp_model.get_piece_size()
+
+    def get_vocab(self):
+        """Returns vocab as a dict"""
+        vocab = {self.convert_ids_to_tokens(i): i for i in range(self.vocab_size)}
+        vocab.update(self.added_tokens_encoder)
+        return vocab
+
+    def _tokenize(self, text):
+        """Returns a tokenized string."""
+        return self.sp_model.encode(text, out_type=str)
+
+    def _convert_token_to_id(self, token):
+        """Converts a token (str) in an id using the vocab."""
+        return self.sp_model.piece_to_id(token)
+
+    def _convert_id_to_token(self, index):
+        """Converts an index (integer) in a token (str) using the vocab."""
+        token = self.sp_model.IdToPiece(index)
+        return token
+
+    def convert_tokens_to_string(self, tokens):
+        """Converts a sequence of tokens (string) in a single string."""
+        current_sub_tokens = []
+        out_string = ""
+        prev_is_special = False
+        for i, token in enumerate(tokens):
+            # make sure that special tokens are not decoded using sentencepiece model
+            if token in self.all_special_tokens:
+                if not prev_is_special and i != 0:
+                    out_string += " "
+                out_string += self.sp_model.decode(current_sub_tokens) + token
+                prev_is_special = True
+                current_sub_tokens = []
+            else:
+                current_sub_tokens.append(token)
+                prev_is_special = False
+        out_string += self.sp_model.decode(current_sub_tokens)
+        return out_string
+
+    def save_vocabulary(self, save_directory, filename_prefix: Optional[str] = None) -> Tuple[str]:
+        """
+        Save the vocabulary and special tokens file to a directory.
+
+        Args:
+            save_directory (`str`):
+                The directory in which to save the vocabulary.
+
+        Returns:
+            `Tuple(str)`: Paths to the files saved.
+        """
+        if not os.path.isdir(save_directory):
+            logger.error(f"Vocabulary path ({save_directory}) should be a directory")
+            return
+        out_vocab_file = os.path.join(
+            save_directory, (filename_prefix + "-" if filename_prefix else "") + VOCAB_FILES_NAMES["vocab_file"]
+        )
+
+        if os.path.abspath(self.vocab_file) != os.path.abspath(out_vocab_file) and os.path.isfile(self.vocab_file):
+            copyfile(self.vocab_file, out_vocab_file)
+        elif not os.path.isfile(self.vocab_file):
+            with open(out_vocab_file, "wb") as fi:
+                content_spiece_model = self.sp_model.serialized_model_proto()
+                fi.write(content_spiece_model)
+
+        return (out_vocab_file,)
+
+    def build_inputs_with_special_tokens(self, token_ids_0, token_ids_1=None):
+        bos_token_id = [self.bos_token_id] if self.add_bos_token else []
+        eos_token_id = [self.eos_token_id] if self.add_eos_token else []
+
+        output = bos_token_id + token_ids_0 + eos_token_id
+
+        if token_ids_1 is not None:
+            output = output + bos_token_id + token_ids_1 + eos_token_id
+
+        return output
+
+    def get_special_tokens_mask(
+        self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None, already_has_special_tokens: bool = False
+    ) -> List[int]:
+        """
+        Retrieve sequence ids from a token list that has no special tokens added. This method is called when adding
+        special tokens using the tokenizer `prepare_for_model` method.
+
+        Args:
+            token_ids_0 (`List[int]`):
+                List of IDs.
+            token_ids_1 (`List[int]`, *optional*):
+                Optional second list of IDs for sequence pairs.
+            already_has_special_tokens (`bool`, *optional*, defaults to `False`):
+                Whether or not the token list is already formatted with special tokens for the model.
+
+        Returns:
+            `List[int]`: A list of integers in the range [0, 1]: 1 for a special token, 0 for a sequence token.
+        """
+        if already_has_special_tokens:
+            return super().get_special_tokens_mask(
+                token_ids_0=token_ids_0, token_ids_1=token_ids_1, already_has_special_tokens=True
+            )
+
+        bos_token_id = [1] if self.add_bos_token else []
+        eos_token_id = [1] if self.add_eos_token else []
+
+        if token_ids_1 is None:
+            return bos_token_id + ([0] * len(token_ids_0)) + eos_token_id
+        return (
+            bos_token_id
+            + ([0] * len(token_ids_0))
+            + eos_token_id
+            + bos_token_id
+            + ([0] * len(token_ids_1))
+            + eos_token_id
+        )
+
+    def create_token_type_ids_from_sequences(
+        self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None
+    ) -> List[int]:
+        """
+        Creates a mask from the two sequences passed to be used in a sequence-pair classification task. An ALBERT
+        sequence pair mask has the following format:
+
+        ```
+        0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1
+        | first sequence    | second sequence |
+        ```
+
+        if token_ids_1 is None, only returns the first portion of the mask (0s).
+
+        Args:
+            token_ids_0 (`List[int]`):
+                List of ids.
+            token_ids_1 (`List[int]`, *optional*):
+                Optional second list of IDs for sequence pairs.
+
+        Returns:
+            `List[int]`: List of [token type IDs](../glossary#token-type-ids) according to the given sequence(s).
+        """
+        bos_token_id = [self.bos_token_id] if self.add_bos_token else []
+        eos_token_id = [self.eos_token_id] if self.add_eos_token else []
+
+        output = [0] * len(bos_token_id + token_ids_0 + eos_token_id)
+
+        if token_ids_1 is not None:
+            output += [1] * len(bos_token_id + token_ids_1 + eos_token_id)
+
+        return output
diff --git a/tokenizer.model b/tokenizer.model
new file mode 100644
index 0000000000000000000000000000000000000000..cd47d1356749ba43803322ca2ca295c2c776b036
--- /dev/null
+++ b/tokenizer.model
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:db947024849f75ec4bd9af5d4c84fa71e96a26971eb353a70acd66194fc7a69b
+size 2197489
diff --git a/tokenizer_config.json b/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..8e8c261678ad76df6865abca3d3250e2ba3c04a7
--- /dev/null
+++ b/tokenizer_config.json
@@ -0,0 +1,39 @@
+{
+  "tokenizer_class": "TelechatTokenizer",
+  "auto_map": {
+    "AutoTokenizer": [
+      "tokenization_telechat.TelechatTokenizer",
+      null
+    ]
+  },
+  "add_bos_token": false,
+  "add_eos_token": false,
+  "use_fast": false,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": {
+    "__type": "AddedToken",
+    "content": "<_end>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": true
+  },
+  "model_max_length": 100000000,
+  "sp_model_kwargs": {},
+  "pad_token": {
+    "__type": "AddedToken",
+    "content": "<_pad>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": true
+  },
+  "unk_token": {
+    "__type": "AddedToken",
+    "content": "<_end>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": true
+  }
+}