imelnyk commited on
Commit
152b38d
1 Parent(s): 2957baf

Model save

Browse files
README.md ADDED
@@ -0,0 +1,79 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: other
3
+ base_model: stabilityai/stablelm-2-zephyr-1_6b
4
+ tags:
5
+ - trl
6
+ - dpo
7
+ - generated_from_trainer
8
+ model-index:
9
+ - name: slm-2-dpo-full
10
+ results: []
11
+ ---
12
+
13
+ <!-- This model card has been generated automatically according to the information the Trainer had access to. You
14
+ should probably proofread and complete it, then remove this comment. -->
15
+
16
+ # slm-2-dpo-full
17
+
18
+ This model is a fine-tuned version of [stabilityai/stablelm-2-zephyr-1_6b](https://huggingface.co/stabilityai/stablelm-2-zephyr-1_6b) on the None dataset.
19
+ It achieves the following results on the evaluation set:
20
+ - Loss: 31.9894
21
+ - Rewards/chosen: 0.0244
22
+ - Rewards/rejected: 0.0188
23
+ - Rewards/accuracies: 0.5234
24
+ - Rewards/margins: 0.0057
25
+ - Logps/rejected: -2491.7576
26
+ - Logps/chosen: -2806.6704
27
+ - Logits/rejected: -1.6239
28
+ - Logits/chosen: -1.6845
29
+
30
+ ## Model description
31
+
32
+ More information needed
33
+
34
+ ## Intended uses & limitations
35
+
36
+ More information needed
37
+
38
+ ## Training and evaluation data
39
+
40
+ More information needed
41
+
42
+ ## Training procedure
43
+
44
+ ### Training hyperparameters
45
+
46
+ The following hyperparameters were used during training:
47
+ - learning_rate: 5e-07
48
+ - train_batch_size: 5
49
+ - eval_batch_size: 8
50
+ - seed: 42
51
+ - distributed_type: multi-GPU
52
+ - num_devices: 8
53
+ - gradient_accumulation_steps: 2
54
+ - total_train_batch_size: 80
55
+ - total_eval_batch_size: 64
56
+ - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
57
+ - lr_scheduler_type: cosine
58
+ - lr_scheduler_warmup_ratio: 0.1
59
+ - num_epochs: 1
60
+
61
+ ### Training results
62
+
63
+ | Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
64
+ |:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
65
+ | 16.8403 | 0.13 | 100 | 19.5118 | 0.0256 | 0.0173 | 0.5273 | 0.0082 | -2491.9011 | -2806.5552 | -1.6068 | -1.6730 |
66
+ | 28.1241 | 0.26 | 200 | 32.5175 | 0.0085 | -0.0039 | 0.5234 | 0.0124 | -2494.0195 | -2808.2581 | -1.6183 | -1.6812 |
67
+ | 84.7591 | 0.39 | 300 | 47.8043 | 0.0297 | 0.0136 | 0.5391 | 0.0161 | -2492.2703 | -2806.1406 | -1.5968 | -1.6601 |
68
+ | 40.7835 | 0.52 | 400 | 30.6722 | 0.0168 | -0.0029 | 0.5547 | 0.0197 | -2493.9204 | -2807.4263 | -1.6288 | -1.6917 |
69
+ | 36.2204 | 0.65 | 500 | 31.2202 | 0.0303 | 0.0209 | 0.5352 | 0.0095 | -2491.5447 | -2806.0762 | -1.6236 | -1.6843 |
70
+ | 99.7738 | 0.78 | 600 | 33.7403 | 0.0476 | 0.0372 | 0.5391 | 0.0104 | -2489.9089 | -2804.3484 | -1.6222 | -1.6827 |
71
+ | 41.8506 | 0.92 | 700 | 32.9133 | 0.0301 | 0.0195 | 0.5547 | 0.0106 | -2491.6851 | -2806.1006 | -1.6211 | -1.6823 |
72
+
73
+
74
+ ### Framework versions
75
+
76
+ - Transformers 4.36.2
77
+ - Pytorch 2.2.0+cu118
78
+ - Datasets 2.14.6
79
+ - Tokenizers 0.15.2
all_results.json ADDED
@@ -0,0 +1,21 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 1.0,
3
+ "eval_logits/chosen": -1.684511661529541,
4
+ "eval_logits/rejected": -1.6239439249038696,
5
+ "eval_logps/chosen": -2806.67041015625,
6
+ "eval_logps/rejected": -2491.757568359375,
7
+ "eval_loss": 31.98944854736328,
8
+ "eval_rewards/accuracies": 0.5234375,
9
+ "eval_rewards/chosen": 0.024405580013990402,
10
+ "eval_rewards/margins": 0.0056530386209487915,
11
+ "eval_rewards/rejected": 0.01875254511833191,
12
+ "eval_runtime": 110.5807,
13
+ "eval_samples": 2000,
14
+ "eval_samples_per_second": 18.086,
15
+ "eval_steps_per_second": 0.289,
16
+ "train_loss": 46.613421885577296,
17
+ "train_runtime": 4597.6924,
18
+ "train_samples": 61135,
19
+ "train_samples_per_second": 13.297,
20
+ "train_steps_per_second": 0.166
21
+ }
configuration_stablelm_epoch.py ADDED
@@ -0,0 +1,117 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Copyright 2023 Stability and The HuggingFace Inc. team. All rights reserved.
2
+ #
3
+ # Licensed under the Apache License, Version 2.0 (the "License");
4
+ # you may not use this file except in compliance with the License.
5
+ # You may obtain a copy of the License at
6
+ #
7
+ # http://www.apache.org/licenses/LICENSE-2.0
8
+ #
9
+ # Unless required by applicable law or agreed to in writing, software
10
+ # distributed under the License is distributed on an "AS IS" BASIS,
11
+ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12
+ # See the License for the specific language governing permissions and
13
+ # limitations under the License.
14
+ """ StableLM Epoch model configuration"""
15
+ from transformers import PretrainedConfig
16
+ from transformers.utils import logging
17
+
18
+
19
+ logger = logging.get_logger(__name__)
20
+
21
+
22
+ class StableLMEpochConfig(PretrainedConfig):
23
+ r"""
24
+ Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
25
+ documentation from [`PretrainedConfig`] for more information.
26
+
27
+ Args:
28
+ vocab_size (`int`, *optional*, defaults to 50_304):
29
+ Vocabulary size of the StableLM model. Defines the number of different tokens that
30
+ can be represented by the `inputs_ids` passed when calling [`StableLMEpochModel`].
31
+ intermediate_size (`int`, *optional*, defaults to 6912):
32
+ Dimension of the MLP representations.
33
+ hidden_size (`int`, *optional*, defaults to 2560):
34
+ Dimension of the decoder layers and the pooler layer.
35
+ num_hidden_layers (`int`, *optional*, defaults to 32):
36
+ Number of hidden layers in the Transformer decoder.
37
+ num_attention_heads (`int`, *optional*, defaults to 32):
38
+ Number of attention heads for each attention layer in the Transformer encoder.
39
+ num_key_value_heads (`int`, *optional*):
40
+ This is the number of key_value heads that should be used to implement Grouped Query Attention. If
41
+ `num_key_value_heads=num_attention_heads`, the model will use Multi Head Attention (MHA), if
42
+ `num_key_value_heads=1 the model will use Multi Query Attention (MQA) otherwise GQA is used. When
43
+ converting a multi-head checkpoint to a GQA checkpoint, each group key and value head should be constructed
44
+ by meanpooling all the original heads within that group. For more details checkout [this
45
+ paper](https://arxiv.org/pdf/2305.13245.pdf). If it is not specified, will default to
46
+ `num_attention_heads`.
47
+ hidden_act (`str` or `function`, *optional*, defaults to `"silu"`):
48
+ The non-linear activation function (function or string).
49
+ rope_pct (`float`, *optional*, defaults to 1.0):
50
+ Percentage of hidden dimensions to allocate to rotary embeddings.
51
+ rope_theta (`float`, *optional*, defaults to 10000.0):
52
+ The base period of the RoPE embeddings.
53
+ max_position_embeddings (`int`, *optional*, defaults to 2048):
54
+ The maximum sequence length that this model might ever be used with.
55
+ Typically set this to something large just in case (e.g., 512 or 1024 or 2048).
56
+ initializer_range (`float`, *optional*, defaults to 1e-5):
57
+ The standard deviation of the truncated_normal_initializer for initializing
58
+ all weight matrices.
59
+ norm_eps (`float`, *optional*, defaults to 1e-8):
60
+ The epsilon used by the normalization layers.
61
+ use_cache (`bool`, *optional*, defaults to `True`):
62
+ Whether or not the model should return the last key/values attentions
63
+ (not used by all models). Only relevant if `config.is_decoder=True`.
64
+ use_qkv_bias (`bool`, *optional*, defaults to `True`):
65
+ Whether or not the model should use bias for qkv layers.
66
+ tie_word_embeddings(`bool`, *optional*, defaults to `False`):
67
+ Whether to tie weight embeddings
68
+ attention_dropout (`float`, *optional*, defaults to 0.0):
69
+ The dropout ratio for the attention probabilities.
70
+ """
71
+ model_type = "stablelm_epoch"
72
+ keys_to_ignore_at_inference = ["past_key_values"]
73
+
74
+ def __init__(
75
+ self,
76
+ vocab_size=50_304,
77
+ intermediate_size=6912,
78
+ hidden_size=2560,
79
+ num_hidden_layers=32,
80
+ num_attention_heads=32,
81
+ num_key_value_heads=32,
82
+ hidden_act="silu",
83
+ rope_pct=0.25,
84
+ rope_theta=10_000,
85
+ max_position_embeddings=4096,
86
+ initializer_range=0.02,
87
+ norm_eps=1.0e-5,
88
+ use_cache=True,
89
+ use_qkv_bias=True,
90
+ bos_token_id=0,
91
+ eos_token_id=2,
92
+ tie_word_embeddings=False,
93
+ attention_dropout: float = 0.0,
94
+ **kwargs,
95
+ ):
96
+ self.vocab_size = vocab_size
97
+ self.max_position_embeddings = max_position_embeddings
98
+ self.intermediate_size = intermediate_size
99
+ self.hidden_size = hidden_size
100
+ self.num_hidden_layers = num_hidden_layers
101
+ self.num_attention_heads = num_attention_heads
102
+ self.num_key_value_heads = num_key_value_heads
103
+ self.hidden_act = hidden_act
104
+ self.rope_pct = rope_pct
105
+ self.rope_theta = rope_theta
106
+ self.initializer_range = initializer_range
107
+ self.norm_eps = norm_eps
108
+ self.use_cache = use_cache
109
+ self.use_qkv_bias = use_qkv_bias
110
+ self.tie_word_embeddings = tie_word_embeddings
111
+ self.attention_dropout = attention_dropout
112
+ super().__init__(
113
+ bos_token_id=bos_token_id,
114
+ eos_token_id=eos_token_id,
115
+ tie_word_embeddings=tie_word_embeddings,
116
+ **kwargs,
117
+ )
eval_results.json ADDED
@@ -0,0 +1,16 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 1.0,
3
+ "eval_logits/chosen": -1.684511661529541,
4
+ "eval_logits/rejected": -1.6239439249038696,
5
+ "eval_logps/chosen": -2806.67041015625,
6
+ "eval_logps/rejected": -2491.757568359375,
7
+ "eval_loss": 31.98944854736328,
8
+ "eval_rewards/accuracies": 0.5234375,
9
+ "eval_rewards/chosen": 0.024405580013990402,
10
+ "eval_rewards/margins": 0.0056530386209487915,
11
+ "eval_rewards/rejected": 0.01875254511833191,
12
+ "eval_runtime": 110.5807,
13
+ "eval_samples": 2000,
14
+ "eval_samples_per_second": 18.086,
15
+ "eval_steps_per_second": 0.289
16
+ }
generation_config.json ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "bos_token_id": 100257,
4
+ "eos_token_id": 100257,
5
+ "transformers_version": "4.36.2"
6
+ }
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:31fe86514c36173de9e3fe7c6dae2ff5aa36b5e17acce167c2c060ea90d78f77
3
  size 3289069520
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:699753fbaa57b9da7029d3c4177187bad5eefe713a604dc0f9e0a2e5757ffe81
3
  size 3289069520
modeling_stablelm_epoch.py ADDED
@@ -0,0 +1,919 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # coding=utf-8
2
+ # Copyright 2023 Stability AI, EleutherAI, 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
+ #
16
+ # This code is based off the following work:
17
+ # https://github.com/huggingface/transformers/blob/main/src/transformers/models/llama/modeling_llama.py
18
+ # https://github.com/huggingface/transformers/blob/main/src/transformers/models/gpt_neox/modeling_gpt_neox.py
19
+ """ PyTorch StableLM Epoch model. """
20
+ from typing import Optional, Tuple, Union
21
+ import math
22
+ import warnings
23
+
24
+ import torch
25
+ import torch.nn.functional as F
26
+ import torch.utils.checkpoint
27
+ from torch import nn
28
+ from torch.nn import CrossEntropyLoss
29
+
30
+ from transformers.cache_utils import Cache
31
+ from transformers.modeling_outputs import (
32
+ BaseModelOutputWithPast,
33
+ CausalLMOutputWithPast,
34
+ )
35
+ from transformers.modeling_utils import PreTrainedModel
36
+ from transformers.utils import logging, is_flash_attn_greater_or_equal_2_10
37
+
38
+ from .configuration_stablelm_epoch import StableLMEpochConfig
39
+
40
+ try:
41
+ from flash_attn import flash_attn_func, flash_attn_varlen_func
42
+ from flash_attn.bert_padding import index_first_axis, pad_input, unpad_input
43
+ except:
44
+ flash_attn_func, flash_attn_varlen_func = None, None
45
+ index_first_axis, pad_input, unpad_input = None, None, None
46
+
47
+
48
+ logger = logging.get_logger(__name__)
49
+
50
+
51
+ # Copied from transformers.models.llama.modeling_llama._get_unpad_data
52
+ def _get_unpad_data(attention_mask):
53
+ seqlens_in_batch = attention_mask.sum(dim=-1, dtype=torch.int32)
54
+ indices = torch.nonzero(attention_mask.flatten(), as_tuple=False).flatten()
55
+ max_seqlen_in_batch = seqlens_in_batch.max().item()
56
+ cu_seqlens = F.pad(torch.cumsum(seqlens_in_batch, dim=0, dtype=torch.torch.int32), (1, 0))
57
+ return (
58
+ indices,
59
+ cu_seqlens,
60
+ max_seqlen_in_batch,
61
+ )
62
+
63
+
64
+ # Copied from transformers.models.bart.modeling_bart._make_causal_mask
65
+ def _make_causal_mask(
66
+ input_ids_shape: torch.Size,
67
+ dtype: torch.dtype,
68
+ device: torch.device,
69
+ past_key_values_length: int = 0,
70
+ ):
71
+ """Make causal mask used for bi-directional self-attention."""
72
+ batch_size, tgt_len = input_ids_shape
73
+ mask = torch.full((tgt_len, tgt_len), torch.finfo(torch.float16).min, device=device)
74
+ mask_cond = torch.arange(mask.size(-1), device=device)
75
+ mask.masked_fill_(mask_cond < (mask_cond + 1).view(mask.size(-1), 1), 0)
76
+ mask = mask.to(dtype)
77
+ if past_key_values_length > 0:
78
+ mask = torch.cat([torch.zeros(tgt_len, past_key_values_length, dtype=dtype, device=device), mask], dim=-1)
79
+ return mask[None, None, :, :].expand(batch_size, 1, tgt_len, tgt_len + past_key_values_length)
80
+
81
+
82
+ # Copied from transformers.models.bart.modeling_bart._expand_mask
83
+ def _expand_mask(mask: torch.Tensor, dtype: torch.dtype, tgt_len: Optional[int] = None):
84
+ """Expands attention_mask from `[batch_size, seq_len]` to `[batch_size, 1, tgt_seq_len, src_seq_len]`."""
85
+ batch_size, src_len = mask.size()
86
+ tgt_len = tgt_len if tgt_len is not None else src_len
87
+
88
+ expanded_mask = mask[:, None, None, :].expand(batch_size, 1, tgt_len, src_len).to(dtype)
89
+ inverted_mask = 1.0 - expanded_mask
90
+
91
+ return inverted_mask.masked_fill(
92
+ inverted_mask.to(torch.bool), torch.finfo(dtype).min
93
+ )
94
+
95
+
96
+ class RotaryEmbedding(nn.Module):
97
+ def __init__(
98
+ self,
99
+ dim: int,
100
+ max_position_embeddings: int,
101
+ base: int = 10_000,
102
+ device: Optional[torch.device] = None,
103
+ ):
104
+ super().__init__()
105
+
106
+ self.dim = dim
107
+ self.max_position_embeddings = max_position_embeddings
108
+ self.base = base
109
+ inv_freq = 1.0 / (self.base ** (torch.arange(0, self.dim, 2, device=device, dtype=torch.float32) / self.dim))
110
+ self.register_buffer("inv_freq", inv_freq, persistent=False)
111
+
112
+ # Build here to make `torch.jit.trace` work.
113
+ self._set_cos_sin_cache(
114
+ seq_len=max_position_embeddings, device=self.inv_freq.device, dtype=torch.get_default_dtype(),
115
+ )
116
+
117
+ def _set_cos_sin_cache(self, seq_len: int, device: torch.device, dtype: torch.dtype):
118
+ self.max_seq_len_cached = seq_len
119
+ t = torch.arange(self.max_seq_len_cached, device=device, dtype=torch.float32)
120
+
121
+ # Don't do einsum, it converts fp32 to fp16 under AMP
122
+ # freqs = torch.einsum("i,j->ij", t, self.inv_freq)
123
+ freqs = torch.outer(t, self.inv_freq)
124
+ # Different from paper, but it uses a different permutation in order to obtain the same calculation
125
+ emb = torch.cat((freqs, freqs), dim=-1)
126
+ self.register_buffer("cos_cached", emb.cos()[None, None, :, :].to(dtype), persistent=False)
127
+ self.register_buffer("sin_cached", emb.sin()[None, None, :, :].to(dtype), persistent=False)
128
+
129
+ def forward(self, x: torch.Tensor, seq_len: Optional[int] = None):
130
+ # x: [batch_size, num_heads, seq_len, head_size]
131
+ if seq_len > self.max_seq_len_cached:
132
+ self._set_cos_sin_cache(seq_len=seq_len, device=x.device, dtype=torch.get_default_dtype())
133
+ return (
134
+ self.cos_cached[:, :, :seq_len, ...].to(dtype=x.dtype),
135
+ self.sin_cached[:, :, :seq_len, ...].to(dtype=x.dtype),
136
+ )
137
+
138
+
139
+ def rotate_half(x: torch.Tensor):
140
+ """Rotates half the hidden dims of the input."""
141
+ x1, x2 = torch.chunk(x, 2, dim=-1)
142
+ return torch.cat((-x2, x1), dim=-1)
143
+
144
+
145
+ def apply_rotary_pos_emb(q, k, cos, sin, position_ids):
146
+ # The first two dimensions of cos and sin are always 1, so we can `squeeze` them.
147
+ cos = cos.squeeze(1).squeeze(0) # [seq_len, dim]
148
+ sin = sin.squeeze(1).squeeze(0) # [seq_len, dim]
149
+ cos = cos[position_ids].unsqueeze(1) # [batch_size, 1, seq_len, dim]
150
+ sin = sin[position_ids].unsqueeze(1) # [batch_size, 1, seq_len, dim]
151
+ q_embed = (q * cos) + (rotate_half(q) * sin)
152
+ k_embed = (k * cos) + (rotate_half(k) * sin)
153
+ return q_embed, k_embed
154
+
155
+
156
+ class MLP(nn.Module):
157
+ def __init__(self, config: StableLMEpochConfig):
158
+ super().__init__()
159
+ self.config = config
160
+ self.hidden_size = config.hidden_size
161
+ self.intermediate_size = config.intermediate_size
162
+ self.gate_proj = nn.Linear(config.hidden_size, config.intermediate_size, bias=False)
163
+ self.up_proj = nn.Linear(config.hidden_size, config.intermediate_size, bias=False)
164
+ self.down_proj = nn.Linear(config.intermediate_size, config.hidden_size, bias=False)
165
+ self.act_fn = nn.SiLU()
166
+
167
+ def forward(self, x: torch.Tensor) -> torch.Tensor:
168
+ return self.down_proj(self.act_fn(self.gate_proj(x)) * self.up_proj(x))
169
+
170
+
171
+ def repeat_kv(hidden_states: torch.Tensor, n_rep: int) -> torch.Tensor:
172
+ """
173
+ This is the equivalent of torch.repeat_interleave(x, dim=1, repeats=n_rep). The hidden states go from (batch,
174
+ num_key_value_heads, seqlen, head_dim) to (batch, num_attention_heads, seqlen, head_dim)
175
+ """
176
+ batch, num_key_value_heads, slen, head_dim = hidden_states.shape
177
+ if n_rep == 1:
178
+ return hidden_states
179
+ hidden_states = hidden_states[:, :, None, :, :].expand(batch, num_key_value_heads, n_rep, slen, head_dim)
180
+ return hidden_states.reshape(batch, num_key_value_heads * n_rep, slen, head_dim)
181
+
182
+
183
+ class Attention(nn.Module):
184
+ def __init__(self, config: StableLMEpochConfig):
185
+ super().__init__()
186
+ self.config = config
187
+ self.hidden_size = config.hidden_size
188
+ self.num_heads = config.num_attention_heads
189
+ self.head_dim = self.hidden_size // self.num_heads
190
+ self.num_key_value_heads = config.num_key_value_heads
191
+ self.num_key_value_groups = self.num_heads // self.num_key_value_heads
192
+ self.max_position_embeddings = config.max_position_embeddings
193
+ self.is_causal = True
194
+ self.attention_dropout = config.attention_dropout
195
+
196
+ if (self.head_dim * self.num_heads) != self.hidden_size:
197
+ raise ValueError(
198
+ f"hidden_size must be divisible by num_heads (got `hidden_size`: {self.hidden_size}"
199
+ f" and `num_heads`: {self.num_heads})."
200
+ )
201
+
202
+ self.q_proj = nn.Linear(self.hidden_size, self.num_heads * self.head_dim, bias=config.use_qkv_bias)
203
+ self.k_proj = nn.Linear(self.hidden_size, self.num_key_value_heads * self.head_dim, bias=config.use_qkv_bias)
204
+ self.v_proj = nn.Linear(self.hidden_size, self.num_key_value_heads * self.head_dim, bias=config.use_qkv_bias)
205
+ self.o_proj = nn.Linear(self.hidden_size, self.hidden_size, bias=False)
206
+
207
+ self._init_rope()
208
+
209
+ def _init_rope(self):
210
+ self.rotary_ndims = int(self.head_dim * self.config.rope_pct)
211
+ self.rotary_emb = RotaryEmbedding(
212
+ self.rotary_ndims,
213
+ max_position_embeddings=self.config.max_position_embeddings,
214
+ base=self.config.rope_theta,
215
+ )
216
+
217
+ def forward(
218
+ self,
219
+ hidden_states: torch.FloatTensor,
220
+ attention_mask: torch.FloatTensor,
221
+ position_ids: torch.LongTensor,
222
+ past_key_value: Optional[Tuple[torch.Tensor]] = None,
223
+ output_attentions: Optional[bool] = False,
224
+ use_cache: Optional[bool] = False,
225
+ ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
226
+ bsz, q_len, _ = hidden_states.size()
227
+
228
+ query_states = self.q_proj(hidden_states)
229
+ key_states = self.k_proj(hidden_states)
230
+ value_states = self.v_proj(hidden_states)
231
+
232
+ query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
233
+ key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
234
+ value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
235
+
236
+ query_rot = query_states[..., : self.rotary_ndims]
237
+ query_pass = query_states[..., self.rotary_ndims :]
238
+ key_rot = key_states[..., : self.rotary_ndims]
239
+ key_pass = key_states[..., self.rotary_ndims :]
240
+
241
+ kv_seq_len = key_states.shape[-2]
242
+ if past_key_value is not None:
243
+ kv_seq_len += past_key_value[0].shape[-2]
244
+ cos, sin = self.rotary_emb(value_states, seq_len=kv_seq_len)
245
+ query_states, key_states = apply_rotary_pos_emb(query_rot, key_rot, cos, sin, position_ids)
246
+
247
+ # [batch_size, num_heads, seq_len, head_dim]
248
+ query_states = torch.cat((query_states, query_pass), dim=-1)
249
+ key_states = torch.cat((key_states, key_pass), dim=-1)
250
+
251
+ if past_key_value is not None:
252
+ # Reuse k, v, self_attention
253
+ key_states = torch.cat((past_key_value[0], key_states), dim=2)
254
+ value_states = torch.cat((past_key_value[1], value_states), dim=2)
255
+
256
+ past_key_value = (key_states, value_states) if use_cache else None
257
+
258
+ # Repeat k/v heads if n_kv_heads < n_heads
259
+ key_states = repeat_kv(key_states, self.num_key_value_groups)
260
+ value_states = repeat_kv(value_states, self.num_key_value_groups)
261
+
262
+ attn_weights = torch.matmul(query_states, key_states.transpose(2, 3)) / math.sqrt(self.head_dim)
263
+
264
+ if attn_weights.size() != (bsz, self.num_heads, q_len, kv_seq_len):
265
+ raise ValueError(
266
+ f"Attention weights should be of size {(bsz, self.num_heads, q_len, kv_seq_len)}, but is"
267
+ f" {attn_weights.size()}"
268
+ )
269
+
270
+ if attention_mask is not None:
271
+ if attention_mask.size() != (bsz, 1, q_len, kv_seq_len):
272
+ raise ValueError(
273
+ f"Attention mask should be of size {(bsz, 1, q_len, kv_seq_len)}, but is {attention_mask.size()}"
274
+ )
275
+ attn_weights = attn_weights + attention_mask
276
+
277
+ # Upcast attention to fp32
278
+ attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(query_states.dtype)
279
+ attn_weights = nn.functional.dropout(attn_weights, p=self.attention_dropout, training=self.training)
280
+ attn_output = torch.matmul(attn_weights, value_states)
281
+
282
+ if attn_output.size() != (bsz, self.num_heads, q_len, self.head_dim):
283
+ raise ValueError(
284
+ f"`attn_output` should be of size {(bsz, self.num_heads, q_len, self.head_dim)}, but is"
285
+ f" {attn_output.size()}"
286
+ )
287
+
288
+ # Merge heads
289
+ attn_output = attn_output.transpose(1, 2).contiguous()
290
+ attn_output = attn_output.reshape(bsz, q_len, self.hidden_size)
291
+
292
+ # Final linear projection
293
+ attn_output = self.o_proj(attn_output)
294
+
295
+ if not output_attentions:
296
+ attn_weights = None
297
+
298
+ return attn_output, attn_weights, past_key_value
299
+
300
+
301
+ class FlashAttention2(Attention):
302
+ """
303
+ Reference: https://github.com/huggingface/transformers/blob/5d36025ca13d05151b7a0c761e90d429c4644a30/src/transformers/models/llama/modeling_llama.py#L456
304
+ """
305
+
306
+ def __init__(self, *args, **kwargs):
307
+ super().__init__(*args, **kwargs)
308
+
309
+ # TODO: Should be removed once Flash Attention for RoCm is bumped to 2.1.
310
+ # flash_attn<2.1 generates top-left aligned causal mask, while what is needed here is bottom-right alignement, that was made default for flash_attn>=2.1. This attribute is used to handle this difference. Reference: https://github.com/Dao-AILab/flash-attention/releases/tag/v2.1.0.
311
+ # Beware that with flash_attn<2.1, using q_seqlen != k_seqlen (except for the case q_seqlen == 1) produces a wrong mask (top-left).
312
+ self._flash_attn_uses_top_left_mask = not is_flash_attn_greater_or_equal_2_10()
313
+
314
+ def forward(
315
+ self,
316
+ hidden_states: torch.Tensor,
317
+ attention_mask: Optional[torch.LongTensor] = None,
318
+ position_ids: Optional[torch.LongTensor] = None,
319
+ past_key_value: Optional[Cache] = None,
320
+ output_attentions: bool = False,
321
+ use_cache: bool = False,
322
+ **kwargs,
323
+ ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
324
+ # FlashAttention2 attention does not support output_attentions
325
+ if "padding_mask" in kwargs:
326
+ warnings.warn(
327
+ "Passing `padding_mask` is deprecated and will be removed in v4.37. Please make sure use `attention_mask` instead.`"
328
+ )
329
+
330
+ # overwrite attention_mask with padding_mask
331
+ attention_mask = kwargs.pop("padding_mask")
332
+
333
+ output_attentions = False
334
+
335
+ bsz, q_len, _ = hidden_states.size()
336
+
337
+ query_states = self.q_proj(hidden_states)
338
+ key_states = self.k_proj(hidden_states)
339
+ value_states = self.v_proj(hidden_states)
340
+
341
+ # Flash attention requires the input to have the shape
342
+ # batch_size x seq_length x head_dim x hidden_dim
343
+ # therefore we just need to keep the original shape
344
+ query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
345
+ key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
346
+ value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
347
+
348
+ query_rot = query_states[..., : self.rotary_ndims]
349
+ query_pass = query_states[..., self.rotary_ndims :]
350
+ key_rot = key_states[..., : self.rotary_ndims]
351
+ key_pass = key_states[..., self.rotary_ndims :]
352
+
353
+ kv_seq_len = key_states.shape[-2]
354
+ if past_key_value is not None:
355
+ kv_seq_len += past_key_value[0].shape[-2]
356
+ cos, sin = self.rotary_emb(value_states, seq_len=kv_seq_len)
357
+ query_states, key_states = apply_rotary_pos_emb(query_rot, key_rot, cos, sin, position_ids)
358
+
359
+ # [batch_size, num_heads, seq_len, head_dim]
360
+ query_states = torch.cat((query_states, query_pass), dim=-1)
361
+ key_states = torch.cat((key_states, key_pass), dim=-1)
362
+
363
+ if past_key_value is not None:
364
+ # Reuse k, v, self_attention
365
+ key_states = torch.cat((past_key_value[0], key_states), dim=2)
366
+ value_states = torch.cat((past_key_value[1], value_states), dim=2)
367
+
368
+ past_key_value = (key_states, value_states) if use_cache else None
369
+
370
+ # TODO: These transpose are quite inefficient but Flash Attention requires the layout [batch_size, sequence_length, num_heads, head_dim]. We would need to refactor the KV cache
371
+ # to be able to avoid many of these transpose/reshape/view.
372
+ query_states = query_states.transpose(1, 2)
373
+ key_states = key_states.transpose(1, 2)
374
+ value_states = value_states.transpose(1, 2)
375
+
376
+ dropout_rate = self.attention_dropout if self.training else 0.0
377
+
378
+ attn_output = self._flash_attention_forward(
379
+ query_states, key_states, value_states, attention_mask, q_len, dropout=dropout_rate
380
+ )
381
+ attn_output = attn_output.reshape(bsz, q_len, self.hidden_size).contiguous()
382
+ attn_output = self.o_proj(attn_output)
383
+
384
+ if not output_attentions:
385
+ attn_weights = None
386
+
387
+ return attn_output, attn_weights, past_key_value
388
+
389
+ def _flash_attention_forward(
390
+ self, query_states, key_states, value_states, attention_mask, query_length, dropout=0.0, softmax_scale=None
391
+ ):
392
+ """
393
+ Calls the forward method of Flash Attention - if the input hidden states contain at least one padding token
394
+ first unpad the input, then computes the attention scores and pad the final attention scores.
395
+
396
+ Args:
397
+ query_states (`torch.Tensor`):
398
+ Input query states to be passed to Flash Attention API
399
+ key_states (`torch.Tensor`):
400
+ Input key states to be passed to Flash Attention API
401
+ value_states (`torch.Tensor`):
402
+ Input value states to be passed to Flash Attention API
403
+ attention_mask (`torch.Tensor`):
404
+ The padding mask - corresponds to a tensor of size `(batch_size, seq_len)` where 0 stands for the
405
+ position of padding tokens and 1 for the position of non-padding tokens.
406
+ dropout (`int`, *optional*):
407
+ Attention dropout
408
+ softmax_scale (`float`, *optional*):
409
+ The scaling of QK^T before applying softmax. Default to 1 / sqrt(head_dim)
410
+ """
411
+ if not self._flash_attn_uses_top_left_mask:
412
+ causal = self.is_causal
413
+ else:
414
+ # TODO: Remove the `query_length != 1` check once Flash Attention for RoCm is bumped to 2.1. For details, please see the comment in FlashAttention2 __init__.
415
+ causal = self.is_causal and query_length != 1
416
+
417
+ # Contains at least one padding token in the sequence
418
+ if attention_mask is not None:
419
+ batch_size = query_states.shape[0]
420
+ query_states, key_states, value_states, indices_q, cu_seq_lens, max_seq_lens = self._upad_input(
421
+ query_states, key_states, value_states, attention_mask, query_length
422
+ )
423
+
424
+ cu_seqlens_q, cu_seqlens_k = cu_seq_lens
425
+ max_seqlen_in_batch_q, max_seqlen_in_batch_k = max_seq_lens
426
+
427
+ attn_output_unpad = flash_attn_varlen_func(
428
+ query_states,
429
+ key_states,
430
+ value_states,
431
+ cu_seqlens_q=cu_seqlens_q,
432
+ cu_seqlens_k=cu_seqlens_k,
433
+ max_seqlen_q=max_seqlen_in_batch_q,
434
+ max_seqlen_k=max_seqlen_in_batch_k,
435
+ dropout_p=dropout,
436
+ softmax_scale=softmax_scale,
437
+ causal=causal,
438
+ )
439
+
440
+ attn_output = pad_input(attn_output_unpad, indices_q, batch_size, query_length)
441
+ else:
442
+ attn_output = flash_attn_func(
443
+ query_states, key_states, value_states, dropout, softmax_scale=softmax_scale, causal=causal
444
+ )
445
+
446
+ return attn_output
447
+
448
+ def _upad_input(self, query_layer, key_layer, value_layer, attention_mask, query_length):
449
+ indices_k, cu_seqlens_k, max_seqlen_in_batch_k = _get_unpad_data(attention_mask)
450
+ batch_size, kv_seq_len, num_key_value_heads, head_dim = key_layer.shape
451
+
452
+ key_layer = index_first_axis(
453
+ key_layer.reshape(batch_size * kv_seq_len, num_key_value_heads, head_dim), indices_k
454
+ )
455
+ value_layer = index_first_axis(
456
+ value_layer.reshape(batch_size * kv_seq_len, num_key_value_heads, head_dim), indices_k
457
+ )
458
+ if query_length == kv_seq_len:
459
+ query_layer = index_first_axis(
460
+ query_layer.reshape(batch_size * kv_seq_len, self.num_heads, head_dim), indices_k
461
+ )
462
+ cu_seqlens_q = cu_seqlens_k
463
+ max_seqlen_in_batch_q = max_seqlen_in_batch_k
464
+ indices_q = indices_k
465
+ elif query_length == 1:
466
+ max_seqlen_in_batch_q = 1
467
+ cu_seqlens_q = torch.arange(
468
+ batch_size + 1, dtype=torch.int32, device=query_layer.device
469
+ ) # There is a memcpy here, that is very bad.
470
+ indices_q = cu_seqlens_q[:-1]
471
+ query_layer = query_layer.squeeze(1)
472
+ else:
473
+ # The -q_len: slice assumes left padding.
474
+ attention_mask = attention_mask[:, -query_length:]
475
+ query_layer, indices_q, cu_seqlens_q, max_seqlen_in_batch_q = unpad_input(query_layer, attention_mask)
476
+
477
+ return (
478
+ query_layer,
479
+ key_layer,
480
+ value_layer,
481
+ indices_q,
482
+ (cu_seqlens_q, cu_seqlens_k),
483
+ (max_seqlen_in_batch_q, max_seqlen_in_batch_k),
484
+ )
485
+
486
+
487
+ ATTENTION_CLASSES = {
488
+ "eager": Attention,
489
+ "flash_attention_2": FlashAttention2,
490
+ }
491
+
492
+
493
+ class DecoderLayer(nn.Module):
494
+ def __init__(self, config: StableLMEpochConfig):
495
+ super().__init__()
496
+ self.self_attn = ATTENTION_CLASSES[config._attn_implementation](config=config)
497
+ self.mlp = MLP(config)
498
+ self.input_layernorm = nn.LayerNorm(config.hidden_size, eps=config.norm_eps)
499
+ self.post_attention_layernorm = nn.LayerNorm(config.hidden_size, eps=config.norm_eps)
500
+
501
+ def forward(
502
+ self,
503
+ hidden_states: Optional[torch.FloatTensor],
504
+ attention_mask: Optional[torch.FloatTensor] = None,
505
+ position_ids: Optional[torch.LongTensor] = None,
506
+ past_key_value: Optional[Tuple[torch.Tensor]] = None,
507
+ output_attentions: Optional[bool] = False,
508
+ use_cache: Optional[bool] = False,
509
+ ) -> Union[Tuple[torch.Tensor], Optional[Tuple[torch.Tensor, Tuple[torch.FloatTensor, ...]]]]:
510
+ residual = hidden_states
511
+
512
+ hidden_states = self.input_layernorm(hidden_states)
513
+
514
+ # Self Attention
515
+ hidden_states, self_attn_weights, present_key_value = self.self_attn(
516
+ hidden_states=hidden_states,
517
+ attention_mask=attention_mask,
518
+ position_ids=position_ids,
519
+ past_key_value=past_key_value,
520
+ output_attentions=output_attentions,
521
+ use_cache=use_cache,
522
+ )
523
+ hidden_states = residual + hidden_states
524
+
525
+ # Fully Connected
526
+ residual = hidden_states
527
+ hidden_states = self.post_attention_layernorm(hidden_states)
528
+ hidden_states = self.mlp(hidden_states)
529
+ hidden_states = residual + hidden_states
530
+
531
+ outputs = (hidden_states,)
532
+
533
+ if output_attentions:
534
+ outputs += (self_attn_weights,)
535
+
536
+ if use_cache:
537
+ outputs += (present_key_value,)
538
+
539
+ return outputs
540
+
541
+
542
+ class StableLMEpochPreTrainedModel(PreTrainedModel):
543
+ """An abstract class to handle weights initialization and a simple interface
544
+ for downloading and loading pretrained models.
545
+ """
546
+
547
+ config_class = StableLMEpochConfig
548
+ base_model_prefix = "model"
549
+ supports_gradient_checkpointing = True
550
+ _no_split_modules = ["DecoderLayer"]
551
+ _skip_keys_device_placement = "past_key_values"
552
+ _supports_flash_attn_2 = True
553
+
554
+ def _init_weights(self, module: nn.Module):
555
+ """Initialize the weights"""
556
+ if isinstance(module, nn.Linear):
557
+ module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
558
+ if module.bias is not None:
559
+ module.bias.data.zero_()
560
+ elif isinstance(module, nn.Embedding):
561
+ module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
562
+ if module.padding_idx is not None:
563
+ module.weight.data[module.padding_idx].zero_()
564
+ elif isinstance(module, nn.LayerNorm):
565
+ module.bias.data.zero_()
566
+ module.weight.data.fill_(1.0)
567
+
568
+ def _set_gradient_checkpointing(self, module: nn.Module, value=False):
569
+ if isinstance(module, StableLMEpochModel):
570
+ module.gradient_checkpointing = value
571
+
572
+
573
+ class StableLMEpochModel(StableLMEpochPreTrainedModel):
574
+ def __init__(self, config: StableLMEpochConfig):
575
+ super().__init__(config)
576
+ self.embed_tokens = nn.Embedding(config.vocab_size, config.hidden_size, config.pad_token_id)
577
+ self.layers = nn.ModuleList([DecoderLayer(config) for _ in range(config.num_hidden_layers)])
578
+ self.norm = nn.LayerNorm(config.hidden_size, eps=config.norm_eps)
579
+
580
+ self._use_flash_attention_2 = config._attn_implementation == "flash_attention_2"
581
+ self.gradient_checkpointing = False
582
+ # Initialize weights and apply final processing
583
+ self.post_init()
584
+
585
+ def get_input_embeddings(self):
586
+ return self.embed_tokens
587
+
588
+ def set_input_embeddings(self, value: nn.Module):
589
+ self.embed_tokens = value
590
+
591
+ # Copied from transformers.models.bart.modeling_bart.BartDecoder._prepare_decoder_attention_mask
592
+ def _prepare_decoder_attention_mask(
593
+ self,
594
+ attention_mask: torch.Tensor,
595
+ input_shape: torch.Size,
596
+ inputs_embeds: torch.Tensor,
597
+ past_key_values_length: int,
598
+ ):
599
+ # Create causal mask
600
+ # [batch_size, seq_len] -> [batch_size, 1, tgt_seq_len, src_seq_len]
601
+ combined_attention_mask = None
602
+ if input_shape[-1] > 1:
603
+ combined_attention_mask = _make_causal_mask(
604
+ input_shape,
605
+ inputs_embeds.dtype,
606
+ device=inputs_embeds.device,
607
+ past_key_values_length=past_key_values_length,
608
+ )
609
+
610
+ if attention_mask is not None:
611
+ # [batch_size, seq_len] -> [batch_size, 1, tgt_seq_len, src_seq_len]
612
+ expanded_attn_mask = _expand_mask(
613
+ attention_mask, inputs_embeds.dtype, tgt_len=input_shape[-1]
614
+ ).to(inputs_embeds.device)
615
+ combined_attention_mask = expanded_attn_mask if combined_attention_mask is None else expanded_attn_mask + combined_attention_mask
616
+
617
+ return combined_attention_mask
618
+
619
+ def forward(
620
+ self,
621
+ input_ids: Optional[torch.LongTensor] = None,
622
+ attention_mask: Optional[torch.FloatTensor] = None,
623
+ position_ids: Optional[torch.LongTensor] = None,
624
+ past_key_values: Optional[Tuple[Tuple[torch.FloatTensor]]] = None,
625
+ inputs_embeds: Optional[torch.FloatTensor] = None,
626
+ use_cache: Optional[bool] = None,
627
+ output_attentions: Optional[bool] = None,
628
+ output_hidden_states: Optional[bool] = None,
629
+ return_dict: Optional[bool] = None,
630
+ ) -> Union[Tuple, BaseModelOutputWithPast]:
631
+ output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
632
+ output_hidden_states = output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
633
+ use_cache = use_cache if use_cache is not None else self.config.use_cache
634
+
635
+ return_dict = return_dict if return_dict is not None else self.config.use_return_dict
636
+
637
+ # Retrieve input_ids and inputs_embeds
638
+ if input_ids is not None and inputs_embeds is not None:
639
+ raise ValueError(
640
+ "You cannot specify both decoder_input_ids and decoder_inputs_embeds at the same time"
641
+ )
642
+ elif input_ids is not None:
643
+ batch_size, seq_length = input_ids.shape
644
+ elif inputs_embeds is not None:
645
+ batch_size, seq_length, _ = inputs_embeds.shape
646
+ else:
647
+ raise ValueError(
648
+ "You have to specify either decoder_input_ids or decoder_inputs_embeds"
649
+ )
650
+
651
+ seq_length_with_past = seq_length
652
+ past_key_values_length = 0
653
+
654
+ if position_ids is None:
655
+ device = input_ids.device if input_ids is not None else inputs_embeds.device
656
+ position_ids = torch.arange(
657
+ past_key_values_length,
658
+ seq_length + past_key_values_length,
659
+ dtype=torch.long,
660
+ device=device,
661
+ )
662
+ position_ids = position_ids.unsqueeze(0).view(-1, seq_length)
663
+ else:
664
+ position_ids = position_ids.view(-1, seq_length).long()
665
+
666
+ if inputs_embeds is None:
667
+ inputs_embeds = self.embed_tokens(input_ids)
668
+ # Embed positions
669
+ if self._use_flash_attention_2:
670
+ # 2d mask is passed through the layers
671
+ attention_mask = attention_mask if (attention_mask is not None and 0 in attention_mask) else None
672
+ else:
673
+ if attention_mask is None:
674
+ attention_mask = torch.ones(
675
+ (batch_size, seq_length_with_past),
676
+ dtype=torch.bool,
677
+ device=inputs_embeds.device,
678
+ )
679
+ attention_mask = self._prepare_decoder_attention_mask(
680
+ attention_mask,
681
+ (batch_size, seq_length),
682
+ inputs_embeds,
683
+ past_key_values_length,
684
+ )
685
+
686
+ hidden_states = inputs_embeds
687
+
688
+ if self.gradient_checkpointing and self.training:
689
+ if use_cache:
690
+ logger.warning(
691
+ "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
692
+ )
693
+ use_cache = False
694
+
695
+ # Decoder layers
696
+ all_hidden_states = () if output_hidden_states else None
697
+ all_self_attns = () if output_attentions else None
698
+ next_decoder_cache = () if use_cache else None
699
+
700
+ for idx, decoder_layer in enumerate(self.layers):
701
+ if output_hidden_states:
702
+ all_hidden_states += (hidden_states,)
703
+
704
+ past_key_value = (
705
+ past_key_values[idx] if past_key_values is not None else None
706
+ )
707
+
708
+ if self.gradient_checkpointing and self.training:
709
+
710
+ def create_custom_forward(module):
711
+ def custom_forward(*inputs):
712
+ # None for past_key_value
713
+ return module(*inputs, past_key_value, output_attentions)
714
+
715
+ return custom_forward
716
+
717
+ layer_outputs = torch.utils.checkpoint.checkpoint(
718
+ create_custom_forward(decoder_layer),
719
+ hidden_states,
720
+ attention_mask,
721
+ position_ids,
722
+ )
723
+ else:
724
+ layer_outputs = decoder_layer(
725
+ hidden_states,
726
+ attention_mask=attention_mask,
727
+ position_ids=position_ids,
728
+ past_key_value=past_key_value,
729
+ output_attentions=output_attentions,
730
+ use_cache=use_cache,
731
+ )
732
+
733
+ hidden_states = layer_outputs[0]
734
+
735
+ if use_cache:
736
+ next_decoder_cache += (layer_outputs[2 if output_attentions else 1],)
737
+
738
+ if output_attentions:
739
+ all_self_attns += (layer_outputs[1],)
740
+
741
+ hidden_states = self.norm(hidden_states)
742
+
743
+ # Add hidden states from the last decoder layer
744
+ if output_hidden_states:
745
+ all_hidden_states += (hidden_states,)
746
+
747
+ next_cache = next_decoder_cache if use_cache else None
748
+ if not return_dict:
749
+ return tuple(
750
+ v
751
+ for v in [hidden_states, next_cache, all_hidden_states, all_self_attns]
752
+ if v is not None
753
+ )
754
+ return BaseModelOutputWithPast(
755
+ last_hidden_state=hidden_states,
756
+ past_key_values=next_cache,
757
+ hidden_states=all_hidden_states,
758
+ attentions=all_self_attns,
759
+ )
760
+
761
+
762
+ class StableLMEpochForCausalLM(StableLMEpochPreTrainedModel):
763
+ _tied_weights_keys = ["lm_head.weight"]
764
+
765
+ def __init__(self, config: StableLMEpochConfig):
766
+ super().__init__(config)
767
+
768
+ self.model = StableLMEpochModel(config)
769
+ self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
770
+
771
+ # Initialize weights and apply final processing
772
+ self.post_init()
773
+
774
+ def get_input_embeddings(self):
775
+ return self.model.embed_tokens
776
+
777
+ def set_input_embeddings(self, value):
778
+ self.model.embed_tokens = value
779
+
780
+ def get_output_embeddings(self):
781
+ return self.lm_head
782
+
783
+ def set_output_embeddings(self, new_embeddings: nn.Module):
784
+ self.lm_head = new_embeddings
785
+
786
+ def get_decoder(self):
787
+ return self.model
788
+
789
+ def set_decoder(self, decoder):
790
+ self.model = decoder
791
+
792
+ def forward(
793
+ self,
794
+ input_ids: Optional[torch.LongTensor] = None,
795
+ attention_mask: Optional[torch.FloatTensor] = None,
796
+ position_ids: Optional[torch.LongTensor] = None,
797
+ past_key_values: Optional[Tuple[Tuple[torch.FloatTensor]]] = None,
798
+ inputs_embeds: Optional[torch.FloatTensor] = None,
799
+ labels: Optional[torch.LongTensor] = None,
800
+ use_cache: Optional[bool] = None,
801
+ output_attentions: Optional[bool] = None,
802
+ output_hidden_states: Optional[bool] = None,
803
+ return_dict: Optional[bool] = None,
804
+ ) -> Union[Tuple, CausalLMOutputWithPast]:
805
+ output_attentions = (
806
+ output_attentions
807
+ if output_attentions is not None
808
+ else self.config.output_attentions
809
+ )
810
+ output_hidden_states = (
811
+ output_hidden_states
812
+ if output_hidden_states is not None
813
+ else self.config.output_hidden_states
814
+ )
815
+ return_dict = (
816
+ return_dict if return_dict is not None else self.config.use_return_dict
817
+ )
818
+
819
+ # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
820
+ outputs = self.model(
821
+ input_ids,
822
+ attention_mask=attention_mask,
823
+ position_ids=position_ids,
824
+ past_key_values=past_key_values,
825
+ inputs_embeds=inputs_embeds,
826
+ use_cache=use_cache,
827
+ output_attentions=output_attentions,
828
+ output_hidden_states=output_hidden_states,
829
+ return_dict=return_dict,
830
+ )
831
+
832
+ hidden_states = outputs[0]
833
+ logits = self.lm_head(hidden_states).float()
834
+
835
+ loss = None
836
+ if labels is not None:
837
+ # Shift so that tokens < n predict n
838
+ shift_logits = logits[..., :-1, :].contiguous()
839
+ shift_labels = labels[..., 1:].contiguous()
840
+ # Flatten the tokens
841
+ loss_fct = CrossEntropyLoss()
842
+ shift_logits = shift_logits.view(-1, self.config.vocab_size)
843
+ shift_labels = shift_labels.view(-1)
844
+ # Enable model parallelism
845
+ shift_labels = shift_labels.to(shift_logits.device)
846
+ loss = loss_fct(shift_logits, shift_labels)
847
+
848
+ if not return_dict:
849
+ output = (logits,) + outputs[1:]
850
+ return (loss,) + output if loss is not None else output
851
+
852
+ return CausalLMOutputWithPast(
853
+ loss=loss,
854
+ logits=logits,
855
+ past_key_values=outputs.past_key_values,
856
+ hidden_states=outputs.hidden_states,
857
+ attentions=outputs.attentions,
858
+ )
859
+
860
+ def prepare_inputs_for_generation(
861
+ self,
862
+ input_ids,
863
+ past_key_values: Optional[torch.Tensor] = None,
864
+ attention_mask: Optional[torch.Tensor] = None,
865
+ inputs_embeds: Optional[torch.Tensor] = None,
866
+ **kwargs,
867
+ ):
868
+ # Trim decoder_input_ids if past is used
869
+ if past_key_values is not None:
870
+ past_length = past_key_values[0][0].shape[2]
871
+
872
+ # Some generation methods already pass only the last input ID
873
+ if input_ids.shape[1] > past_length:
874
+ remove_prefix_length = past_length
875
+ else:
876
+ # Default to old behavior: keep only final ID
877
+ remove_prefix_length = input_ids.shape[1] - 1
878
+
879
+ input_ids = input_ids[:, remove_prefix_length:]
880
+
881
+ position_ids = kwargs.get("position_ids", None)
882
+ if attention_mask is not None and position_ids is None:
883
+ # Create position_ids on the fly for batch generation
884
+ position_ids = attention_mask.long().cumsum(-1) - 1
885
+ position_ids.masked_fill_(attention_mask == 0, 1)
886
+ if past_key_values:
887
+ position_ids = position_ids[:, -1].unsqueeze(-1)
888
+
889
+ # If `inputs_embeds` are passed, we only want to use them in the 1st generation step
890
+ if inputs_embeds is not None and past_key_values is None:
891
+ model_inputs = {"inputs_embeds": inputs_embeds}
892
+ else:
893
+ model_inputs = {"input_ids": input_ids}
894
+
895
+ model_inputs.update(
896
+ {
897
+ "attention_mask": attention_mask,
898
+ "past_key_values": past_key_values,
899
+ "use_cache": kwargs.get("use_cache"),
900
+ "position_ids": position_ids,
901
+ }
902
+ )
903
+ return model_inputs
904
+
905
+ @staticmethod
906
+ def _reorder_cache(past_key_values, beam_idx):
907
+ reordered_past = ()
908
+ for layer_past in past_key_values:
909
+ reordered_past += (
910
+ tuple(
911
+ past_state.index_select(0, beam_idx.to(past_state.device))
912
+ for past_state in layer_past
913
+ ),
914
+ )
915
+ return reordered_past
916
+
917
+
918
+ StableLMEpochConfig.register_for_auto_class()
919
+ StableLMEpochForCausalLM.register_for_auto_class("AutoModelForCausalLM")
runs/Feb19_23-08-06_cccxc542/events.out.tfevents.1708402191.cccxc542.93482.0 CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:a780bbc9c5101256ec2c7ee35478510a10fe4ac9ebcbb721ac1923eb04936be7
3
- size 54855
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1065ec687b6baaf0f460adf547c466cb2226a5e05ee5483d5bc2a5db5e551894
3
+ size 59013
runs/Feb19_23-08-06_cccxc542/events.out.tfevents.1708406899.cccxc542.93482.1 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b880746d534205b79b0a80157a8b930699e5b566b7a8e0db0a8cf7a7a483a0f7
3
+ size 828
train_results.json ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 1.0,
3
+ "train_loss": 46.613421885577296,
4
+ "train_runtime": 4597.6924,
5
+ "train_samples": 61135,
6
+ "train_samples_per_second": 13.297,
7
+ "train_steps_per_second": 0.166
8
+ }
trainer_state.json ADDED
@@ -0,0 +1,1220 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": null,
3
+ "best_model_checkpoint": null,
4
+ "epoch": 0.999345977763244,
5
+ "eval_steps": 100,
6
+ "global_step": 764,
7
+ "is_hyper_param_search": false,
8
+ "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
+ "log_history": [
11
+ {
12
+ "epoch": 0.0,
13
+ "learning_rate": 6.493506493506494e-09,
14
+ "logits/chosen": -2.0615594387054443,
15
+ "logits/rejected": -1.9222214221954346,
16
+ "logps/chosen": -3380.6083984375,
17
+ "logps/rejected": -2521.2978515625,
18
+ "loss": 0.0001,
19
+ "rewards/accuracies": 0.0,
20
+ "rewards/chosen": 0.0,
21
+ "rewards/margins": 0.0,
22
+ "rewards/rejected": 0.0,
23
+ "step": 1
24
+ },
25
+ {
26
+ "epoch": 0.01,
27
+ "learning_rate": 6.493506493506492e-08,
28
+ "logits/chosen": -1.674426555633545,
29
+ "logits/rejected": -1.637134313583374,
30
+ "logps/chosen": -2549.3515625,
31
+ "logps/rejected": -2319.4013671875,
32
+ "loss": 10.0505,
33
+ "rewards/accuracies": 0.4333333373069763,
34
+ "rewards/chosen": 0.0008169158827513456,
35
+ "rewards/margins": 0.0011402772506698966,
36
+ "rewards/rejected": -0.00032336192089132965,
37
+ "step": 10
38
+ },
39
+ {
40
+ "epoch": 0.03,
41
+ "learning_rate": 1.2987012987012984e-07,
42
+ "logits/chosen": -1.6043205261230469,
43
+ "logits/rejected": -1.5535523891448975,
44
+ "logps/chosen": -2340.101318359375,
45
+ "logps/rejected": -2224.145263671875,
46
+ "loss": 7.4843,
47
+ "rewards/accuracies": 0.5200000405311584,
48
+ "rewards/chosen": 0.00018432810611557215,
49
+ "rewards/margins": 0.0009077669237740338,
50
+ "rewards/rejected": -0.0007234388613142073,
51
+ "step": 20
52
+ },
53
+ {
54
+ "epoch": 0.04,
55
+ "learning_rate": 1.948051948051948e-07,
56
+ "logits/chosen": -1.6847426891326904,
57
+ "logits/rejected": -1.6577625274658203,
58
+ "logps/chosen": -2983.23681640625,
59
+ "logps/rejected": -2513.237060546875,
60
+ "loss": 9.1379,
61
+ "rewards/accuracies": 0.48000001907348633,
62
+ "rewards/chosen": 0.010261936113238335,
63
+ "rewards/margins": 0.004135974682867527,
64
+ "rewards/rejected": 0.006125961430370808,
65
+ "step": 30
66
+ },
67
+ {
68
+ "epoch": 0.05,
69
+ "learning_rate": 2.597402597402597e-07,
70
+ "logits/chosen": -1.859400987625122,
71
+ "logits/rejected": -1.8100417852401733,
72
+ "logps/chosen": -2879.57470703125,
73
+ "logps/rejected": -2273.878173828125,
74
+ "loss": 12.271,
75
+ "rewards/accuracies": 0.5,
76
+ "rewards/chosen": 0.012033696286380291,
77
+ "rewards/margins": 0.005555520299822092,
78
+ "rewards/rejected": 0.006478174589574337,
79
+ "step": 40
80
+ },
81
+ {
82
+ "epoch": 0.07,
83
+ "learning_rate": 3.2467532467532465e-07,
84
+ "logits/chosen": -1.828608751296997,
85
+ "logits/rejected": -1.805625319480896,
86
+ "logps/chosen": -2893.784423828125,
87
+ "logps/rejected": -2551.77294921875,
88
+ "loss": 8.7411,
89
+ "rewards/accuracies": 0.5600000023841858,
90
+ "rewards/chosen": 0.02166888490319252,
91
+ "rewards/margins": 0.007775471545755863,
92
+ "rewards/rejected": 0.013893413357436657,
93
+ "step": 50
94
+ },
95
+ {
96
+ "epoch": 0.08,
97
+ "learning_rate": 3.896103896103896e-07,
98
+ "logits/chosen": -1.7459495067596436,
99
+ "logits/rejected": -1.6628999710083008,
100
+ "logps/chosen": -3231.689697265625,
101
+ "logps/rejected": -2554.42919921875,
102
+ "loss": 9.758,
103
+ "rewards/accuracies": 0.559999942779541,
104
+ "rewards/chosen": 0.027519574388861656,
105
+ "rewards/margins": 0.008895651437342167,
106
+ "rewards/rejected": 0.018623923882842064,
107
+ "step": 60
108
+ },
109
+ {
110
+ "epoch": 0.09,
111
+ "learning_rate": 4.545454545454545e-07,
112
+ "logits/chosen": -1.8072433471679688,
113
+ "logits/rejected": -1.7838470935821533,
114
+ "logps/chosen": -2829.386474609375,
115
+ "logps/rejected": -2542.68701171875,
116
+ "loss": 11.0017,
117
+ "rewards/accuracies": 0.5,
118
+ "rewards/chosen": 0.024034958332777023,
119
+ "rewards/margins": 0.006175906863063574,
120
+ "rewards/rejected": 0.017859051004052162,
121
+ "step": 70
122
+ },
123
+ {
124
+ "epoch": 0.1,
125
+ "learning_rate": 4.99976474872689e-07,
126
+ "logits/chosen": -1.7730411291122437,
127
+ "logits/rejected": -1.7399647235870361,
128
+ "logps/chosen": -2769.705322265625,
129
+ "logps/rejected": -2476.75634765625,
130
+ "loss": 15.623,
131
+ "rewards/accuracies": 0.5400000214576721,
132
+ "rewards/chosen": 0.008623984642326832,
133
+ "rewards/margins": 0.008157819509506226,
134
+ "rewards/rejected": 0.0004661638231482357,
135
+ "step": 80
136
+ },
137
+ {
138
+ "epoch": 0.12,
139
+ "learning_rate": 4.995583735427465e-07,
140
+ "logits/chosen": -1.790204644203186,
141
+ "logits/rejected": -1.7226215600967407,
142
+ "logps/chosen": -2688.0732421875,
143
+ "logps/rejected": -2436.649658203125,
144
+ "loss": 11.9811,
145
+ "rewards/accuracies": 0.6100000143051147,
146
+ "rewards/chosen": 0.017978714779019356,
147
+ "rewards/margins": 0.017238261178135872,
148
+ "rewards/rejected": 0.0007404519128613174,
149
+ "step": 90
150
+ },
151
+ {
152
+ "epoch": 0.13,
153
+ "learning_rate": 4.986184978516146e-07,
154
+ "logits/chosen": -1.7211675643920898,
155
+ "logits/rejected": -1.6991230249404907,
156
+ "logps/chosen": -2611.177001953125,
157
+ "logps/rejected": -2212.4033203125,
158
+ "loss": 16.8403,
159
+ "rewards/accuracies": 0.5200001001358032,
160
+ "rewards/chosen": 0.024822045117616653,
161
+ "rewards/margins": 0.00336282467469573,
162
+ "rewards/rejected": 0.021459218114614487,
163
+ "step": 100
164
+ },
165
+ {
166
+ "epoch": 0.13,
167
+ "eval_logits/chosen": -1.6729556322097778,
168
+ "eval_logits/rejected": -1.6068017482757568,
169
+ "eval_logps/chosen": -2806.55517578125,
170
+ "eval_logps/rejected": -2491.901123046875,
171
+ "eval_loss": 19.51178741455078,
172
+ "eval_rewards/accuracies": 0.52734375,
173
+ "eval_rewards/chosen": 0.025559017434716225,
174
+ "eval_rewards/margins": 0.008243386633694172,
175
+ "eval_rewards/rejected": 0.017315629869699478,
176
+ "eval_runtime": 115.2508,
177
+ "eval_samples_per_second": 17.353,
178
+ "eval_steps_per_second": 0.278,
179
+ "step": 100
180
+ },
181
+ {
182
+ "epoch": 0.14,
183
+ "learning_rate": 4.971588128827782e-07,
184
+ "logits/chosen": -1.7473026514053345,
185
+ "logits/rejected": -1.6806236505508423,
186
+ "logps/chosen": -3125.757080078125,
187
+ "logps/rejected": -2645.337158203125,
188
+ "loss": 26.9149,
189
+ "rewards/accuracies": 0.6000000238418579,
190
+ "rewards/chosen": 0.016931097954511642,
191
+ "rewards/margins": 0.002864243695512414,
192
+ "rewards/rejected": 0.014066850766539574,
193
+ "step": 110
194
+ },
195
+ {
196
+ "epoch": 0.16,
197
+ "learning_rate": 4.951823705321981e-07,
198
+ "logits/chosen": -1.7069530487060547,
199
+ "logits/rejected": -1.6579583883285522,
200
+ "logps/chosen": -2828.78662109375,
201
+ "logps/rejected": -2442.76416015625,
202
+ "loss": 33.872,
203
+ "rewards/accuracies": 0.5600000023841858,
204
+ "rewards/chosen": 0.013961514458060265,
205
+ "rewards/margins": 0.00896529946476221,
206
+ "rewards/rejected": 0.004996216390281916,
207
+ "step": 120
208
+ },
209
+ {
210
+ "epoch": 0.17,
211
+ "learning_rate": 4.926933031274343e-07,
212
+ "logits/chosen": -1.7224699258804321,
213
+ "logits/rejected": -1.6934731006622314,
214
+ "logps/chosen": -2923.9306640625,
215
+ "logps/rejected": -2566.210693359375,
216
+ "loss": 39.0757,
217
+ "rewards/accuracies": 0.550000011920929,
218
+ "rewards/chosen": 0.038237668573856354,
219
+ "rewards/margins": 0.006029448006302118,
220
+ "rewards/rejected": 0.03220822289586067,
221
+ "step": 130
222
+ },
223
+ {
224
+ "epoch": 0.18,
225
+ "learning_rate": 4.896968147878145e-07,
226
+ "logits/chosen": -1.7280409336090088,
227
+ "logits/rejected": -1.7070726156234741,
228
+ "logps/chosen": -2737.75927734375,
229
+ "logps/rejected": -2486.45751953125,
230
+ "loss": 18.5231,
231
+ "rewards/accuracies": 0.6299999952316284,
232
+ "rewards/chosen": 0.031596291810274124,
233
+ "rewards/margins": 0.04216960817575455,
234
+ "rewards/rejected": -0.010573318228125572,
235
+ "step": 140
236
+ },
237
+ {
238
+ "epoch": 0.2,
239
+ "learning_rate": 4.861991705437081e-07,
240
+ "logits/chosen": -1.7859830856323242,
241
+ "logits/rejected": -1.7191492319107056,
242
+ "logps/chosen": -2743.43310546875,
243
+ "logps/rejected": -2297.162109375,
244
+ "loss": 20.7835,
245
+ "rewards/accuracies": 0.5800000429153442,
246
+ "rewards/chosen": 0.0336376316845417,
247
+ "rewards/margins": 0.011832155287265778,
248
+ "rewards/rejected": 0.021805476397275925,
249
+ "step": 150
250
+ },
251
+ {
252
+ "epoch": 0.21,
253
+ "learning_rate": 4.822076832376586e-07,
254
+ "logits/chosen": -1.8132251501083374,
255
+ "logits/rejected": -1.7665789127349854,
256
+ "logps/chosen": -2841.165771484375,
257
+ "logps/rejected": -2748.486572265625,
258
+ "loss": 57.9401,
259
+ "rewards/accuracies": 0.5099999904632568,
260
+ "rewards/chosen": 0.006118610501289368,
261
+ "rewards/margins": 0.0022155127953737974,
262
+ "rewards/rejected": 0.0039030970074236393,
263
+ "step": 160
264
+ },
265
+ {
266
+ "epoch": 0.22,
267
+ "learning_rate": 4.777306982347594e-07,
268
+ "logits/chosen": -1.6557657718658447,
269
+ "logits/rejected": -1.5996118783950806,
270
+ "logps/chosen": -3055.95361328125,
271
+ "logps/rejected": -2603.83642578125,
272
+ "loss": 23.1296,
273
+ "rewards/accuracies": 0.6200000047683716,
274
+ "rewards/chosen": 0.028251701965928078,
275
+ "rewards/margins": 0.020935241132974625,
276
+ "rewards/rejected": 0.007316464092582464,
277
+ "step": 170
278
+ },
279
+ {
280
+ "epoch": 0.24,
281
+ "learning_rate": 4.7277757597424075e-07,
282
+ "logits/chosen": -1.8335905075073242,
283
+ "logits/rejected": -1.7595329284667969,
284
+ "logps/chosen": -2963.73779296875,
285
+ "logps/rejected": -2540.163818359375,
286
+ "loss": 40.5046,
287
+ "rewards/accuracies": 0.5400000810623169,
288
+ "rewards/chosen": 0.018591446802020073,
289
+ "rewards/margins": -0.0024230503477156162,
290
+ "rewards/rejected": 0.02101449854671955,
291
+ "step": 180
292
+ },
293
+ {
294
+ "epoch": 0.25,
295
+ "learning_rate": 4.6735867239874904e-07,
296
+ "logits/chosen": -1.8637840747833252,
297
+ "logits/rejected": -1.7640159130096436,
298
+ "logps/chosen": -3237.434814453125,
299
+ "logps/rejected": -2429.197998046875,
300
+ "loss": 36.3042,
301
+ "rewards/accuracies": 0.6200000047683716,
302
+ "rewards/chosen": 0.04794805496931076,
303
+ "rewards/margins": 0.019117821007966995,
304
+ "rewards/rejected": 0.028830235823988914,
305
+ "step": 190
306
+ },
307
+ {
308
+ "epoch": 0.26,
309
+ "learning_rate": 4.6148531730223733e-07,
310
+ "logits/chosen": -1.6909841299057007,
311
+ "logits/rejected": -1.6915366649627686,
312
+ "logps/chosen": -2649.89404296875,
313
+ "logps/rejected": -2436.87353515625,
314
+ "loss": 28.1241,
315
+ "rewards/accuracies": 0.5300000309944153,
316
+ "rewards/chosen": 0.007661645300686359,
317
+ "rewards/margins": 0.0055509163066744804,
318
+ "rewards/rejected": 0.0021107294596731663,
319
+ "step": 200
320
+ },
321
+ {
322
+ "epoch": 0.26,
323
+ "eval_logits/chosen": -1.681164264678955,
324
+ "eval_logits/rejected": -1.618328332901001,
325
+ "eval_logps/chosen": -2808.258056640625,
326
+ "eval_logps/rejected": -2494.01953125,
327
+ "eval_loss": 32.517486572265625,
328
+ "eval_rewards/accuracies": 0.5234375,
329
+ "eval_rewards/chosen": 0.008527392521500587,
330
+ "eval_rewards/margins": 0.012391308322548866,
331
+ "eval_rewards/rejected": -0.0038639232516288757,
332
+ "eval_runtime": 113.682,
333
+ "eval_samples_per_second": 17.593,
334
+ "eval_steps_per_second": 0.281,
335
+ "step": 200
336
+ },
337
+ {
338
+ "epoch": 0.27,
339
+ "learning_rate": 4.5516979064173524e-07,
340
+ "logits/chosen": -1.749903917312622,
341
+ "logits/rejected": -1.7615283727645874,
342
+ "logps/chosen": -2285.7451171875,
343
+ "logps/rejected": -2269.229736328125,
344
+ "loss": 25.9535,
345
+ "rewards/accuracies": 0.6100000143051147,
346
+ "rewards/chosen": 0.011981850489974022,
347
+ "rewards/margins": 0.014764687046408653,
348
+ "rewards/rejected": -0.0027828349266201258,
349
+ "step": 210
350
+ },
351
+ {
352
+ "epoch": 0.29,
353
+ "learning_rate": 4.484252968625277e-07,
354
+ "logits/chosen": -1.716509222984314,
355
+ "logits/rejected": -1.6396989822387695,
356
+ "logps/chosen": -2435.95556640625,
357
+ "logps/rejected": -1922.770751953125,
358
+ "loss": 28.3739,
359
+ "rewards/accuracies": 0.6200000047683716,
360
+ "rewards/chosen": 0.004359879065304995,
361
+ "rewards/margins": 0.007711753249168396,
362
+ "rewards/rejected": -0.0033518739510327578,
363
+ "step": 220
364
+ },
365
+ {
366
+ "epoch": 0.3,
367
+ "learning_rate": 4.4126593729042193e-07,
368
+ "logits/chosen": -1.799469232559204,
369
+ "logits/rejected": -1.757004737854004,
370
+ "logps/chosen": -3254.6396484375,
371
+ "logps/rejected": -2515.59619140625,
372
+ "loss": 39.4707,
373
+ "rewards/accuracies": 0.5900000333786011,
374
+ "rewards/chosen": 0.03561704605817795,
375
+ "rewards/margins": 0.019830647855997086,
376
+ "rewards/rejected": 0.015786398202180862,
377
+ "step": 230
378
+ },
379
+ {
380
+ "epoch": 0.31,
381
+ "learning_rate": 4.3370668064882397e-07,
382
+ "logits/chosen": -1.7325947284698486,
383
+ "logits/rejected": -1.7474550008773804,
384
+ "logps/chosen": -2579.47412109375,
385
+ "logps/rejected": -2328.500732421875,
386
+ "loss": 44.2727,
387
+ "rewards/accuracies": 0.5100000500679016,
388
+ "rewards/chosen": 0.04269097000360489,
389
+ "rewards/margins": 0.02060030959546566,
390
+ "rewards/rejected": 0.02209065482020378,
391
+ "step": 240
392
+ },
393
+ {
394
+ "epoch": 0.33,
395
+ "learning_rate": 4.2576333176226944e-07,
396
+ "logits/chosen": -1.7366650104522705,
397
+ "logits/rejected": -1.706789255142212,
398
+ "logps/chosen": -2479.5576171875,
399
+ "logps/rejected": -2277.726318359375,
400
+ "loss": 29.5758,
401
+ "rewards/accuracies": 0.5300000309944153,
402
+ "rewards/chosen": 0.1058274507522583,
403
+ "rewards/margins": 0.013660475611686707,
404
+ "rewards/rejected": 0.0921669602394104,
405
+ "step": 250
406
+ },
407
+ {
408
+ "epoch": 0.34,
409
+ "learning_rate": 4.17452498511841e-07,
410
+ "logits/chosen": -1.7807962894439697,
411
+ "logits/rejected": -1.7134149074554443,
412
+ "logps/chosen": -2989.12841796875,
413
+ "logps/rejected": -2354.25830078125,
414
+ "loss": 38.7316,
415
+ "rewards/accuracies": 0.5200000405311584,
416
+ "rewards/chosen": 0.023859605193138123,
417
+ "rewards/margins": 0.005521018523722887,
418
+ "rewards/rejected": 0.018338587135076523,
419
+ "step": 260
420
+ },
421
+ {
422
+ "epoch": 0.35,
423
+ "learning_rate": 4.087915571115629e-07,
424
+ "logits/chosen": -1.8165556192398071,
425
+ "logits/rejected": -1.7687098979949951,
426
+ "logps/chosen": -2833.55859375,
427
+ "logps/rejected": -2183.32470703125,
428
+ "loss": 330.4642,
429
+ "rewards/accuracies": 0.5699999928474426,
430
+ "rewards/chosen": 0.031318746507167816,
431
+ "rewards/margins": 0.029996121302247047,
432
+ "rewards/rejected": 0.0013226259034126997,
433
+ "step": 270
434
+ },
435
+ {
436
+ "epoch": 0.37,
437
+ "learning_rate": 3.997986157783715e-07,
438
+ "logits/chosen": -1.6980018615722656,
439
+ "logits/rejected": -1.589050531387329,
440
+ "logps/chosen": -3510.792236328125,
441
+ "logps/rejected": -2689.208984375,
442
+ "loss": 58.1646,
443
+ "rewards/accuracies": 0.5200000405311584,
444
+ "rewards/chosen": 0.014776378870010376,
445
+ "rewards/margins": 0.011294273659586906,
446
+ "rewards/rejected": 0.0034821047447621822,
447
+ "step": 280
448
+ },
449
+ {
450
+ "epoch": 0.38,
451
+ "learning_rate": 3.9049247687162155e-07,
452
+ "logits/chosen": -1.7791646718978882,
453
+ "logits/rejected": -1.7399044036865234,
454
+ "logps/chosen": -2478.590576171875,
455
+ "logps/rejected": -2269.01416015625,
456
+ "loss": 31.6725,
457
+ "rewards/accuracies": 0.5699999928474426,
458
+ "rewards/chosen": 0.04884537309408188,
459
+ "rewards/margins": 0.0339895561337471,
460
+ "rewards/rejected": 0.014855814166367054,
461
+ "step": 290
462
+ },
463
+ {
464
+ "epoch": 0.39,
465
+ "learning_rate": 3.8089259758128543e-07,
466
+ "logits/chosen": -1.670789361000061,
467
+ "logits/rejected": -1.6030629873275757,
468
+ "logps/chosen": -2726.465576171875,
469
+ "logps/rejected": -2119.26123046875,
470
+ "loss": 84.7591,
471
+ "rewards/accuracies": 0.5699999928474426,
472
+ "rewards/chosen": 0.021672677248716354,
473
+ "rewards/margins": -0.010785548016428947,
474
+ "rewards/rejected": 0.03245822712779045,
475
+ "step": 300
476
+ },
477
+ {
478
+ "epoch": 0.39,
479
+ "eval_logits/chosen": -1.660080909729004,
480
+ "eval_logits/rejected": -1.596778154373169,
481
+ "eval_logps/chosen": -2806.140625,
482
+ "eval_logps/rejected": -2492.270263671875,
483
+ "eval_loss": 47.80431365966797,
484
+ "eval_rewards/accuracies": 0.5390625,
485
+ "eval_rewards/chosen": 0.029702020809054375,
486
+ "eval_rewards/margins": 0.01607733778655529,
487
+ "eval_rewards/rejected": 0.013624681159853935,
488
+ "eval_runtime": 116.3019,
489
+ "eval_samples_per_second": 17.197,
490
+ "eval_steps_per_second": 0.275,
491
+ "step": 300
492
+ },
493
+ {
494
+ "epoch": 0.41,
495
+ "learning_rate": 3.710190492470386e-07,
496
+ "logits/chosen": -1.6620228290557861,
497
+ "logits/rejected": -1.7311099767684937,
498
+ "logps/chosen": -2315.977294921875,
499
+ "logps/rejected": -2199.08251953125,
500
+ "loss": 43.6013,
501
+ "rewards/accuracies": 0.5400000214576721,
502
+ "rewards/chosen": 0.032384876161813736,
503
+ "rewards/margins": 0.008921505883336067,
504
+ "rewards/rejected": 0.02346337027847767,
505
+ "step": 310
506
+ },
507
+ {
508
+ "epoch": 0.42,
509
+ "learning_rate": 3.6089247539328616e-07,
510
+ "logits/chosen": -1.7675050497055054,
511
+ "logits/rejected": -1.7156997919082642,
512
+ "logps/chosen": -2859.810791015625,
513
+ "logps/rejected": -2569.75537109375,
514
+ "loss": 38.8904,
515
+ "rewards/accuracies": 0.559999942779541,
516
+ "rewards/chosen": 0.020630866289138794,
517
+ "rewards/margins": 0.0018306337296962738,
518
+ "rewards/rejected": 0.01880022883415222,
519
+ "step": 320
520
+ },
521
+ {
522
+ "epoch": 0.43,
523
+ "learning_rate": 3.5053404856787166e-07,
524
+ "logits/chosen": -1.6446609497070312,
525
+ "logits/rejected": -1.5918724536895752,
526
+ "logps/chosen": -3104.72802734375,
527
+ "logps/rejected": -2430.239013671875,
528
+ "loss": 84.9753,
529
+ "rewards/accuracies": 0.47999995946884155,
530
+ "rewards/chosen": 0.053660690784454346,
531
+ "rewards/margins": -0.005831834394484758,
532
+ "rewards/rejected": 0.05949252098798752,
533
+ "step": 330
534
+ },
535
+ {
536
+ "epoch": 0.44,
537
+ "learning_rate": 3.399654260747078e-07,
538
+ "logits/chosen": -1.699196219444275,
539
+ "logits/rejected": -1.7045748233795166,
540
+ "logps/chosen": -2584.699462890625,
541
+ "logps/rejected": -2263.678466796875,
542
+ "loss": 38.1532,
543
+ "rewards/accuracies": 0.5300000309944153,
544
+ "rewards/chosen": 0.02709970250725746,
545
+ "rewards/margins": 0.01412280835211277,
546
+ "rewards/rejected": 0.012976895086467266,
547
+ "step": 340
548
+ },
549
+ {
550
+ "epoch": 0.46,
551
+ "learning_rate": 3.2920870469288373e-07,
552
+ "logits/chosen": -1.7267532348632812,
553
+ "logits/rejected": -1.6659395694732666,
554
+ "logps/chosen": -2935.341796875,
555
+ "logps/rejected": -2503.583984375,
556
+ "loss": 47.1836,
557
+ "rewards/accuracies": 0.5199999809265137,
558
+ "rewards/chosen": 0.031215447932481766,
559
+ "rewards/margins": 0.022346725687384605,
560
+ "rewards/rejected": 0.008868719451129436,
561
+ "step": 350
562
+ },
563
+ {
564
+ "epoch": 0.47,
565
+ "learning_rate": 3.182863744769218e-07,
566
+ "logits/chosen": -1.7288787364959717,
567
+ "logits/rejected": -1.6928844451904297,
568
+ "logps/chosen": -2811.489501953125,
569
+ "logps/rejected": -2596.68310546875,
570
+ "loss": 36.8176,
571
+ "rewards/accuracies": 0.5099999904632568,
572
+ "rewards/chosen": 0.1375296413898468,
573
+ "rewards/margins": 0.0820910781621933,
574
+ "rewards/rejected": 0.05543852597475052,
575
+ "step": 360
576
+ },
577
+ {
578
+ "epoch": 0.48,
579
+ "learning_rate": 3.072212717347776e-07,
580
+ "logits/chosen": -1.7680120468139648,
581
+ "logits/rejected": -1.6781940460205078,
582
+ "logps/chosen": -3101.98583984375,
583
+ "logps/rejected": -2426.4716796875,
584
+ "loss": 36.7837,
585
+ "rewards/accuracies": 0.5199999809265137,
586
+ "rewards/chosen": 0.022122934460639954,
587
+ "rewards/margins": 0.011781491339206696,
588
+ "rewards/rejected": 0.010341441258788109,
589
+ "step": 370
590
+ },
591
+ {
592
+ "epoch": 0.5,
593
+ "learning_rate": 2.9603653128189665e-07,
594
+ "logits/chosen": -1.6812299489974976,
595
+ "logits/rejected": -1.7215496301651,
596
+ "logps/chosen": -2823.8291015625,
597
+ "logps/rejected": -2762.53076171875,
598
+ "loss": 42.732,
599
+ "rewards/accuracies": 0.5,
600
+ "rewards/chosen": 0.028385426849126816,
601
+ "rewards/margins": -0.006515379063785076,
602
+ "rewards/rejected": 0.03490080684423447,
603
+ "step": 380
604
+ },
605
+ {
606
+ "epoch": 0.51,
607
+ "learning_rate": 2.8475553807115387e-07,
608
+ "logits/chosen": -1.8070951700210571,
609
+ "logits/rejected": -1.7426990270614624,
610
+ "logps/chosen": -2697.833251953125,
611
+ "logps/rejected": -2263.9990234375,
612
+ "loss": 55.8683,
613
+ "rewards/accuracies": 0.5099999904632568,
614
+ "rewards/chosen": 0.012206131592392921,
615
+ "rewards/margins": 0.011461116373538971,
616
+ "rewards/rejected": 0.0007450145785696805,
617
+ "step": 390
618
+ },
619
+ {
620
+ "epoch": 0.52,
621
+ "learning_rate": 2.7340187829980883e-07,
622
+ "logits/chosen": -1.8249183893203735,
623
+ "logits/rejected": -1.7130759954452515,
624
+ "logps/chosen": -2940.11181640625,
625
+ "logps/rejected": -2463.068359375,
626
+ "loss": 40.7835,
627
+ "rewards/accuracies": 0.6000000238418579,
628
+ "rewards/chosen": 0.0059137181378901005,
629
+ "rewards/margins": 0.01795141212642193,
630
+ "rewards/rejected": -0.012037692591547966,
631
+ "step": 400
632
+ },
633
+ {
634
+ "epoch": 0.52,
635
+ "eval_logits/chosen": -1.6917269229888916,
636
+ "eval_logits/rejected": -1.628839373588562,
637
+ "eval_logps/chosen": -2807.42626953125,
638
+ "eval_logps/rejected": -2493.92041015625,
639
+ "eval_loss": 30.672218322753906,
640
+ "eval_rewards/accuracies": 0.5546875,
641
+ "eval_rewards/chosen": 0.016848012804985046,
642
+ "eval_rewards/margins": 0.019721925258636475,
643
+ "eval_rewards/rejected": -0.0028739143162965775,
644
+ "eval_runtime": 110.0303,
645
+ "eval_samples_per_second": 18.177,
646
+ "eval_steps_per_second": 0.291,
647
+ "step": 400
648
+ },
649
+ {
650
+ "epoch": 0.54,
651
+ "learning_rate": 2.6199929009569996e-07,
652
+ "logits/chosen": -1.7034717798233032,
653
+ "logits/rejected": -1.707564353942871,
654
+ "logps/chosen": -2599.38330078125,
655
+ "logps/rejected": -2273.864990234375,
656
+ "loss": 43.9981,
657
+ "rewards/accuracies": 0.5600000023841858,
658
+ "rewards/chosen": 0.02160579524934292,
659
+ "rewards/margins": 0.0038177832029759884,
660
+ "rewards/rejected": 0.017788011580705643,
661
+ "step": 410
662
+ },
663
+ {
664
+ "epoch": 0.55,
665
+ "learning_rate": 2.5057161388578505e-07,
666
+ "logits/chosen": -1.7964134216308594,
667
+ "logits/rejected": -1.730661392211914,
668
+ "logps/chosen": -3038.08740234375,
669
+ "logps/rejected": -2405.333740234375,
670
+ "loss": 31.4477,
671
+ "rewards/accuracies": 0.5600000619888306,
672
+ "rewards/chosen": 0.05013390630483627,
673
+ "rewards/margins": 0.029845798388123512,
674
+ "rewards/rejected": 0.02028810977935791,
675
+ "step": 420
676
+ },
677
+ {
678
+ "epoch": 0.56,
679
+ "learning_rate": 2.391427425507943e-07,
680
+ "logits/chosen": -1.6959331035614014,
681
+ "logits/rejected": -1.6784296035766602,
682
+ "logps/chosen": -2696.2236328125,
683
+ "logps/rejected": -2173.48583984375,
684
+ "loss": 32.2174,
685
+ "rewards/accuracies": 0.5600000619888306,
686
+ "rewards/chosen": 0.01727980561554432,
687
+ "rewards/margins": 0.013602805323898792,
688
+ "rewards/rejected": 0.0036770000588148832,
689
+ "step": 430
690
+ },
691
+ {
692
+ "epoch": 0.58,
693
+ "learning_rate": 2.2773657147021465e-07,
694
+ "logits/chosen": -1.8469693660736084,
695
+ "logits/rejected": -1.7459551095962524,
696
+ "logps/chosen": -3117.762451171875,
697
+ "logps/rejected": -2390.564208984375,
698
+ "loss": 37.6526,
699
+ "rewards/accuracies": 0.5,
700
+ "rewards/chosen": 0.01079155970364809,
701
+ "rewards/margins": 0.011436818167567253,
702
+ "rewards/rejected": -0.00064525764901191,
703
+ "step": 440
704
+ },
705
+ {
706
+ "epoch": 0.59,
707
+ "learning_rate": 2.1637694856204885e-07,
708
+ "logits/chosen": -1.751587152481079,
709
+ "logits/rejected": -1.6395552158355713,
710
+ "logps/chosen": -2887.770751953125,
711
+ "logps/rejected": -2129.771728515625,
712
+ "loss": 53.6906,
713
+ "rewards/accuracies": 0.550000011920929,
714
+ "rewards/chosen": 0.004514098167419434,
715
+ "rewards/margins": 0.00042482782737351954,
716
+ "rewards/rejected": 0.004089272115379572,
717
+ "step": 450
718
+ },
719
+ {
720
+ "epoch": 0.6,
721
+ "learning_rate": 2.0508762442180743e-07,
722
+ "logits/chosen": -1.8443762063980103,
723
+ "logits/rejected": -1.792295217514038,
724
+ "logps/chosen": -2964.06494140625,
725
+ "logps/rejected": -2577.50244140625,
726
+ "loss": 62.4137,
727
+ "rewards/accuracies": 0.5600000023841858,
728
+ "rewards/chosen": 0.04290894791483879,
729
+ "rewards/margins": 0.01085699163377285,
730
+ "rewards/rejected": 0.03205195814371109,
731
+ "step": 460
732
+ },
733
+ {
734
+ "epoch": 0.61,
735
+ "learning_rate": 1.93892202664981e-07,
736
+ "logits/chosen": -1.6403262615203857,
737
+ "logits/rejected": -1.712969183921814,
738
+ "logps/chosen": -2689.706787109375,
739
+ "logps/rejected": -2513.2998046875,
740
+ "loss": 31.7885,
741
+ "rewards/accuracies": 0.5100000500679016,
742
+ "rewards/chosen": 0.01223880797624588,
743
+ "rewards/margins": 0.011182873509824276,
744
+ "rewards/rejected": 0.0010559323709458113,
745
+ "step": 470
746
+ },
747
+ {
748
+ "epoch": 0.63,
749
+ "learning_rate": 1.8281409057681686e-07,
750
+ "logits/chosen": -1.651449203491211,
751
+ "logits/rejected": -1.5920675992965698,
752
+ "logps/chosen": -3211.50341796875,
753
+ "logps/rejected": -2753.0322265625,
754
+ "loss": 103.2519,
755
+ "rewards/accuracies": 0.550000011920929,
756
+ "rewards/chosen": 0.0340498685836792,
757
+ "rewards/margins": 0.005449384916573763,
758
+ "rewards/rejected": 0.028600484132766724,
759
+ "step": 480
760
+ },
761
+ {
762
+ "epoch": 0.64,
763
+ "learning_rate": 1.7187645017258195e-07,
764
+ "logits/chosen": -1.823428750038147,
765
+ "logits/rejected": -1.7740917205810547,
766
+ "logps/chosen": -2745.991455078125,
767
+ "logps/rejected": -2407.27978515625,
768
+ "loss": 48.2582,
769
+ "rewards/accuracies": 0.5300000309944153,
770
+ "rewards/chosen": 0.03057839907705784,
771
+ "rewards/margins": 0.003420495195314288,
772
+ "rewards/rejected": 0.027157902717590332,
773
+ "step": 490
774
+ },
775
+ {
776
+ "epoch": 0.65,
777
+ "learning_rate": 1.6110214977063343e-07,
778
+ "logits/chosen": -1.7967636585235596,
779
+ "logits/rejected": -1.7410199642181396,
780
+ "logps/chosen": -2905.251708984375,
781
+ "logps/rejected": -2435.401611328125,
782
+ "loss": 36.2204,
783
+ "rewards/accuracies": 0.5,
784
+ "rewards/chosen": 0.013903191313147545,
785
+ "rewards/margins": 0.00013892585411667824,
786
+ "rewards/rejected": 0.013764267787337303,
787
+ "step": 500
788
+ },
789
+ {
790
+ "epoch": 0.65,
791
+ "eval_logits/chosen": -1.6842743158340454,
792
+ "eval_logits/rejected": -1.6236169338226318,
793
+ "eval_logps/chosen": -2806.076171875,
794
+ "eval_logps/rejected": -2491.544677734375,
795
+ "eval_loss": 31.220157623291016,
796
+ "eval_rewards/accuracies": 0.53515625,
797
+ "eval_rewards/chosen": 0.030346479266881943,
798
+ "eval_rewards/margins": 0.009465347044169903,
799
+ "eval_rewards/rejected": 0.020881133154034615,
800
+ "eval_runtime": 112.3374,
801
+ "eval_samples_per_second": 17.804,
802
+ "eval_steps_per_second": 0.285,
803
+ "step": 500
804
+ },
805
+ {
806
+ "epoch": 0.67,
807
+ "learning_rate": 1.5051371617954777e-07,
808
+ "logits/chosen": -1.6810442209243774,
809
+ "logits/rejected": -1.6596931219100952,
810
+ "logps/chosen": -2559.396728515625,
811
+ "logps/rejected": -2228.120361328125,
812
+ "loss": 44.2046,
813
+ "rewards/accuracies": 0.5399999618530273,
814
+ "rewards/chosen": 0.016220757737755775,
815
+ "rewards/margins": 0.007862111553549767,
816
+ "rewards/rejected": 0.00835864432156086,
817
+ "step": 510
818
+ },
819
+ {
820
+ "epoch": 0.68,
821
+ "learning_rate": 1.4013328759927622e-07,
822
+ "logits/chosen": -1.6315361261367798,
823
+ "logits/rejected": -1.6191142797470093,
824
+ "logps/chosen": -2893.280029296875,
825
+ "logps/rejected": -2805.344970703125,
826
+ "loss": 31.2508,
827
+ "rewards/accuracies": 0.6299999952316284,
828
+ "rewards/chosen": 0.02805119752883911,
829
+ "rewards/margins": 0.01223050244152546,
830
+ "rewards/rejected": 0.0158206969499588,
831
+ "step": 520
832
+ },
833
+ {
834
+ "epoch": 0.69,
835
+ "learning_rate": 1.2998256733479896e-07,
836
+ "logits/chosen": -1.810739278793335,
837
+ "logits/rejected": -1.8173195123672485,
838
+ "logps/chosen": -2332.383056640625,
839
+ "logps/rejected": -1922.5443115234375,
840
+ "loss": 226.0483,
841
+ "rewards/accuracies": 0.5600000023841858,
842
+ "rewards/chosen": 0.020228227600455284,
843
+ "rewards/margins": 0.009395391680300236,
844
+ "rewards/rejected": 0.010832836851477623,
845
+ "step": 530
846
+ },
847
+ {
848
+ "epoch": 0.71,
849
+ "learning_rate": 1.200827784190537e-07,
850
+ "logits/chosen": -1.6795597076416016,
851
+ "logits/rejected": -1.6883628368377686,
852
+ "logps/chosen": -3027.91796875,
853
+ "logps/rejected": -2619.313232421875,
854
+ "loss": 29.3654,
855
+ "rewards/accuracies": 0.5600000619888306,
856
+ "rewards/chosen": 0.01968817412853241,
857
+ "rewards/margins": 0.008831174112856388,
858
+ "rewards/rejected": 0.010857000946998596,
859
+ "step": 540
860
+ },
861
+ {
862
+ "epoch": 0.72,
863
+ "learning_rate": 1.1045461924001323e-07,
864
+ "logits/chosen": -1.791738748550415,
865
+ "logits/rejected": -1.8031442165374756,
866
+ "logps/chosen": -2852.6904296875,
867
+ "logps/rejected": -2462.853271484375,
868
+ "loss": 45.3966,
869
+ "rewards/accuracies": 0.46000003814697266,
870
+ "rewards/chosen": 0.010838394984602928,
871
+ "rewards/margins": 0.003685446921736002,
872
+ "rewards/rejected": 0.007152946200221777,
873
+ "step": 550
874
+ },
875
+ {
876
+ "epoch": 0.73,
877
+ "learning_rate": 1.0111822026468514e-07,
878
+ "logits/chosen": -1.7872514724731445,
879
+ "logits/rejected": -1.658860445022583,
880
+ "logps/chosen": -2903.530029296875,
881
+ "logps/rejected": -2319.64599609375,
882
+ "loss": 67.4473,
883
+ "rewards/accuracies": 0.5700000524520874,
884
+ "rewards/chosen": 0.009608490392565727,
885
+ "rewards/margins": 0.004570655524730682,
886
+ "rewards/rejected": 0.0050378344021737576,
887
+ "step": 560
888
+ },
889
+ {
890
+ "epoch": 0.75,
891
+ "learning_rate": 9.209310195051581e-08,
892
+ "logits/chosen": -1.8252109289169312,
893
+ "logits/rejected": -1.6855742931365967,
894
+ "logps/chosen": -2538.860107421875,
895
+ "logps/rejected": -1955.6048583984375,
896
+ "loss": 63.0174,
897
+ "rewards/accuracies": 0.6299999952316284,
898
+ "rewards/chosen": 0.04829864576458931,
899
+ "rewards/margins": 0.022980675101280212,
900
+ "rewards/rejected": 0.025317972525954247,
901
+ "step": 570
902
+ },
903
+ {
904
+ "epoch": 0.76,
905
+ "learning_rate": 8.339813393219713e-08,
906
+ "logits/chosen": -1.739793062210083,
907
+ "logits/rejected": -1.641005516052246,
908
+ "logps/chosen": -2791.561767578125,
909
+ "logps/rejected": -2475.72998046875,
910
+ "loss": 59.8369,
911
+ "rewards/accuracies": 0.5699999928474426,
912
+ "rewards/chosen": 0.05231914669275284,
913
+ "rewards/margins": 0.021858692169189453,
914
+ "rewards/rejected": 0.030460450798273087,
915
+ "step": 580
916
+ },
917
+ {
918
+ "epoch": 0.77,
919
+ "learning_rate": 7.505149556920698e-08,
920
+ "logits/chosen": -1.8431494235992432,
921
+ "logits/rejected": -1.7774893045425415,
922
+ "logps/chosen": -2542.13427734375,
923
+ "logps/rejected": -2193.825927734375,
924
+ "loss": 29.3999,
925
+ "rewards/accuracies": 0.5800000429153442,
926
+ "rewards/chosen": 0.04752471297979355,
927
+ "rewards/margins": 0.017444033175706863,
928
+ "rewards/rejected": 0.030080681666731834,
929
+ "step": 590
930
+ },
931
+ {
932
+ "epoch": 0.78,
933
+ "learning_rate": 6.707063793657064e-08,
934
+ "logits/chosen": -1.7773969173431396,
935
+ "logits/rejected": -1.6891686916351318,
936
+ "logps/chosen": -2942.21240234375,
937
+ "logps/rejected": -2429.352294921875,
938
+ "loss": 99.7738,
939
+ "rewards/accuracies": 0.6200000047683716,
940
+ "rewards/chosen": 0.03306427597999573,
941
+ "rewards/margins": 0.01405587512999773,
942
+ "rewards/rejected": 0.019008399918675423,
943
+ "step": 600
944
+ },
945
+ {
946
+ "epoch": 0.78,
947
+ "eval_logits/chosen": -1.6827195882797241,
948
+ "eval_logits/rejected": -1.6222153902053833,
949
+ "eval_logps/chosen": -2804.348388671875,
950
+ "eval_logps/rejected": -2489.908935546875,
951
+ "eval_loss": 33.74028778076172,
952
+ "eval_rewards/accuracies": 0.5390625,
953
+ "eval_rewards/chosen": 0.04762275516986847,
954
+ "eval_rewards/margins": 0.010385587811470032,
955
+ "eval_rewards/rejected": 0.037237171083688736,
956
+ "eval_runtime": 106.3716,
957
+ "eval_samples_per_second": 18.802,
958
+ "eval_steps_per_second": 0.301,
959
+ "step": 600
960
+ },
961
+ {
962
+ "epoch": 0.8,
963
+ "learning_rate": 5.947224733831363e-08,
964
+ "logits/chosen": -1.759399175643921,
965
+ "logits/rejected": -1.7431520223617554,
966
+ "logps/chosen": -2756.701416015625,
967
+ "logps/rejected": -2470.905029296875,
968
+ "loss": 51.5387,
969
+ "rewards/accuracies": 0.5199999809265137,
970
+ "rewards/chosen": 0.011415710672736168,
971
+ "rewards/margins": 0.009652274660766125,
972
+ "rewards/rejected": 0.0017634350806474686,
973
+ "step": 610
974
+ },
975
+ {
976
+ "epoch": 0.81,
977
+ "learning_rate": 5.227221041988955e-08,
978
+ "logits/chosen": -1.7857062816619873,
979
+ "logits/rejected": -1.725630760192871,
980
+ "logps/chosen": -2520.410400390625,
981
+ "logps/rejected": -2319.78564453125,
982
+ "loss": 28.3912,
983
+ "rewards/accuracies": 0.5900000333786011,
984
+ "rewards/chosen": 0.01811736635863781,
985
+ "rewards/margins": 0.01297797542065382,
986
+ "rewards/rejected": 0.005139390472322702,
987
+ "step": 620
988
+ },
989
+ {
990
+ "epoch": 0.82,
991
+ "learning_rate": 4.548558095252758e-08,
992
+ "logits/chosen": -1.6374757289886475,
993
+ "logits/rejected": -1.673044204711914,
994
+ "logps/chosen": -2845.2119140625,
995
+ "logps/rejected": -2698.294677734375,
996
+ "loss": 42.0619,
997
+ "rewards/accuracies": 0.48000001907348633,
998
+ "rewards/chosen": 0.030720695853233337,
999
+ "rewards/margins": 0.020541973412036896,
1000
+ "rewards/rejected": 0.010178723372519016,
1001
+ "step": 630
1002
+ },
1003
+ {
1004
+ "epoch": 0.84,
1005
+ "learning_rate": 3.9126548358945635e-08,
1006
+ "logits/chosen": -1.7063062191009521,
1007
+ "logits/rejected": -1.6988853216171265,
1008
+ "logps/chosen": -3136.520263671875,
1009
+ "logps/rejected": -2731.25927734375,
1010
+ "loss": 46.0608,
1011
+ "rewards/accuracies": 0.5600000619888306,
1012
+ "rewards/chosen": 0.029350021854043007,
1013
+ "rewards/margins": 0.010425332933664322,
1014
+ "rewards/rejected": 0.018924688920378685,
1015
+ "step": 640
1016
+ },
1017
+ {
1018
+ "epoch": 0.85,
1019
+ "learning_rate": 3.3208408046234896e-08,
1020
+ "logits/chosen": -1.8164535760879517,
1021
+ "logits/rejected": -1.7656440734863281,
1022
+ "logps/chosen": -2538.8046875,
1023
+ "logps/rejected": -2061.81591796875,
1024
+ "loss": 40.0838,
1025
+ "rewards/accuracies": 0.550000011920929,
1026
+ "rewards/chosen": 0.01678399369120598,
1027
+ "rewards/margins": 0.009666666388511658,
1028
+ "rewards/rejected": 0.007117328234016895,
1029
+ "step": 650
1030
+ },
1031
+ {
1032
+ "epoch": 0.86,
1033
+ "learning_rate": 2.774353360794493e-08,
1034
+ "logits/chosen": -1.7155154943466187,
1035
+ "logits/rejected": -1.7442939281463623,
1036
+ "logps/chosen": -2761.740966796875,
1037
+ "logps/rejected": -2534.80419921875,
1038
+ "loss": 36.8374,
1039
+ "rewards/accuracies": 0.64000004529953,
1040
+ "rewards/chosen": 0.03588343411684036,
1041
+ "rewards/margins": 0.037469957023859024,
1042
+ "rewards/rejected": -0.0015865217428654432,
1043
+ "step": 660
1044
+ },
1045
+ {
1046
+ "epoch": 0.88,
1047
+ "learning_rate": 2.2743350953487422e-08,
1048
+ "logits/chosen": -1.6992709636688232,
1049
+ "logits/rejected": -1.7416222095489502,
1050
+ "logps/chosen": -2850.97705078125,
1051
+ "logps/rejected": -2569.76611328125,
1052
+ "loss": 86.4259,
1053
+ "rewards/accuracies": 0.5300000905990601,
1054
+ "rewards/chosen": 0.018378589302301407,
1055
+ "rewards/margins": 0.004103804472833872,
1056
+ "rewards/rejected": 0.014274786226451397,
1057
+ "step": 670
1058
+ },
1059
+ {
1060
+ "epoch": 0.89,
1061
+ "learning_rate": 1.8218314418949387e-08,
1062
+ "logits/chosen": -1.718764305114746,
1063
+ "logits/rejected": -1.6741740703582764,
1064
+ "logps/chosen": -2353.093017578125,
1065
+ "logps/rejected": -2190.77001953125,
1066
+ "loss": 46.1473,
1067
+ "rewards/accuracies": 0.5699999928474426,
1068
+ "rewards/chosen": 0.0031641994137316942,
1069
+ "rewards/margins": 0.0004996396601200104,
1070
+ "rewards/rejected": 0.002664559753611684,
1071
+ "step": 680
1072
+ },
1073
+ {
1074
+ "epoch": 0.9,
1075
+ "learning_rate": 1.4177884909263277e-08,
1076
+ "logits/chosen": -1.6867101192474365,
1077
+ "logits/rejected": -1.652515172958374,
1078
+ "logps/chosen": -2937.97412109375,
1079
+ "logps/rejected": -2552.07177734375,
1080
+ "loss": 37.0029,
1081
+ "rewards/accuracies": 0.5199999809265137,
1082
+ "rewards/chosen": 0.007288885302841663,
1083
+ "rewards/margins": -0.0001541988895041868,
1084
+ "rewards/rejected": 0.007443083915859461,
1085
+ "step": 690
1086
+ },
1087
+ {
1088
+ "epoch": 0.92,
1089
+ "learning_rate": 1.063051011743335e-08,
1090
+ "logits/chosen": -1.7554800510406494,
1091
+ "logits/rejected": -1.7419729232788086,
1092
+ "logps/chosen": -2755.109619140625,
1093
+ "logps/rejected": -2368.246337890625,
1094
+ "loss": 41.8506,
1095
+ "rewards/accuracies": 0.46000003814697266,
1096
+ "rewards/chosen": 0.006798497401177883,
1097
+ "rewards/margins": 0.010020612739026546,
1098
+ "rewards/rejected": -0.0032221146393567324,
1099
+ "step": 700
1100
+ },
1101
+ {
1102
+ "epoch": 0.92,
1103
+ "eval_logits/chosen": -1.682308554649353,
1104
+ "eval_logits/rejected": -1.6210675239562988,
1105
+ "eval_logps/chosen": -2806.1005859375,
1106
+ "eval_logps/rejected": -2491.68505859375,
1107
+ "eval_loss": 32.91334915161133,
1108
+ "eval_rewards/accuracies": 0.5546875,
1109
+ "eval_rewards/chosen": 0.030103469267487526,
1110
+ "eval_rewards/margins": 0.01062812004238367,
1111
+ "eval_rewards/rejected": 0.01947534643113613,
1112
+ "eval_runtime": 110.9725,
1113
+ "eval_samples_per_second": 18.022,
1114
+ "eval_steps_per_second": 0.288,
1115
+ "step": 700
1116
+ },
1117
+ {
1118
+ "epoch": 0.93,
1119
+ "learning_rate": 7.58360686217671e-09,
1120
+ "logits/chosen": -1.7843902111053467,
1121
+ "logits/rejected": -1.6857761144638062,
1122
+ "logps/chosen": -2821.310546875,
1123
+ "logps/rejected": -2445.586669921875,
1124
+ "loss": 44.0616,
1125
+ "rewards/accuracies": 0.6000000238418579,
1126
+ "rewards/chosen": 0.025277357548475266,
1127
+ "rewards/margins": 0.014521745964884758,
1128
+ "rewards/rejected": 0.010755611583590508,
1129
+ "step": 710
1130
+ },
1131
+ {
1132
+ "epoch": 0.94,
1133
+ "learning_rate": 5.043545580906694e-09,
1134
+ "logits/chosen": -1.7102206945419312,
1135
+ "logits/rejected": -1.6009165048599243,
1136
+ "logps/chosen": -2682.336669921875,
1137
+ "logps/rejected": -2234.36279296875,
1138
+ "loss": 44.9558,
1139
+ "rewards/accuracies": 0.6299999952316284,
1140
+ "rewards/chosen": 0.018401915207505226,
1141
+ "rewards/margins": 0.030992329120635986,
1142
+ "rewards/rejected": -0.012590417638421059,
1143
+ "step": 720
1144
+ },
1145
+ {
1146
+ "epoch": 0.95,
1147
+ "learning_rate": 3.015637010480576e-09,
1148
+ "logits/chosen": -1.7701480388641357,
1149
+ "logits/rejected": -1.7436597347259521,
1150
+ "logps/chosen": -3042.88330078125,
1151
+ "logps/rejected": -2501.55615234375,
1152
+ "loss": 35.3819,
1153
+ "rewards/accuracies": 0.5199999809265137,
1154
+ "rewards/chosen": 0.012798592448234558,
1155
+ "rewards/margins": -0.0020437492057681084,
1156
+ "rewards/rejected": 0.014842341654002666,
1157
+ "step": 730
1158
+ },
1159
+ {
1160
+ "epoch": 0.97,
1161
+ "learning_rate": 1.5041210835596285e-09,
1162
+ "logits/chosen": -1.703537940979004,
1163
+ "logits/rejected": -1.6897165775299072,
1164
+ "logps/chosen": -2817.1484375,
1165
+ "logps/rejected": -2390.23095703125,
1166
+ "loss": 55.7217,
1167
+ "rewards/accuracies": 0.5800000429153442,
1168
+ "rewards/chosen": 0.022753870114684105,
1169
+ "rewards/margins": 0.017634030431509018,
1170
+ "rewards/rejected": 0.005119838751852512,
1171
+ "step": 740
1172
+ },
1173
+ {
1174
+ "epoch": 0.98,
1175
+ "learning_rate": 5.121580637968137e-10,
1176
+ "logits/chosen": -1.7322509288787842,
1177
+ "logits/rejected": -1.6345192193984985,
1178
+ "logps/chosen": -2836.63623046875,
1179
+ "logps/rejected": -2363.93505859375,
1180
+ "loss": 67.2692,
1181
+ "rewards/accuracies": 0.6300000548362732,
1182
+ "rewards/chosen": 0.02889620140194893,
1183
+ "rewards/margins": 0.014323192648589611,
1184
+ "rewards/rejected": 0.014573007822036743,
1185
+ "step": 750
1186
+ },
1187
+ {
1188
+ "epoch": 0.99,
1189
+ "learning_rate": 4.1821938386477075e-11,
1190
+ "logits/chosen": -1.7962977886199951,
1191
+ "logits/rejected": -1.7127879858016968,
1192
+ "logps/chosen": -2801.806640625,
1193
+ "logps/rejected": -2341.844970703125,
1194
+ "loss": 46.0674,
1195
+ "rewards/accuracies": 0.5700000524520874,
1196
+ "rewards/chosen": 0.009056088514626026,
1197
+ "rewards/margins": 0.013271180912852287,
1198
+ "rewards/rejected": -0.004215092398226261,
1199
+ "step": 760
1200
+ },
1201
+ {
1202
+ "epoch": 1.0,
1203
+ "step": 764,
1204
+ "total_flos": 0.0,
1205
+ "train_loss": 46.613421885577296,
1206
+ "train_runtime": 4597.6924,
1207
+ "train_samples_per_second": 13.297,
1208
+ "train_steps_per_second": 0.166
1209
+ }
1210
+ ],
1211
+ "logging_steps": 10,
1212
+ "max_steps": 764,
1213
+ "num_input_tokens_seen": 0,
1214
+ "num_train_epochs": 1,
1215
+ "save_steps": 100,
1216
+ "total_flos": 0.0,
1217
+ "train_batch_size": 5,
1218
+ "trial_name": null,
1219
+ "trial_params": null
1220
+ }