blockblockblock commited on
Commit
f53693f
1 Parent(s): 4557cd0

Upload folder using huggingface_hub

Browse files
README.md ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ ---
2
+ datasets:
3
+ - open-web-math/open-web-math
4
+ ---
5
+
6
+ Mistral-7b with continued pretraining using Quiet-STaR (https://arxiv.org/abs/2403.09629) for generating 8 thought tokens before each output token.
added_tokens.json ADDED
@@ -0,0 +1,4 @@
 
 
 
 
 
1
+ {
2
+ "<|endthought|>": 32000,
3
+ "<|startthought|>": 32001
4
+ }
config.json ADDED
@@ -0,0 +1,52 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "Crystalcareai/Quiet-Star-Custom",
3
+ "architectures": [
4
+ "QuietForCausalLM"
5
+ ],
6
+ "attention_dropout": 0.0,
7
+ "auto_map": {
8
+ "AutoConfig": "configuration_quiet.QuietConfig",
9
+ "AutoModel": "modeling_quiet.QuietModel",
10
+ "AutoModelForCausalLM": "modeling_quiet.QuietForCausalLM"
11
+ },
12
+ "bos_token_id": 1,
13
+ "eos_token_id": 2,
14
+ "hidden_act": "silu",
15
+ "hidden_size": 4096,
16
+ "initializer_range": 0.02,
17
+ "intermediate_size": 14336,
18
+ "max_position_embeddings": 32768,
19
+ "max_thoughts": 10,
20
+ "merged_lm_and_talk_heads": false,
21
+ "merged_lm_and_think_heads": true,
22
+ "merged_talk_heads": true,
23
+ "model_type": "quiet",
24
+ "num_attention_heads": 32,
25
+ "num_hidden_layers": 32,
26
+ "num_key_value_heads": 8,
27
+ "rms_norm_eps": 1e-05,
28
+ "rope_theta": 10000.0,
29
+ "sliding_window": 4096,
30
+ "tie_word_embeddings": false,
31
+ "torch_dtype": "bfloat16",
32
+ "transformers_version": "4.37.0.dev0",
33
+ "use_cache": true,
34
+ "use_complex_talk_head": true,
35
+ "use_complex_think_head": false,
36
+ "use_concat_talk_head": true,
37
+ "use_shallow_talk": false,
38
+ "use_shallow_think": true,
39
+ "use_weighted_talk_head": true,
40
+ "vocab_size": 32002,
41
+ "quantization_config": {
42
+ "quant_method": "exl2",
43
+ "version": "0.0.15",
44
+ "bits": 2.5,
45
+ "head_bits": 6,
46
+ "calibration": {
47
+ "rows": 100,
48
+ "length": 2048,
49
+ "dataset": "(default)"
50
+ }
51
+ }
52
+ }
configuration_quiet.py ADDED
@@ -0,0 +1,172 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # coding=utf-8
2
+ # Copyright 2023 Quiet AI and the HuggingFace Inc. team. All rights reserved.
3
+ #
4
+ # Licensed under the Apache License, Version 2.0 (the "License");
5
+ # you may not use this file except in compliance with the License.
6
+ # You may obtain a copy of the License at
7
+ #
8
+ # http://www.apache.org/licenses/LICENSE-2.0
9
+ #
10
+ # Unless required by applicable law or agreed to in writing, software
11
+ # distributed under the License is distributed on an "AS IS" BASIS,
12
+ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13
+ # See the License for the specific language governing permissions and
14
+ # limitations under the License.
15
+ """ Quiet model configuration"""
16
+
17
+ from transformers.configuration_utils import PretrainedConfig
18
+ from transformers.utils import logging
19
+
20
+
21
+ logger = logging.get_logger(__name__)
22
+
23
+ QUIET_PRETRAINED_CONFIG_ARCHIVE_MAP = {
24
+ "quietai/Quiet-7B-v0.1": "https://huggingface.co/quietai/Quiet-7B-v0.1/resolve/main/config.json",
25
+ "quietai/Quiet-7B-Instruct-v0.1": "https://huggingface.co/quietai/Quiet-7B-Instruct-v0.1/resolve/main/config.json",
26
+ }
27
+
28
+
29
+ class QuietConfig(PretrainedConfig):
30
+ r"""
31
+ This is the configuration class to store the configuration of a [`QuietModel`]. It is used to instantiate an
32
+ Quiet model according to the specified arguments, defining the model architecture. Instantiating a configuration
33
+ with the defaults will yield a similar configuration to that of the Quiet-7B-v0.1 or Quiet-7B-Instruct-v0.1.
34
+
35
+ [quietai/Quiet-7B-v0.1](https://huggingface.co/quietai/Quiet-7B-v0.1)
36
+ [quietai/Quiet-7B-Instruct-v0.1](https://huggingface.co/quietai/Quiet-7B-Instruct-v0.1)
37
+
38
+ Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
39
+ documentation from [`PretrainedConfig`] for more information.
40
+
41
+
42
+ Args:
43
+ vocab_size (`int`, *optional*, defaults to 32000):
44
+ Vocabulary size of the Quiet model. Defines the number of different tokens that can be represented by the
45
+ `inputs_ids` passed when calling [`QuietModel`]
46
+ hidden_size (`int`, *optional*, defaults to 4096):
47
+ Dimension of the hidden representations.
48
+ intermediate_size (`int`, *optional*, defaults to 14336):
49
+ Dimension of the MLP representations.
50
+ num_hidden_layers (`int`, *optional*, defaults to 32):
51
+ Number of hidden layers in the Transformer encoder.
52
+ num_attention_heads (`int`, *optional*, defaults to 32):
53
+ Number of attention heads for each attention layer in the Transformer encoder.
54
+ num_key_value_heads (`int`, *optional*, defaults to 8):
55
+ This is the number of key_value heads that should be used to implement Grouped Query Attention. If
56
+ `num_key_value_heads=num_attention_heads`, the model will use Multi Head Attention (MHA), if
57
+ `num_key_value_heads=1 the model will use Multi Query Attention (MQA) otherwise GQA is used. When
58
+ converting a multi-head checkpoint to a GQA checkpoint, each group key and value head should be constructed
59
+ by meanpooling all the original heads within that group. For more details checkout [this
60
+ paper](https://arxiv.org/pdf/2305.13245.pdf). If it is not specified, will default to `8`.
61
+ hidden_act (`str` or `function`, *optional*, defaults to `"silu"`):
62
+ The non-linear activation function (function or string) in the decoder.
63
+ max_position_embeddings (`int`, *optional*, defaults to `4096*32`):
64
+ The maximum sequence length that this model might ever be used with. Quiet's sliding window attention
65
+ allows sequence of up to 4096*32 tokens.
66
+ initializer_range (`float`, *optional*, defaults to 0.02):
67
+ The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
68
+ rms_norm_eps (`float`, *optional*, defaults to 1e-06):
69
+ The epsilon used by the rms normalization layers.
70
+ use_cache (`bool`, *optional*, defaults to `True`):
71
+ Whether or not the model should return the last key/values attentions (not used by all models). Only
72
+ relevant if `config.is_decoder=True`.
73
+ pad_token_id (`int`, *optional*):
74
+ The id of the padding token.
75
+ bos_token_id (`int`, *optional*, defaults to 1):
76
+ The id of the "beginning-of-sequence" token.
77
+ eos_token_id (`int`, *optional*, defaults to 2):
78
+ The id of the "end-of-sequence" token.
79
+ tie_word_embeddings (`bool`, *optional*, defaults to `False`):
80
+ Whether the model's input and output word embeddings should be tied.
81
+ rope_theta (`float`, *optional*, defaults to 10000.0):
82
+ The base period of the RoPE embeddings.
83
+ sliding_window (`int`, *optional*, defaults to 4096):
84
+ Sliding window attention window size. If not specified, will default to `4096`.
85
+ attention_dropout (`float`, *optional*, defaults to 0.0):
86
+ The dropout ratio for the attention probabilities.
87
+
88
+ ```python
89
+ >>> from transformers import QuietModel, QuietConfig
90
+
91
+ >>> # Initializing a Quiet 7B style configuration
92
+ >>> configuration = QuietConfig()
93
+
94
+ >>> # Initializing a model from the Quiet 7B style configuration
95
+ >>> model = QuietModel(configuration)
96
+
97
+ >>> # Accessing the model configuration
98
+ >>> configuration = model.config
99
+ ```"""
100
+
101
+ model_type = "quiet"
102
+ keys_to_ignore_at_inference = ["past_key_values"]
103
+
104
+ def __init__(
105
+ self,
106
+ vocab_size=32000,
107
+ hidden_size=4096,
108
+ intermediate_size=14336,
109
+ num_hidden_layers=32,
110
+ num_attention_heads=32,
111
+ num_key_value_heads=8,
112
+ hidden_act="silu",
113
+ max_position_embeddings=4096 * 32,
114
+ initializer_range=0.02,
115
+ rms_norm_eps=1e-6,
116
+ use_cache=True,
117
+ pad_token_id=None,
118
+ bos_token_id=1,
119
+ eos_token_id=2,
120
+ tie_word_embeddings=False,
121
+ rope_theta=10000.0,
122
+ sliding_window=4096,
123
+ attention_dropout=0.0,
124
+ max_thoughts=16,
125
+ merged_talk_heads=True,
126
+ merged_lm_and_talk_heads=False,
127
+ merged_lm_and_think_heads=True,
128
+ use_concat_talk_head=True,
129
+ use_shallow_think=True,
130
+ use_shallow_talk=False,
131
+ use_complex_think_head=False,
132
+ use_complex_talk_head=True,
133
+ use_weighted_talk_head=True,
134
+ **kwargs,
135
+ ):
136
+ self.vocab_size = vocab_size
137
+ self.max_position_embeddings = max_position_embeddings
138
+ self.hidden_size = hidden_size
139
+ self.intermediate_size = intermediate_size
140
+ self.num_hidden_layers = num_hidden_layers
141
+ self.num_attention_heads = num_attention_heads
142
+ self.sliding_window = sliding_window
143
+
144
+ # for backward compatibility
145
+ if num_key_value_heads is None:
146
+ num_key_value_heads = num_attention_heads
147
+
148
+ self.num_key_value_heads = num_key_value_heads
149
+ self.hidden_act = hidden_act
150
+ self.initializer_range = initializer_range
151
+ self.rms_norm_eps = rms_norm_eps
152
+ self.use_cache = use_cache
153
+ self.rope_theta = rope_theta
154
+ self.attention_dropout = attention_dropout
155
+ self.max_thoughts = max_thoughts
156
+ self.merged_talk_heads = merged_talk_heads
157
+ self.merged_lm_and_talk_heads = merged_lm_and_talk_heads
158
+ self.merged_lm_and_think_heads = merged_lm_and_think_heads
159
+ self.use_concat_talk_head = use_concat_talk_head
160
+ self.use_shallow_think = use_shallow_think
161
+ self.use_shallow_talk = use_shallow_talk
162
+ self.use_complex_think_head = use_complex_think_head
163
+ self.use_complex_talk_head = use_complex_talk_head
164
+ self.use_weighted_talk_head = use_weighted_talk_head
165
+
166
+ super().__init__(
167
+ pad_token_id=pad_token_id,
168
+ bos_token_id=bos_token_id,
169
+ eos_token_id=eos_token_id,
170
+ tie_word_embeddings=tie_word_embeddings,
171
+ **kwargs,
172
+ )
generation_config.json ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "bos_token_id": 1,
4
+ "eos_token_id": 2,
5
+ "transformers_version": "4.37.0.dev0"
6
+ }
inference.py ADDED
@@ -0,0 +1,36 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import torch
2
+ from transformers import AutoTokenizer, TextStreamer, AutoModelForCausalLM
3
+
4
+ model_path = "Crystalcareai/GemMoE-Medium-v0.4"
5
+
6
+ # Load model
7
+ model = AutoModelForCausalLM.from_pretrained(
8
+ model_path,
9
+ device_map="auto",
10
+ low_cpu_mem_usage=True,
11
+ torch_dtype=torch.float16,
12
+ attn_implementation="flash_attention_2"
13
+ trust_remote_code=True,
14
+ )
15
+
16
+ tokenizer = AutoTokenizer.from_pretrained(model_path)
17
+ streamer = TextStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)
18
+
19
+ # Convert prompt to tokens
20
+ prompt_template = "[INST] {prompt} [/INST]"
21
+
22
+ prompt = "You're standing on the surface of the Earth. "\
23
+ "You walk one mile south, one mile west and one mile north. "\
24
+ "You end up exactly where you started. Where are you?"
25
+
26
+ tokens = tokenizer(
27
+ prompt_template.format(prompt=prompt),
28
+ return_tensors='pt'
29
+ ).input_ids.cuda()
30
+
31
+ # Generate output
32
+ generation_output = model.generate(
33
+ tokens,
34
+ streamer=streamer,
35
+ max_new_tokens=512
36
+ )
model.safetensors.index.json ADDED
@@ -0,0 +1,305 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "metadata": {
3
+ "total_size": 14584217600
4
+ },
5
+ "weight_map": {
6
+ "end_embedding": "model-00001-of-00003.safetensors",
7
+ "lm_head.weight": "model-00003-of-00003.safetensors",
8
+ "model.embed_tokens.weight": "model-00001-of-00003.safetensors",
9
+ "model.layers.0.input_layernorm.weight": "model-00001-of-00003.safetensors",
10
+ "model.layers.0.mlp.down_proj.weight": "model-00001-of-00003.safetensors",
11
+ "model.layers.0.mlp.gate_proj.weight": "model-00001-of-00003.safetensors",
12
+ "model.layers.0.mlp.up_proj.weight": "model-00001-of-00003.safetensors",
13
+ "model.layers.0.post_attention_layernorm.weight": "model-00001-of-00003.safetensors",
14
+ "model.layers.0.self_attn.k_proj.weight": "model-00001-of-00003.safetensors",
15
+ "model.layers.0.self_attn.o_proj.weight": "model-00001-of-00003.safetensors",
16
+ "model.layers.0.self_attn.q_proj.weight": "model-00001-of-00003.safetensors",
17
+ "model.layers.0.self_attn.v_proj.weight": "model-00001-of-00003.safetensors",
18
+ "model.layers.1.input_layernorm.weight": "model-00001-of-00003.safetensors",
19
+ "model.layers.1.mlp.down_proj.weight": "model-00001-of-00003.safetensors",
20
+ "model.layers.1.mlp.gate_proj.weight": "model-00001-of-00003.safetensors",
21
+ "model.layers.1.mlp.up_proj.weight": "model-00001-of-00003.safetensors",
22
+ "model.layers.1.post_attention_layernorm.weight": "model-00001-of-00003.safetensors",
23
+ "model.layers.1.self_attn.k_proj.weight": "model-00001-of-00003.safetensors",
24
+ "model.layers.1.self_attn.o_proj.weight": "model-00001-of-00003.safetensors",
25
+ "model.layers.1.self_attn.q_proj.weight": "model-00001-of-00003.safetensors",
26
+ "model.layers.1.self_attn.v_proj.weight": "model-00001-of-00003.safetensors",
27
+ "model.layers.10.input_layernorm.weight": "model-00002-of-00003.safetensors",
28
+ "model.layers.10.mlp.down_proj.weight": "model-00002-of-00003.safetensors",
29
+ "model.layers.10.mlp.gate_proj.weight": "model-00001-of-00003.safetensors",
30
+ "model.layers.10.mlp.up_proj.weight": "model-00001-of-00003.safetensors",
31
+ "model.layers.10.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
32
+ "model.layers.10.self_attn.k_proj.weight": "model-00001-of-00003.safetensors",
33
+ "model.layers.10.self_attn.o_proj.weight": "model-00001-of-00003.safetensors",
34
+ "model.layers.10.self_attn.q_proj.weight": "model-00001-of-00003.safetensors",
35
+ "model.layers.10.self_attn.v_proj.weight": "model-00001-of-00003.safetensors",
36
+ "model.layers.11.input_layernorm.weight": "model-00002-of-00003.safetensors",
37
+ "model.layers.11.mlp.down_proj.weight": "model-00002-of-00003.safetensors",
38
+ "model.layers.11.mlp.gate_proj.weight": "model-00002-of-00003.safetensors",
39
+ "model.layers.11.mlp.up_proj.weight": "model-00002-of-00003.safetensors",
40
+ "model.layers.11.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
41
+ "model.layers.11.self_attn.k_proj.weight": "model-00002-of-00003.safetensors",
42
+ "model.layers.11.self_attn.o_proj.weight": "model-00002-of-00003.safetensors",
43
+ "model.layers.11.self_attn.q_proj.weight": "model-00002-of-00003.safetensors",
44
+ "model.layers.11.self_attn.v_proj.weight": "model-00002-of-00003.safetensors",
45
+ "model.layers.12.input_layernorm.weight": "model-00002-of-00003.safetensors",
46
+ "model.layers.12.mlp.down_proj.weight": "model-00002-of-00003.safetensors",
47
+ "model.layers.12.mlp.gate_proj.weight": "model-00002-of-00003.safetensors",
48
+ "model.layers.12.mlp.up_proj.weight": "model-00002-of-00003.safetensors",
49
+ "model.layers.12.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
50
+ "model.layers.12.self_attn.k_proj.weight": "model-00002-of-00003.safetensors",
51
+ "model.layers.12.self_attn.o_proj.weight": "model-00002-of-00003.safetensors",
52
+ "model.layers.12.self_attn.q_proj.weight": "model-00002-of-00003.safetensors",
53
+ "model.layers.12.self_attn.v_proj.weight": "model-00002-of-00003.safetensors",
54
+ "model.layers.13.input_layernorm.weight": "model-00002-of-00003.safetensors",
55
+ "model.layers.13.mlp.down_proj.weight": "model-00002-of-00003.safetensors",
56
+ "model.layers.13.mlp.gate_proj.weight": "model-00002-of-00003.safetensors",
57
+ "model.layers.13.mlp.up_proj.weight": "model-00002-of-00003.safetensors",
58
+ "model.layers.13.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
59
+ "model.layers.13.self_attn.k_proj.weight": "model-00002-of-00003.safetensors",
60
+ "model.layers.13.self_attn.o_proj.weight": "model-00002-of-00003.safetensors",
61
+ "model.layers.13.self_attn.q_proj.weight": "model-00002-of-00003.safetensors",
62
+ "model.layers.13.self_attn.v_proj.weight": "model-00002-of-00003.safetensors",
63
+ "model.layers.14.input_layernorm.weight": "model-00002-of-00003.safetensors",
64
+ "model.layers.14.mlp.down_proj.weight": "model-00002-of-00003.safetensors",
65
+ "model.layers.14.mlp.gate_proj.weight": "model-00002-of-00003.safetensors",
66
+ "model.layers.14.mlp.up_proj.weight": "model-00002-of-00003.safetensors",
67
+ "model.layers.14.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
68
+ "model.layers.14.self_attn.k_proj.weight": "model-00002-of-00003.safetensors",
69
+ "model.layers.14.self_attn.o_proj.weight": "model-00002-of-00003.safetensors",
70
+ "model.layers.14.self_attn.q_proj.weight": "model-00002-of-00003.safetensors",
71
+ "model.layers.14.self_attn.v_proj.weight": "model-00002-of-00003.safetensors",
72
+ "model.layers.15.input_layernorm.weight": "model-00002-of-00003.safetensors",
73
+ "model.layers.15.mlp.down_proj.weight": "model-00002-of-00003.safetensors",
74
+ "model.layers.15.mlp.gate_proj.weight": "model-00002-of-00003.safetensors",
75
+ "model.layers.15.mlp.up_proj.weight": "model-00002-of-00003.safetensors",
76
+ "model.layers.15.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
77
+ "model.layers.15.self_attn.k_proj.weight": "model-00002-of-00003.safetensors",
78
+ "model.layers.15.self_attn.o_proj.weight": "model-00002-of-00003.safetensors",
79
+ "model.layers.15.self_attn.q_proj.weight": "model-00002-of-00003.safetensors",
80
+ "model.layers.15.self_attn.v_proj.weight": "model-00002-of-00003.safetensors",
81
+ "model.layers.16.input_layernorm.weight": "model-00002-of-00003.safetensors",
82
+ "model.layers.16.mlp.down_proj.weight": "model-00002-of-00003.safetensors",
83
+ "model.layers.16.mlp.gate_proj.weight": "model-00002-of-00003.safetensors",
84
+ "model.layers.16.mlp.up_proj.weight": "model-00002-of-00003.safetensors",
85
+ "model.layers.16.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
86
+ "model.layers.16.self_attn.k_proj.weight": "model-00002-of-00003.safetensors",
87
+ "model.layers.16.self_attn.o_proj.weight": "model-00002-of-00003.safetensors",
88
+ "model.layers.16.self_attn.q_proj.weight": "model-00002-of-00003.safetensors",
89
+ "model.layers.16.self_attn.v_proj.weight": "model-00002-of-00003.safetensors",
90
+ "model.layers.17.input_layernorm.weight": "model-00002-of-00003.safetensors",
91
+ "model.layers.17.mlp.down_proj.weight": "model-00002-of-00003.safetensors",
92
+ "model.layers.17.mlp.gate_proj.weight": "model-00002-of-00003.safetensors",
93
+ "model.layers.17.mlp.up_proj.weight": "model-00002-of-00003.safetensors",
94
+ "model.layers.17.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
95
+ "model.layers.17.self_attn.k_proj.weight": "model-00002-of-00003.safetensors",
96
+ "model.layers.17.self_attn.o_proj.weight": "model-00002-of-00003.safetensors",
97
+ "model.layers.17.self_attn.q_proj.weight": "model-00002-of-00003.safetensors",
98
+ "model.layers.17.self_attn.v_proj.weight": "model-00002-of-00003.safetensors",
99
+ "model.layers.18.input_layernorm.weight": "model-00002-of-00003.safetensors",
100
+ "model.layers.18.mlp.down_proj.weight": "model-00002-of-00003.safetensors",
101
+ "model.layers.18.mlp.gate_proj.weight": "model-00002-of-00003.safetensors",
102
+ "model.layers.18.mlp.up_proj.weight": "model-00002-of-00003.safetensors",
103
+ "model.layers.18.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
104
+ "model.layers.18.self_attn.k_proj.weight": "model-00002-of-00003.safetensors",
105
+ "model.layers.18.self_attn.o_proj.weight": "model-00002-of-00003.safetensors",
106
+ "model.layers.18.self_attn.q_proj.weight": "model-00002-of-00003.safetensors",
107
+ "model.layers.18.self_attn.v_proj.weight": "model-00002-of-00003.safetensors",
108
+ "model.layers.19.input_layernorm.weight": "model-00002-of-00003.safetensors",
109
+ "model.layers.19.mlp.down_proj.weight": "model-00002-of-00003.safetensors",
110
+ "model.layers.19.mlp.gate_proj.weight": "model-00002-of-00003.safetensors",
111
+ "model.layers.19.mlp.up_proj.weight": "model-00002-of-00003.safetensors",
112
+ "model.layers.19.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
113
+ "model.layers.19.self_attn.k_proj.weight": "model-00002-of-00003.safetensors",
114
+ "model.layers.19.self_attn.o_proj.weight": "model-00002-of-00003.safetensors",
115
+ "model.layers.19.self_attn.q_proj.weight": "model-00002-of-00003.safetensors",
116
+ "model.layers.19.self_attn.v_proj.weight": "model-00002-of-00003.safetensors",
117
+ "model.layers.2.input_layernorm.weight": "model-00001-of-00003.safetensors",
118
+ "model.layers.2.mlp.down_proj.weight": "model-00001-of-00003.safetensors",
119
+ "model.layers.2.mlp.gate_proj.weight": "model-00001-of-00003.safetensors",
120
+ "model.layers.2.mlp.up_proj.weight": "model-00001-of-00003.safetensors",
121
+ "model.layers.2.post_attention_layernorm.weight": "model-00001-of-00003.safetensors",
122
+ "model.layers.2.self_attn.k_proj.weight": "model-00001-of-00003.safetensors",
123
+ "model.layers.2.self_attn.o_proj.weight": "model-00001-of-00003.safetensors",
124
+ "model.layers.2.self_attn.q_proj.weight": "model-00001-of-00003.safetensors",
125
+ "model.layers.2.self_attn.v_proj.weight": "model-00001-of-00003.safetensors",
126
+ "model.layers.20.input_layernorm.weight": "model-00002-of-00003.safetensors",
127
+ "model.layers.20.mlp.down_proj.weight": "model-00002-of-00003.safetensors",
128
+ "model.layers.20.mlp.gate_proj.weight": "model-00002-of-00003.safetensors",
129
+ "model.layers.20.mlp.up_proj.weight": "model-00002-of-00003.safetensors",
130
+ "model.layers.20.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
131
+ "model.layers.20.self_attn.k_proj.weight": "model-00002-of-00003.safetensors",
132
+ "model.layers.20.self_attn.o_proj.weight": "model-00002-of-00003.safetensors",
133
+ "model.layers.20.self_attn.q_proj.weight": "model-00002-of-00003.safetensors",
134
+ "model.layers.20.self_attn.v_proj.weight": "model-00002-of-00003.safetensors",
135
+ "model.layers.21.input_layernorm.weight": "model-00002-of-00003.safetensors",
136
+ "model.layers.21.mlp.down_proj.weight": "model-00002-of-00003.safetensors",
137
+ "model.layers.21.mlp.gate_proj.weight": "model-00002-of-00003.safetensors",
138
+ "model.layers.21.mlp.up_proj.weight": "model-00002-of-00003.safetensors",
139
+ "model.layers.21.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
140
+ "model.layers.21.self_attn.k_proj.weight": "model-00002-of-00003.safetensors",
141
+ "model.layers.21.self_attn.o_proj.weight": "model-00002-of-00003.safetensors",
142
+ "model.layers.21.self_attn.q_proj.weight": "model-00002-of-00003.safetensors",
143
+ "model.layers.21.self_attn.v_proj.weight": "model-00002-of-00003.safetensors",
144
+ "model.layers.22.input_layernorm.weight": "model-00003-of-00003.safetensors",
145
+ "model.layers.22.mlp.down_proj.weight": "model-00003-of-00003.safetensors",
146
+ "model.layers.22.mlp.gate_proj.weight": "model-00003-of-00003.safetensors",
147
+ "model.layers.22.mlp.up_proj.weight": "model-00003-of-00003.safetensors",
148
+ "model.layers.22.post_attention_layernorm.weight": "model-00003-of-00003.safetensors",
149
+ "model.layers.22.self_attn.k_proj.weight": "model-00002-of-00003.safetensors",
150
+ "model.layers.22.self_attn.o_proj.weight": "model-00002-of-00003.safetensors",
151
+ "model.layers.22.self_attn.q_proj.weight": "model-00002-of-00003.safetensors",
152
+ "model.layers.22.self_attn.v_proj.weight": "model-00002-of-00003.safetensors",
153
+ "model.layers.23.input_layernorm.weight": "model-00003-of-00003.safetensors",
154
+ "model.layers.23.mlp.down_proj.weight": "model-00003-of-00003.safetensors",
155
+ "model.layers.23.mlp.gate_proj.weight": "model-00003-of-00003.safetensors",
156
+ "model.layers.23.mlp.up_proj.weight": "model-00003-of-00003.safetensors",
157
+ "model.layers.23.post_attention_layernorm.weight": "model-00003-of-00003.safetensors",
158
+ "model.layers.23.self_attn.k_proj.weight": "model-00003-of-00003.safetensors",
159
+ "model.layers.23.self_attn.o_proj.weight": "model-00003-of-00003.safetensors",
160
+ "model.layers.23.self_attn.q_proj.weight": "model-00003-of-00003.safetensors",
161
+ "model.layers.23.self_attn.v_proj.weight": "model-00003-of-00003.safetensors",
162
+ "model.layers.24.input_layernorm.weight": "model-00003-of-00003.safetensors",
163
+ "model.layers.24.mlp.down_proj.weight": "model-00003-of-00003.safetensors",
164
+ "model.layers.24.mlp.gate_proj.weight": "model-00003-of-00003.safetensors",
165
+ "model.layers.24.mlp.up_proj.weight": "model-00003-of-00003.safetensors",
166
+ "model.layers.24.post_attention_layernorm.weight": "model-00003-of-00003.safetensors",
167
+ "model.layers.24.self_attn.k_proj.weight": "model-00003-of-00003.safetensors",
168
+ "model.layers.24.self_attn.o_proj.weight": "model-00003-of-00003.safetensors",
169
+ "model.layers.24.self_attn.q_proj.weight": "model-00003-of-00003.safetensors",
170
+ "model.layers.24.self_attn.v_proj.weight": "model-00003-of-00003.safetensors",
171
+ "model.layers.25.input_layernorm.weight": "model-00003-of-00003.safetensors",
172
+ "model.layers.25.mlp.down_proj.weight": "model-00003-of-00003.safetensors",
173
+ "model.layers.25.mlp.gate_proj.weight": "model-00003-of-00003.safetensors",
174
+ "model.layers.25.mlp.up_proj.weight": "model-00003-of-00003.safetensors",
175
+ "model.layers.25.post_attention_layernorm.weight": "model-00003-of-00003.safetensors",
176
+ "model.layers.25.self_attn.k_proj.weight": "model-00003-of-00003.safetensors",
177
+ "model.layers.25.self_attn.o_proj.weight": "model-00003-of-00003.safetensors",
178
+ "model.layers.25.self_attn.q_proj.weight": "model-00003-of-00003.safetensors",
179
+ "model.layers.25.self_attn.v_proj.weight": "model-00003-of-00003.safetensors",
180
+ "model.layers.26.input_layernorm.weight": "model-00003-of-00003.safetensors",
181
+ "model.layers.26.mlp.down_proj.weight": "model-00003-of-00003.safetensors",
182
+ "model.layers.26.mlp.gate_proj.weight": "model-00003-of-00003.safetensors",
183
+ "model.layers.26.mlp.up_proj.weight": "model-00003-of-00003.safetensors",
184
+ "model.layers.26.post_attention_layernorm.weight": "model-00003-of-00003.safetensors",
185
+ "model.layers.26.self_attn.k_proj.weight": "model-00003-of-00003.safetensors",
186
+ "model.layers.26.self_attn.o_proj.weight": "model-00003-of-00003.safetensors",
187
+ "model.layers.26.self_attn.q_proj.weight": "model-00003-of-00003.safetensors",
188
+ "model.layers.26.self_attn.v_proj.weight": "model-00003-of-00003.safetensors",
189
+ "model.layers.27.input_layernorm.weight": "model-00003-of-00003.safetensors",
190
+ "model.layers.27.mlp.down_proj.weight": "model-00003-of-00003.safetensors",
191
+ "model.layers.27.mlp.gate_proj.weight": "model-00003-of-00003.safetensors",
192
+ "model.layers.27.mlp.up_proj.weight": "model-00003-of-00003.safetensors",
193
+ "model.layers.27.post_attention_layernorm.weight": "model-00003-of-00003.safetensors",
194
+ "model.layers.27.self_attn.k_proj.weight": "model-00003-of-00003.safetensors",
195
+ "model.layers.27.self_attn.o_proj.weight": "model-00003-of-00003.safetensors",
196
+ "model.layers.27.self_attn.q_proj.weight": "model-00003-of-00003.safetensors",
197
+ "model.layers.27.self_attn.v_proj.weight": "model-00003-of-00003.safetensors",
198
+ "model.layers.28.input_layernorm.weight": "model-00003-of-00003.safetensors",
199
+ "model.layers.28.mlp.down_proj.weight": "model-00003-of-00003.safetensors",
200
+ "model.layers.28.mlp.gate_proj.weight": "model-00003-of-00003.safetensors",
201
+ "model.layers.28.mlp.up_proj.weight": "model-00003-of-00003.safetensors",
202
+ "model.layers.28.post_attention_layernorm.weight": "model-00003-of-00003.safetensors",
203
+ "model.layers.28.self_attn.k_proj.weight": "model-00003-of-00003.safetensors",
204
+ "model.layers.28.self_attn.o_proj.weight": "model-00003-of-00003.safetensors",
205
+ "model.layers.28.self_attn.q_proj.weight": "model-00003-of-00003.safetensors",
206
+ "model.layers.28.self_attn.v_proj.weight": "model-00003-of-00003.safetensors",
207
+ "model.layers.29.input_layernorm.weight": "model-00003-of-00003.safetensors",
208
+ "model.layers.29.mlp.down_proj.weight": "model-00003-of-00003.safetensors",
209
+ "model.layers.29.mlp.gate_proj.weight": "model-00003-of-00003.safetensors",
210
+ "model.layers.29.mlp.up_proj.weight": "model-00003-of-00003.safetensors",
211
+ "model.layers.29.post_attention_layernorm.weight": "model-00003-of-00003.safetensors",
212
+ "model.layers.29.self_attn.k_proj.weight": "model-00003-of-00003.safetensors",
213
+ "model.layers.29.self_attn.o_proj.weight": "model-00003-of-00003.safetensors",
214
+ "model.layers.29.self_attn.q_proj.weight": "model-00003-of-00003.safetensors",
215
+ "model.layers.29.self_attn.v_proj.weight": "model-00003-of-00003.safetensors",
216
+ "model.layers.3.input_layernorm.weight": "model-00001-of-00003.safetensors",
217
+ "model.layers.3.mlp.down_proj.weight": "model-00001-of-00003.safetensors",
218
+ "model.layers.3.mlp.gate_proj.weight": "model-00001-of-00003.safetensors",
219
+ "model.layers.3.mlp.up_proj.weight": "model-00001-of-00003.safetensors",
220
+ "model.layers.3.post_attention_layernorm.weight": "model-00001-of-00003.safetensors",
221
+ "model.layers.3.self_attn.k_proj.weight": "model-00001-of-00003.safetensors",
222
+ "model.layers.3.self_attn.o_proj.weight": "model-00001-of-00003.safetensors",
223
+ "model.layers.3.self_attn.q_proj.weight": "model-00001-of-00003.safetensors",
224
+ "model.layers.3.self_attn.v_proj.weight": "model-00001-of-00003.safetensors",
225
+ "model.layers.30.input_layernorm.weight": "model-00003-of-00003.safetensors",
226
+ "model.layers.30.mlp.down_proj.weight": "model-00003-of-00003.safetensors",
227
+ "model.layers.30.mlp.gate_proj.weight": "model-00003-of-00003.safetensors",
228
+ "model.layers.30.mlp.up_proj.weight": "model-00003-of-00003.safetensors",
229
+ "model.layers.30.post_attention_layernorm.weight": "model-00003-of-00003.safetensors",
230
+ "model.layers.30.self_attn.k_proj.weight": "model-00003-of-00003.safetensors",
231
+ "model.layers.30.self_attn.o_proj.weight": "model-00003-of-00003.safetensors",
232
+ "model.layers.30.self_attn.q_proj.weight": "model-00003-of-00003.safetensors",
233
+ "model.layers.30.self_attn.v_proj.weight": "model-00003-of-00003.safetensors",
234
+ "model.layers.31.input_layernorm.weight": "model-00003-of-00003.safetensors",
235
+ "model.layers.31.mlp.down_proj.weight": "model-00003-of-00003.safetensors",
236
+ "model.layers.31.mlp.gate_proj.weight": "model-00003-of-00003.safetensors",
237
+ "model.layers.31.mlp.up_proj.weight": "model-00003-of-00003.safetensors",
238
+ "model.layers.31.post_attention_layernorm.weight": "model-00003-of-00003.safetensors",
239
+ "model.layers.31.self_attn.k_proj.weight": "model-00003-of-00003.safetensors",
240
+ "model.layers.31.self_attn.o_proj.weight": "model-00003-of-00003.safetensors",
241
+ "model.layers.31.self_attn.q_proj.weight": "model-00003-of-00003.safetensors",
242
+ "model.layers.31.self_attn.v_proj.weight": "model-00003-of-00003.safetensors",
243
+ "model.layers.4.input_layernorm.weight": "model-00001-of-00003.safetensors",
244
+ "model.layers.4.mlp.down_proj.weight": "model-00001-of-00003.safetensors",
245
+ "model.layers.4.mlp.gate_proj.weight": "model-00001-of-00003.safetensors",
246
+ "model.layers.4.mlp.up_proj.weight": "model-00001-of-00003.safetensors",
247
+ "model.layers.4.post_attention_layernorm.weight": "model-00001-of-00003.safetensors",
248
+ "model.layers.4.self_attn.k_proj.weight": "model-00001-of-00003.safetensors",
249
+ "model.layers.4.self_attn.o_proj.weight": "model-00001-of-00003.safetensors",
250
+ "model.layers.4.self_attn.q_proj.weight": "model-00001-of-00003.safetensors",
251
+ "model.layers.4.self_attn.v_proj.weight": "model-00001-of-00003.safetensors",
252
+ "model.layers.5.input_layernorm.weight": "model-00001-of-00003.safetensors",
253
+ "model.layers.5.mlp.down_proj.weight": "model-00001-of-00003.safetensors",
254
+ "model.layers.5.mlp.gate_proj.weight": "model-00001-of-00003.safetensors",
255
+ "model.layers.5.mlp.up_proj.weight": "model-00001-of-00003.safetensors",
256
+ "model.layers.5.post_attention_layernorm.weight": "model-00001-of-00003.safetensors",
257
+ "model.layers.5.self_attn.k_proj.weight": "model-00001-of-00003.safetensors",
258
+ "model.layers.5.self_attn.o_proj.weight": "model-00001-of-00003.safetensors",
259
+ "model.layers.5.self_attn.q_proj.weight": "model-00001-of-00003.safetensors",
260
+ "model.layers.5.self_attn.v_proj.weight": "model-00001-of-00003.safetensors",
261
+ "model.layers.6.input_layernorm.weight": "model-00001-of-00003.safetensors",
262
+ "model.layers.6.mlp.down_proj.weight": "model-00001-of-00003.safetensors",
263
+ "model.layers.6.mlp.gate_proj.weight": "model-00001-of-00003.safetensors",
264
+ "model.layers.6.mlp.up_proj.weight": "model-00001-of-00003.safetensors",
265
+ "model.layers.6.post_attention_layernorm.weight": "model-00001-of-00003.safetensors",
266
+ "model.layers.6.self_attn.k_proj.weight": "model-00001-of-00003.safetensors",
267
+ "model.layers.6.self_attn.o_proj.weight": "model-00001-of-00003.safetensors",
268
+ "model.layers.6.self_attn.q_proj.weight": "model-00001-of-00003.safetensors",
269
+ "model.layers.6.self_attn.v_proj.weight": "model-00001-of-00003.safetensors",
270
+ "model.layers.7.input_layernorm.weight": "model-00001-of-00003.safetensors",
271
+ "model.layers.7.mlp.down_proj.weight": "model-00001-of-00003.safetensors",
272
+ "model.layers.7.mlp.gate_proj.weight": "model-00001-of-00003.safetensors",
273
+ "model.layers.7.mlp.up_proj.weight": "model-00001-of-00003.safetensors",
274
+ "model.layers.7.post_attention_layernorm.weight": "model-00001-of-00003.safetensors",
275
+ "model.layers.7.self_attn.k_proj.weight": "model-00001-of-00003.safetensors",
276
+ "model.layers.7.self_attn.o_proj.weight": "model-00001-of-00003.safetensors",
277
+ "model.layers.7.self_attn.q_proj.weight": "model-00001-of-00003.safetensors",
278
+ "model.layers.7.self_attn.v_proj.weight": "model-00001-of-00003.safetensors",
279
+ "model.layers.8.input_layernorm.weight": "model-00001-of-00003.safetensors",
280
+ "model.layers.8.mlp.down_proj.weight": "model-00001-of-00003.safetensors",
281
+ "model.layers.8.mlp.gate_proj.weight": "model-00001-of-00003.safetensors",
282
+ "model.layers.8.mlp.up_proj.weight": "model-00001-of-00003.safetensors",
283
+ "model.layers.8.post_attention_layernorm.weight": "model-00001-of-00003.safetensors",
284
+ "model.layers.8.self_attn.k_proj.weight": "model-00001-of-00003.safetensors",
285
+ "model.layers.8.self_attn.o_proj.weight": "model-00001-of-00003.safetensors",
286
+ "model.layers.8.self_attn.q_proj.weight": "model-00001-of-00003.safetensors",
287
+ "model.layers.8.self_attn.v_proj.weight": "model-00001-of-00003.safetensors",
288
+ "model.layers.9.input_layernorm.weight": "model-00001-of-00003.safetensors",
289
+ "model.layers.9.mlp.down_proj.weight": "model-00001-of-00003.safetensors",
290
+ "model.layers.9.mlp.gate_proj.weight": "model-00001-of-00003.safetensors",
291
+ "model.layers.9.mlp.up_proj.weight": "model-00001-of-00003.safetensors",
292
+ "model.layers.9.post_attention_layernorm.weight": "model-00001-of-00003.safetensors",
293
+ "model.layers.9.self_attn.k_proj.weight": "model-00001-of-00003.safetensors",
294
+ "model.layers.9.self_attn.o_proj.weight": "model-00001-of-00003.safetensors",
295
+ "model.layers.9.self_attn.q_proj.weight": "model-00001-of-00003.safetensors",
296
+ "model.layers.9.self_attn.v_proj.weight": "model-00001-of-00003.safetensors",
297
+ "model.norm.weight": "model-00003-of-00003.safetensors",
298
+ "start_embedding": "model-00001-of-00003.safetensors",
299
+ "talk_head.0.0.bias": "model-00003-of-00003.safetensors",
300
+ "talk_head.0.0.weight": "model-00003-of-00003.safetensors",
301
+ "talk_head.0.2.bias": "model-00003-of-00003.safetensors",
302
+ "talk_head.0.2.weight": "model-00003-of-00003.safetensors",
303
+ "talk_head.0.4.weight": "model-00003-of-00003.safetensors"
304
+ }
305
+ }
modeling_quiet.py ADDED
The diff for this file is too large to render. See raw diff
 
output.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2c584cf56712170bba5022de2d9352b52da15c439b5c21e7ea3ba3e87a57b2d5
3
+ size 2547859492
special_tokens_map.json ADDED
@@ -0,0 +1,40 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "additional_special_tokens": [
3
+ {
4
+ "content": "<|endthought|>",
5
+ "lstrip": false,
6
+ "normalized": false,
7
+ "rstrip": false,
8
+ "single_word": false
9
+ },
10
+ {
11
+ "content": "<|startthought|>",
12
+ "lstrip": false,
13
+ "normalized": false,
14
+ "rstrip": false,
15
+ "single_word": false
16
+ }
17
+ ],
18
+ "bos_token": {
19
+ "content": "<s>",
20
+ "lstrip": false,
21
+ "normalized": false,
22
+ "rstrip": false,
23
+ "single_word": false
24
+ },
25
+ "eos_token": {
26
+ "content": "</s>",
27
+ "lstrip": false,
28
+ "normalized": false,
29
+ "rstrip": false,
30
+ "single_word": false
31
+ },
32
+ "pad_token": "</s>",
33
+ "unk_token": {
34
+ "content": "<unk>",
35
+ "lstrip": false,
36
+ "normalized": false,
37
+ "rstrip": false,
38
+ "single_word": false
39
+ }
40
+ }
tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
tokenizer.model ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:dadfd56d766715c61d2ef780a525ab43b8e6da4de6865bda3d95fdef5e134055
3
+ size 493443
tokenizer_config.json ADDED
@@ -0,0 +1,61 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_bos_token": true,
3
+ "add_eos_token": false,
4
+ "added_tokens_decoder": {
5
+ "0": {
6
+ "content": "<unk>",
7
+ "lstrip": false,
8
+ "normalized": false,
9
+ "rstrip": false,
10
+ "single_word": false,
11
+ "special": true
12
+ },
13
+ "1": {
14
+ "content": "<s>",
15
+ "lstrip": false,
16
+ "normalized": false,
17
+ "rstrip": false,
18
+ "single_word": false,
19
+ "special": true
20
+ },
21
+ "2": {
22
+ "content": "</s>",
23
+ "lstrip": false,
24
+ "normalized": false,
25
+ "rstrip": false,
26
+ "single_word": false,
27
+ "special": true
28
+ },
29
+ "32000": {
30
+ "content": "<|endthought|>",
31
+ "lstrip": false,
32
+ "normalized": false,
33
+ "rstrip": false,
34
+ "single_word": false,
35
+ "special": true
36
+ },
37
+ "32001": {
38
+ "content": "<|startthought|>",
39
+ "lstrip": false,
40
+ "normalized": false,
41
+ "rstrip": false,
42
+ "single_word": false,
43
+ "special": true
44
+ }
45
+ },
46
+ "additional_special_tokens": [
47
+ "<|endthought|>",
48
+ "<|startthought|>"
49
+ ],
50
+ "bos_token": "<s>",
51
+ "clean_up_tokenization_spaces": false,
52
+ "eos_token": "</s>",
53
+ "legacy": true,
54
+ "model_max_length": 1000000000000000019884624838656,
55
+ "pad_token": "</s>",
56
+ "sp_model_kwargs": {},
57
+ "spaces_between_special_tokens": false,
58
+ "tokenizer_class": "LlamaTokenizer",
59
+ "unk_token": "<unk>",
60
+ "use_default_system_prompt": false
61
+ }
train.py ADDED
@@ -0,0 +1,144 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import torch
2
+ torch.backends.cuda.matmul.allow_tf32 = True
3
+ import random
4
+ from transformers import AutoTokenizer, AutoModelForCausalLM
5
+ from datasets import load_dataset
6
+ from transformers import TrainingArguments
7
+ from trl import SFTTrainer
8
+ from peft import LoraConfig
9
+
10
+ import time
11
+ random_seed = 42
12
+ torch.manual_seed(random_seed)
13
+ random.seed(random_seed)
14
+
15
+ dataset = load_dataset("HuggingFaceH4/deita-10k-v0-sft", split="train_sft")
16
+
17
+ n_ahead_talk_global = 2
18
+ n_passes_global = 2
19
+ n_ahead_global = 2
20
+ n_examples = 0
21
+ full_batch_size = 2
22
+ eval_and_logging_steps = 2
23
+ save_steps = 100
24
+
25
+
26
+ def model_init(params):
27
+ original = False
28
+ if params is None:
29
+ params = {}
30
+ else:
31
+ params = params.params
32
+ # save params to file
33
+ n_ahead = params.get("n_ahead", n_ahead_global if not original else 1)
34
+ n_ahead_talk = params.get("n_ahead_talk", n_ahead_talk_global if not original else 1)
35
+ n_passes = params.get("n_passes", n_passes_global if not original else 1)
36
+ gumbel_temperature = params.get("gumbel_temperature", 1)
37
+ use_start_thought_token = params.get("use_start_thought_token", True)
38
+ use_end_thought_token = params.get("use_end_thought_token", True)
39
+ include_policy_loss = params.get("include_policy_loss", True)
40
+ gumbel_detach = params.get("gumbel_detach", True)
41
+ merged_talk_heads = params.get("merged_talk_heads", True)
42
+ gradient_accumulation_steps = params.get("gradient_accumulation_steps", global_gradient_accumulation_steps)
43
+ residual_think_head = params.get("residual_think_head", False)
44
+ optimize_lm_head_only_at_start = params.get("optimize_lm_head_only_at_start", False)
45
+
46
+ model_id = "Crystalcareai/Quiet-Star-Custom"
47
+ tokenizer_id = model_id
48
+ print("Loading model")
49
+ model = AutoModelForCausalLM.from_pretrained(
50
+ model_id,
51
+ torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
52
+ max_thoughts=n_ahead + n_ahead_talk + 1,
53
+ merged_talk_heads=merged_talk_heads,
54
+ merged_lm_and_talk_heads=False,
55
+ merged_lm_and_think_heads=True,
56
+ use_concat_talk_head=True,
57
+ use_shallow_think=True,
58
+ use_shallow_talk=False,
59
+ use_complex_think_head=False,
60
+ use_complex_talk_head=True,
61
+ use_weighted_talk_head=True,
62
+ trust_remote_code=True,
63
+ device_map="auto",
64
+ )
65
+ print("Loaded model")
66
+
67
+ tokenizer = AutoTokenizer.from_pretrained(tokenizer_id,padding=False,truncation=True)
68
+ tokenizer.pad_token_id = tokenizer.eos_token_id
69
+
70
+ special_tokens_to_add = []
71
+ if model.use_start_thought_token:
72
+ special_tokens_to_add.append("<|startthought|>")
73
+ if model.use_end_thought_token:
74
+ special_tokens_to_add.append("<|endthought|>")
75
+ if special_tokens_to_add:
76
+ tokenizer.add_special_tokens({"additional_special_tokens": special_tokens_to_add})
77
+ model.resize_token_embeddings(len(tokenizer))
78
+ model.tokenizer = tokenizer
79
+ model.gumbel_detach = gumbel_detach
80
+ model.include_policy_loss = include_policy_loss
81
+ model.use_end_thought_token = use_end_thought_token
82
+ model.use_start_thought_token = use_start_thought_token
83
+ model.n_ahead = n_ahead
84
+ model.n_ahead_talk = n_ahead_talk
85
+ model.n_passes = n_passes
86
+ model.n_tokens_print = gradient_accumulation_steps
87
+ model.gradient_accumulation_steps = gradient_accumulation_steps
88
+ model.residual_think_head = residual_think_head
89
+ model.optimize_lm_head_only_at_start = optimize_lm_head_only_at_start
90
+ model.gumbel_temperature = gumbel_temperature
91
+ model.original_mode = original
92
+ model.config_params = params
93
+ model.run_start = int(time.time())
94
+ model.kill_after = 100
95
+ model.train()
96
+ return model
97
+
98
+
99
+ batch_size = full_batch_size // n_passes_global
100
+ global_gradient_accumulation_steps = full_batch_size // batch_size
101
+ run_id = int(time.time())
102
+ training_args = TrainingArguments(
103
+ output_dir="./out",
104
+ num_train_epochs=3,
105
+ per_device_train_batch_size=1,
106
+ gradient_checkpointing=False,
107
+ gradient_accumulation_steps=4,
108
+ optim="adamw_torch_fused",
109
+ logging_steps=1,
110
+ save_strategy="steps",
111
+ save_steps=300,
112
+ bf16=True,
113
+ tf32=False,
114
+ # auto_find_batch_size=True
115
+ learning_rate=2e-07,
116
+ max_grad_norm=1.0, # Gradient clipping with a maximum gradient norm of 0.3
117
+ warmup_steps=100,
118
+ lr_scheduler_type="cosine",
119
+ push_to_hub=False,
120
+ )
121
+
122
+ # peft_config = LoraConfig(
123
+ # r = 16, # Choose any number > 0 ! Suggested 8, 16, 32, 64, 128
124
+ # target_modules = ["q_proj", "k_proj", "v_proj", "o_proj",
125
+ # "gate_proj", "up_proj", "down_proj",],
126
+ # lora_alpha = 16,
127
+ # lora_dropout = 0, # Supports any, but = 0 is optimized
128
+ # bias = "none", # Enable Dora method
129
+ # use_dora=True,
130
+ # )
131
+
132
+ torch.autograd.set_detect_anomaly(True)
133
+ model = model_init(None) # Initialize the model
134
+ tokenizer = model.tokenizer
135
+
136
+ trainer = SFTTrainer(
137
+ args=training_args,
138
+ train_dataset=dataset,
139
+ model=model,
140
+ # peft_config=peft_config,
141
+ tokenizer=tokenizer,
142
+ )
143
+
144
+ trainer.train()