Hugging Face's logo

Datasets:
michaelchenkj
/
test-models

Dataset card Files
xet
 Community
michaelchenkj commited on
Commit
52d5459
·
verified ·
1 Parent(s): c1ed7a6

Add files using upload-large-folder tool

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. runs/t0/t0_S3_lr0.0003/checkpoint-864/chat_template.jinja +85 -0
  2. runs/t0/t0_S3_lr0.0003/checkpoint-864/config.json +32 -0
  3. runs/t0/t0_S3_lr0.0003/checkpoint-864/generation_config.json +11 -0
  4. runs/t0/t0_S3_lr0.0003/checkpoint-864/tokenizer_config.json +19 -0
  5. runs/t0/t0_S3_lr0.0003/checkpoint-864/trainer_state.json +739 -0
  6. runs/t0/t0_S3_lr0.0003/train.log +0 -0
  7. runs/t0/t0_S3_lr0.0006/checkpoint-864/generation_config.json +11 -0
  8. runs/t0/t0_S3_lr0.0006/train.log +0 -0
  9. runs/t0/t0_S3_lr0.0012/checkpoint-864/chat_template.jinja +85 -0
  10. runs/t0/t0_S3_lr0.0012/checkpoint-864/config.json +32 -0
  11. runs/t0/t0_S3_lr0.0012/checkpoint-864/generation_config.json +11 -0
  12. runs/t0/t0_S3_lr0.0012/checkpoint-864/tokenizer_config.json +19 -0
  13. runs/t0/t0_S3_lr0.0012/checkpoint-864/trainer_state.json +739 -0
  14. runs/t0/t0_S3_lr0.0012/train.log +0 -0
  15. runs/t0/t0_S4_lr0.0006/checkpoint-864/chat_template.jinja +85 -0
  16. runs/t0/t0_S4_lr0.0006/checkpoint-864/config.json +32 -0
  17. runs/t0/t0_S4_lr0.0006/checkpoint-864/generation_config.json +11 -0
  18. runs/t0/t0_S4_lr0.0006/checkpoint-864/tokenizer_config.json +19 -0
  19. runs/t0/t0_S4_lr0.0006/checkpoint-864/trainer_state.json +739 -0
  20. runs/t0/t0_S4_lr0.0006/train.log +0 -0
  21. runs/t0/t0_S4_lr0.0012/checkpoint-864/chat_template.jinja +85 -0
  22. runs/t0/t0_S4_lr0.0012/checkpoint-864/config.json +32 -0
  23. runs/t0/t0_S4_lr0.0012/checkpoint-864/generation_config.json +11 -0
  24. runs/t0/t0_S4_lr0.0012/checkpoint-864/tokenizer_config.json +19 -0
  25. runs/t0/t0_S4_lr0.0012/checkpoint-864/trainer_state.json +739 -0
  26. runs/t0/t0_S4_lr0.0012/train.log +0 -0
  27. runs/t0/t0_S5_lr0.0003/checkpoint-720/chat_template.jinja +85 -0
  28. runs/t0/t0_S5_lr0.0003/checkpoint-720/config.json +32 -0
  29. runs/t0/t0_S5_lr0.0003/checkpoint-720/generation_config.json +11 -0
  30. runs/t0/t0_S5_lr0.0003/checkpoint-720/tokenizer_config.json +19 -0
  31. runs/t0/t0_S5_lr0.0003/checkpoint-720/trainer_state.json +625 -0
  32. runs/t0/t0_S5_lr0.0003/train.log +0 -0
  33. runs/t0/t0_S5_lr0.0012/checkpoint-720/chat_template.jinja +85 -0
  34. runs/t0/t0_S5_lr0.0012/checkpoint-720/config.json +32 -0
  35. runs/t0/t0_S5_lr0.0012/checkpoint-720/generation_config.json +11 -0
  36. runs/t0/t0_S5_lr0.0012/checkpoint-720/tokenizer_config.json +19 -0
  37. runs/t0/t0_S5_lr0.0012/checkpoint-720/trainer_state.json +625 -0
  38. runs/t0/t0_S5_lr0.0012/train.log +0 -0
  39. runs/t2a/t2a_150M_C0_baseline/checkpoint-2232/chat_template.jinja +85 -0
  40. runs/t2a/t2a_150M_C0_baseline/checkpoint-2232/config.json +32 -0
  41. runs/t2a/t2a_150M_C0_baseline/checkpoint-2232/generation_config.json +11 -0
  42. runs/t2a/t2a_150M_C0_baseline/checkpoint-2232/tokenizer_config.json +19 -0
  43. runs/t2a/t2a_150M_C0_baseline/checkpoint-2232/trainer_state.json +1850 -0
  44. runs/t2a/t2a_150M_C0_baseline/train.log +0 -0
  45. runs/t2a/t2a_19M_C0_baseline/checkpoint-576/chat_template.jinja +85 -0
  46. runs/t2a/t2a_19M_C0_baseline/checkpoint-576/config.json +32 -0
  47. runs/t2a/t2a_19M_C0_baseline/checkpoint-576/generation_config.json +11 -0
  48. runs/t2a/t2a_19M_C0_baseline/checkpoint-576/tokenizer_config.json +19 -0
  49. runs/t2a/t2a_19M_C0_baseline/checkpoint-576/trainer_state.json +504 -0
  50. runs/t2a/t2a_19M_C0_baseline/train.log +0 -0
runs/t0/t0_S3_lr0.0003/checkpoint-864/chat_template.jinja ADDED
@@ -0,0 +1,85 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {%- if tools %}
2
+ {{- '<|im_start|>system\n' }}
3
+ {%- if messages[0].role == 'system' %}
4
+ {{- messages[0].content + '\n\n' }}
5
+ {%- endif %}
6
+ {{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
7
+ {%- for tool in tools %}
8
+ {{- "\n" }}
9
+ {{- tool | tojson }}
10
+ {%- endfor %}
11
+ {{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
12
+ {%- else %}
13
+ {%- if messages[0].role == 'system' %}
14
+ {{- '<|im_start|>system\n' + messages[0].content + '<|im_end|>\n' }}
15
+ {%- endif %}
16
+ {%- endif %}
17
+ {%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
18
+ {%- for message in messages[::-1] %}
19
+ {%- set index = (messages|length - 1) - loop.index0 %}
20
+ {%- if ns.multi_step_tool and message.role == "user" and not(message.content.startswith('<tool_response>') and message.content.endswith('</tool_response>')) %}
21
+ {%- set ns.multi_step_tool = false %}
22
+ {%- set ns.last_query_index = index %}
23
+ {%- endif %}
24
+ {%- endfor %}
25
+ {%- for message in messages %}
26
+ {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
27
+ {{- '<|im_start|>' + message.role + '\n' + message.content + '<|im_end|>' + '\n' }}
28
+ {%- elif message.role == "assistant" %}
29
+ {%- set content = message.content %}
30
+ {%- set reasoning_content = '' %}
31
+ {%- if message.reasoning_content is defined and message.reasoning_content is not none %}
32
+ {%- set reasoning_content = message.reasoning_content %}
33
+ {%- else %}
34
+ {%- if '</think>' in message.content %}
35
+ {%- set content = message.content.split('</think>')[-1].lstrip('\n') %}
36
+ {%- set reasoning_content = message.content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
37
+ {%- endif %}
38
+ {%- endif %}
39
+ {%- if loop.index0 > ns.last_query_index %}
40
+ {%- if loop.last or (not loop.last and reasoning_content) %}
41
+ {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
42
+ {%- else %}
43
+ {{- '<|im_start|>' + message.role + '\n' + content }}
44
+ {%- endif %}
45
+ {%- else %}
46
+ {{- '<|im_start|>' + message.role + '\n' + content }}
47
+ {%- endif %}
48
+ {%- if message.tool_calls %}
49
+ {%- for tool_call in message.tool_calls %}
50
+ {%- if (loop.first and content) or (not loop.first) %}
51
+ {{- '\n' }}
52
+ {%- endif %}
53
+ {%- if tool_call.function %}
54
+ {%- set tool_call = tool_call.function %}
55
+ {%- endif %}
56
+ {{- '<tool_call>\n{"name": "' }}
57
+ {{- tool_call.name }}
58
+ {{- '", "arguments": ' }}
59
+ {%- if tool_call.arguments is string %}
60
+ {{- tool_call.arguments }}
61
+ {%- else %}
62
+ {{- tool_call.arguments | tojson }}
63
+ {%- endif %}
64
+ {{- '}\n</tool_call>' }}
65
+ {%- endfor %}
66
+ {%- endif %}
67
+ {{- '<|im_end|>\n' }}
68
+ {%- elif message.role == "tool" %}
69
+ {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
70
+ {{- '<|im_start|>user' }}
71
+ {%- endif %}
72
+ {{- '\n<tool_response>\n' }}
73
+ {{- message.content }}
74
+ {{- '\n</tool_response>' }}
75
+ {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
76
+ {{- '<|im_end|>\n' }}
77
+ {%- endif %}
78
+ {%- endif %}
79
+ {%- endfor %}
80
+ {%- if add_generation_prompt %}
81
+ {{- '<|im_start|>assistant\n' }}
82
+ {%- if enable_thinking is defined and enable_thinking is false %}
83
+ {{- '<think>\n\n</think>\n\n' }}
84
+ {%- endif %}
85
+ {%- endif %}
runs/t0/t0_S3_lr0.0003/checkpoint-864/config.json ADDED
@@ -0,0 +1,32 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "LlamaForCausalLM"
4
+ ],
5
+ "attention_bias": false,
6
+ "attention_dropout": 0.0,
7
+ "bos_token_id": null,
8
+ "dtype": "float32",
9
+ "eos_token_id": 151645,
10
+ "head_dim": 128,
11
+ "hidden_act": "silu",
12
+ "hidden_size": 768,
13
+ "initializer_range": 0.02,
14
+ "intermediate_size": 2048,
15
+ "max_position_embeddings": 2048,
16
+ "mlp_bias": false,
17
+ "model_type": "llama",
18
+ "num_attention_heads": 6,
19
+ "num_hidden_layers": 20,
20
+ "num_key_value_heads": 6,
21
+ "pad_token_id": 151645,
22
+ "pretraining_tp": 1,
23
+ "rms_norm_eps": 1e-06,
24
+ "rope_parameters": {
25
+ "rope_theta": 10000.0,
26
+ "rope_type": "default"
27
+ },
28
+ "tie_word_embeddings": true,
29
+ "transformers_version": "5.15.0",
30
+ "use_cache": false,
31
+ "vocab_size": 151671
32
+ }
runs/t0/t0_S3_lr0.0003/checkpoint-864/generation_config.json ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "eos_token_id": [
4
+ 151645
5
+ ],
6
+ "output_attentions": false,
7
+ "output_hidden_states": false,
8
+ "pad_token_id": 151645,
9
+ "transformers_version": "5.15.0",
10
+ "use_cache": true
11
+ }
runs/t0/t0_S3_lr0.0003/checkpoint-864/tokenizer_config.json ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "backend": "tokenizers",
4
+ "bos_token": null,
5
+ "clean_up_tokenization_spaces": false,
6
+ "eos_token": "<|im_end|>",
7
+ "errors": "replace",
8
+ "extra_special_tokens": [
9
+ "<|l2r_pred|>",
10
+ "<|r2l_pred|>"
11
+ ],
12
+ "is_local": false,
13
+ "local_files_only": false,
14
+ "model_max_length": 131072,
15
+ "pad_token": "<|im_end|>",
16
+ "split_special_tokens": false,
17
+ "tokenizer_class": "Qwen2Tokenizer",
18
+ "unk_token": null
19
+ }
runs/t0/t0_S3_lr0.0003/checkpoint-864/trainer_state.json ADDED
@@ -0,0 +1,739 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 12.0,
6
+ "eval_steps": 72,
7
+ "global_step": 864,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "epoch": 0.013961605584642234,
14
+ "grad_norm": 2.5511720180511475,
15
+ "learning_rate": 0.0,
16
+ "loss": 12.086503982543945,
17
+ "step": 1
18
+ },
19
+ {
20
+ "epoch": 0.13961605584642234,
21
+ "grad_norm": 2.48952317237854,
22
+ "learning_rate": 2.6999999999999996e-05,
23
+ "loss": 11.761650933159721,
24
+ "step": 10
25
+ },
26
+ {
27
+ "epoch": 0.2792321116928447,
28
+ "grad_norm": 1.994715690612793,
29
+ "learning_rate": 5.6999999999999996e-05,
30
+ "loss": 10.831330871582031,
31
+ "step": 20
32
+ },
33
+ {
34
+ "epoch": 0.418848167539267,
35
+ "grad_norm": 1.869005560874939,
36
+ "learning_rate": 8.699999999999999e-05,
37
+ "loss": 10.210093688964843,
38
+ "step": 30
39
+ },
40
+ {
41
+ "epoch": 0.5584642233856894,
42
+ "grad_norm": 1.5961251258850098,
43
+ "learning_rate": 0.000117,
44
+ "loss": 9.370370483398437,
45
+ "step": 40
46
+ },
47
+ {
48
+ "epoch": 0.6980802792321117,
49
+ "grad_norm": 1.033909797668457,
50
+ "learning_rate": 0.000147,
51
+ "loss": 8.43478012084961,
52
+ "step": 50
53
+ },
54
+ {
55
+ "epoch": 0.837696335078534,
56
+ "grad_norm": 0.7044252157211304,
57
+ "learning_rate": 0.00017699999999999997,
58
+ "loss": 7.778045654296875,
59
+ "step": 60
60
+ },
61
+ {
62
+ "epoch": 0.9773123909249564,
63
+ "grad_norm": 0.4867978096008301,
64
+ "learning_rate": 0.00020699999999999996,
65
+ "loss": 7.52720947265625,
66
+ "step": 70
67
+ },
68
+ {
69
+ "epoch": 1.0,
70
+ "eval_loss": 7.445277214050293,
71
+ "eval_runtime": 28.509,
72
+ "eval_samples_per_second": 85.622,
73
+ "eval_steps_per_second": 1.368,
74
+ "step": 72
75
+ },
76
+ {
77
+ "epoch": 1.111692844677138,
78
+ "grad_norm": 0.6972123384475708,
79
+ "learning_rate": 0.000237,
80
+ "loss": 7.379924011230469,
81
+ "step": 80
82
+ },
83
+ {
84
+ "epoch": 1.2513089005235603,
85
+ "grad_norm": 0.6060797572135925,
86
+ "learning_rate": 0.000267,
87
+ "loss": 7.171945190429687,
88
+ "step": 90
89
+ },
90
+ {
91
+ "epoch": 1.3909249563699826,
92
+ "grad_norm": 0.8712506890296936,
93
+ "learning_rate": 0.00029699999999999996,
94
+ "loss": 7.012107849121094,
95
+ "step": 100
96
+ },
97
+ {
98
+ "epoch": 1.530541012216405,
99
+ "grad_norm": 0.5367537140846252,
100
+ "learning_rate": 0.0003,
101
+ "loss": 6.8590141296386715,
102
+ "step": 110
103
+ },
104
+ {
105
+ "epoch": 1.6701570680628273,
106
+ "grad_norm": 0.5023939609527588,
107
+ "learning_rate": 0.0003,
108
+ "loss": 6.714488220214844,
109
+ "step": 120
110
+ },
111
+ {
112
+ "epoch": 1.8097731239092496,
113
+ "grad_norm": 0.7802119255065918,
114
+ "learning_rate": 0.0003,
115
+ "loss": 6.580387878417969,
116
+ "step": 130
117
+ },
118
+ {
119
+ "epoch": 1.949389179755672,
120
+ "grad_norm": 1.0640617609024048,
121
+ "learning_rate": 0.0003,
122
+ "loss": 6.485602569580078,
123
+ "step": 140
124
+ },
125
+ {
126
+ "epoch": 2.0,
127
+ "eval_loss": 6.399908542633057,
128
+ "eval_runtime": 4.1035,
129
+ "eval_samples_per_second": 594.865,
130
+ "eval_steps_per_second": 9.504,
131
+ "step": 144
132
+ },
133
+ {
134
+ "epoch": 2.0837696335078535,
135
+ "grad_norm": 0.6536300778388977,
136
+ "learning_rate": 0.0003,
137
+ "loss": 6.3652080535888675,
138
+ "step": 150
139
+ },
140
+ {
141
+ "epoch": 2.223385689354276,
142
+ "grad_norm": 0.861472487449646,
143
+ "learning_rate": 0.0003,
144
+ "loss": 6.283423233032226,
145
+ "step": 160
146
+ },
147
+ {
148
+ "epoch": 2.363001745200698,
149
+ "grad_norm": 0.48216626048088074,
150
+ "learning_rate": 0.0003,
151
+ "loss": 6.193089294433594,
152
+ "step": 170
153
+ },
154
+ {
155
+ "epoch": 2.5026178010471205,
156
+ "grad_norm": 1.4223753213882446,
157
+ "learning_rate": 0.0003,
158
+ "loss": 6.143649291992188,
159
+ "step": 180
160
+ },
161
+ {
162
+ "epoch": 2.642233856893543,
163
+ "grad_norm": 0.5395786762237549,
164
+ "learning_rate": 0.0003,
165
+ "loss": 6.06384506225586,
166
+ "step": 190
167
+ },
168
+ {
169
+ "epoch": 2.781849912739965,
170
+ "grad_norm": 1.3055977821350098,
171
+ "learning_rate": 0.0003,
172
+ "loss": 5.971584320068359,
173
+ "step": 200
174
+ },
175
+ {
176
+ "epoch": 2.9214659685863875,
177
+ "grad_norm": 0.7041425704956055,
178
+ "learning_rate": 0.0003,
179
+ "loss": 5.910428619384765,
180
+ "step": 210
181
+ },
182
+ {
183
+ "epoch": 3.0,
184
+ "eval_loss": 5.867927551269531,
185
+ "eval_runtime": 4.1161,
186
+ "eval_samples_per_second": 593.031,
187
+ "eval_steps_per_second": 9.475,
188
+ "step": 216
189
+ },
190
+ {
191
+ "epoch": 3.055846422338569,
192
+ "grad_norm": 1.0085630416870117,
193
+ "learning_rate": 0.0003,
194
+ "loss": 5.828436279296875,
195
+ "step": 220
196
+ },
197
+ {
198
+ "epoch": 3.195462478184991,
199
+ "grad_norm": 0.7435064315795898,
200
+ "learning_rate": 0.0003,
201
+ "loss": 5.782387161254883,
202
+ "step": 230
203
+ },
204
+ {
205
+ "epoch": 3.3350785340314135,
206
+ "grad_norm": 0.4216879606246948,
207
+ "learning_rate": 0.0003,
208
+ "loss": 5.703854751586914,
209
+ "step": 240
210
+ },
211
+ {
212
+ "epoch": 3.474694589877836,
213
+ "grad_norm": 0.734950065612793,
214
+ "learning_rate": 0.0003,
215
+ "loss": 5.653765487670898,
216
+ "step": 250
217
+ },
218
+ {
219
+ "epoch": 3.614310645724258,
220
+ "grad_norm": 0.7231405973434448,
221
+ "learning_rate": 0.0003,
222
+ "loss": 5.640114593505859,
223
+ "step": 260
224
+ },
225
+ {
226
+ "epoch": 3.7539267015706805,
227
+ "grad_norm": 0.43395647406578064,
228
+ "learning_rate": 0.0003,
229
+ "loss": 5.542626953125,
230
+ "step": 270
231
+ },
232
+ {
233
+ "epoch": 3.893542757417103,
234
+ "grad_norm": 0.6494223475456238,
235
+ "learning_rate": 0.0003,
236
+ "loss": 5.490487670898437,
237
+ "step": 280
238
+ },
239
+ {
240
+ "epoch": 4.0,
241
+ "eval_loss": 5.469155788421631,
242
+ "eval_runtime": 4.1041,
243
+ "eval_samples_per_second": 594.77,
244
+ "eval_steps_per_second": 9.503,
245
+ "step": 288
246
+ },
247
+ {
248
+ "epoch": 4.027923211169284,
249
+ "grad_norm": 0.40678805112838745,
250
+ "learning_rate": 0.0003,
251
+ "loss": 5.421612548828125,
252
+ "step": 290
253
+ },
254
+ {
255
+ "epoch": 4.167539267015707,
256
+ "grad_norm": 0.948244035243988,
257
+ "learning_rate": 0.0003,
258
+ "loss": 5.366499328613282,
259
+ "step": 300
260
+ },
261
+ {
262
+ "epoch": 4.307155322862129,
263
+ "grad_norm": 0.7340747714042664,
264
+ "learning_rate": 0.0003,
265
+ "loss": 5.337801742553711,
266
+ "step": 310
267
+ },
268
+ {
269
+ "epoch": 4.446771378708552,
270
+ "grad_norm": 0.5626649856567383,
271
+ "learning_rate": 0.0003,
272
+ "loss": 5.294442749023437,
273
+ "step": 320
274
+ },
275
+ {
276
+ "epoch": 4.5863874345549736,
277
+ "grad_norm": 0.6366444230079651,
278
+ "learning_rate": 0.0003,
279
+ "loss": 5.244319915771484,
280
+ "step": 330
281
+ },
282
+ {
283
+ "epoch": 4.726003490401396,
284
+ "grad_norm": 0.48262909054756165,
285
+ "learning_rate": 0.0003,
286
+ "loss": 5.197024154663086,
287
+ "step": 340
288
+ },
289
+ {
290
+ "epoch": 4.865619546247818,
291
+ "grad_norm": 0.5314050912857056,
292
+ "learning_rate": 0.0003,
293
+ "loss": 5.135079193115234,
294
+ "step": 350
295
+ },
296
+ {
297
+ "epoch": 5.0,
298
+ "grad_norm": 1.139448881149292,
299
+ "learning_rate": 0.0003,
300
+ "loss": 5.122449493408203,
301
+ "step": 360
302
+ },
303
+ {
304
+ "epoch": 5.0,
305
+ "eval_loss": 5.215811252593994,
306
+ "eval_runtime": 4.0976,
307
+ "eval_samples_per_second": 595.708,
308
+ "eval_steps_per_second": 9.518,
309
+ "step": 360
310
+ },
311
+ {
312
+ "epoch": 5.139616055846423,
313
+ "grad_norm": 1.0207905769348145,
314
+ "learning_rate": 0.0003,
315
+ "loss": 5.087126922607422,
316
+ "step": 370
317
+ },
318
+ {
319
+ "epoch": 5.279232111692845,
320
+ "grad_norm": 0.45449236035346985,
321
+ "learning_rate": 0.0003,
322
+ "loss": 5.048321914672852,
323
+ "step": 380
324
+ },
325
+ {
326
+ "epoch": 5.418848167539267,
327
+ "grad_norm": 0.411428838968277,
328
+ "learning_rate": 0.0003,
329
+ "loss": 4.9702400207519535,
330
+ "step": 390
331
+ },
332
+ {
333
+ "epoch": 5.558464223385689,
334
+ "grad_norm": 0.77912837266922,
335
+ "learning_rate": 0.0003,
336
+ "loss": 4.928448486328125,
337
+ "step": 400
338
+ },
339
+ {
340
+ "epoch": 5.698080279232112,
341
+ "grad_norm": 0.6806897521018982,
342
+ "learning_rate": 0.0003,
343
+ "loss": 4.942894744873047,
344
+ "step": 410
345
+ },
346
+ {
347
+ "epoch": 5.837696335078534,
348
+ "grad_norm": 0.6517318487167358,
349
+ "learning_rate": 0.0003,
350
+ "loss": 4.901599884033203,
351
+ "step": 420
352
+ },
353
+ {
354
+ "epoch": 5.977312390924956,
355
+ "grad_norm": 0.33704403042793274,
356
+ "learning_rate": 0.0003,
357
+ "loss": 4.843351364135742,
358
+ "step": 430
359
+ },
360
+ {
361
+ "epoch": 6.0,
362
+ "eval_loss": 4.90324068069458,
363
+ "eval_runtime": 4.1113,
364
+ "eval_samples_per_second": 593.724,
365
+ "eval_steps_per_second": 9.486,
366
+ "step": 432
367
+ },
368
+ {
369
+ "epoch": 6.111692844677138,
370
+ "grad_norm": 0.4563899040222168,
371
+ "learning_rate": 0.0003,
372
+ "loss": 4.772877883911133,
373
+ "step": 440
374
+ },
375
+ {
376
+ "epoch": 6.2513089005235605,
377
+ "grad_norm": 0.8754525780677795,
378
+ "learning_rate": 0.0003,
379
+ "loss": 4.744700241088867,
380
+ "step": 450
381
+ },
382
+ {
383
+ "epoch": 6.390924956369982,
384
+ "grad_norm": 0.958769679069519,
385
+ "learning_rate": 0.0003,
386
+ "loss": 4.739905548095703,
387
+ "step": 460
388
+ },
389
+ {
390
+ "epoch": 6.530541012216405,
391
+ "grad_norm": 0.4947556257247925,
392
+ "learning_rate": 0.0003,
393
+ "loss": 4.723302841186523,
394
+ "step": 470
395
+ },
396
+ {
397
+ "epoch": 6.670157068062827,
398
+ "grad_norm": 0.32547932863235474,
399
+ "learning_rate": 0.0003,
400
+ "loss": 4.660966873168945,
401
+ "step": 480
402
+ },
403
+ {
404
+ "epoch": 6.80977312390925,
405
+ "grad_norm": 0.3814781606197357,
406
+ "learning_rate": 0.0003,
407
+ "loss": 4.627614593505859,
408
+ "step": 490
409
+ },
410
+ {
411
+ "epoch": 6.949389179755672,
412
+ "grad_norm": 0.8448502421379089,
413
+ "learning_rate": 0.0003,
414
+ "loss": 4.592163848876953,
415
+ "step": 500
416
+ },
417
+ {
418
+ "epoch": 7.0,
419
+ "eval_loss": 4.711507797241211,
420
+ "eval_runtime": 4.1149,
421
+ "eval_samples_per_second": 593.208,
422
+ "eval_steps_per_second": 9.478,
423
+ "step": 504
424
+ },
425
+ {
426
+ "epoch": 7.0837696335078535,
427
+ "grad_norm": 0.7609332203865051,
428
+ "learning_rate": 0.0003,
429
+ "loss": 4.566059875488281,
430
+ "step": 510
431
+ },
432
+ {
433
+ "epoch": 7.223385689354275,
434
+ "grad_norm": 0.6690367460250854,
435
+ "learning_rate": 0.0003,
436
+ "loss": 4.540019989013672,
437
+ "step": 520
438
+ },
439
+ {
440
+ "epoch": 7.363001745200698,
441
+ "grad_norm": 0.7389764785766602,
442
+ "learning_rate": 0.0003,
443
+ "loss": 4.510246276855469,
444
+ "step": 530
445
+ },
446
+ {
447
+ "epoch": 7.50261780104712,
448
+ "grad_norm": 0.3664371371269226,
449
+ "learning_rate": 0.0003,
450
+ "loss": 4.489366149902343,
451
+ "step": 540
452
+ },
453
+ {
454
+ "epoch": 7.642233856893543,
455
+ "grad_norm": 0.35561150312423706,
456
+ "learning_rate": 0.0003,
457
+ "loss": 4.4513092041015625,
458
+ "step": 550
459
+ },
460
+ {
461
+ "epoch": 7.781849912739965,
462
+ "grad_norm": 0.3935790956020355,
463
+ "learning_rate": 0.0003,
464
+ "loss": 4.422682189941407,
465
+ "step": 560
466
+ },
467
+ {
468
+ "epoch": 7.9214659685863875,
469
+ "grad_norm": 0.46314138174057007,
470
+ "learning_rate": 0.0003,
471
+ "loss": 4.4081573486328125,
472
+ "step": 570
473
+ },
474
+ {
475
+ "epoch": 8.0,
476
+ "eval_loss": 4.537301540374756,
477
+ "eval_runtime": 4.0993,
478
+ "eval_samples_per_second": 595.464,
479
+ "eval_steps_per_second": 9.514,
480
+ "step": 576
481
+ },
482
+ {
483
+ "epoch": 8.055846422338568,
484
+ "grad_norm": 0.7809078097343445,
485
+ "learning_rate": 0.0003,
486
+ "loss": 4.366108703613281,
487
+ "step": 580
488
+ },
489
+ {
490
+ "epoch": 8.195462478184991,
491
+ "grad_norm": 0.8947575092315674,
492
+ "learning_rate": 0.0003,
493
+ "loss": 4.3550163269042965,
494
+ "step": 590
495
+ },
496
+ {
497
+ "epoch": 8.335078534031414,
498
+ "grad_norm": 0.7821335196495056,
499
+ "learning_rate": 0.0003,
500
+ "loss": 4.352532196044922,
501
+ "step": 600
502
+ },
503
+ {
504
+ "epoch": 8.474694589877837,
505
+ "grad_norm": 0.37841781973838806,
506
+ "learning_rate": 0.0003,
507
+ "loss": 4.314500427246093,
508
+ "step": 610
509
+ },
510
+ {
511
+ "epoch": 8.614310645724258,
512
+ "grad_norm": 0.35384002327919006,
513
+ "learning_rate": 0.0003,
514
+ "loss": 4.2818855285644535,
515
+ "step": 620
516
+ },
517
+ {
518
+ "epoch": 8.75392670157068,
519
+ "grad_norm": 0.32511261105537415,
520
+ "learning_rate": 0.0003,
521
+ "loss": 4.258856964111328,
522
+ "step": 630
523
+ },
524
+ {
525
+ "epoch": 8.893542757417103,
526
+ "grad_norm": 0.5073819756507874,
527
+ "learning_rate": 0.0003,
528
+ "loss": 4.258666610717773,
529
+ "step": 640
530
+ },
531
+ {
532
+ "epoch": 9.0,
533
+ "eval_loss": 4.416695594787598,
534
+ "eval_runtime": 4.1007,
535
+ "eval_samples_per_second": 595.27,
536
+ "eval_steps_per_second": 9.511,
537
+ "step": 648
538
+ },
539
+ {
540
+ "epoch": 9.027923211169284,
541
+ "grad_norm": 0.5689071416854858,
542
+ "learning_rate": 0.0003,
543
+ "loss": 4.231290817260742,
544
+ "step": 650
545
+ },
546
+ {
547
+ "epoch": 9.167539267015707,
548
+ "grad_norm": 0.9789859056472778,
549
+ "learning_rate": 0.0003,
550
+ "loss": 4.191323852539062,
551
+ "step": 660
552
+ },
553
+ {
554
+ "epoch": 9.30715532286213,
555
+ "grad_norm": 0.8916944265365601,
556
+ "learning_rate": 0.0003,
557
+ "loss": 4.21032943725586,
558
+ "step": 670
559
+ },
560
+ {
561
+ "epoch": 9.44677137870855,
562
+ "grad_norm": 0.5173555612564087,
563
+ "learning_rate": 0.0003,
564
+ "loss": 4.184156036376953,
565
+ "step": 680
566
+ },
567
+ {
568
+ "epoch": 9.586387434554974,
569
+ "grad_norm": 0.34253525733947754,
570
+ "learning_rate": 0.0003,
571
+ "loss": 4.148382568359375,
572
+ "step": 690
573
+ },
574
+ {
575
+ "epoch": 9.726003490401396,
576
+ "grad_norm": 0.42179620265960693,
577
+ "learning_rate": 0.0003,
578
+ "loss": 4.139811706542969,
579
+ "step": 700
580
+ },
581
+ {
582
+ "epoch": 9.86561954624782,
583
+ "grad_norm": 0.504753053188324,
584
+ "learning_rate": 0.0003,
585
+ "loss": 4.118508529663086,
586
+ "step": 710
587
+ },
588
+ {
589
+ "epoch": 10.0,
590
+ "grad_norm": 0.5721378922462463,
591
+ "learning_rate": 0.0003,
592
+ "loss": 4.104060363769531,
593
+ "step": 720
594
+ },
595
+ {
596
+ "epoch": 10.0,
597
+ "eval_loss": 4.33306360244751,
598
+ "eval_runtime": 4.1147,
599
+ "eval_samples_per_second": 593.236,
600
+ "eval_steps_per_second": 9.478,
601
+ "step": 720
602
+ },
603
+ {
604
+ "epoch": 10.139616055846423,
605
+ "grad_norm": 0.841262698173523,
606
+ "learning_rate": 0.0003,
607
+ "loss": 4.05061149597168,
608
+ "step": 730
609
+ },
610
+ {
611
+ "epoch": 10.279232111692846,
612
+ "grad_norm": 0.8539153933525085,
613
+ "learning_rate": 0.0003,
614
+ "loss": 4.062906265258789,
615
+ "step": 740
616
+ },
617
+ {
618
+ "epoch": 10.418848167539267,
619
+ "grad_norm": 0.5237502455711365,
620
+ "learning_rate": 0.0003,
621
+ "loss": 4.045460510253906,
622
+ "step": 750
623
+ },
624
+ {
625
+ "epoch": 10.55846422338569,
626
+ "grad_norm": 0.38326072692871094,
627
+ "learning_rate": 0.0003,
628
+ "loss": 4.043621444702149,
629
+ "step": 760
630
+ },
631
+ {
632
+ "epoch": 10.698080279232112,
633
+ "grad_norm": 0.4053055942058563,
634
+ "learning_rate": 0.0003,
635
+ "loss": 4.020166015625,
636
+ "step": 770
637
+ },
638
+ {
639
+ "epoch": 10.837696335078533,
640
+ "grad_norm": 0.46810540556907654,
641
+ "learning_rate": 0.0003,
642
+ "loss": 4.012229919433594,
643
+ "step": 780
644
+ },
645
+ {
646
+ "epoch": 10.977312390924956,
647
+ "grad_norm": 0.45282575488090515,
648
+ "learning_rate": 0.0003,
649
+ "loss": 3.9989597320556642,
650
+ "step": 790
651
+ },
652
+ {
653
+ "epoch": 11.0,
654
+ "eval_loss": 4.255781650543213,
655
+ "eval_runtime": 4.0982,
656
+ "eval_samples_per_second": 595.63,
657
+ "eval_steps_per_second": 9.516,
658
+ "step": 792
659
+ },
660
+ {
661
+ "epoch": 11.111692844677139,
662
+ "grad_norm": 0.812175989151001,
663
+ "learning_rate": 0.0003,
664
+ "loss": 3.9566085815429686,
665
+ "step": 800
666
+ },
667
+ {
668
+ "epoch": 11.25130890052356,
669
+ "grad_norm": 0.7589244842529297,
670
+ "learning_rate": 0.0003,
671
+ "loss": 3.9253917694091798,
672
+ "step": 810
673
+ },
674
+ {
675
+ "epoch": 11.390924956369982,
676
+ "grad_norm": 0.6504644751548767,
677
+ "learning_rate": 0.0003,
678
+ "loss": 3.927887725830078,
679
+ "step": 820
680
+ },
681
+ {
682
+ "epoch": 11.530541012216405,
683
+ "grad_norm": 0.47677385807037354,
684
+ "learning_rate": 0.0003,
685
+ "loss": 3.9283016204833983,
686
+ "step": 830
687
+ },
688
+ {
689
+ "epoch": 11.670157068062828,
690
+ "grad_norm": 0.4261974096298218,
691
+ "learning_rate": 0.0003,
692
+ "loss": 3.9170894622802734,
693
+ "step": 840
694
+ },
695
+ {
696
+ "epoch": 11.809773123909249,
697
+ "grad_norm": 0.437930166721344,
698
+ "learning_rate": 0.0003,
699
+ "loss": 3.906512451171875,
700
+ "step": 850
701
+ },
702
+ {
703
+ "epoch": 11.949389179755672,
704
+ "grad_norm": 0.3658443093299866,
705
+ "learning_rate": 0.0003,
706
+ "loss": 3.906256103515625,
707
+ "step": 860
708
+ },
709
+ {
710
+ "epoch": 12.0,
711
+ "eval_loss": 4.190188407897949,
712
+ "eval_runtime": 4.0969,
713
+ "eval_samples_per_second": 595.819,
714
+ "eval_steps_per_second": 9.519,
715
+ "step": 864
716
+ }
717
+ ],
718
+ "logging_steps": 10,
719
+ "max_steps": 864,
720
+ "num_input_tokens_seen": 0,
721
+ "num_train_epochs": 12,
722
+ "save_steps": 500,
723
+ "stateful_callbacks": {
724
+ "TrainerControl": {
725
+ "args": {
726
+ "should_epoch_stop": false,
727
+ "should_evaluate": false,
728
+ "should_log": false,
729
+ "should_save": false,
730
+ "should_training_stop": true
731
+ },
732
+ "attributes": {}
733
+ }
734
+ },
735
+ "total_flos": 7.656447700673495e+17,
736
+ "train_batch_size": 8,
737
+ "trial_name": null,
738
+ "trial_params": null
739
+ }
runs/t0/t0_S3_lr0.0003/train.log ADDED
The diff for this file is too large to render. See raw diff
 
runs/t0/t0_S3_lr0.0006/checkpoint-864/generation_config.json ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "eos_token_id": [
4
+ 151645
5
+ ],
6
+ "output_attentions": false,
7
+ "output_hidden_states": false,
8
+ "pad_token_id": 151645,
9
+ "transformers_version": "5.15.0",
10
+ "use_cache": true
11
+ }
runs/t0/t0_S3_lr0.0006/train.log ADDED
The diff for this file is too large to render. See raw diff
 
runs/t0/t0_S3_lr0.0012/checkpoint-864/chat_template.jinja ADDED
@@ -0,0 +1,85 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {%- if tools %}
2
+ {{- '<|im_start|>system\n' }}
3
+ {%- if messages[0].role == 'system' %}
4
+ {{- messages[0].content + '\n\n' }}
5
+ {%- endif %}
6
+ {{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
7
+ {%- for tool in tools %}
8
+ {{- "\n" }}
9
+ {{- tool | tojson }}
10
+ {%- endfor %}
11
+ {{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
12
+ {%- else %}
13
+ {%- if messages[0].role == 'system' %}
14
+ {{- '<|im_start|>system\n' + messages[0].content + '<|im_end|>\n' }}
15
+ {%- endif %}
16
+ {%- endif %}
17
+ {%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
18
+ {%- for message in messages[::-1] %}
19
+ {%- set index = (messages|length - 1) - loop.index0 %}
20
+ {%- if ns.multi_step_tool and message.role == "user" and not(message.content.startswith('<tool_response>') and message.content.endswith('</tool_response>')) %}
21
+ {%- set ns.multi_step_tool = false %}
22
+ {%- set ns.last_query_index = index %}
23
+ {%- endif %}
24
+ {%- endfor %}
25
+ {%- for message in messages %}
26
+ {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
27
+ {{- '<|im_start|>' + message.role + '\n' + message.content + '<|im_end|>' + '\n' }}
28
+ {%- elif message.role == "assistant" %}
29
+ {%- set content = message.content %}
30
+ {%- set reasoning_content = '' %}
31
+ {%- if message.reasoning_content is defined and message.reasoning_content is not none %}
32
+ {%- set reasoning_content = message.reasoning_content %}
33
+ {%- else %}
34
+ {%- if '</think>' in message.content %}
35
+ {%- set content = message.content.split('</think>')[-1].lstrip('\n') %}
36
+ {%- set reasoning_content = message.content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
37
+ {%- endif %}
38
+ {%- endif %}
39
+ {%- if loop.index0 > ns.last_query_index %}
40
+ {%- if loop.last or (not loop.last and reasoning_content) %}
41
+ {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
42
+ {%- else %}
43
+ {{- '<|im_start|>' + message.role + '\n' + content }}
44
+ {%- endif %}
45
+ {%- else %}
46
+ {{- '<|im_start|>' + message.role + '\n' + content }}
47
+ {%- endif %}
48
+ {%- if message.tool_calls %}
49
+ {%- for tool_call in message.tool_calls %}
50
+ {%- if (loop.first and content) or (not loop.first) %}
51
+ {{- '\n' }}
52
+ {%- endif %}
53
+ {%- if tool_call.function %}
54
+ {%- set tool_call = tool_call.function %}
55
+ {%- endif %}
56
+ {{- '<tool_call>\n{"name": "' }}
57
+ {{- tool_call.name }}
58
+ {{- '", "arguments": ' }}
59
+ {%- if tool_call.arguments is string %}
60
+ {{- tool_call.arguments }}
61
+ {%- else %}
62
+ {{- tool_call.arguments | tojson }}
63
+ {%- endif %}
64
+ {{- '}\n</tool_call>' }}
65
+ {%- endfor %}
66
+ {%- endif %}
67
+ {{- '<|im_end|>\n' }}
68
+ {%- elif message.role == "tool" %}
69
+ {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
70
+ {{- '<|im_start|>user' }}
71
+ {%- endif %}
72
+ {{- '\n<tool_response>\n' }}
73
+ {{- message.content }}
74
+ {{- '\n</tool_response>' }}
75
+ {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
76
+ {{- '<|im_end|>\n' }}
77
+ {%- endif %}
78
+ {%- endif %}
79
+ {%- endfor %}
80
+ {%- if add_generation_prompt %}
81
+ {{- '<|im_start|>assistant\n' }}
82
+ {%- if enable_thinking is defined and enable_thinking is false %}
83
+ {{- '<think>\n\n</think>\n\n' }}
84
+ {%- endif %}
85
+ {%- endif %}
runs/t0/t0_S3_lr0.0012/checkpoint-864/config.json ADDED
@@ -0,0 +1,32 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "LlamaForCausalLM"
4
+ ],
5
+ "attention_bias": false,
6
+ "attention_dropout": 0.0,
7
+ "bos_token_id": null,
8
+ "dtype": "float32",
9
+ "eos_token_id": 151645,
10
+ "head_dim": 128,
11
+ "hidden_act": "silu",
12
+ "hidden_size": 768,
13
+ "initializer_range": 0.02,
14
+ "intermediate_size": 2048,
15
+ "max_position_embeddings": 2048,
16
+ "mlp_bias": false,
17
+ "model_type": "llama",
18
+ "num_attention_heads": 6,
19
+ "num_hidden_layers": 20,
20
+ "num_key_value_heads": 6,
21
+ "pad_token_id": 151645,
22
+ "pretraining_tp": 1,
23
+ "rms_norm_eps": 1e-06,
24
+ "rope_parameters": {
25
+ "rope_theta": 10000.0,
26
+ "rope_type": "default"
27
+ },
28
+ "tie_word_embeddings": true,
29
+ "transformers_version": "5.15.0",
30
+ "use_cache": false,
31
+ "vocab_size": 151671
32
+ }
runs/t0/t0_S3_lr0.0012/checkpoint-864/generation_config.json ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "eos_token_id": [
4
+ 151645
5
+ ],
6
+ "output_attentions": false,
7
+ "output_hidden_states": false,
8
+ "pad_token_id": 151645,
9
+ "transformers_version": "5.15.0",
10
+ "use_cache": true
11
+ }
runs/t0/t0_S3_lr0.0012/checkpoint-864/tokenizer_config.json ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "backend": "tokenizers",
4
+ "bos_token": null,
5
+ "clean_up_tokenization_spaces": false,
6
+ "eos_token": "<|im_end|>",
7
+ "errors": "replace",
8
+ "extra_special_tokens": [
9
+ "<|l2r_pred|>",
10
+ "<|r2l_pred|>"
11
+ ],
12
+ "is_local": false,
13
+ "local_files_only": false,
14
+ "model_max_length": 131072,
15
+ "pad_token": "<|im_end|>",
16
+ "split_special_tokens": false,
17
+ "tokenizer_class": "Qwen2Tokenizer",
18
+ "unk_token": null
19
+ }
runs/t0/t0_S3_lr0.0012/checkpoint-864/trainer_state.json ADDED
@@ -0,0 +1,739 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 12.0,
6
+ "eval_steps": 72,
7
+ "global_step": 864,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "epoch": 0.013961605584642234,
14
+ "grad_norm": 2.625096082687378,
15
+ "learning_rate": 0.0,
16
+ "loss": 12.076139450073242,
17
+ "step": 1
18
+ },
19
+ {
20
+ "epoch": 0.13961605584642234,
21
+ "grad_norm": 1.9982500076293945,
22
+ "learning_rate": 0.00010799999999999998,
23
+ "loss": 11.33876207139757,
24
+ "step": 10
25
+ },
26
+ {
27
+ "epoch": 0.2792321116928447,
28
+ "grad_norm": 1.7737500667572021,
29
+ "learning_rate": 0.00022799999999999999,
30
+ "loss": 9.99845199584961,
31
+ "step": 20
32
+ },
33
+ {
34
+ "epoch": 0.418848167539267,
35
+ "grad_norm": 0.6299463510513306,
36
+ "learning_rate": 0.00034799999999999995,
37
+ "loss": 8.386837768554688,
38
+ "step": 30
39
+ },
40
+ {
41
+ "epoch": 0.5584642233856894,
42
+ "grad_norm": 0.4135228395462036,
43
+ "learning_rate": 0.000468,
44
+ "loss": 7.641131591796875,
45
+ "step": 40
46
+ },
47
+ {
48
+ "epoch": 0.6980802792321117,
49
+ "grad_norm": 0.6167853474617004,
50
+ "learning_rate": 0.000588,
51
+ "loss": 7.482666015625,
52
+ "step": 50
53
+ },
54
+ {
55
+ "epoch": 0.837696335078534,
56
+ "grad_norm": 0.7763964533805847,
57
+ "learning_rate": 0.0007079999999999999,
58
+ "loss": 7.347389984130859,
59
+ "step": 60
60
+ },
61
+ {
62
+ "epoch": 0.9773123909249564,
63
+ "grad_norm": 0.4353961646556854,
64
+ "learning_rate": 0.0008279999999999999,
65
+ "loss": 7.112811279296875,
66
+ "step": 70
67
+ },
68
+ {
69
+ "epoch": 1.0,
70
+ "eval_loss": 6.975900173187256,
71
+ "eval_runtime": 27.9985,
72
+ "eval_samples_per_second": 87.183,
73
+ "eval_steps_per_second": 1.393,
74
+ "step": 72
75
+ },
76
+ {
77
+ "epoch": 1.111692844677138,
78
+ "grad_norm": 0.6767817735671997,
79
+ "learning_rate": 0.000948,
80
+ "loss": 6.936407470703125,
81
+ "step": 80
82
+ },
83
+ {
84
+ "epoch": 1.2513089005235603,
85
+ "grad_norm": 0.5563651919364929,
86
+ "learning_rate": 0.001068,
87
+ "loss": 6.7925537109375,
88
+ "step": 90
89
+ },
90
+ {
91
+ "epoch": 1.3909249563699826,
92
+ "grad_norm": 0.9051397442817688,
93
+ "learning_rate": 0.0011879999999999998,
94
+ "loss": 6.682518768310547,
95
+ "step": 100
96
+ },
97
+ {
98
+ "epoch": 1.530541012216405,
99
+ "grad_norm": 0.3846557140350342,
100
+ "learning_rate": 0.0012,
101
+ "loss": 6.563172912597656,
102
+ "step": 110
103
+ },
104
+ {
105
+ "epoch": 1.6701570680628273,
106
+ "grad_norm": 0.413596510887146,
107
+ "learning_rate": 0.0012,
108
+ "loss": 6.416682434082031,
109
+ "step": 120
110
+ },
111
+ {
112
+ "epoch": 1.8097731239092496,
113
+ "grad_norm": 0.6242079138755798,
114
+ "learning_rate": 0.0012,
115
+ "loss": 6.320872497558594,
116
+ "step": 130
117
+ },
118
+ {
119
+ "epoch": 1.949389179755672,
120
+ "grad_norm": 0.4003734290599823,
121
+ "learning_rate": 0.0012,
122
+ "loss": 6.239190292358399,
123
+ "step": 140
124
+ },
125
+ {
126
+ "epoch": 2.0,
127
+ "eval_loss": 6.174596786499023,
128
+ "eval_runtime": 4.094,
129
+ "eval_samples_per_second": 596.236,
130
+ "eval_steps_per_second": 9.526,
131
+ "step": 144
132
+ },
133
+ {
134
+ "epoch": 2.0837696335078535,
135
+ "grad_norm": 0.3320383131504059,
136
+ "learning_rate": 0.0012,
137
+ "loss": 6.135129928588867,
138
+ "step": 150
139
+ },
140
+ {
141
+ "epoch": 2.223385689354276,
142
+ "grad_norm": 0.4379850924015045,
143
+ "learning_rate": 0.0012,
144
+ "loss": 6.063224029541016,
145
+ "step": 160
146
+ },
147
+ {
148
+ "epoch": 2.363001745200698,
149
+ "grad_norm": 0.4536810517311096,
150
+ "learning_rate": 0.0012,
151
+ "loss": 5.9813385009765625,
152
+ "step": 170
153
+ },
154
+ {
155
+ "epoch": 2.5026178010471205,
156
+ "grad_norm": 0.8055464029312134,
157
+ "learning_rate": 0.0012,
158
+ "loss": 5.904870986938477,
159
+ "step": 180
160
+ },
161
+ {
162
+ "epoch": 2.642233856893543,
163
+ "grad_norm": 0.41387513279914856,
164
+ "learning_rate": 0.0012,
165
+ "loss": 5.888151550292969,
166
+ "step": 190
167
+ },
168
+ {
169
+ "epoch": 2.781849912739965,
170
+ "grad_norm": 0.30246642231941223,
171
+ "learning_rate": 0.0012,
172
+ "loss": 5.781609725952149,
173
+ "step": 200
174
+ },
175
+ {
176
+ "epoch": 2.9214659685863875,
177
+ "grad_norm": 0.2841012477874756,
178
+ "learning_rate": 0.0012,
179
+ "loss": 5.709811401367188,
180
+ "step": 210
181
+ },
182
+ {
183
+ "epoch": 3.0,
184
+ "eval_loss": 5.691598892211914,
185
+ "eval_runtime": 4.097,
186
+ "eval_samples_per_second": 595.796,
187
+ "eval_steps_per_second": 9.519,
188
+ "step": 216
189
+ },
190
+ {
191
+ "epoch": 3.055846422338569,
192
+ "grad_norm": 0.38285037875175476,
193
+ "learning_rate": 0.0012,
194
+ "loss": 5.628893661499023,
195
+ "step": 220
196
+ },
197
+ {
198
+ "epoch": 3.195462478184991,
199
+ "grad_norm": 0.46441373229026794,
200
+ "learning_rate": 0.0012,
201
+ "loss": 5.541646575927734,
202
+ "step": 230
203
+ },
204
+ {
205
+ "epoch": 3.3350785340314135,
206
+ "grad_norm": 0.5794033408164978,
207
+ "learning_rate": 0.0012,
208
+ "loss": 5.48505744934082,
209
+ "step": 240
210
+ },
211
+ {
212
+ "epoch": 3.474694589877836,
213
+ "grad_norm": 0.560783326625824,
214
+ "learning_rate": 0.0012,
215
+ "loss": 5.4691825866699215,
216
+ "step": 250
217
+ },
218
+ {
219
+ "epoch": 3.614310645724258,
220
+ "grad_norm": 0.2417001873254776,
221
+ "learning_rate": 0.0012,
222
+ "loss": 5.374152755737304,
223
+ "step": 260
224
+ },
225
+ {
226
+ "epoch": 3.7539267015706805,
227
+ "grad_norm": 0.7059624791145325,
228
+ "learning_rate": 0.0012,
229
+ "loss": 5.307765197753906,
230
+ "step": 270
231
+ },
232
+ {
233
+ "epoch": 3.893542757417103,
234
+ "grad_norm": 0.26051831245422363,
235
+ "learning_rate": 0.0012,
236
+ "loss": 5.307967758178711,
237
+ "step": 280
238
+ },
239
+ {
240
+ "epoch": 4.0,
241
+ "eval_loss": 5.226171016693115,
242
+ "eval_runtime": 4.087,
243
+ "eval_samples_per_second": 597.255,
244
+ "eval_steps_per_second": 9.542,
245
+ "step": 288
246
+ },
247
+ {
248
+ "epoch": 4.027923211169284,
249
+ "grad_norm": 0.20769786834716797,
250
+ "learning_rate": 0.0012,
251
+ "loss": 5.167432022094727,
252
+ "step": 290
253
+ },
254
+ {
255
+ "epoch": 4.167539267015707,
256
+ "grad_norm": 0.5713754892349243,
257
+ "learning_rate": 0.0012,
258
+ "loss": 5.101666259765625,
259
+ "step": 300
260
+ },
261
+ {
262
+ "epoch": 4.307155322862129,
263
+ "grad_norm": 0.23590517044067383,
264
+ "learning_rate": 0.0012,
265
+ "loss": 5.0780693054199215,
266
+ "step": 310
267
+ },
268
+ {
269
+ "epoch": 4.446771378708552,
270
+ "grad_norm": 0.17983147501945496,
271
+ "learning_rate": 0.0012,
272
+ "loss": 4.969886016845703,
273
+ "step": 320
274
+ },
275
+ {
276
+ "epoch": 4.5863874345549736,
277
+ "grad_norm": 0.6250120997428894,
278
+ "learning_rate": 0.0012,
279
+ "loss": 4.961678314208984,
280
+ "step": 330
281
+ },
282
+ {
283
+ "epoch": 4.726003490401396,
284
+ "grad_norm": 0.5018834471702576,
285
+ "learning_rate": 0.0012,
286
+ "loss": 4.9425300598144535,
287
+ "step": 340
288
+ },
289
+ {
290
+ "epoch": 4.865619546247818,
291
+ "grad_norm": 0.1806376427412033,
292
+ "learning_rate": 0.0012,
293
+ "loss": 4.830913543701172,
294
+ "step": 350
295
+ },
296
+ {
297
+ "epoch": 5.0,
298
+ "grad_norm": 0.42157575488090515,
299
+ "learning_rate": 0.0012,
300
+ "loss": 4.768266296386718,
301
+ "step": 360
302
+ },
303
+ {
304
+ "epoch": 5.0,
305
+ "eval_loss": 4.85431432723999,
306
+ "eval_runtime": 4.0892,
307
+ "eval_samples_per_second": 596.936,
308
+ "eval_steps_per_second": 9.537,
309
+ "step": 360
310
+ },
311
+ {
312
+ "epoch": 5.139616055846423,
313
+ "grad_norm": 0.4851687252521515,
314
+ "learning_rate": 0.0012,
315
+ "loss": 4.7232307434082035,
316
+ "step": 370
317
+ },
318
+ {
319
+ "epoch": 5.279232111692845,
320
+ "grad_norm": 0.20989617705345154,
321
+ "learning_rate": 0.0012,
322
+ "loss": 4.6745655059814455,
323
+ "step": 380
324
+ },
325
+ {
326
+ "epoch": 5.418848167539267,
327
+ "grad_norm": 0.18929427862167358,
328
+ "learning_rate": 0.0012,
329
+ "loss": 4.5781105041503904,
330
+ "step": 390
331
+ },
332
+ {
333
+ "epoch": 5.558464223385689,
334
+ "grad_norm": 0.5501760244369507,
335
+ "learning_rate": 0.0012,
336
+ "loss": 4.528529357910156,
337
+ "step": 400
338
+ },
339
+ {
340
+ "epoch": 5.698080279232112,
341
+ "grad_norm": 0.33675816655158997,
342
+ "learning_rate": 0.0012,
343
+ "loss": 4.550762176513672,
344
+ "step": 410
345
+ },
346
+ {
347
+ "epoch": 5.837696335078534,
348
+ "grad_norm": 0.1552075892686844,
349
+ "learning_rate": 0.0012,
350
+ "loss": 4.4934234619140625,
351
+ "step": 420
352
+ },
353
+ {
354
+ "epoch": 5.977312390924956,
355
+ "grad_norm": 0.3856748640537262,
356
+ "learning_rate": 0.0012,
357
+ "loss": 4.433892822265625,
358
+ "step": 430
359
+ },
360
+ {
361
+ "epoch": 6.0,
362
+ "eval_loss": 4.608406066894531,
363
+ "eval_runtime": 4.0882,
364
+ "eval_samples_per_second": 597.088,
365
+ "eval_steps_per_second": 9.54,
366
+ "step": 432
367
+ },
368
+ {
369
+ "epoch": 6.111692844677138,
370
+ "grad_norm": 0.4819994270801544,
371
+ "learning_rate": 0.0012,
372
+ "loss": 4.424343872070312,
373
+ "step": 440
374
+ },
375
+ {
376
+ "epoch": 6.2513089005235605,
377
+ "grad_norm": 0.20761343836784363,
378
+ "learning_rate": 0.0012,
379
+ "loss": 4.38978271484375,
380
+ "step": 450
381
+ },
382
+ {
383
+ "epoch": 6.390924956369982,
384
+ "grad_norm": 0.15534761548042297,
385
+ "learning_rate": 0.0012,
386
+ "loss": 4.303800201416015,
387
+ "step": 460
388
+ },
389
+ {
390
+ "epoch": 6.530541012216405,
391
+ "grad_norm": 0.16674354672431946,
392
+ "learning_rate": 0.0012,
393
+ "loss": 4.259825134277344,
394
+ "step": 470
395
+ },
396
+ {
397
+ "epoch": 6.670157068062827,
398
+ "grad_norm": 0.40141749382019043,
399
+ "learning_rate": 0.0012,
400
+ "loss": 4.243598937988281,
401
+ "step": 480
402
+ },
403
+ {
404
+ "epoch": 6.80977312390925,
405
+ "grad_norm": 0.16742125153541565,
406
+ "learning_rate": 0.0012,
407
+ "loss": 4.259889221191406,
408
+ "step": 490
409
+ },
410
+ {
411
+ "epoch": 6.949389179755672,
412
+ "grad_norm": 0.19535250961780548,
413
+ "learning_rate": 0.0012,
414
+ "loss": 4.196575927734375,
415
+ "step": 500
416
+ },
417
+ {
418
+ "epoch": 7.0,
419
+ "eval_loss": 4.354514122009277,
420
+ "eval_runtime": 4.0881,
421
+ "eval_samples_per_second": 597.094,
422
+ "eval_steps_per_second": 9.54,
423
+ "step": 504
424
+ },
425
+ {
426
+ "epoch": 7.0837696335078535,
427
+ "grad_norm": 0.433318555355072,
428
+ "learning_rate": 0.0012,
429
+ "loss": 4.1356159210205075,
430
+ "step": 510
431
+ },
432
+ {
433
+ "epoch": 7.223385689354275,
434
+ "grad_norm": 0.4595310688018799,
435
+ "learning_rate": 0.0012,
436
+ "loss": 4.177924728393554,
437
+ "step": 520
438
+ },
439
+ {
440
+ "epoch": 7.363001745200698,
441
+ "grad_norm": 0.18393999338150024,
442
+ "learning_rate": 0.0012,
443
+ "loss": 4.169436645507813,
444
+ "step": 530
445
+ },
446
+ {
447
+ "epoch": 7.50261780104712,
448
+ "grad_norm": 0.12354253232479095,
449
+ "learning_rate": 0.0012,
450
+ "loss": 4.10106086730957,
451
+ "step": 540
452
+ },
453
+ {
454
+ "epoch": 7.642233856893543,
455
+ "grad_norm": 0.11428206413984299,
456
+ "learning_rate": 0.0012,
457
+ "loss": 4.0617938995361325,
458
+ "step": 550
459
+ },
460
+ {
461
+ "epoch": 7.781849912739965,
462
+ "grad_norm": 0.24703778326511383,
463
+ "learning_rate": 0.0012,
464
+ "loss": 4.040187072753906,
465
+ "step": 560
466
+ },
467
+ {
468
+ "epoch": 7.9214659685863875,
469
+ "grad_norm": 0.1985350400209427,
470
+ "learning_rate": 0.0012,
471
+ "loss": 4.034548950195313,
472
+ "step": 570
473
+ },
474
+ {
475
+ "epoch": 8.0,
476
+ "eval_loss": 4.230696201324463,
477
+ "eval_runtime": 4.0967,
478
+ "eval_samples_per_second": 595.847,
479
+ "eval_steps_per_second": 9.52,
480
+ "step": 576
481
+ },
482
+ {
483
+ "epoch": 8.055846422338568,
484
+ "grad_norm": 0.22547073662281036,
485
+ "learning_rate": 0.0012,
486
+ "loss": 3.9681182861328126,
487
+ "step": 580
488
+ },
489
+ {
490
+ "epoch": 8.195462478184991,
491
+ "grad_norm": 0.17060305178165436,
492
+ "learning_rate": 0.0012,
493
+ "loss": 3.8946533203125,
494
+ "step": 590
495
+ },
496
+ {
497
+ "epoch": 8.335078534031414,
498
+ "grad_norm": 0.15548421442508698,
499
+ "learning_rate": 0.0012,
500
+ "loss": 3.884337615966797,
501
+ "step": 600
502
+ },
503
+ {
504
+ "epoch": 8.474694589877837,
505
+ "grad_norm": 0.16989755630493164,
506
+ "learning_rate": 0.0012,
507
+ "loss": 3.867962646484375,
508
+ "step": 610
509
+ },
510
+ {
511
+ "epoch": 8.614310645724258,
512
+ "grad_norm": 0.3586825132369995,
513
+ "learning_rate": 0.0012,
514
+ "loss": 3.874079132080078,
515
+ "step": 620
516
+ },
517
+ {
518
+ "epoch": 8.75392670157068,
519
+ "grad_norm": 0.2342267632484436,
520
+ "learning_rate": 0.0012,
521
+ "loss": 3.8905319213867187,
522
+ "step": 630
523
+ },
524
+ {
525
+ "epoch": 8.893542757417103,
526
+ "grad_norm": 0.18740172684192657,
527
+ "learning_rate": 0.0012,
528
+ "loss": 3.8916366577148436,
529
+ "step": 640
530
+ },
531
+ {
532
+ "epoch": 9.0,
533
+ "eval_loss": 4.121691703796387,
534
+ "eval_runtime": 4.1052,
535
+ "eval_samples_per_second": 594.617,
536
+ "eval_steps_per_second": 9.5,
537
+ "step": 648
538
+ },
539
+ {
540
+ "epoch": 9.027923211169284,
541
+ "grad_norm": 0.17139847576618195,
542
+ "learning_rate": 0.0012,
543
+ "loss": 3.832661819458008,
544
+ "step": 650
545
+ },
546
+ {
547
+ "epoch": 9.167539267015707,
548
+ "grad_norm": 0.4771718382835388,
549
+ "learning_rate": 0.0012,
550
+ "loss": 3.7312690734863283,
551
+ "step": 660
552
+ },
553
+ {
554
+ "epoch": 9.30715532286213,
555
+ "grad_norm": 0.3208262026309967,
556
+ "learning_rate": 0.0012,
557
+ "loss": 3.7902557373046877,
558
+ "step": 670
559
+ },
560
+ {
561
+ "epoch": 9.44677137870855,
562
+ "grad_norm": 0.269268274307251,
563
+ "learning_rate": 0.0012,
564
+ "loss": 3.79544677734375,
565
+ "step": 680
566
+ },
567
+ {
568
+ "epoch": 9.586387434554974,
569
+ "grad_norm": 0.16161218285560608,
570
+ "learning_rate": 0.0012,
571
+ "loss": 3.7524093627929687,
572
+ "step": 690
573
+ },
574
+ {
575
+ "epoch": 9.726003490401396,
576
+ "grad_norm": 0.13752730190753937,
577
+ "learning_rate": 0.0012,
578
+ "loss": 3.7424781799316404,
579
+ "step": 700
580
+ },
581
+ {
582
+ "epoch": 9.86561954624782,
583
+ "grad_norm": 0.12593114376068115,
584
+ "learning_rate": 0.0012,
585
+ "loss": 3.7211395263671876,
586
+ "step": 710
587
+ },
588
+ {
589
+ "epoch": 10.0,
590
+ "grad_norm": 0.13294491171836853,
591
+ "learning_rate": 0.0012,
592
+ "loss": 3.708177185058594,
593
+ "step": 720
594
+ },
595
+ {
596
+ "epoch": 10.0,
597
+ "eval_loss": 4.056910037994385,
598
+ "eval_runtime": 4.0907,
599
+ "eval_samples_per_second": 596.717,
600
+ "eval_steps_per_second": 9.534,
601
+ "step": 720
602
+ },
603
+ {
604
+ "epoch": 10.139616055846423,
605
+ "grad_norm": 0.5313000082969666,
606
+ "learning_rate": 0.0012,
607
+ "loss": 3.5885597229003907,
608
+ "step": 730
609
+ },
610
+ {
611
+ "epoch": 10.279232111692846,
612
+ "grad_norm": 0.37508344650268555,
613
+ "learning_rate": 0.0012,
614
+ "loss": 3.6925918579101564,
615
+ "step": 740
616
+ },
617
+ {
618
+ "epoch": 10.418848167539267,
619
+ "grad_norm": 0.1845429688692093,
620
+ "learning_rate": 0.0012,
621
+ "loss": 3.656546783447266,
622
+ "step": 750
623
+ },
624
+ {
625
+ "epoch": 10.55846422338569,
626
+ "grad_norm": 0.13196681439876556,
627
+ "learning_rate": 0.0012,
628
+ "loss": 3.6393321990966796,
629
+ "step": 760
630
+ },
631
+ {
632
+ "epoch": 10.698080279232112,
633
+ "grad_norm": 0.12762899696826935,
634
+ "learning_rate": 0.0012,
635
+ "loss": 3.6187034606933595,
636
+ "step": 770
637
+ },
638
+ {
639
+ "epoch": 10.837696335078533,
640
+ "grad_norm": 0.12318549305200577,
641
+ "learning_rate": 0.0012,
642
+ "loss": 3.6179458618164064,
643
+ "step": 780
644
+ },
645
+ {
646
+ "epoch": 10.977312390924956,
647
+ "grad_norm": 0.12958157062530518,
648
+ "learning_rate": 0.0012,
649
+ "loss": 3.6080657958984377,
650
+ "step": 790
651
+ },
652
+ {
653
+ "epoch": 11.0,
654
+ "eval_loss": 4.0241265296936035,
655
+ "eval_runtime": 4.0952,
656
+ "eval_samples_per_second": 596.068,
657
+ "eval_steps_per_second": 9.523,
658
+ "step": 792
659
+ },
660
+ {
661
+ "epoch": 11.111692844677139,
662
+ "grad_norm": 0.28613024950027466,
663
+ "learning_rate": 0.0012,
664
+ "loss": 3.4843353271484374,
665
+ "step": 800
666
+ },
667
+ {
668
+ "epoch": 11.25130890052356,
669
+ "grad_norm": 0.3105561137199402,
670
+ "learning_rate": 0.0012,
671
+ "loss": 3.458019256591797,
672
+ "step": 810
673
+ },
674
+ {
675
+ "epoch": 11.390924956369982,
676
+ "grad_norm": 0.2634027302265167,
677
+ "learning_rate": 0.0012,
678
+ "loss": 3.488126754760742,
679
+ "step": 820
680
+ },
681
+ {
682
+ "epoch": 11.530541012216405,
683
+ "grad_norm": 0.14770400524139404,
684
+ "learning_rate": 0.0012,
685
+ "loss": 3.4990020751953126,
686
+ "step": 830
687
+ },
688
+ {
689
+ "epoch": 11.670157068062828,
690
+ "grad_norm": 0.13869911432266235,
691
+ "learning_rate": 0.0012,
692
+ "loss": 3.4969722747802736,
693
+ "step": 840
694
+ },
695
+ {
696
+ "epoch": 11.809773123909249,
697
+ "grad_norm": 0.12742026150226593,
698
+ "learning_rate": 0.0012,
699
+ "loss": 3.502721405029297,
700
+ "step": 850
701
+ },
702
+ {
703
+ "epoch": 11.949389179755672,
704
+ "grad_norm": 0.1485251933336258,
705
+ "learning_rate": 0.0012,
706
+ "loss": 3.5136146545410156,
707
+ "step": 860
708
+ },
709
+ {
710
+ "epoch": 12.0,
711
+ "eval_loss": 4.010306358337402,
712
+ "eval_runtime": 4.1105,
713
+ "eval_samples_per_second": 593.846,
714
+ "eval_steps_per_second": 9.488,
715
+ "step": 864
716
+ }
717
+ ],
718
+ "logging_steps": 10,
719
+ "max_steps": 864,
720
+ "num_input_tokens_seen": 0,
721
+ "num_train_epochs": 12,
722
+ "save_steps": 500,
723
+ "stateful_callbacks": {
724
+ "TrainerControl": {
725
+ "args": {
726
+ "should_epoch_stop": false,
727
+ "should_evaluate": false,
728
+ "should_log": false,
729
+ "should_save": false,
730
+ "should_training_stop": true
731
+ },
732
+ "attributes": {}
733
+ }
734
+ },
735
+ "total_flos": 7.656447700673495e+17,
736
+ "train_batch_size": 8,
737
+ "trial_name": null,
738
+ "trial_params": null
739
+ }
runs/t0/t0_S3_lr0.0012/train.log ADDED
The diff for this file is too large to render. See raw diff
 
runs/t0/t0_S4_lr0.0006/checkpoint-864/chat_template.jinja ADDED
@@ -0,0 +1,85 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {%- if tools %}
2
+ {{- '<|im_start|>system\n' }}
3
+ {%- if messages[0].role == 'system' %}
4
+ {{- messages[0].content + '\n\n' }}
5
+ {%- endif %}
6
+ {{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
7
+ {%- for tool in tools %}
8
+ {{- "\n" }}
9
+ {{- tool | tojson }}
10
+ {%- endfor %}
11
+ {{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
12
+ {%- else %}
13
+ {%- if messages[0].role == 'system' %}
14
+ {{- '<|im_start|>system\n' + messages[0].content + '<|im_end|>\n' }}
15
+ {%- endif %}
16
+ {%- endif %}
17
+ {%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
18
+ {%- for message in messages[::-1] %}
19
+ {%- set index = (messages|length - 1) - loop.index0 %}
20
+ {%- if ns.multi_step_tool and message.role == "user" and not(message.content.startswith('<tool_response>') and message.content.endswith('</tool_response>')) %}
21
+ {%- set ns.multi_step_tool = false %}
22
+ {%- set ns.last_query_index = index %}
23
+ {%- endif %}
24
+ {%- endfor %}
25
+ {%- for message in messages %}
26
+ {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
27
+ {{- '<|im_start|>' + message.role + '\n' + message.content + '<|im_end|>' + '\n' }}
28
+ {%- elif message.role == "assistant" %}
29
+ {%- set content = message.content %}
30
+ {%- set reasoning_content = '' %}
31
+ {%- if message.reasoning_content is defined and message.reasoning_content is not none %}
32
+ {%- set reasoning_content = message.reasoning_content %}
33
+ {%- else %}
34
+ {%- if '</think>' in message.content %}
35
+ {%- set content = message.content.split('</think>')[-1].lstrip('\n') %}
36
+ {%- set reasoning_content = message.content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
37
+ {%- endif %}
38
+ {%- endif %}
39
+ {%- if loop.index0 > ns.last_query_index %}
40
+ {%- if loop.last or (not loop.last and reasoning_content) %}
41
+ {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
42
+ {%- else %}
43
+ {{- '<|im_start|>' + message.role + '\n' + content }}
44
+ {%- endif %}
45
+ {%- else %}
46
+ {{- '<|im_start|>' + message.role + '\n' + content }}
47
+ {%- endif %}
48
+ {%- if message.tool_calls %}
49
+ {%- for tool_call in message.tool_calls %}
50
+ {%- if (loop.first and content) or (not loop.first) %}
51
+ {{- '\n' }}
52
+ {%- endif %}
53
+ {%- if tool_call.function %}
54
+ {%- set tool_call = tool_call.function %}
55
+ {%- endif %}
56
+ {{- '<tool_call>\n{"name": "' }}
57
+ {{- tool_call.name }}
58
+ {{- '", "arguments": ' }}
59
+ {%- if tool_call.arguments is string %}
60
+ {{- tool_call.arguments }}
61
+ {%- else %}
62
+ {{- tool_call.arguments | tojson }}
63
+ {%- endif %}
64
+ {{- '}\n</tool_call>' }}
65
+ {%- endfor %}
66
+ {%- endif %}
67
+ {{- '<|im_end|>\n' }}
68
+ {%- elif message.role == "tool" %}
69
+ {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
70
+ {{- '<|im_start|>user' }}
71
+ {%- endif %}
72
+ {{- '\n<tool_response>\n' }}
73
+ {{- message.content }}
74
+ {{- '\n</tool_response>' }}
75
+ {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
76
+ {{- '<|im_end|>\n' }}
77
+ {%- endif %}
78
+ {%- endif %}
79
+ {%- endfor %}
80
+ {%- if add_generation_prompt %}
81
+ {{- '<|im_start|>assistant\n' }}
82
+ {%- if enable_thinking is defined and enable_thinking is false %}
83
+ {{- '<think>\n\n</think>\n\n' }}
84
+ {%- endif %}
85
+ {%- endif %}
runs/t0/t0_S4_lr0.0006/checkpoint-864/config.json ADDED
@@ -0,0 +1,32 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "LlamaForCausalLM"
4
+ ],
5
+ "attention_bias": false,
6
+ "attention_dropout": 0.0,
7
+ "bos_token_id": null,
8
+ "dtype": "float32",
9
+ "eos_token_id": 151645,
10
+ "head_dim": 128,
11
+ "hidden_act": "silu",
12
+ "hidden_size": 1024,
13
+ "initializer_range": 0.02,
14
+ "intermediate_size": 2816,
15
+ "max_position_embeddings": 2048,
16
+ "mlp_bias": false,
17
+ "model_type": "llama",
18
+ "num_attention_heads": 8,
19
+ "num_hidden_layers": 24,
20
+ "num_key_value_heads": 8,
21
+ "pad_token_id": 151645,
22
+ "pretraining_tp": 1,
23
+ "rms_norm_eps": 1e-06,
24
+ "rope_parameters": {
25
+ "rope_theta": 10000.0,
26
+ "rope_type": "default"
27
+ },
28
+ "tie_word_embeddings": true,
29
+ "transformers_version": "5.15.0",
30
+ "use_cache": false,
31
+ "vocab_size": 151671
32
+ }
runs/t0/t0_S4_lr0.0006/checkpoint-864/generation_config.json ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "eos_token_id": [
4
+ 151645
5
+ ],
6
+ "output_attentions": false,
7
+ "output_hidden_states": false,
8
+ "pad_token_id": 151645,
9
+ "transformers_version": "5.15.0",
10
+ "use_cache": true
11
+ }
runs/t0/t0_S4_lr0.0006/checkpoint-864/tokenizer_config.json ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "backend": "tokenizers",
4
+ "bos_token": null,
5
+ "clean_up_tokenization_spaces": false,
6
+ "eos_token": "<|im_end|>",
7
+ "errors": "replace",
8
+ "extra_special_tokens": [
9
+ "<|l2r_pred|>",
10
+ "<|r2l_pred|>"
11
+ ],
12
+ "is_local": false,
13
+ "local_files_only": false,
14
+ "model_max_length": 131072,
15
+ "pad_token": "<|im_end|>",
16
+ "split_special_tokens": false,
17
+ "tokenizer_class": "Qwen2Tokenizer",
18
+ "unk_token": null
19
+ }
runs/t0/t0_S4_lr0.0006/checkpoint-864/trainer_state.json ADDED
@@ -0,0 +1,739 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 12.0,
6
+ "eval_steps": 72,
7
+ "global_step": 864,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "epoch": 0.013961605584642234,
14
+ "grad_norm": 2.9206132888793945,
15
+ "learning_rate": 0.0,
16
+ "loss": 12.153533935546875,
17
+ "step": 1
18
+ },
19
+ {
20
+ "epoch": 0.13961605584642234,
21
+ "grad_norm": 2.1237220764160156,
22
+ "learning_rate": 5.399999999999999e-05,
23
+ "loss": 11.390463087293837,
24
+ "step": 10
25
+ },
26
+ {
27
+ "epoch": 0.2792321116928447,
28
+ "grad_norm": 1.830725073814392,
29
+ "learning_rate": 0.00011399999999999999,
30
+ "loss": 10.077462768554687,
31
+ "step": 20
32
+ },
33
+ {
34
+ "epoch": 0.418848167539267,
35
+ "grad_norm": 1.0628323554992676,
36
+ "learning_rate": 0.00017399999999999997,
37
+ "loss": 8.793680572509766,
38
+ "step": 30
39
+ },
40
+ {
41
+ "epoch": 0.5584642233856894,
42
+ "grad_norm": 0.45776990056037903,
43
+ "learning_rate": 0.000234,
44
+ "loss": 7.745941162109375,
45
+ "step": 40
46
+ },
47
+ {
48
+ "epoch": 0.6980802792321117,
49
+ "grad_norm": 1.1509424448013306,
50
+ "learning_rate": 0.000294,
51
+ "loss": 7.502726745605469,
52
+ "step": 50
53
+ },
54
+ {
55
+ "epoch": 0.837696335078534,
56
+ "grad_norm": 0.8295908570289612,
57
+ "learning_rate": 0.00035399999999999993,
58
+ "loss": 7.317385864257813,
59
+ "step": 60
60
+ },
61
+ {
62
+ "epoch": 0.9773123909249564,
63
+ "grad_norm": 0.8668380975723267,
64
+ "learning_rate": 0.0004139999999999999,
65
+ "loss": 7.089698028564453,
66
+ "step": 70
67
+ },
68
+ {
69
+ "epoch": 1.0,
70
+ "eval_loss": 6.985106468200684,
71
+ "eval_runtime": 29.6678,
72
+ "eval_samples_per_second": 82.278,
73
+ "eval_steps_per_second": 1.315,
74
+ "step": 72
75
+ },
76
+ {
77
+ "epoch": 1.111692844677138,
78
+ "grad_norm": 0.6774813532829285,
79
+ "learning_rate": 0.000474,
80
+ "loss": 6.9334571838378904,
81
+ "step": 80
82
+ },
83
+ {
84
+ "epoch": 1.2513089005235603,
85
+ "grad_norm": 0.6552468538284302,
86
+ "learning_rate": 0.000534,
87
+ "loss": 6.771307373046875,
88
+ "step": 90
89
+ },
90
+ {
91
+ "epoch": 1.3909249563699826,
92
+ "grad_norm": 1.0316975116729736,
93
+ "learning_rate": 0.0005939999999999999,
94
+ "loss": 6.63712158203125,
95
+ "step": 100
96
+ },
97
+ {
98
+ "epoch": 1.530541012216405,
99
+ "grad_norm": 0.45376425981521606,
100
+ "learning_rate": 0.0006,
101
+ "loss": 6.486009979248047,
102
+ "step": 110
103
+ },
104
+ {
105
+ "epoch": 1.6701570680628273,
106
+ "grad_norm": 0.7860228419303894,
107
+ "learning_rate": 0.0006,
108
+ "loss": 6.39349594116211,
109
+ "step": 120
110
+ },
111
+ {
112
+ "epoch": 1.8097731239092496,
113
+ "grad_norm": 0.4239625632762909,
114
+ "learning_rate": 0.0006,
115
+ "loss": 6.256306457519531,
116
+ "step": 130
117
+ },
118
+ {
119
+ "epoch": 1.949389179755672,
120
+ "grad_norm": 0.6127851605415344,
121
+ "learning_rate": 0.0006,
122
+ "loss": 6.163679885864258,
123
+ "step": 140
124
+ },
125
+ {
126
+ "epoch": 2.0,
127
+ "eval_loss": 6.093886852264404,
128
+ "eval_runtime": 5.4342,
129
+ "eval_samples_per_second": 449.188,
130
+ "eval_steps_per_second": 7.177,
131
+ "step": 144
132
+ },
133
+ {
134
+ "epoch": 2.0837696335078535,
135
+ "grad_norm": 0.47936490178108215,
136
+ "learning_rate": 0.0006,
137
+ "loss": 6.04110107421875,
138
+ "step": 150
139
+ },
140
+ {
141
+ "epoch": 2.223385689354276,
142
+ "grad_norm": 0.5541312098503113,
143
+ "learning_rate": 0.0006,
144
+ "loss": 5.978498077392578,
145
+ "step": 160
146
+ },
147
+ {
148
+ "epoch": 2.363001745200698,
149
+ "grad_norm": 0.3599136471748352,
150
+ "learning_rate": 0.0006,
151
+ "loss": 5.869022369384766,
152
+ "step": 170
153
+ },
154
+ {
155
+ "epoch": 2.5026178010471205,
156
+ "grad_norm": 0.42818182706832886,
157
+ "learning_rate": 0.0006,
158
+ "loss": 5.82703742980957,
159
+ "step": 180
160
+ },
161
+ {
162
+ "epoch": 2.642233856893543,
163
+ "grad_norm": 0.6170725226402283,
164
+ "learning_rate": 0.0006,
165
+ "loss": 5.707303237915039,
166
+ "step": 190
167
+ },
168
+ {
169
+ "epoch": 2.781849912739965,
170
+ "grad_norm": 0.5079529881477356,
171
+ "learning_rate": 0.0006,
172
+ "loss": 5.721559524536133,
173
+ "step": 200
174
+ },
175
+ {
176
+ "epoch": 2.9214659685863875,
177
+ "grad_norm": 0.22502170503139496,
178
+ "learning_rate": 0.0006,
179
+ "loss": 5.599060821533203,
180
+ "step": 210
181
+ },
182
+ {
183
+ "epoch": 3.0,
184
+ "eval_loss": 5.532996654510498,
185
+ "eval_runtime": 5.4393,
186
+ "eval_samples_per_second": 448.773,
187
+ "eval_steps_per_second": 7.17,
188
+ "step": 216
189
+ },
190
+ {
191
+ "epoch": 3.055846422338569,
192
+ "grad_norm": 0.4988436996936798,
193
+ "learning_rate": 0.0006,
194
+ "loss": 5.476123046875,
195
+ "step": 220
196
+ },
197
+ {
198
+ "epoch": 3.195462478184991,
199
+ "grad_norm": 0.47404739260673523,
200
+ "learning_rate": 0.0006,
201
+ "loss": 5.449727630615234,
202
+ "step": 230
203
+ },
204
+ {
205
+ "epoch": 3.3350785340314135,
206
+ "grad_norm": 0.29689666628837585,
207
+ "learning_rate": 0.0006,
208
+ "loss": 5.350494003295898,
209
+ "step": 240
210
+ },
211
+ {
212
+ "epoch": 3.474694589877836,
213
+ "grad_norm": 0.5025255680084229,
214
+ "learning_rate": 0.0006,
215
+ "loss": 5.306208801269531,
216
+ "step": 250
217
+ },
218
+ {
219
+ "epoch": 3.614310645724258,
220
+ "grad_norm": 0.3087793290615082,
221
+ "learning_rate": 0.0006,
222
+ "loss": 5.252265167236328,
223
+ "step": 260
224
+ },
225
+ {
226
+ "epoch": 3.7539267015706805,
227
+ "grad_norm": 0.24143578112125397,
228
+ "learning_rate": 0.0006,
229
+ "loss": 5.147513580322266,
230
+ "step": 270
231
+ },
232
+ {
233
+ "epoch": 3.893542757417103,
234
+ "grad_norm": 0.6470257639884949,
235
+ "learning_rate": 0.0006,
236
+ "loss": 5.087617492675781,
237
+ "step": 280
238
+ },
239
+ {
240
+ "epoch": 4.0,
241
+ "eval_loss": 5.172598361968994,
242
+ "eval_runtime": 5.4512,
243
+ "eval_samples_per_second": 447.792,
244
+ "eval_steps_per_second": 7.154,
245
+ "step": 288
246
+ },
247
+ {
248
+ "epoch": 4.027923211169284,
249
+ "grad_norm": 0.6637756824493408,
250
+ "learning_rate": 0.0006,
251
+ "loss": 5.070102691650391,
252
+ "step": 290
253
+ },
254
+ {
255
+ "epoch": 4.167539267015707,
256
+ "grad_norm": 0.25461071729660034,
257
+ "learning_rate": 0.0006,
258
+ "loss": 4.9610851287841795,
259
+ "step": 300
260
+ },
261
+ {
262
+ "epoch": 4.307155322862129,
263
+ "grad_norm": 0.5295916795730591,
264
+ "learning_rate": 0.0006,
265
+ "loss": 4.859738159179687,
266
+ "step": 310
267
+ },
268
+ {
269
+ "epoch": 4.446771378708552,
270
+ "grad_norm": 0.5386121869087219,
271
+ "learning_rate": 0.0006,
272
+ "loss": 4.826572418212891,
273
+ "step": 320
274
+ },
275
+ {
276
+ "epoch": 4.5863874345549736,
277
+ "grad_norm": 0.42201587557792664,
278
+ "learning_rate": 0.0006,
279
+ "loss": 4.807993316650391,
280
+ "step": 330
281
+ },
282
+ {
283
+ "epoch": 4.726003490401396,
284
+ "grad_norm": 0.42141446471214294,
285
+ "learning_rate": 0.0006,
286
+ "loss": 4.7262310028076175,
287
+ "step": 340
288
+ },
289
+ {
290
+ "epoch": 4.865619546247818,
291
+ "grad_norm": 0.25011977553367615,
292
+ "learning_rate": 0.0006,
293
+ "loss": 4.645610046386719,
294
+ "step": 350
295
+ },
296
+ {
297
+ "epoch": 5.0,
298
+ "grad_norm": 0.40314558148384094,
299
+ "learning_rate": 0.0006,
300
+ "loss": 4.607339477539062,
301
+ "step": 360
302
+ },
303
+ {
304
+ "epoch": 5.0,
305
+ "eval_loss": 4.691513538360596,
306
+ "eval_runtime": 5.4559,
307
+ "eval_samples_per_second": 447.402,
308
+ "eval_steps_per_second": 7.148,
309
+ "step": 360
310
+ },
311
+ {
312
+ "epoch": 5.139616055846423,
313
+ "grad_norm": 0.35727429389953613,
314
+ "learning_rate": 0.0006,
315
+ "loss": 4.493846130371094,
316
+ "step": 370
317
+ },
318
+ {
319
+ "epoch": 5.279232111692845,
320
+ "grad_norm": 0.579684317111969,
321
+ "learning_rate": 0.0006,
322
+ "loss": 4.477697372436523,
323
+ "step": 380
324
+ },
325
+ {
326
+ "epoch": 5.418848167539267,
327
+ "grad_norm": 0.4042253792285919,
328
+ "learning_rate": 0.0006,
329
+ "loss": 4.435718154907226,
330
+ "step": 390
331
+ },
332
+ {
333
+ "epoch": 5.558464223385689,
334
+ "grad_norm": 0.2034163773059845,
335
+ "learning_rate": 0.0006,
336
+ "loss": 4.361290740966797,
337
+ "step": 400
338
+ },
339
+ {
340
+ "epoch": 5.698080279232112,
341
+ "grad_norm": 0.21729686856269836,
342
+ "learning_rate": 0.0006,
343
+ "loss": 4.318072509765625,
344
+ "step": 410
345
+ },
346
+ {
347
+ "epoch": 5.837696335078534,
348
+ "grad_norm": 0.45357993245124817,
349
+ "learning_rate": 0.0006,
350
+ "loss": 4.285897827148437,
351
+ "step": 420
352
+ },
353
+ {
354
+ "epoch": 5.977312390924956,
355
+ "grad_norm": 0.4423562288284302,
356
+ "learning_rate": 0.0006,
357
+ "loss": 4.292565155029297,
358
+ "step": 430
359
+ },
360
+ {
361
+ "epoch": 6.0,
362
+ "eval_loss": 4.448753356933594,
363
+ "eval_runtime": 5.4351,
364
+ "eval_samples_per_second": 449.117,
365
+ "eval_steps_per_second": 7.176,
366
+ "step": 432
367
+ },
368
+ {
369
+ "epoch": 6.111692844677138,
370
+ "grad_norm": 0.6937609910964966,
371
+ "learning_rate": 0.0006,
372
+ "loss": 4.254433441162109,
373
+ "step": 440
374
+ },
375
+ {
376
+ "epoch": 6.2513089005235605,
377
+ "grad_norm": 0.501941442489624,
378
+ "learning_rate": 0.0006,
379
+ "loss": 4.290074157714844,
380
+ "step": 450
381
+ },
382
+ {
383
+ "epoch": 6.390924956369982,
384
+ "grad_norm": 0.2516627907752991,
385
+ "learning_rate": 0.0006,
386
+ "loss": 4.194451522827149,
387
+ "step": 460
388
+ },
389
+ {
390
+ "epoch": 6.530541012216405,
391
+ "grad_norm": 0.14753806591033936,
392
+ "learning_rate": 0.0006,
393
+ "loss": 4.124375915527343,
394
+ "step": 470
395
+ },
396
+ {
397
+ "epoch": 6.670157068062827,
398
+ "grad_norm": 0.14808453619480133,
399
+ "learning_rate": 0.0006,
400
+ "loss": 4.078779602050782,
401
+ "step": 480
402
+ },
403
+ {
404
+ "epoch": 6.80977312390925,
405
+ "grad_norm": 0.20621934533119202,
406
+ "learning_rate": 0.0006,
407
+ "loss": 4.063456344604492,
408
+ "step": 490
409
+ },
410
+ {
411
+ "epoch": 6.949389179755672,
412
+ "grad_norm": 0.19101426005363464,
413
+ "learning_rate": 0.0006,
414
+ "loss": 4.029528045654297,
415
+ "step": 500
416
+ },
417
+ {
418
+ "epoch": 7.0,
419
+ "eval_loss": 4.217341899871826,
420
+ "eval_runtime": 5.4543,
421
+ "eval_samples_per_second": 447.538,
422
+ "eval_steps_per_second": 7.15,
423
+ "step": 504
424
+ },
425
+ {
426
+ "epoch": 7.0837696335078535,
427
+ "grad_norm": 0.34916263818740845,
428
+ "learning_rate": 0.0006,
429
+ "loss": 3.9608695983886717,
430
+ "step": 510
431
+ },
432
+ {
433
+ "epoch": 7.223385689354275,
434
+ "grad_norm": 0.4681544005870819,
435
+ "learning_rate": 0.0006,
436
+ "loss": 3.9381080627441407,
437
+ "step": 520
438
+ },
439
+ {
440
+ "epoch": 7.363001745200698,
441
+ "grad_norm": 0.2090049833059311,
442
+ "learning_rate": 0.0006,
443
+ "loss": 3.907509994506836,
444
+ "step": 530
445
+ },
446
+ {
447
+ "epoch": 7.50261780104712,
448
+ "grad_norm": 0.17446690797805786,
449
+ "learning_rate": 0.0006,
450
+ "loss": 3.882748031616211,
451
+ "step": 540
452
+ },
453
+ {
454
+ "epoch": 7.642233856893543,
455
+ "grad_norm": 0.20346012711524963,
456
+ "learning_rate": 0.0006,
457
+ "loss": 3.867753601074219,
458
+ "step": 550
459
+ },
460
+ {
461
+ "epoch": 7.781849912739965,
462
+ "grad_norm": 0.24103151261806488,
463
+ "learning_rate": 0.0006,
464
+ "loss": 3.861027145385742,
465
+ "step": 560
466
+ },
467
+ {
468
+ "epoch": 7.9214659685863875,
469
+ "grad_norm": 0.22708670794963837,
470
+ "learning_rate": 0.0006,
471
+ "loss": 3.848551940917969,
472
+ "step": 570
473
+ },
474
+ {
475
+ "epoch": 8.0,
476
+ "eval_loss": 4.099960803985596,
477
+ "eval_runtime": 5.4497,
478
+ "eval_samples_per_second": 447.913,
479
+ "eval_steps_per_second": 7.156,
480
+ "step": 576
481
+ },
482
+ {
483
+ "epoch": 8.055846422338568,
484
+ "grad_norm": 0.3998148739337921,
485
+ "learning_rate": 0.0006,
486
+ "loss": 3.7770992279052735,
487
+ "step": 580
488
+ },
489
+ {
490
+ "epoch": 8.195462478184991,
491
+ "grad_norm": 0.7326574921607971,
492
+ "learning_rate": 0.0006,
493
+ "loss": 3.76361083984375,
494
+ "step": 590
495
+ },
496
+ {
497
+ "epoch": 8.335078534031414,
498
+ "grad_norm": 1.4837658405303955,
499
+ "learning_rate": 0.0006,
500
+ "loss": 4.08636245727539,
501
+ "step": 600
502
+ },
503
+ {
504
+ "epoch": 8.474694589877837,
505
+ "grad_norm": 1.0409595966339111,
506
+ "learning_rate": 0.0006,
507
+ "loss": 4.354316711425781,
508
+ "step": 610
509
+ },
510
+ {
511
+ "epoch": 8.614310645724258,
512
+ "grad_norm": 0.8818002343177795,
513
+ "learning_rate": 0.0006,
514
+ "loss": 4.53700065612793,
515
+ "step": 620
516
+ },
517
+ {
518
+ "epoch": 8.75392670157068,
519
+ "grad_norm": 0.197915181517601,
520
+ "learning_rate": 0.0006,
521
+ "loss": 4.277849578857422,
522
+ "step": 630
523
+ },
524
+ {
525
+ "epoch": 8.893542757417103,
526
+ "grad_norm": 0.13980717957019806,
527
+ "learning_rate": 0.0006,
528
+ "loss": 4.076632690429688,
529
+ "step": 640
530
+ },
531
+ {
532
+ "epoch": 9.0,
533
+ "eval_loss": 4.183110237121582,
534
+ "eval_runtime": 5.4533,
535
+ "eval_samples_per_second": 447.615,
536
+ "eval_steps_per_second": 7.152,
537
+ "step": 648
538
+ },
539
+ {
540
+ "epoch": 9.027923211169284,
541
+ "grad_norm": 0.12084736675024033,
542
+ "learning_rate": 0.0006,
543
+ "loss": 3.934686279296875,
544
+ "step": 650
545
+ },
546
+ {
547
+ "epoch": 9.167539267015707,
548
+ "grad_norm": 0.14144660532474518,
549
+ "learning_rate": 0.0006,
550
+ "loss": 3.8070610046386717,
551
+ "step": 660
552
+ },
553
+ {
554
+ "epoch": 9.30715532286213,
555
+ "grad_norm": 0.13604827225208282,
556
+ "learning_rate": 0.0006,
557
+ "loss": 3.7739948272705077,
558
+ "step": 670
559
+ },
560
+ {
561
+ "epoch": 9.44677137870855,
562
+ "grad_norm": 0.13959047198295593,
563
+ "learning_rate": 0.0006,
564
+ "loss": 3.7471954345703127,
565
+ "step": 680
566
+ },
567
+ {
568
+ "epoch": 9.586387434554974,
569
+ "grad_norm": 0.12865500152111053,
570
+ "learning_rate": 0.0006,
571
+ "loss": 3.7092533111572266,
572
+ "step": 690
573
+ },
574
+ {
575
+ "epoch": 9.726003490401396,
576
+ "grad_norm": 0.17941144108772278,
577
+ "learning_rate": 0.0006,
578
+ "loss": 3.7057247161865234,
579
+ "step": 700
580
+ },
581
+ {
582
+ "epoch": 9.86561954624782,
583
+ "grad_norm": 0.1411111056804657,
584
+ "learning_rate": 0.0006,
585
+ "loss": 3.679571533203125,
586
+ "step": 710
587
+ },
588
+ {
589
+ "epoch": 10.0,
590
+ "grad_norm": 0.16089625656604767,
591
+ "learning_rate": 0.0006,
592
+ "loss": 3.6586647033691406,
593
+ "step": 720
594
+ },
595
+ {
596
+ "epoch": 10.0,
597
+ "eval_loss": 4.012387752532959,
598
+ "eval_runtime": 5.4629,
599
+ "eval_samples_per_second": 446.835,
600
+ "eval_steps_per_second": 7.139,
601
+ "step": 720
602
+ },
603
+ {
604
+ "epoch": 10.139616055846423,
605
+ "grad_norm": 0.3115384876728058,
606
+ "learning_rate": 0.0006,
607
+ "loss": 3.541545867919922,
608
+ "step": 730
609
+ },
610
+ {
611
+ "epoch": 10.279232111692846,
612
+ "grad_norm": 0.27592477202415466,
613
+ "learning_rate": 0.0006,
614
+ "loss": 3.5572509765625,
615
+ "step": 740
616
+ },
617
+ {
618
+ "epoch": 10.418848167539267,
619
+ "grad_norm": 0.17261362075805664,
620
+ "learning_rate": 0.0006,
621
+ "loss": 3.5457481384277343,
622
+ "step": 750
623
+ },
624
+ {
625
+ "epoch": 10.55846422338569,
626
+ "grad_norm": 0.1407252699136734,
627
+ "learning_rate": 0.0006,
628
+ "loss": 3.5450862884521483,
629
+ "step": 760
630
+ },
631
+ {
632
+ "epoch": 10.698080279232112,
633
+ "grad_norm": 0.12504896521568298,
634
+ "learning_rate": 0.0006,
635
+ "loss": 3.5322113037109375,
636
+ "step": 770
637
+ },
638
+ {
639
+ "epoch": 10.837696335078533,
640
+ "grad_norm": 0.134646475315094,
641
+ "learning_rate": 0.0006,
642
+ "loss": 3.5337203979492187,
643
+ "step": 780
644
+ },
645
+ {
646
+ "epoch": 10.977312390924956,
647
+ "grad_norm": 0.14491793513298035,
648
+ "learning_rate": 0.0006,
649
+ "loss": 3.5220596313476564,
650
+ "step": 790
651
+ },
652
+ {
653
+ "epoch": 11.0,
654
+ "eval_loss": 3.970649480819702,
655
+ "eval_runtime": 5.4496,
656
+ "eval_samples_per_second": 447.925,
657
+ "eval_steps_per_second": 7.157,
658
+ "step": 792
659
+ },
660
+ {
661
+ "epoch": 11.111692844677139,
662
+ "grad_norm": 0.3893614709377289,
663
+ "learning_rate": 0.0006,
664
+ "loss": 3.3977737426757812,
665
+ "step": 800
666
+ },
667
+ {
668
+ "epoch": 11.25130890052356,
669
+ "grad_norm": 0.24212592840194702,
670
+ "learning_rate": 0.0006,
671
+ "loss": 3.3481903076171875,
672
+ "step": 810
673
+ },
674
+ {
675
+ "epoch": 11.390924956369982,
676
+ "grad_norm": 0.2399962693452835,
677
+ "learning_rate": 0.0006,
678
+ "loss": 3.3614341735839846,
679
+ "step": 820
680
+ },
681
+ {
682
+ "epoch": 11.530541012216405,
683
+ "grad_norm": 0.18185333907604218,
684
+ "learning_rate": 0.0006,
685
+ "loss": 3.377635192871094,
686
+ "step": 830
687
+ },
688
+ {
689
+ "epoch": 11.670157068062828,
690
+ "grad_norm": 0.1585046797990799,
691
+ "learning_rate": 0.0006,
692
+ "loss": 3.3790657043457033,
693
+ "step": 840
694
+ },
695
+ {
696
+ "epoch": 11.809773123909249,
697
+ "grad_norm": 0.17206858098506927,
698
+ "learning_rate": 0.0006,
699
+ "loss": 3.387134552001953,
700
+ "step": 850
701
+ },
702
+ {
703
+ "epoch": 11.949389179755672,
704
+ "grad_norm": 0.16680264472961426,
705
+ "learning_rate": 0.0006,
706
+ "loss": 3.397046661376953,
707
+ "step": 860
708
+ },
709
+ {
710
+ "epoch": 12.0,
711
+ "eval_loss": 3.9641687870025635,
712
+ "eval_runtime": 5.4473,
713
+ "eval_samples_per_second": 448.114,
714
+ "eval_steps_per_second": 7.16,
715
+ "step": 864
716
+ }
717
+ ],
718
+ "logging_steps": 10,
719
+ "max_steps": 864,
720
+ "num_input_tokens_seen": 0,
721
+ "num_train_epochs": 12,
722
+ "save_steps": 500,
723
+ "stateful_callbacks": {
724
+ "TrainerControl": {
725
+ "args": {
726
+ "should_epoch_stop": false,
727
+ "should_evaluate": false,
728
+ "should_log": false,
729
+ "should_save": false,
730
+ "should_training_stop": true
731
+ },
732
+ "attributes": {}
733
+ }
734
+ },
735
+ "total_flos": 1.6673046800042557e+18,
736
+ "train_batch_size": 8,
737
+ "trial_name": null,
738
+ "trial_params": null
739
+ }
runs/t0/t0_S4_lr0.0006/train.log ADDED
The diff for this file is too large to render. See raw diff
 
runs/t0/t0_S4_lr0.0012/checkpoint-864/chat_template.jinja ADDED
@@ -0,0 +1,85 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {%- if tools %}
2
+ {{- '<|im_start|>system\n' }}
3
+ {%- if messages[0].role == 'system' %}
4
+ {{- messages[0].content + '\n\n' }}
5
+ {%- endif %}
6
+ {{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
7
+ {%- for tool in tools %}
8
+ {{- "\n" }}
9
+ {{- tool | tojson }}
10
+ {%- endfor %}
11
+ {{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
12
+ {%- else %}
13
+ {%- if messages[0].role == 'system' %}
14
+ {{- '<|im_start|>system\n' + messages[0].content + '<|im_end|>\n' }}
15
+ {%- endif %}
16
+ {%- endif %}
17
+ {%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
18
+ {%- for message in messages[::-1] %}
19
+ {%- set index = (messages|length - 1) - loop.index0 %}
20
+ {%- if ns.multi_step_tool and message.role == "user" and not(message.content.startswith('<tool_response>') and message.content.endswith('</tool_response>')) %}
21
+ {%- set ns.multi_step_tool = false %}
22
+ {%- set ns.last_query_index = index %}
23
+ {%- endif %}
24
+ {%- endfor %}
25
+ {%- for message in messages %}
26
+ {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
27
+ {{- '<|im_start|>' + message.role + '\n' + message.content + '<|im_end|>' + '\n' }}
28
+ {%- elif message.role == "assistant" %}
29
+ {%- set content = message.content %}
30
+ {%- set reasoning_content = '' %}
31
+ {%- if message.reasoning_content is defined and message.reasoning_content is not none %}
32
+ {%- set reasoning_content = message.reasoning_content %}
33
+ {%- else %}
34
+ {%- if '</think>' in message.content %}
35
+ {%- set content = message.content.split('</think>')[-1].lstrip('\n') %}
36
+ {%- set reasoning_content = message.content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
37
+ {%- endif %}
38
+ {%- endif %}
39
+ {%- if loop.index0 > ns.last_query_index %}
40
+ {%- if loop.last or (not loop.last and reasoning_content) %}
41
+ {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
42
+ {%- else %}
43
+ {{- '<|im_start|>' + message.role + '\n' + content }}
44
+ {%- endif %}
45
+ {%- else %}
46
+ {{- '<|im_start|>' + message.role + '\n' + content }}
47
+ {%- endif %}
48
+ {%- if message.tool_calls %}
49
+ {%- for tool_call in message.tool_calls %}
50
+ {%- if (loop.first and content) or (not loop.first) %}
51
+ {{- '\n' }}
52
+ {%- endif %}
53
+ {%- if tool_call.function %}
54
+ {%- set tool_call = tool_call.function %}
55
+ {%- endif %}
56
+ {{- '<tool_call>\n{"name": "' }}
57
+ {{- tool_call.name }}
58
+ {{- '", "arguments": ' }}
59
+ {%- if tool_call.arguments is string %}
60
+ {{- tool_call.arguments }}
61
+ {%- else %}
62
+ {{- tool_call.arguments | tojson }}
63
+ {%- endif %}
64
+ {{- '}\n</tool_call>' }}
65
+ {%- endfor %}
66
+ {%- endif %}
67
+ {{- '<|im_end|>\n' }}
68
+ {%- elif message.role == "tool" %}
69
+ {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
70
+ {{- '<|im_start|>user' }}
71
+ {%- endif %}
72
+ {{- '\n<tool_response>\n' }}
73
+ {{- message.content }}
74
+ {{- '\n</tool_response>' }}
75
+ {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
76
+ {{- '<|im_end|>\n' }}
77
+ {%- endif %}
78
+ {%- endif %}
79
+ {%- endfor %}
80
+ {%- if add_generation_prompt %}
81
+ {{- '<|im_start|>assistant\n' }}
82
+ {%- if enable_thinking is defined and enable_thinking is false %}
83
+ {{- '<think>\n\n</think>\n\n' }}
84
+ {%- endif %}
85
+ {%- endif %}
runs/t0/t0_S4_lr0.0012/checkpoint-864/config.json ADDED
@@ -0,0 +1,32 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "LlamaForCausalLM"
4
+ ],
5
+ "attention_bias": false,
6
+ "attention_dropout": 0.0,
7
+ "bos_token_id": null,
8
+ "dtype": "float32",
9
+ "eos_token_id": 151645,
10
+ "head_dim": 128,
11
+ "hidden_act": "silu",
12
+ "hidden_size": 1024,
13
+ "initializer_range": 0.02,
14
+ "intermediate_size": 2816,
15
+ "max_position_embeddings": 2048,
16
+ "mlp_bias": false,
17
+ "model_type": "llama",
18
+ "num_attention_heads": 8,
19
+ "num_hidden_layers": 24,
20
+ "num_key_value_heads": 8,
21
+ "pad_token_id": 151645,
22
+ "pretraining_tp": 1,
23
+ "rms_norm_eps": 1e-06,
24
+ "rope_parameters": {
25
+ "rope_theta": 10000.0,
26
+ "rope_type": "default"
27
+ },
28
+ "tie_word_embeddings": true,
29
+ "transformers_version": "5.15.0",
30
+ "use_cache": false,
31
+ "vocab_size": 151671
32
+ }
runs/t0/t0_S4_lr0.0012/checkpoint-864/generation_config.json ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "eos_token_id": [
4
+ 151645
5
+ ],
6
+ "output_attentions": false,
7
+ "output_hidden_states": false,
8
+ "pad_token_id": 151645,
9
+ "transformers_version": "5.15.0",
10
+ "use_cache": true
11
+ }
runs/t0/t0_S4_lr0.0012/checkpoint-864/tokenizer_config.json ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "backend": "tokenizers",
4
+ "bos_token": null,
5
+ "clean_up_tokenization_spaces": false,
6
+ "eos_token": "<|im_end|>",
7
+ "errors": "replace",
8
+ "extra_special_tokens": [
9
+ "<|l2r_pred|>",
10
+ "<|r2l_pred|>"
11
+ ],
12
+ "is_local": false,
13
+ "local_files_only": false,
14
+ "model_max_length": 131072,
15
+ "pad_token": "<|im_end|>",
16
+ "split_special_tokens": false,
17
+ "tokenizer_class": "Qwen2Tokenizer",
18
+ "unk_token": null
19
+ }
runs/t0/t0_S4_lr0.0012/checkpoint-864/trainer_state.json ADDED
@@ -0,0 +1,739 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 12.0,
6
+ "eval_steps": 72,
7
+ "global_step": 864,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "epoch": 0.013961605584642234,
14
+ "grad_norm": 2.7830731868743896,
15
+ "learning_rate": 0.0,
16
+ "loss": 12.132288932800293,
17
+ "step": 1
18
+ },
19
+ {
20
+ "epoch": 0.13961605584642234,
21
+ "grad_norm": 2.0695760250091553,
22
+ "learning_rate": 0.00010799999999999998,
23
+ "loss": 11.117290920681423,
24
+ "step": 10
25
+ },
26
+ {
27
+ "epoch": 0.2792321116928447,
28
+ "grad_norm": 1.3685357570648193,
29
+ "learning_rate": 0.00022799999999999999,
30
+ "loss": 9.446217346191407,
31
+ "step": 20
32
+ },
33
+ {
34
+ "epoch": 0.418848167539267,
35
+ "grad_norm": 0.3884938955307007,
36
+ "learning_rate": 0.00034799999999999995,
37
+ "loss": 7.895368957519532,
38
+ "step": 30
39
+ },
40
+ {
41
+ "epoch": 0.5584642233856894,
42
+ "grad_norm": 0.487929105758667,
43
+ "learning_rate": 0.000468,
44
+ "loss": 7.554646301269531,
45
+ "step": 40
46
+ },
47
+ {
48
+ "epoch": 0.6980802792321117,
49
+ "grad_norm": 0.8012627363204956,
50
+ "learning_rate": 0.000588,
51
+ "loss": 7.401921081542969,
52
+ "step": 50
53
+ },
54
+ {
55
+ "epoch": 0.837696335078534,
56
+ "grad_norm": 0.6215198636054993,
57
+ "learning_rate": 0.0007079999999999999,
58
+ "loss": 7.233555603027344,
59
+ "step": 60
60
+ },
61
+ {
62
+ "epoch": 0.9773123909249564,
63
+ "grad_norm": 0.9021868705749512,
64
+ "learning_rate": 0.0008279999999999999,
65
+ "loss": 7.0121917724609375,
66
+ "step": 70
67
+ },
68
+ {
69
+ "epoch": 1.0,
70
+ "eval_loss": 6.884247779846191,
71
+ "eval_runtime": 29.6155,
72
+ "eval_samples_per_second": 82.423,
73
+ "eval_steps_per_second": 1.317,
74
+ "step": 72
75
+ },
76
+ {
77
+ "epoch": 1.111692844677138,
78
+ "grad_norm": 0.7726740837097168,
79
+ "learning_rate": 0.000948,
80
+ "loss": 6.848681640625,
81
+ "step": 80
82
+ },
83
+ {
84
+ "epoch": 1.2513089005235603,
85
+ "grad_norm": 0.39272984862327576,
86
+ "learning_rate": 0.001068,
87
+ "loss": 6.674762725830078,
88
+ "step": 90
89
+ },
90
+ {
91
+ "epoch": 1.3909249563699826,
92
+ "grad_norm": 0.6953481435775757,
93
+ "learning_rate": 0.0011879999999999998,
94
+ "loss": 6.53614501953125,
95
+ "step": 100
96
+ },
97
+ {
98
+ "epoch": 1.530541012216405,
99
+ "grad_norm": 0.566683292388916,
100
+ "learning_rate": 0.0012,
101
+ "loss": 6.513034057617188,
102
+ "step": 110
103
+ },
104
+ {
105
+ "epoch": 1.6701570680628273,
106
+ "grad_norm": 0.29915207624435425,
107
+ "learning_rate": 0.0012,
108
+ "loss": 6.323231887817383,
109
+ "step": 120
110
+ },
111
+ {
112
+ "epoch": 1.8097731239092496,
113
+ "grad_norm": 0.4436410665512085,
114
+ "learning_rate": 0.0012,
115
+ "loss": 6.200662994384766,
116
+ "step": 130
117
+ },
118
+ {
119
+ "epoch": 1.949389179755672,
120
+ "grad_norm": 0.3148385286331177,
121
+ "learning_rate": 0.0012,
122
+ "loss": 6.173930740356445,
123
+ "step": 140
124
+ },
125
+ {
126
+ "epoch": 2.0,
127
+ "eval_loss": 6.08213472366333,
128
+ "eval_runtime": 5.4403,
129
+ "eval_samples_per_second": 448.691,
130
+ "eval_steps_per_second": 7.169,
131
+ "step": 144
132
+ },
133
+ {
134
+ "epoch": 2.0837696335078535,
135
+ "grad_norm": 0.5374071002006531,
136
+ "learning_rate": 0.0012,
137
+ "loss": 6.055215835571289,
138
+ "step": 150
139
+ },
140
+ {
141
+ "epoch": 2.223385689354276,
142
+ "grad_norm": 0.2520872950553894,
143
+ "learning_rate": 0.0012,
144
+ "loss": 5.946698760986328,
145
+ "step": 160
146
+ },
147
+ {
148
+ "epoch": 2.363001745200698,
149
+ "grad_norm": 0.44598719477653503,
150
+ "learning_rate": 0.0012,
151
+ "loss": 5.896925735473633,
152
+ "step": 170
153
+ },
154
+ {
155
+ "epoch": 2.5026178010471205,
156
+ "grad_norm": 0.22086943686008453,
157
+ "learning_rate": 0.0012,
158
+ "loss": 5.833065795898437,
159
+ "step": 180
160
+ },
161
+ {
162
+ "epoch": 2.642233856893543,
163
+ "grad_norm": 0.599107563495636,
164
+ "learning_rate": 0.0012,
165
+ "loss": 5.761848068237304,
166
+ "step": 190
167
+ },
168
+ {
169
+ "epoch": 2.781849912739965,
170
+ "grad_norm": 0.2769874930381775,
171
+ "learning_rate": 0.0012,
172
+ "loss": 5.68456802368164,
173
+ "step": 200
174
+ },
175
+ {
176
+ "epoch": 2.9214659685863875,
177
+ "grad_norm": 0.7282533049583435,
178
+ "learning_rate": 0.0012,
179
+ "loss": 5.649662780761719,
180
+ "step": 210
181
+ },
182
+ {
183
+ "epoch": 3.0,
184
+ "eval_loss": 5.649172782897949,
185
+ "eval_runtime": 5.4657,
186
+ "eval_samples_per_second": 446.6,
187
+ "eval_steps_per_second": 7.135,
188
+ "step": 216
189
+ },
190
+ {
191
+ "epoch": 3.055846422338569,
192
+ "grad_norm": 0.28119179606437683,
193
+ "learning_rate": 0.0012,
194
+ "loss": 5.5973163604736325,
195
+ "step": 220
196
+ },
197
+ {
198
+ "epoch": 3.195462478184991,
199
+ "grad_norm": 0.1549271196126938,
200
+ "learning_rate": 0.0012,
201
+ "loss": 5.466296005249023,
202
+ "step": 230
203
+ },
204
+ {
205
+ "epoch": 3.3350785340314135,
206
+ "grad_norm": 0.6101787090301514,
207
+ "learning_rate": 0.0012,
208
+ "loss": 5.418326568603516,
209
+ "step": 240
210
+ },
211
+ {
212
+ "epoch": 3.474694589877836,
213
+ "grad_norm": 0.27038276195526123,
214
+ "learning_rate": 0.0012,
215
+ "loss": 5.406241989135742,
216
+ "step": 250
217
+ },
218
+ {
219
+ "epoch": 3.614310645724258,
220
+ "grad_norm": 0.20545881986618042,
221
+ "learning_rate": 0.0012,
222
+ "loss": 5.286199951171875,
223
+ "step": 260
224
+ },
225
+ {
226
+ "epoch": 3.7539267015706805,
227
+ "grad_norm": 0.5039545297622681,
228
+ "learning_rate": 0.0012,
229
+ "loss": 5.312792587280273,
230
+ "step": 270
231
+ },
232
+ {
233
+ "epoch": 3.893542757417103,
234
+ "grad_norm": 0.21515797078609467,
235
+ "learning_rate": 0.0012,
236
+ "loss": 5.265902709960938,
237
+ "step": 280
238
+ },
239
+ {
240
+ "epoch": 4.0,
241
+ "eval_loss": 5.197559356689453,
242
+ "eval_runtime": 5.439,
243
+ "eval_samples_per_second": 448.799,
244
+ "eval_steps_per_second": 7.17,
245
+ "step": 288
246
+ },
247
+ {
248
+ "epoch": 4.027923211169284,
249
+ "grad_norm": 0.31526851654052734,
250
+ "learning_rate": 0.0012,
251
+ "loss": 5.1200202941894535,
252
+ "step": 290
253
+ },
254
+ {
255
+ "epoch": 4.167539267015707,
256
+ "grad_norm": 0.40778130292892456,
257
+ "learning_rate": 0.0012,
258
+ "loss": 5.063777923583984,
259
+ "step": 300
260
+ },
261
+ {
262
+ "epoch": 4.307155322862129,
263
+ "grad_norm": 0.16718265414237976,
264
+ "learning_rate": 0.0012,
265
+ "loss": 4.9782966613769535,
266
+ "step": 310
267
+ },
268
+ {
269
+ "epoch": 4.446771378708552,
270
+ "grad_norm": 0.257915735244751,
271
+ "learning_rate": 0.0012,
272
+ "loss": 4.8924823760986325,
273
+ "step": 320
274
+ },
275
+ {
276
+ "epoch": 4.5863874345549736,
277
+ "grad_norm": 0.5843560099601746,
278
+ "learning_rate": 0.0012,
279
+ "loss": 4.866177368164062,
280
+ "step": 330
281
+ },
282
+ {
283
+ "epoch": 4.726003490401396,
284
+ "grad_norm": 0.27052536606788635,
285
+ "learning_rate": 0.0012,
286
+ "loss": 4.885505676269531,
287
+ "step": 340
288
+ },
289
+ {
290
+ "epoch": 4.865619546247818,
291
+ "grad_norm": 0.128607839345932,
292
+ "learning_rate": 0.0012,
293
+ "loss": 4.7557823181152346,
294
+ "step": 350
295
+ },
296
+ {
297
+ "epoch": 5.0,
298
+ "grad_norm": 0.3044080436229706,
299
+ "learning_rate": 0.0012,
300
+ "loss": 4.6753181457519535,
301
+ "step": 360
302
+ },
303
+ {
304
+ "epoch": 5.0,
305
+ "eval_loss": 4.7488484382629395,
306
+ "eval_runtime": 5.4378,
307
+ "eval_samples_per_second": 448.891,
308
+ "eval_steps_per_second": 7.172,
309
+ "step": 360
310
+ },
311
+ {
312
+ "epoch": 5.139616055846423,
313
+ "grad_norm": 0.49882882833480835,
314
+ "learning_rate": 0.0012,
315
+ "loss": 4.591573333740234,
316
+ "step": 370
317
+ },
318
+ {
319
+ "epoch": 5.279232111692845,
320
+ "grad_norm": 0.22627392411231995,
321
+ "learning_rate": 0.0012,
322
+ "loss": 4.567809677124023,
323
+ "step": 380
324
+ },
325
+ {
326
+ "epoch": 5.418848167539267,
327
+ "grad_norm": 0.17165814340114594,
328
+ "learning_rate": 0.0012,
329
+ "loss": 4.4708507537841795,
330
+ "step": 390
331
+ },
332
+ {
333
+ "epoch": 5.558464223385689,
334
+ "grad_norm": 0.21155276894569397,
335
+ "learning_rate": 0.0012,
336
+ "loss": 4.3975975036621096,
337
+ "step": 400
338
+ },
339
+ {
340
+ "epoch": 5.698080279232112,
341
+ "grad_norm": 0.29032468795776367,
342
+ "learning_rate": 0.0012,
343
+ "loss": 4.414227676391602,
344
+ "step": 410
345
+ },
346
+ {
347
+ "epoch": 5.837696335078534,
348
+ "grad_norm": 0.1890392005443573,
349
+ "learning_rate": 0.0012,
350
+ "loss": 4.374155807495117,
351
+ "step": 420
352
+ },
353
+ {
354
+ "epoch": 5.977312390924956,
355
+ "grad_norm": 0.17251808941364288,
356
+ "learning_rate": 0.0012,
357
+ "loss": 4.308379745483398,
358
+ "step": 430
359
+ },
360
+ {
361
+ "epoch": 6.0,
362
+ "eval_loss": 4.4369707107543945,
363
+ "eval_runtime": 5.4537,
364
+ "eval_samples_per_second": 447.585,
365
+ "eval_steps_per_second": 7.151,
366
+ "step": 432
367
+ },
368
+ {
369
+ "epoch": 6.111692844677138,
370
+ "grad_norm": 0.2730535566806793,
371
+ "learning_rate": 0.0012,
372
+ "loss": 4.220690536499023,
373
+ "step": 440
374
+ },
375
+ {
376
+ "epoch": 6.2513089005235605,
377
+ "grad_norm": 0.35956087708473206,
378
+ "learning_rate": 0.0012,
379
+ "loss": 4.216027450561524,
380
+ "step": 450
381
+ },
382
+ {
383
+ "epoch": 6.390924956369982,
384
+ "grad_norm": 0.17444954812526703,
385
+ "learning_rate": 0.0012,
386
+ "loss": 4.1690925598144535,
387
+ "step": 460
388
+ },
389
+ {
390
+ "epoch": 6.530541012216405,
391
+ "grad_norm": 0.14641860127449036,
392
+ "learning_rate": 0.0012,
393
+ "loss": 4.113532257080078,
394
+ "step": 470
395
+ },
396
+ {
397
+ "epoch": 6.670157068062827,
398
+ "grad_norm": 0.1628427356481552,
399
+ "learning_rate": 0.0012,
400
+ "loss": 4.076828384399414,
401
+ "step": 480
402
+ },
403
+ {
404
+ "epoch": 6.80977312390925,
405
+ "grad_norm": 0.18194392323493958,
406
+ "learning_rate": 0.0012,
407
+ "loss": 4.0723823547363285,
408
+ "step": 490
409
+ },
410
+ {
411
+ "epoch": 6.949389179755672,
412
+ "grad_norm": 0.18049317598342896,
413
+ "learning_rate": 0.0012,
414
+ "loss": 4.0408069610595705,
415
+ "step": 500
416
+ },
417
+ {
418
+ "epoch": 7.0,
419
+ "eval_loss": 4.229814052581787,
420
+ "eval_runtime": 5.4496,
421
+ "eval_samples_per_second": 447.926,
422
+ "eval_steps_per_second": 7.157,
423
+ "step": 504
424
+ },
425
+ {
426
+ "epoch": 7.0837696335078535,
427
+ "grad_norm": 0.33020293712615967,
428
+ "learning_rate": 0.0012,
429
+ "loss": 3.9618515014648437,
430
+ "step": 510
431
+ },
432
+ {
433
+ "epoch": 7.223385689354275,
434
+ "grad_norm": 0.25742554664611816,
435
+ "learning_rate": 0.0012,
436
+ "loss": 3.991061782836914,
437
+ "step": 520
438
+ },
439
+ {
440
+ "epoch": 7.363001745200698,
441
+ "grad_norm": 0.15890459716320038,
442
+ "learning_rate": 0.0012,
443
+ "loss": 3.946040725708008,
444
+ "step": 530
445
+ },
446
+ {
447
+ "epoch": 7.50261780104712,
448
+ "grad_norm": 0.11660423129796982,
449
+ "learning_rate": 0.0012,
450
+ "loss": 3.8959976196289063,
451
+ "step": 540
452
+ },
453
+ {
454
+ "epoch": 7.642233856893543,
455
+ "grad_norm": 0.11851505190134048,
456
+ "learning_rate": 0.0012,
457
+ "loss": 3.8700828552246094,
458
+ "step": 550
459
+ },
460
+ {
461
+ "epoch": 7.781849912739965,
462
+ "grad_norm": 0.24351678788661957,
463
+ "learning_rate": 0.0012,
464
+ "loss": 3.8552093505859375,
465
+ "step": 560
466
+ },
467
+ {
468
+ "epoch": 7.9214659685863875,
469
+ "grad_norm": 0.15510977804660797,
470
+ "learning_rate": 0.0012,
471
+ "loss": 3.8503486633300783,
472
+ "step": 570
473
+ },
474
+ {
475
+ "epoch": 8.0,
476
+ "eval_loss": 4.088932514190674,
477
+ "eval_runtime": 5.4365,
478
+ "eval_samples_per_second": 449.004,
479
+ "eval_steps_per_second": 7.174,
480
+ "step": 576
481
+ },
482
+ {
483
+ "epoch": 8.055846422338568,
484
+ "grad_norm": 0.27433669567108154,
485
+ "learning_rate": 0.0012,
486
+ "loss": 3.7618026733398438,
487
+ "step": 580
488
+ },
489
+ {
490
+ "epoch": 8.195462478184991,
491
+ "grad_norm": 0.3087034821510315,
492
+ "learning_rate": 0.0012,
493
+ "loss": 3.7158279418945312,
494
+ "step": 590
495
+ },
496
+ {
497
+ "epoch": 8.335078534031414,
498
+ "grad_norm": 0.19240885972976685,
499
+ "learning_rate": 0.0012,
500
+ "loss": 3.7282508850097655,
501
+ "step": 600
502
+ },
503
+ {
504
+ "epoch": 8.474694589877837,
505
+ "grad_norm": 0.12315202504396439,
506
+ "learning_rate": 0.0012,
507
+ "loss": 3.6883113861083983,
508
+ "step": 610
509
+ },
510
+ {
511
+ "epoch": 8.614310645724258,
512
+ "grad_norm": 0.134666308760643,
513
+ "learning_rate": 0.0012,
514
+ "loss": 3.6693538665771483,
515
+ "step": 620
516
+ },
517
+ {
518
+ "epoch": 8.75392670157068,
519
+ "grad_norm": 0.11143336445093155,
520
+ "learning_rate": 0.0012,
521
+ "loss": 3.6612457275390624,
522
+ "step": 630
523
+ },
524
+ {
525
+ "epoch": 8.893542757417103,
526
+ "grad_norm": 0.18448519706726074,
527
+ "learning_rate": 0.0012,
528
+ "loss": 3.672064208984375,
529
+ "step": 640
530
+ },
531
+ {
532
+ "epoch": 9.0,
533
+ "eval_loss": 4.005725860595703,
534
+ "eval_runtime": 5.4368,
535
+ "eval_samples_per_second": 448.974,
536
+ "eval_steps_per_second": 7.173,
537
+ "step": 648
538
+ },
539
+ {
540
+ "epoch": 9.027923211169284,
541
+ "grad_norm": 0.18339823186397552,
542
+ "learning_rate": 0.0012,
543
+ "loss": 3.6225852966308594,
544
+ "step": 650
545
+ },
546
+ {
547
+ "epoch": 9.167539267015707,
548
+ "grad_norm": 0.49299097061157227,
549
+ "learning_rate": 0.0012,
550
+ "loss": 3.556019592285156,
551
+ "step": 660
552
+ },
553
+ {
554
+ "epoch": 9.30715532286213,
555
+ "grad_norm": 1.2851951122283936,
556
+ "learning_rate": 0.0012,
557
+ "loss": 4.133036804199219,
558
+ "step": 670
559
+ },
560
+ {
561
+ "epoch": 9.44677137870855,
562
+ "grad_norm": 1.5363609790802002,
563
+ "learning_rate": 0.0012,
564
+ "loss": 5.060147857666015,
565
+ "step": 680
566
+ },
567
+ {
568
+ "epoch": 9.586387434554974,
569
+ "grad_norm": 0.5289552211761475,
570
+ "learning_rate": 0.0012,
571
+ "loss": 5.2387336730957035,
572
+ "step": 690
573
+ },
574
+ {
575
+ "epoch": 9.726003490401396,
576
+ "grad_norm": 0.1746184378862381,
577
+ "learning_rate": 0.0012,
578
+ "loss": 4.586937332153321,
579
+ "step": 700
580
+ },
581
+ {
582
+ "epoch": 9.86561954624782,
583
+ "grad_norm": 0.27418044209480286,
584
+ "learning_rate": 0.0012,
585
+ "loss": 4.650239181518555,
586
+ "step": 710
587
+ },
588
+ {
589
+ "epoch": 10.0,
590
+ "grad_norm": 0.15791937708854675,
591
+ "learning_rate": 0.0012,
592
+ "loss": 4.335153198242187,
593
+ "step": 720
594
+ },
595
+ {
596
+ "epoch": 10.0,
597
+ "eval_loss": 4.460577487945557,
598
+ "eval_runtime": 5.4539,
599
+ "eval_samples_per_second": 447.567,
600
+ "eval_steps_per_second": 7.151,
601
+ "step": 720
602
+ },
603
+ {
604
+ "epoch": 10.139616055846423,
605
+ "grad_norm": 0.1130480244755745,
606
+ "learning_rate": 0.0012,
607
+ "loss": 4.100911712646484,
608
+ "step": 730
609
+ },
610
+ {
611
+ "epoch": 10.279232111692846,
612
+ "grad_norm": 0.13805827498435974,
613
+ "learning_rate": 0.0012,
614
+ "loss": 3.986064147949219,
615
+ "step": 740
616
+ },
617
+ {
618
+ "epoch": 10.418848167539267,
619
+ "grad_norm": 0.12214994430541992,
620
+ "learning_rate": 0.0012,
621
+ "loss": 3.9097930908203127,
622
+ "step": 750
623
+ },
624
+ {
625
+ "epoch": 10.55846422338569,
626
+ "grad_norm": 0.10231558978557587,
627
+ "learning_rate": 0.0012,
628
+ "loss": 3.8575531005859376,
629
+ "step": 760
630
+ },
631
+ {
632
+ "epoch": 10.698080279232112,
633
+ "grad_norm": 0.2263040393590927,
634
+ "learning_rate": 0.0012,
635
+ "loss": 3.8134490966796877,
636
+ "step": 770
637
+ },
638
+ {
639
+ "epoch": 10.837696335078533,
640
+ "grad_norm": 0.11031533032655716,
641
+ "learning_rate": 0.0012,
642
+ "loss": 3.78343505859375,
643
+ "step": 780
644
+ },
645
+ {
646
+ "epoch": 10.977312390924956,
647
+ "grad_norm": 0.10530456155538559,
648
+ "learning_rate": 0.0012,
649
+ "loss": 3.733928680419922,
650
+ "step": 790
651
+ },
652
+ {
653
+ "epoch": 11.0,
654
+ "eval_loss": 4.046730041503906,
655
+ "eval_runtime": 5.4443,
656
+ "eval_samples_per_second": 448.359,
657
+ "eval_steps_per_second": 7.163,
658
+ "step": 792
659
+ },
660
+ {
661
+ "epoch": 11.111692844677139,
662
+ "grad_norm": 0.2062959522008896,
663
+ "learning_rate": 0.0012,
664
+ "loss": 3.6298500061035157,
665
+ "step": 800
666
+ },
667
+ {
668
+ "epoch": 11.25130890052356,
669
+ "grad_norm": 0.11892785876989365,
670
+ "learning_rate": 0.0012,
671
+ "loss": 3.5670894622802733,
672
+ "step": 810
673
+ },
674
+ {
675
+ "epoch": 11.390924956369982,
676
+ "grad_norm": 0.09811964631080627,
677
+ "learning_rate": 0.0012,
678
+ "loss": 3.5630123138427736,
679
+ "step": 820
680
+ },
681
+ {
682
+ "epoch": 11.530541012216405,
683
+ "grad_norm": 0.09262749552726746,
684
+ "learning_rate": 0.0012,
685
+ "loss": 3.5660240173339846,
686
+ "step": 830
687
+ },
688
+ {
689
+ "epoch": 11.670157068062828,
690
+ "grad_norm": 0.10465727001428604,
691
+ "learning_rate": 0.0012,
692
+ "loss": 3.560059356689453,
693
+ "step": 840
694
+ },
695
+ {
696
+ "epoch": 11.809773123909249,
697
+ "grad_norm": 0.11480753123760223,
698
+ "learning_rate": 0.0012,
699
+ "loss": 3.555225372314453,
700
+ "step": 850
701
+ },
702
+ {
703
+ "epoch": 11.949389179755672,
704
+ "grad_norm": 0.09398435801267624,
705
+ "learning_rate": 0.0012,
706
+ "loss": 3.5535049438476562,
707
+ "step": 860
708
+ },
709
+ {
710
+ "epoch": 12.0,
711
+ "eval_loss": 3.960569381713867,
712
+ "eval_runtime": 5.433,
713
+ "eval_samples_per_second": 449.291,
714
+ "eval_steps_per_second": 7.178,
715
+ "step": 864
716
+ }
717
+ ],
718
+ "logging_steps": 10,
719
+ "max_steps": 864,
720
+ "num_input_tokens_seen": 0,
721
+ "num_train_epochs": 12,
722
+ "save_steps": 500,
723
+ "stateful_callbacks": {
724
+ "TrainerControl": {
725
+ "args": {
726
+ "should_epoch_stop": false,
727
+ "should_evaluate": false,
728
+ "should_log": false,
729
+ "should_save": false,
730
+ "should_training_stop": true
731
+ },
732
+ "attributes": {}
733
+ }
734
+ },
735
+ "total_flos": 1.6673046802726912e+18,
736
+ "train_batch_size": 8,
737
+ "trial_name": null,
738
+ "trial_params": null
739
+ }
runs/t0/t0_S4_lr0.0012/train.log ADDED
The diff for this file is too large to render. See raw diff
 
runs/t0/t0_S5_lr0.0003/checkpoint-720/chat_template.jinja ADDED
@@ -0,0 +1,85 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {%- if tools %}
2
+ {{- '<|im_start|>system\n' }}
3
+ {%- if messages[0].role == 'system' %}
4
+ {{- messages[0].content + '\n\n' }}
5
+ {%- endif %}
6
+ {{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
7
+ {%- for tool in tools %}
8
+ {{- "\n" }}
9
+ {{- tool | tojson }}
10
+ {%- endfor %}
11
+ {{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
12
+ {%- else %}
13
+ {%- if messages[0].role == 'system' %}
14
+ {{- '<|im_start|>system\n' + messages[0].content + '<|im_end|>\n' }}
15
+ {%- endif %}
16
+ {%- endif %}
17
+ {%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
18
+ {%- for message in messages[::-1] %}
19
+ {%- set index = (messages|length - 1) - loop.index0 %}
20
+ {%- if ns.multi_step_tool and message.role == "user" and not(message.content.startswith('<tool_response>') and message.content.endswith('</tool_response>')) %}
21
+ {%- set ns.multi_step_tool = false %}
22
+ {%- set ns.last_query_index = index %}
23
+ {%- endif %}
24
+ {%- endfor %}
25
+ {%- for message in messages %}
26
+ {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
27
+ {{- '<|im_start|>' + message.role + '\n' + message.content + '<|im_end|>' + '\n' }}
28
+ {%- elif message.role == "assistant" %}
29
+ {%- set content = message.content %}
30
+ {%- set reasoning_content = '' %}
31
+ {%- if message.reasoning_content is defined and message.reasoning_content is not none %}
32
+ {%- set reasoning_content = message.reasoning_content %}
33
+ {%- else %}
34
+ {%- if '</think>' in message.content %}
35
+ {%- set content = message.content.split('</think>')[-1].lstrip('\n') %}
36
+ {%- set reasoning_content = message.content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
37
+ {%- endif %}
38
+ {%- endif %}
39
+ {%- if loop.index0 > ns.last_query_index %}
40
+ {%- if loop.last or (not loop.last and reasoning_content) %}
41
+ {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
42
+ {%- else %}
43
+ {{- '<|im_start|>' + message.role + '\n' + content }}
44
+ {%- endif %}
45
+ {%- else %}
46
+ {{- '<|im_start|>' + message.role + '\n' + content }}
47
+ {%- endif %}
48
+ {%- if message.tool_calls %}
49
+ {%- for tool_call in message.tool_calls %}
50
+ {%- if (loop.first and content) or (not loop.first) %}
51
+ {{- '\n' }}
52
+ {%- endif %}
53
+ {%- if tool_call.function %}
54
+ {%- set tool_call = tool_call.function %}
55
+ {%- endif %}
56
+ {{- '<tool_call>\n{"name": "' }}
57
+ {{- tool_call.name }}
58
+ {{- '", "arguments": ' }}
59
+ {%- if tool_call.arguments is string %}
60
+ {{- tool_call.arguments }}
61
+ {%- else %}
62
+ {{- tool_call.arguments | tojson }}
63
+ {%- endif %}
64
+ {{- '}\n</tool_call>' }}
65
+ {%- endfor %}
66
+ {%- endif %}
67
+ {{- '<|im_end|>\n' }}
68
+ {%- elif message.role == "tool" %}
69
+ {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
70
+ {{- '<|im_start|>user' }}
71
+ {%- endif %}
72
+ {{- '\n<tool_response>\n' }}
73
+ {{- message.content }}
74
+ {{- '\n</tool_response>' }}
75
+ {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
76
+ {{- '<|im_end|>\n' }}
77
+ {%- endif %}
78
+ {%- endif %}
79
+ {%- endfor %}
80
+ {%- if add_generation_prompt %}
81
+ {{- '<|im_start|>assistant\n' }}
82
+ {%- if enable_thinking is defined and enable_thinking is false %}
83
+ {{- '<think>\n\n</think>\n\n' }}
84
+ {%- endif %}
85
+ {%- endif %}
runs/t0/t0_S5_lr0.0003/checkpoint-720/config.json ADDED
@@ -0,0 +1,32 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "LlamaForCausalLM"
4
+ ],
5
+ "attention_bias": false,
6
+ "attention_dropout": 0.0,
7
+ "bos_token_id": null,
8
+ "dtype": "float32",
9
+ "eos_token_id": 151645,
10
+ "head_dim": 128,
11
+ "hidden_act": "silu",
12
+ "hidden_size": 1536,
13
+ "initializer_range": 0.02,
14
+ "intermediate_size": 4096,
15
+ "max_position_embeddings": 2048,
16
+ "mlp_bias": false,
17
+ "model_type": "llama",
18
+ "num_attention_heads": 12,
19
+ "num_hidden_layers": 24,
20
+ "num_key_value_heads": 12,
21
+ "pad_token_id": 151645,
22
+ "pretraining_tp": 1,
23
+ "rms_norm_eps": 1e-06,
24
+ "rope_parameters": {
25
+ "rope_theta": 10000.0,
26
+ "rope_type": "default"
27
+ },
28
+ "tie_word_embeddings": true,
29
+ "transformers_version": "5.15.0",
30
+ "use_cache": false,
31
+ "vocab_size": 151671
32
+ }
runs/t0/t0_S5_lr0.0003/checkpoint-720/generation_config.json ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "eos_token_id": [
4
+ 151645
5
+ ],
6
+ "output_attentions": false,
7
+ "output_hidden_states": false,
8
+ "pad_token_id": 151645,
9
+ "transformers_version": "5.15.0",
10
+ "use_cache": true
11
+ }
runs/t0/t0_S5_lr0.0003/checkpoint-720/tokenizer_config.json ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "backend": "tokenizers",
4
+ "bos_token": null,
5
+ "clean_up_tokenization_spaces": false,
6
+ "eos_token": "<|im_end|>",
7
+ "errors": "replace",
8
+ "extra_special_tokens": [
9
+ "<|l2r_pred|>",
10
+ "<|r2l_pred|>"
11
+ ],
12
+ "is_local": false,
13
+ "local_files_only": false,
14
+ "model_max_length": 131072,
15
+ "pad_token": "<|im_end|>",
16
+ "split_special_tokens": false,
17
+ "tokenizer_class": "Qwen2Tokenizer",
18
+ "unk_token": null
19
+ }
runs/t0/t0_S5_lr0.0003/checkpoint-720/trainer_state.json ADDED
@@ -0,0 +1,625 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 10.0,
6
+ "eval_steps": 72,
7
+ "global_step": 720,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "epoch": 0.013973799126637555,
14
+ "grad_norm": 3.2618095874786377,
15
+ "learning_rate": 0.0,
16
+ "loss": 12.214674949645996,
17
+ "step": 1
18
+ },
19
+ {
20
+ "epoch": 0.13973799126637554,
21
+ "grad_norm": 2.4582679271698,
22
+ "learning_rate": 2.6999999999999996e-05,
23
+ "loss": 11.440606011284721,
24
+ "step": 10
25
+ },
26
+ {
27
+ "epoch": 0.2794759825327511,
28
+ "grad_norm": 1.7859525680541992,
29
+ "learning_rate": 5.6999999999999996e-05,
30
+ "loss": 9.868122863769532,
31
+ "step": 20
32
+ },
33
+ {
34
+ "epoch": 0.4192139737991266,
35
+ "grad_norm": 1.0912766456604004,
36
+ "learning_rate": 8.699999999999999e-05,
37
+ "loss": 8.853494262695312,
38
+ "step": 30
39
+ },
40
+ {
41
+ "epoch": 0.5589519650655022,
42
+ "grad_norm": 0.8212459683418274,
43
+ "learning_rate": 0.000117,
44
+ "loss": 7.825148010253907,
45
+ "step": 40
46
+ },
47
+ {
48
+ "epoch": 0.6986899563318777,
49
+ "grad_norm": 1.0021402835845947,
50
+ "learning_rate": 0.000147,
51
+ "loss": 7.4507301330566404,
52
+ "step": 50
53
+ },
54
+ {
55
+ "epoch": 0.8384279475982532,
56
+ "grad_norm": 0.9833669066429138,
57
+ "learning_rate": 0.00017699999999999997,
58
+ "loss": 7.263099670410156,
59
+ "step": 60
60
+ },
61
+ {
62
+ "epoch": 0.9781659388646288,
63
+ "grad_norm": 1.023337483406067,
64
+ "learning_rate": 0.00020699999999999996,
65
+ "loss": 7.0315185546875,
66
+ "step": 70
67
+ },
68
+ {
69
+ "epoch": 1.0,
70
+ "eval_loss": 6.922646999359131,
71
+ "eval_runtime": 32.1357,
72
+ "eval_samples_per_second": 75.959,
73
+ "eval_steps_per_second": 2.396,
74
+ "step": 72
75
+ },
76
+ {
77
+ "epoch": 1.1117903930131003,
78
+ "grad_norm": 1.203994631767273,
79
+ "learning_rate": 0.000237,
80
+ "loss": 6.820037841796875,
81
+ "step": 80
82
+ },
83
+ {
84
+ "epoch": 1.251528384279476,
85
+ "grad_norm": 1.0269978046417236,
86
+ "learning_rate": 0.000267,
87
+ "loss": 6.610933685302735,
88
+ "step": 90
89
+ },
90
+ {
91
+ "epoch": 1.3912663755458516,
92
+ "grad_norm": 0.7635436058044434,
93
+ "learning_rate": 0.00029699999999999996,
94
+ "loss": 6.47060546875,
95
+ "step": 100
96
+ },
97
+ {
98
+ "epoch": 1.531004366812227,
99
+ "grad_norm": 0.9571962952613831,
100
+ "learning_rate": 0.0003,
101
+ "loss": 6.356604766845703,
102
+ "step": 110
103
+ },
104
+ {
105
+ "epoch": 1.6707423580786025,
106
+ "grad_norm": 0.47234103083610535,
107
+ "learning_rate": 0.0003,
108
+ "loss": 6.199187469482422,
109
+ "step": 120
110
+ },
111
+ {
112
+ "epoch": 1.8104803493449781,
113
+ "grad_norm": 0.6224037408828735,
114
+ "learning_rate": 0.0003,
115
+ "loss": 6.105295562744141,
116
+ "step": 130
117
+ },
118
+ {
119
+ "epoch": 1.9502183406113538,
120
+ "grad_norm": 0.7163454294204712,
121
+ "learning_rate": 0.0003,
122
+ "loss": 6.005741119384766,
123
+ "step": 140
124
+ },
125
+ {
126
+ "epoch": 2.0,
127
+ "eval_loss": 5.9517598152160645,
128
+ "eval_runtime": 7.9148,
129
+ "eval_samples_per_second": 308.408,
130
+ "eval_steps_per_second": 9.729,
131
+ "step": 144
132
+ },
133
+ {
134
+ "epoch": 2.0838427947598253,
135
+ "grad_norm": 0.5856307744979858,
136
+ "learning_rate": 0.0003,
137
+ "loss": 5.904288482666016,
138
+ "step": 150
139
+ },
140
+ {
141
+ "epoch": 2.2235807860262007,
142
+ "grad_norm": 0.6896644234657288,
143
+ "learning_rate": 0.0003,
144
+ "loss": 5.783623123168946,
145
+ "step": 160
146
+ },
147
+ {
148
+ "epoch": 2.3633187772925766,
149
+ "grad_norm": 1.0071693658828735,
150
+ "learning_rate": 0.0003,
151
+ "loss": 5.70606575012207,
152
+ "step": 170
153
+ },
154
+ {
155
+ "epoch": 2.503056768558952,
156
+ "grad_norm": 0.8469743728637695,
157
+ "learning_rate": 0.0003,
158
+ "loss": 5.638371276855469,
159
+ "step": 180
160
+ },
161
+ {
162
+ "epoch": 2.6427947598253274,
163
+ "grad_norm": 0.4349379241466522,
164
+ "learning_rate": 0.0003,
165
+ "loss": 5.536750793457031,
166
+ "step": 190
167
+ },
168
+ {
169
+ "epoch": 2.7825327510917033,
170
+ "grad_norm": 0.838222324848175,
171
+ "learning_rate": 0.0003,
172
+ "loss": 5.431597518920898,
173
+ "step": 200
174
+ },
175
+ {
176
+ "epoch": 2.9222707423580787,
177
+ "grad_norm": 0.76561439037323,
178
+ "learning_rate": 0.0003,
179
+ "loss": 5.4217266082763675,
180
+ "step": 210
181
+ },
182
+ {
183
+ "epoch": 3.0,
184
+ "eval_loss": 5.362564563751221,
185
+ "eval_runtime": 7.9146,
186
+ "eval_samples_per_second": 308.419,
187
+ "eval_steps_per_second": 9.729,
188
+ "step": 216
189
+ },
190
+ {
191
+ "epoch": 3.05589519650655,
192
+ "grad_norm": 0.4211660921573639,
193
+ "learning_rate": 0.0003,
194
+ "loss": 5.302315521240234,
195
+ "step": 220
196
+ },
197
+ {
198
+ "epoch": 3.1956331877729256,
199
+ "grad_norm": 0.32755231857299805,
200
+ "learning_rate": 0.0003,
201
+ "loss": 5.182415008544922,
202
+ "step": 230
203
+ },
204
+ {
205
+ "epoch": 3.3353711790393015,
206
+ "grad_norm": 1.0064932107925415,
207
+ "learning_rate": 0.0003,
208
+ "loss": 5.131160736083984,
209
+ "step": 240
210
+ },
211
+ {
212
+ "epoch": 3.475109170305677,
213
+ "grad_norm": 0.39928022027015686,
214
+ "learning_rate": 0.0003,
215
+ "loss": 5.130668258666992,
216
+ "step": 250
217
+ },
218
+ {
219
+ "epoch": 3.6148471615720523,
220
+ "grad_norm": 0.24958354234695435,
221
+ "learning_rate": 0.0003,
222
+ "loss": 5.007334136962891,
223
+ "step": 260
224
+ },
225
+ {
226
+ "epoch": 3.7545851528384278,
227
+ "grad_norm": 0.624622106552124,
228
+ "learning_rate": 0.0003,
229
+ "loss": 4.949510192871093,
230
+ "step": 270
231
+ },
232
+ {
233
+ "epoch": 3.8943231441048036,
234
+ "grad_norm": 1.041094422340393,
235
+ "learning_rate": 0.0003,
236
+ "loss": 4.952127075195312,
237
+ "step": 280
238
+ },
239
+ {
240
+ "epoch": 4.0,
241
+ "eval_loss": 5.030168056488037,
242
+ "eval_runtime": 7.9196,
243
+ "eval_samples_per_second": 308.221,
244
+ "eval_steps_per_second": 9.723,
245
+ "step": 288
246
+ },
247
+ {
248
+ "epoch": 4.0279475982532755,
249
+ "grad_norm": 0.591354250907898,
250
+ "learning_rate": 0.0003,
251
+ "loss": 4.971962356567383,
252
+ "step": 290
253
+ },
254
+ {
255
+ "epoch": 4.1676855895196505,
256
+ "grad_norm": 0.23682217299938202,
257
+ "learning_rate": 0.0003,
258
+ "loss": 4.798876953125,
259
+ "step": 300
260
+ },
261
+ {
262
+ "epoch": 4.307423580786026,
263
+ "grad_norm": 0.2569108009338379,
264
+ "learning_rate": 0.0003,
265
+ "loss": 4.705326843261719,
266
+ "step": 310
267
+ },
268
+ {
269
+ "epoch": 4.447161572052401,
270
+ "grad_norm": 0.46430492401123047,
271
+ "learning_rate": 0.0003,
272
+ "loss": 4.6480354309082035,
273
+ "step": 320
274
+ },
275
+ {
276
+ "epoch": 4.586899563318777,
277
+ "grad_norm": 0.5473710894584656,
278
+ "learning_rate": 0.0003,
279
+ "loss": 4.645205688476563,
280
+ "step": 330
281
+ },
282
+ {
283
+ "epoch": 4.726637554585153,
284
+ "grad_norm": 0.36366525292396545,
285
+ "learning_rate": 0.0003,
286
+ "loss": 4.587823486328125,
287
+ "step": 340
288
+ },
289
+ {
290
+ "epoch": 4.866375545851528,
291
+ "grad_norm": 0.39248207211494446,
292
+ "learning_rate": 0.0003,
293
+ "loss": 4.511217498779297,
294
+ "step": 350
295
+ },
296
+ {
297
+ "epoch": 5.0,
298
+ "grad_norm": 0.4145806133747101,
299
+ "learning_rate": 0.0003,
300
+ "loss": 4.47677001953125,
301
+ "step": 360
302
+ },
303
+ {
304
+ "epoch": 5.0,
305
+ "eval_loss": 4.589375972747803,
306
+ "eval_runtime": 7.9133,
307
+ "eval_samples_per_second": 308.469,
308
+ "eval_steps_per_second": 9.73,
309
+ "step": 360
310
+ },
311
+ {
312
+ "epoch": 5.139737991266376,
313
+ "grad_norm": 0.5570223927497864,
314
+ "learning_rate": 0.0003,
315
+ "loss": 4.355342864990234,
316
+ "step": 370
317
+ },
318
+ {
319
+ "epoch": 5.279475982532751,
320
+ "grad_norm": 0.6994290351867676,
321
+ "learning_rate": 0.0003,
322
+ "loss": 4.36234130859375,
323
+ "step": 380
324
+ },
325
+ {
326
+ "epoch": 5.419213973799127,
327
+ "grad_norm": 0.5207724571228027,
328
+ "learning_rate": 0.0003,
329
+ "loss": 4.337489318847656,
330
+ "step": 390
331
+ },
332
+ {
333
+ "epoch": 5.558951965065502,
334
+ "grad_norm": 0.23361016809940338,
335
+ "learning_rate": 0.0003,
336
+ "loss": 4.260697937011718,
337
+ "step": 400
338
+ },
339
+ {
340
+ "epoch": 5.698689956331878,
341
+ "grad_norm": 0.267966091632843,
342
+ "learning_rate": 0.0003,
343
+ "loss": 4.221217346191406,
344
+ "step": 410
345
+ },
346
+ {
347
+ "epoch": 5.8384279475982535,
348
+ "grad_norm": 0.3150366544723511,
349
+ "learning_rate": 0.0003,
350
+ "loss": 4.19616470336914,
351
+ "step": 420
352
+ },
353
+ {
354
+ "epoch": 5.978165938864628,
355
+ "grad_norm": 0.35929128527641296,
356
+ "learning_rate": 0.0003,
357
+ "loss": 4.170941925048828,
358
+ "step": 430
359
+ },
360
+ {
361
+ "epoch": 6.0,
362
+ "eval_loss": 4.342624187469482,
363
+ "eval_runtime": 7.9157,
364
+ "eval_samples_per_second": 308.376,
365
+ "eval_steps_per_second": 9.728,
366
+ "step": 432
367
+ },
368
+ {
369
+ "epoch": 6.1117903930131,
370
+ "grad_norm": 0.8794248700141907,
371
+ "learning_rate": 0.0003,
372
+ "loss": 4.087378692626953,
373
+ "step": 440
374
+ },
375
+ {
376
+ "epoch": 6.251528384279476,
377
+ "grad_norm": 0.7270575165748596,
378
+ "learning_rate": 0.0003,
379
+ "loss": 4.194872283935547,
380
+ "step": 450
381
+ },
382
+ {
383
+ "epoch": 6.391266375545851,
384
+ "grad_norm": 1.0183628797531128,
385
+ "learning_rate": 0.0003,
386
+ "loss": 4.237209320068359,
387
+ "step": 460
388
+ },
389
+ {
390
+ "epoch": 6.531004366812227,
391
+ "grad_norm": 0.31741416454315186,
392
+ "learning_rate": 0.0003,
393
+ "loss": 4.177013778686524,
394
+ "step": 470
395
+ },
396
+ {
397
+ "epoch": 6.670742358078603,
398
+ "grad_norm": 0.1903829276561737,
399
+ "learning_rate": 0.0003,
400
+ "loss": 4.066764831542969,
401
+ "step": 480
402
+ },
403
+ {
404
+ "epoch": 6.810480349344978,
405
+ "grad_norm": 0.17531758546829224,
406
+ "learning_rate": 0.0003,
407
+ "loss": 4.023290634155273,
408
+ "step": 490
409
+ },
410
+ {
411
+ "epoch": 6.950218340611354,
412
+ "grad_norm": 0.2047526091337204,
413
+ "learning_rate": 0.0003,
414
+ "loss": 3.9738792419433593,
415
+ "step": 500
416
+ },
417
+ {
418
+ "epoch": 7.0,
419
+ "eval_loss": 4.2179388999938965,
420
+ "eval_runtime": 7.9122,
421
+ "eval_samples_per_second": 308.509,
422
+ "eval_steps_per_second": 9.732,
423
+ "step": 504
424
+ },
425
+ {
426
+ "epoch": 7.083842794759826,
427
+ "grad_norm": 0.22795313596725464,
428
+ "learning_rate": 0.0003,
429
+ "loss": 3.88709716796875,
430
+ "step": 510
431
+ },
432
+ {
433
+ "epoch": 7.223580786026201,
434
+ "grad_norm": 0.43237408995628357,
435
+ "learning_rate": 0.0003,
436
+ "loss": 3.8286155700683593,
437
+ "step": 520
438
+ },
439
+ {
440
+ "epoch": 7.3633187772925766,
441
+ "grad_norm": 0.2796283960342407,
442
+ "learning_rate": 0.0003,
443
+ "loss": 3.820693588256836,
444
+ "step": 530
445
+ },
446
+ {
447
+ "epoch": 7.503056768558952,
448
+ "grad_norm": 0.2597295939922333,
449
+ "learning_rate": 0.0003,
450
+ "loss": 3.8072124481201173,
451
+ "step": 540
452
+ },
453
+ {
454
+ "epoch": 7.642794759825327,
455
+ "grad_norm": 0.2075442373752594,
456
+ "learning_rate": 0.0003,
457
+ "loss": 3.7986984252929688,
458
+ "step": 550
459
+ },
460
+ {
461
+ "epoch": 7.782532751091703,
462
+ "grad_norm": 0.2210642695426941,
463
+ "learning_rate": 0.0003,
464
+ "loss": 3.789369583129883,
465
+ "step": 560
466
+ },
467
+ {
468
+ "epoch": 7.922270742358078,
469
+ "grad_norm": 0.2791687548160553,
470
+ "learning_rate": 0.0003,
471
+ "loss": 3.78194580078125,
472
+ "step": 570
473
+ },
474
+ {
475
+ "epoch": 8.0,
476
+ "eval_loss": 4.121953010559082,
477
+ "eval_runtime": 7.9121,
478
+ "eval_samples_per_second": 308.516,
479
+ "eval_steps_per_second": 9.732,
480
+ "step": 576
481
+ },
482
+ {
483
+ "epoch": 8.055895196506551,
484
+ "grad_norm": 0.6974738836288452,
485
+ "learning_rate": 0.0003,
486
+ "loss": 3.7090476989746093,
487
+ "step": 580
488
+ },
489
+ {
490
+ "epoch": 8.195633187772925,
491
+ "grad_norm": 0.8548513650894165,
492
+ "learning_rate": 0.0003,
493
+ "loss": 3.745708465576172,
494
+ "step": 590
495
+ },
496
+ {
497
+ "epoch": 8.335371179039301,
498
+ "grad_norm": 0.8406635522842407,
499
+ "learning_rate": 0.0003,
500
+ "loss": 3.828490447998047,
501
+ "step": 600
502
+ },
503
+ {
504
+ "epoch": 8.475109170305677,
505
+ "grad_norm": 0.43541979789733887,
506
+ "learning_rate": 0.0003,
507
+ "loss": 3.771062469482422,
508
+ "step": 610
509
+ },
510
+ {
511
+ "epoch": 8.614847161572053,
512
+ "grad_norm": 0.22296181321144104,
513
+ "learning_rate": 0.0003,
514
+ "loss": 3.6843185424804688,
515
+ "step": 620
516
+ },
517
+ {
518
+ "epoch": 8.754585152838429,
519
+ "grad_norm": 0.2070133239030838,
520
+ "learning_rate": 0.0003,
521
+ "loss": 3.6420669555664062,
522
+ "step": 630
523
+ },
524
+ {
525
+ "epoch": 8.894323144104803,
526
+ "grad_norm": 0.20196694135665894,
527
+ "learning_rate": 0.0003,
528
+ "loss": 3.6326332092285156,
529
+ "step": 640
530
+ },
531
+ {
532
+ "epoch": 9.0,
533
+ "eval_loss": 4.078823566436768,
534
+ "eval_runtime": 7.9041,
535
+ "eval_samples_per_second": 308.828,
536
+ "eval_steps_per_second": 9.742,
537
+ "step": 648
538
+ },
539
+ {
540
+ "epoch": 9.027947598253276,
541
+ "grad_norm": 0.3818998336791992,
542
+ "learning_rate": 0.0003,
543
+ "loss": 3.5768112182617187,
544
+ "step": 650
545
+ },
546
+ {
547
+ "epoch": 9.167685589519651,
548
+ "grad_norm": 0.6515381932258606,
549
+ "learning_rate": 0.0003,
550
+ "loss": 3.441845703125,
551
+ "step": 660
552
+ },
553
+ {
554
+ "epoch": 9.307423580786025,
555
+ "grad_norm": 0.41331514716148376,
556
+ "learning_rate": 0.0003,
557
+ "loss": 3.4659652709960938,
558
+ "step": 670
559
+ },
560
+ {
561
+ "epoch": 9.447161572052401,
562
+ "grad_norm": 0.3088191747665405,
563
+ "learning_rate": 0.0003,
564
+ "loss": 3.4546302795410155,
565
+ "step": 680
566
+ },
567
+ {
568
+ "epoch": 9.586899563318777,
569
+ "grad_norm": 0.2400057464838028,
570
+ "learning_rate": 0.0003,
571
+ "loss": 3.431840515136719,
572
+ "step": 690
573
+ },
574
+ {
575
+ "epoch": 9.726637554585153,
576
+ "grad_norm": 0.26047268509864807,
577
+ "learning_rate": 0.0003,
578
+ "loss": 3.4436744689941405,
579
+ "step": 700
580
+ },
581
+ {
582
+ "epoch": 9.866375545851529,
583
+ "grad_norm": 0.24156703054904938,
584
+ "learning_rate": 0.0003,
585
+ "loss": 3.435311508178711,
586
+ "step": 710
587
+ },
588
+ {
589
+ "epoch": 10.0,
590
+ "grad_norm": 0.290781170129776,
591
+ "learning_rate": 0.0003,
592
+ "loss": 3.428261566162109,
593
+ "step": 720
594
+ },
595
+ {
596
+ "epoch": 10.0,
597
+ "eval_loss": 4.061178207397461,
598
+ "eval_runtime": 7.9111,
599
+ "eval_samples_per_second": 308.552,
600
+ "eval_steps_per_second": 9.733,
601
+ "step": 720
602
+ }
603
+ ],
604
+ "logging_steps": 10,
605
+ "max_steps": 864,
606
+ "num_input_tokens_seen": 0,
607
+ "num_train_epochs": 12,
608
+ "save_steps": 500,
609
+ "stateful_callbacks": {
610
+ "TrainerControl": {
611
+ "args": {
612
+ "should_epoch_stop": false,
613
+ "should_evaluate": false,
614
+ "should_log": false,
615
+ "should_save": false,
616
+ "should_training_stop": false
617
+ },
618
+ "attributes": {}
619
+ }
620
+ },
621
+ "total_flos": 3.059564909802029e+18,
622
+ "train_batch_size": 4,
623
+ "trial_name": null,
624
+ "trial_params": null
625
+ }
runs/t0/t0_S5_lr0.0003/train.log ADDED
The diff for this file is too large to render. See raw diff
 
runs/t0/t0_S5_lr0.0012/checkpoint-720/chat_template.jinja ADDED
@@ -0,0 +1,85 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {%- if tools %}
2
+ {{- '<|im_start|>system\n' }}
3
+ {%- if messages[0].role == 'system' %}
4
+ {{- messages[0].content + '\n\n' }}
5
+ {%- endif %}
6
+ {{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
7
+ {%- for tool in tools %}
8
+ {{- "\n" }}
9
+ {{- tool | tojson }}
10
+ {%- endfor %}
11
+ {{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
12
+ {%- else %}
13
+ {%- if messages[0].role == 'system' %}
14
+ {{- '<|im_start|>system\n' + messages[0].content + '<|im_end|>\n' }}
15
+ {%- endif %}
16
+ {%- endif %}
17
+ {%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
18
+ {%- for message in messages[::-1] %}
19
+ {%- set index = (messages|length - 1) - loop.index0 %}
20
+ {%- if ns.multi_step_tool and message.role == "user" and not(message.content.startswith('<tool_response>') and message.content.endswith('</tool_response>')) %}
21
+ {%- set ns.multi_step_tool = false %}
22
+ {%- set ns.last_query_index = index %}
23
+ {%- endif %}
24
+ {%- endfor %}
25
+ {%- for message in messages %}
26
+ {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
27
+ {{- '<|im_start|>' + message.role + '\n' + message.content + '<|im_end|>' + '\n' }}
28
+ {%- elif message.role == "assistant" %}
29
+ {%- set content = message.content %}
30
+ {%- set reasoning_content = '' %}
31
+ {%- if message.reasoning_content is defined and message.reasoning_content is not none %}
32
+ {%- set reasoning_content = message.reasoning_content %}
33
+ {%- else %}
34
+ {%- if '</think>' in message.content %}
35
+ {%- set content = message.content.split('</think>')[-1].lstrip('\n') %}
36
+ {%- set reasoning_content = message.content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
37
+ {%- endif %}
38
+ {%- endif %}
39
+ {%- if loop.index0 > ns.last_query_index %}
40
+ {%- if loop.last or (not loop.last and reasoning_content) %}
41
+ {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
42
+ {%- else %}
43
+ {{- '<|im_start|>' + message.role + '\n' + content }}
44
+ {%- endif %}
45
+ {%- else %}
46
+ {{- '<|im_start|>' + message.role + '\n' + content }}
47
+ {%- endif %}
48
+ {%- if message.tool_calls %}
49
+ {%- for tool_call in message.tool_calls %}
50
+ {%- if (loop.first and content) or (not loop.first) %}
51
+ {{- '\n' }}
52
+ {%- endif %}
53
+ {%- if tool_call.function %}
54
+ {%- set tool_call = tool_call.function %}
55
+ {%- endif %}
56
+ {{- '<tool_call>\n{"name": "' }}
57
+ {{- tool_call.name }}
58
+ {{- '", "arguments": ' }}
59
+ {%- if tool_call.arguments is string %}
60
+ {{- tool_call.arguments }}
61
+ {%- else %}
62
+ {{- tool_call.arguments | tojson }}
63
+ {%- endif %}
64
+ {{- '}\n</tool_call>' }}
65
+ {%- endfor %}
66
+ {%- endif %}
67
+ {{- '<|im_end|>\n' }}
68
+ {%- elif message.role == "tool" %}
69
+ {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
70
+ {{- '<|im_start|>user' }}
71
+ {%- endif %}
72
+ {{- '\n<tool_response>\n' }}
73
+ {{- message.content }}
74
+ {{- '\n</tool_response>' }}
75
+ {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
76
+ {{- '<|im_end|>\n' }}
77
+ {%- endif %}
78
+ {%- endif %}
79
+ {%- endfor %}
80
+ {%- if add_generation_prompt %}
81
+ {{- '<|im_start|>assistant\n' }}
82
+ {%- if enable_thinking is defined and enable_thinking is false %}
83
+ {{- '<think>\n\n</think>\n\n' }}
84
+ {%- endif %}
85
+ {%- endif %}
runs/t0/t0_S5_lr0.0012/checkpoint-720/config.json ADDED
@@ -0,0 +1,32 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "LlamaForCausalLM"
4
+ ],
5
+ "attention_bias": false,
6
+ "attention_dropout": 0.0,
7
+ "bos_token_id": null,
8
+ "dtype": "float32",
9
+ "eos_token_id": 151645,
10
+ "head_dim": 128,
11
+ "hidden_act": "silu",
12
+ "hidden_size": 1536,
13
+ "initializer_range": 0.02,
14
+ "intermediate_size": 4096,
15
+ "max_position_embeddings": 2048,
16
+ "mlp_bias": false,
17
+ "model_type": "llama",
18
+ "num_attention_heads": 12,
19
+ "num_hidden_layers": 24,
20
+ "num_key_value_heads": 12,
21
+ "pad_token_id": 151645,
22
+ "pretraining_tp": 1,
23
+ "rms_norm_eps": 1e-06,
24
+ "rope_parameters": {
25
+ "rope_theta": 10000.0,
26
+ "rope_type": "default"
27
+ },
28
+ "tie_word_embeddings": true,
29
+ "transformers_version": "5.15.0",
30
+ "use_cache": false,
31
+ "vocab_size": 151671
32
+ }
runs/t0/t0_S5_lr0.0012/checkpoint-720/generation_config.json ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "eos_token_id": [
4
+ 151645
5
+ ],
6
+ "output_attentions": false,
7
+ "output_hidden_states": false,
8
+ "pad_token_id": 151645,
9
+ "transformers_version": "5.15.0",
10
+ "use_cache": true
11
+ }
runs/t0/t0_S5_lr0.0012/checkpoint-720/tokenizer_config.json ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "backend": "tokenizers",
4
+ "bos_token": null,
5
+ "clean_up_tokenization_spaces": false,
6
+ "eos_token": "<|im_end|>",
7
+ "errors": "replace",
8
+ "extra_special_tokens": [
9
+ "<|l2r_pred|>",
10
+ "<|r2l_pred|>"
11
+ ],
12
+ "is_local": false,
13
+ "local_files_only": false,
14
+ "model_max_length": 131072,
15
+ "pad_token": "<|im_end|>",
16
+ "split_special_tokens": false,
17
+ "tokenizer_class": "Qwen2Tokenizer",
18
+ "unk_token": null
19
+ }
runs/t0/t0_S5_lr0.0012/checkpoint-720/trainer_state.json ADDED
@@ -0,0 +1,625 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 10.0,
6
+ "eval_steps": 72,
7
+ "global_step": 720,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "epoch": 0.013973799126637555,
14
+ "grad_norm": 3.2854130268096924,
15
+ "learning_rate": 0.0,
16
+ "loss": 12.239278793334961,
17
+ "step": 1
18
+ },
19
+ {
20
+ "epoch": 0.13973799126637554,
21
+ "grad_norm": 2.8914339542388916,
22
+ "learning_rate": 0.00010799999999999998,
23
+ "loss": 10.772604200575087,
24
+ "step": 10
25
+ },
26
+ {
27
+ "epoch": 0.2794759825327511,
28
+ "grad_norm": 0.8951712846755981,
29
+ "learning_rate": 0.00022799999999999999,
30
+ "loss": 8.708420562744141,
31
+ "step": 20
32
+ },
33
+ {
34
+ "epoch": 0.4192139737991266,
35
+ "grad_norm": 0.8144592642784119,
36
+ "learning_rate": 0.00034799999999999995,
37
+ "loss": 7.609435272216797,
38
+ "step": 30
39
+ },
40
+ {
41
+ "epoch": 0.5589519650655022,
42
+ "grad_norm": 0.7392408847808838,
43
+ "learning_rate": 0.000468,
44
+ "loss": 7.431675720214844,
45
+ "step": 40
46
+ },
47
+ {
48
+ "epoch": 0.6986899563318777,
49
+ "grad_norm": 1.823486089706421,
50
+ "learning_rate": 0.000588,
51
+ "loss": 7.193846130371094,
52
+ "step": 50
53
+ },
54
+ {
55
+ "epoch": 0.8384279475982532,
56
+ "grad_norm": 1.0612823963165283,
57
+ "learning_rate": 0.0007079999999999999,
58
+ "loss": 6.98529052734375,
59
+ "step": 60
60
+ },
61
+ {
62
+ "epoch": 0.9781659388646288,
63
+ "grad_norm": 0.8120266795158386,
64
+ "learning_rate": 0.0008279999999999999,
65
+ "loss": 6.868883514404297,
66
+ "step": 70
67
+ },
68
+ {
69
+ "epoch": 1.0,
70
+ "eval_loss": 6.711178302764893,
71
+ "eval_runtime": 32.1685,
72
+ "eval_samples_per_second": 75.882,
73
+ "eval_steps_per_second": 2.394,
74
+ "step": 72
75
+ },
76
+ {
77
+ "epoch": 1.1117903930131003,
78
+ "grad_norm": 0.6914841532707214,
79
+ "learning_rate": 0.000948,
80
+ "loss": 6.636103820800781,
81
+ "step": 80
82
+ },
83
+ {
84
+ "epoch": 1.251528384279476,
85
+ "grad_norm": 0.780213475227356,
86
+ "learning_rate": 0.001068,
87
+ "loss": 6.488282775878906,
88
+ "step": 90
89
+ },
90
+ {
91
+ "epoch": 1.3912663755458516,
92
+ "grad_norm": 0.9557600021362305,
93
+ "learning_rate": 0.0011879999999999998,
94
+ "loss": 6.354541778564453,
95
+ "step": 100
96
+ },
97
+ {
98
+ "epoch": 1.531004366812227,
99
+ "grad_norm": 0.7227393984794617,
100
+ "learning_rate": 0.0012,
101
+ "loss": 6.275666046142578,
102
+ "step": 110
103
+ },
104
+ {
105
+ "epoch": 1.6707423580786025,
106
+ "grad_norm": 2.0571846961975098,
107
+ "learning_rate": 0.0012,
108
+ "loss": 6.149044036865234,
109
+ "step": 120
110
+ },
111
+ {
112
+ "epoch": 1.8104803493449781,
113
+ "grad_norm": 0.3692125380039215,
114
+ "learning_rate": 0.0012,
115
+ "loss": 6.129460144042969,
116
+ "step": 130
117
+ },
118
+ {
119
+ "epoch": 1.9502183406113538,
120
+ "grad_norm": 0.35232868790626526,
121
+ "learning_rate": 0.0012,
122
+ "loss": 5.982979583740234,
123
+ "step": 140
124
+ },
125
+ {
126
+ "epoch": 2.0,
127
+ "eval_loss": 6.093167304992676,
128
+ "eval_runtime": 7.9136,
129
+ "eval_samples_per_second": 308.458,
130
+ "eval_steps_per_second": 9.73,
131
+ "step": 144
132
+ },
133
+ {
134
+ "epoch": 2.0838427947598253,
135
+ "grad_norm": 0.40879613161087036,
136
+ "learning_rate": 0.0012,
137
+ "loss": 5.902141571044922,
138
+ "step": 150
139
+ },
140
+ {
141
+ "epoch": 2.2235807860262007,
142
+ "grad_norm": 0.44311410188674927,
143
+ "learning_rate": 0.0012,
144
+ "loss": 5.77630844116211,
145
+ "step": 160
146
+ },
147
+ {
148
+ "epoch": 2.3633187772925766,
149
+ "grad_norm": 0.5734341144561768,
150
+ "learning_rate": 0.0012,
151
+ "loss": 5.730416107177734,
152
+ "step": 170
153
+ },
154
+ {
155
+ "epoch": 2.503056768558952,
156
+ "grad_norm": 0.3401224613189697,
157
+ "learning_rate": 0.0012,
158
+ "loss": 5.651076126098633,
159
+ "step": 180
160
+ },
161
+ {
162
+ "epoch": 2.6427947598253274,
163
+ "grad_norm": 0.7364341020584106,
164
+ "learning_rate": 0.0012,
165
+ "loss": 5.60278549194336,
166
+ "step": 190
167
+ },
168
+ {
169
+ "epoch": 2.7825327510917033,
170
+ "grad_norm": 0.27776798605918884,
171
+ "learning_rate": 0.0012,
172
+ "loss": 5.503947830200195,
173
+ "step": 200
174
+ },
175
+ {
176
+ "epoch": 2.9222707423580787,
177
+ "grad_norm": 0.5733747482299805,
178
+ "learning_rate": 0.0012,
179
+ "loss": 5.419222259521485,
180
+ "step": 210
181
+ },
182
+ {
183
+ "epoch": 3.0,
184
+ "eval_loss": 5.410129070281982,
185
+ "eval_runtime": 7.9145,
186
+ "eval_samples_per_second": 308.42,
187
+ "eval_steps_per_second": 9.729,
188
+ "step": 216
189
+ },
190
+ {
191
+ "epoch": 3.05589519650655,
192
+ "grad_norm": 0.22739177942276,
193
+ "learning_rate": 0.0012,
194
+ "loss": 5.340640640258789,
195
+ "step": 220
196
+ },
197
+ {
198
+ "epoch": 3.1956331877729256,
199
+ "grad_norm": 0.3654351830482483,
200
+ "learning_rate": 0.0012,
201
+ "loss": 5.246836090087891,
202
+ "step": 230
203
+ },
204
+ {
205
+ "epoch": 3.3353711790393015,
206
+ "grad_norm": 0.30522432923316956,
207
+ "learning_rate": 0.0012,
208
+ "loss": 5.183464813232422,
209
+ "step": 240
210
+ },
211
+ {
212
+ "epoch": 3.475109170305677,
213
+ "grad_norm": 0.7293229103088379,
214
+ "learning_rate": 0.0012,
215
+ "loss": 5.146945571899414,
216
+ "step": 250
217
+ },
218
+ {
219
+ "epoch": 3.6148471615720523,
220
+ "grad_norm": 0.27008751034736633,
221
+ "learning_rate": 0.0012,
222
+ "loss": 5.120269012451172,
223
+ "step": 260
224
+ },
225
+ {
226
+ "epoch": 3.7545851528384278,
227
+ "grad_norm": 0.25352606177330017,
228
+ "learning_rate": 0.0012,
229
+ "loss": 4.999597549438477,
230
+ "step": 270
231
+ },
232
+ {
233
+ "epoch": 3.8943231441048036,
234
+ "grad_norm": 0.622519314289093,
235
+ "learning_rate": 0.0012,
236
+ "loss": 4.967436218261719,
237
+ "step": 280
238
+ },
239
+ {
240
+ "epoch": 4.0,
241
+ "eval_loss": 5.000211238861084,
242
+ "eval_runtime": 7.9158,
243
+ "eval_samples_per_second": 308.369,
244
+ "eval_steps_per_second": 9.727,
245
+ "step": 288
246
+ },
247
+ {
248
+ "epoch": 4.0279475982532755,
249
+ "grad_norm": 0.33289751410484314,
250
+ "learning_rate": 0.0012,
251
+ "loss": 4.923779678344727,
252
+ "step": 290
253
+ },
254
+ {
255
+ "epoch": 4.1676855895196505,
256
+ "grad_norm": 0.2035096287727356,
257
+ "learning_rate": 0.0012,
258
+ "loss": 4.7559326171875,
259
+ "step": 300
260
+ },
261
+ {
262
+ "epoch": 4.307423580786026,
263
+ "grad_norm": 0.4597044587135315,
264
+ "learning_rate": 0.0012,
265
+ "loss": 4.681830596923828,
266
+ "step": 310
267
+ },
268
+ {
269
+ "epoch": 4.447161572052401,
270
+ "grad_norm": 0.39535853266716003,
271
+ "learning_rate": 0.0012,
272
+ "loss": 4.705263900756836,
273
+ "step": 320
274
+ },
275
+ {
276
+ "epoch": 4.586899563318777,
277
+ "grad_norm": 0.18191099166870117,
278
+ "learning_rate": 0.0012,
279
+ "loss": 4.623991394042969,
280
+ "step": 330
281
+ },
282
+ {
283
+ "epoch": 4.726637554585153,
284
+ "grad_norm": 0.3703513741493225,
285
+ "learning_rate": 0.0012,
286
+ "loss": 4.537522506713867,
287
+ "step": 340
288
+ },
289
+ {
290
+ "epoch": 4.866375545851528,
291
+ "grad_norm": 0.4233785569667816,
292
+ "learning_rate": 0.0012,
293
+ "loss": 4.576454162597656,
294
+ "step": 350
295
+ },
296
+ {
297
+ "epoch": 5.0,
298
+ "grad_norm": 0.17962193489074707,
299
+ "learning_rate": 0.0012,
300
+ "loss": 4.482110214233399,
301
+ "step": 360
302
+ },
303
+ {
304
+ "epoch": 5.0,
305
+ "eval_loss": 4.558745384216309,
306
+ "eval_runtime": 7.9108,
307
+ "eval_samples_per_second": 308.566,
308
+ "eval_steps_per_second": 9.734,
309
+ "step": 360
310
+ },
311
+ {
312
+ "epoch": 5.139737991266376,
313
+ "grad_norm": 0.33794084191322327,
314
+ "learning_rate": 0.0012,
315
+ "loss": 4.3163707733154295,
316
+ "step": 370
317
+ },
318
+ {
319
+ "epoch": 5.279475982532751,
320
+ "grad_norm": 0.21855460107326508,
321
+ "learning_rate": 0.0012,
322
+ "loss": 4.3009895324707035,
323
+ "step": 380
324
+ },
325
+ {
326
+ "epoch": 5.419213973799127,
327
+ "grad_norm": 0.15166562795639038,
328
+ "learning_rate": 0.0012,
329
+ "loss": 4.227708435058593,
330
+ "step": 390
331
+ },
332
+ {
333
+ "epoch": 5.558951965065502,
334
+ "grad_norm": 0.29830625653266907,
335
+ "learning_rate": 0.0012,
336
+ "loss": 4.207986068725586,
337
+ "step": 400
338
+ },
339
+ {
340
+ "epoch": 5.698689956331878,
341
+ "grad_norm": 0.24197085201740265,
342
+ "learning_rate": 0.0012,
343
+ "loss": 4.186326599121093,
344
+ "step": 410
345
+ },
346
+ {
347
+ "epoch": 5.8384279475982535,
348
+ "grad_norm": 0.17020633816719055,
349
+ "learning_rate": 0.0012,
350
+ "loss": 4.153177642822266,
351
+ "step": 420
352
+ },
353
+ {
354
+ "epoch": 5.978165938864628,
355
+ "grad_norm": 0.23463435471057892,
356
+ "learning_rate": 0.0012,
357
+ "loss": 4.124895477294922,
358
+ "step": 430
359
+ },
360
+ {
361
+ "epoch": 6.0,
362
+ "eval_loss": 4.291956424713135,
363
+ "eval_runtime": 7.9141,
364
+ "eval_samples_per_second": 308.435,
365
+ "eval_steps_per_second": 9.729,
366
+ "step": 432
367
+ },
368
+ {
369
+ "epoch": 6.1117903930131,
370
+ "grad_norm": 0.33610767126083374,
371
+ "learning_rate": 0.0012,
372
+ "loss": 4.008880615234375,
373
+ "step": 440
374
+ },
375
+ {
376
+ "epoch": 6.251528384279476,
377
+ "grad_norm": 0.2952611744403839,
378
+ "learning_rate": 0.0012,
379
+ "loss": 4.0130470275878904,
380
+ "step": 450
381
+ },
382
+ {
383
+ "epoch": 6.391266375545851,
384
+ "grad_norm": 0.1474909633398056,
385
+ "learning_rate": 0.0012,
386
+ "loss": 3.9605411529541015,
387
+ "step": 460
388
+ },
389
+ {
390
+ "epoch": 6.531004366812227,
391
+ "grad_norm": 0.17840559780597687,
392
+ "learning_rate": 0.0012,
393
+ "loss": 3.9215927124023438,
394
+ "step": 470
395
+ },
396
+ {
397
+ "epoch": 6.670742358078603,
398
+ "grad_norm": 0.22309419512748718,
399
+ "learning_rate": 0.0012,
400
+ "loss": 3.9179962158203123,
401
+ "step": 480
402
+ },
403
+ {
404
+ "epoch": 6.810480349344978,
405
+ "grad_norm": 0.17490209639072418,
406
+ "learning_rate": 0.0012,
407
+ "loss": 3.9208335876464844,
408
+ "step": 490
409
+ },
410
+ {
411
+ "epoch": 6.950218340611354,
412
+ "grad_norm": 0.16962862014770508,
413
+ "learning_rate": 0.0012,
414
+ "loss": 3.8781494140625,
415
+ "step": 500
416
+ },
417
+ {
418
+ "epoch": 7.0,
419
+ "eval_loss": 4.123973369598389,
420
+ "eval_runtime": 7.9124,
421
+ "eval_samples_per_second": 308.502,
422
+ "eval_steps_per_second": 9.732,
423
+ "step": 504
424
+ },
425
+ {
426
+ "epoch": 7.083842794759826,
427
+ "grad_norm": 0.2539425194263458,
428
+ "learning_rate": 0.0012,
429
+ "loss": 3.77655029296875,
430
+ "step": 510
431
+ },
432
+ {
433
+ "epoch": 7.223580786026201,
434
+ "grad_norm": 0.26208364963531494,
435
+ "learning_rate": 0.0012,
436
+ "loss": 3.745014953613281,
437
+ "step": 520
438
+ },
439
+ {
440
+ "epoch": 7.3633187772925766,
441
+ "grad_norm": 0.1574983447790146,
442
+ "learning_rate": 0.0012,
443
+ "loss": 3.7169628143310547,
444
+ "step": 530
445
+ },
446
+ {
447
+ "epoch": 7.503056768558952,
448
+ "grad_norm": 0.18709805607795715,
449
+ "learning_rate": 0.0012,
450
+ "loss": 3.7050544738769533,
451
+ "step": 540
452
+ },
453
+ {
454
+ "epoch": 7.642794759825327,
455
+ "grad_norm": 0.14072749018669128,
456
+ "learning_rate": 0.0012,
457
+ "loss": 3.6973304748535156,
458
+ "step": 550
459
+ },
460
+ {
461
+ "epoch": 7.782532751091703,
462
+ "grad_norm": 0.13279055058956146,
463
+ "learning_rate": 0.0012,
464
+ "loss": 3.686246871948242,
465
+ "step": 560
466
+ },
467
+ {
468
+ "epoch": 7.922270742358078,
469
+ "grad_norm": 0.2193097621202469,
470
+ "learning_rate": 0.0012,
471
+ "loss": 3.6956031799316404,
472
+ "step": 570
473
+ },
474
+ {
475
+ "epoch": 8.0,
476
+ "eval_loss": 4.036094665527344,
477
+ "eval_runtime": 7.9031,
478
+ "eval_samples_per_second": 308.867,
479
+ "eval_steps_per_second": 9.743,
480
+ "step": 576
481
+ },
482
+ {
483
+ "epoch": 8.055895196506551,
484
+ "grad_norm": 0.31165802478790283,
485
+ "learning_rate": 0.0012,
486
+ "loss": 3.6125083923339845,
487
+ "step": 580
488
+ },
489
+ {
490
+ "epoch": 8.195633187772925,
491
+ "grad_norm": 0.4581586420536041,
492
+ "learning_rate": 0.0012,
493
+ "loss": 3.610798645019531,
494
+ "step": 590
495
+ },
496
+ {
497
+ "epoch": 8.335371179039301,
498
+ "grad_norm": 0.28279274702072144,
499
+ "learning_rate": 0.0012,
500
+ "loss": 3.702981948852539,
501
+ "step": 600
502
+ },
503
+ {
504
+ "epoch": 8.475109170305677,
505
+ "grad_norm": 0.15088902413845062,
506
+ "learning_rate": 0.0012,
507
+ "loss": 3.6176616668701174,
508
+ "step": 610
509
+ },
510
+ {
511
+ "epoch": 8.614847161572053,
512
+ "grad_norm": 0.10666224360466003,
513
+ "learning_rate": 0.0012,
514
+ "loss": 3.558070755004883,
515
+ "step": 620
516
+ },
517
+ {
518
+ "epoch": 8.754585152838429,
519
+ "grad_norm": 0.11449340730905533,
520
+ "learning_rate": 0.0012,
521
+ "loss": 3.536176300048828,
522
+ "step": 630
523
+ },
524
+ {
525
+ "epoch": 8.894323144104803,
526
+ "grad_norm": 0.11072063446044922,
527
+ "learning_rate": 0.0012,
528
+ "loss": 3.5396492004394533,
529
+ "step": 640
530
+ },
531
+ {
532
+ "epoch": 9.0,
533
+ "eval_loss": 3.974152088165283,
534
+ "eval_runtime": 7.9166,
535
+ "eval_samples_per_second": 308.338,
536
+ "eval_steps_per_second": 9.726,
537
+ "step": 648
538
+ },
539
+ {
540
+ "epoch": 9.027947598253276,
541
+ "grad_norm": 0.20924493670463562,
542
+ "learning_rate": 0.0012,
543
+ "loss": 3.4809524536132814,
544
+ "step": 650
545
+ },
546
+ {
547
+ "epoch": 9.167685589519651,
548
+ "grad_norm": 0.27390941977500916,
549
+ "learning_rate": 0.0012,
550
+ "loss": 3.3156246185302733,
551
+ "step": 660
552
+ },
553
+ {
554
+ "epoch": 9.307423580786025,
555
+ "grad_norm": 0.20567722618579865,
556
+ "learning_rate": 0.0012,
557
+ "loss": 3.3422481536865236,
558
+ "step": 670
559
+ },
560
+ {
561
+ "epoch": 9.447161572052401,
562
+ "grad_norm": 0.1480301320552826,
563
+ "learning_rate": 0.0012,
564
+ "loss": 3.34552001953125,
565
+ "step": 680
566
+ },
567
+ {
568
+ "epoch": 9.586899563318777,
569
+ "grad_norm": 0.11175384372472763,
570
+ "learning_rate": 0.0012,
571
+ "loss": 3.3335189819335938,
572
+ "step": 690
573
+ },
574
+ {
575
+ "epoch": 9.726637554585153,
576
+ "grad_norm": 0.11760509014129639,
577
+ "learning_rate": 0.0012,
578
+ "loss": 3.355607604980469,
579
+ "step": 700
580
+ },
581
+ {
582
+ "epoch": 9.866375545851529,
583
+ "grad_norm": 0.11177218705415726,
584
+ "learning_rate": 0.0012,
585
+ "loss": 3.355641174316406,
586
+ "step": 710
587
+ },
588
+ {
589
+ "epoch": 10.0,
590
+ "grad_norm": 0.13623695075511932,
591
+ "learning_rate": 0.0012,
592
+ "loss": 3.355918121337891,
593
+ "step": 720
594
+ },
595
+ {
596
+ "epoch": 10.0,
597
+ "eval_loss": 3.9539742469787598,
598
+ "eval_runtime": 7.9182,
599
+ "eval_samples_per_second": 308.278,
600
+ "eval_steps_per_second": 9.724,
601
+ "step": 720
602
+ }
603
+ ],
604
+ "logging_steps": 10,
605
+ "max_steps": 864,
606
+ "num_input_tokens_seen": 0,
607
+ "num_train_epochs": 12,
608
+ "save_steps": 500,
609
+ "stateful_callbacks": {
610
+ "TrainerControl": {
611
+ "args": {
612
+ "should_epoch_stop": false,
613
+ "should_evaluate": false,
614
+ "should_log": false,
615
+ "should_save": false,
616
+ "should_training_stop": false
617
+ },
618
+ "attributes": {}
619
+ }
620
+ },
621
+ "total_flos": 3.059564909802029e+18,
622
+ "train_batch_size": 4,
623
+ "trial_name": null,
624
+ "trial_params": null
625
+ }
runs/t0/t0_S5_lr0.0012/train.log ADDED
The diff for this file is too large to render. See raw diff
 
runs/t2a/t2a_150M_C0_baseline/checkpoint-2232/chat_template.jinja ADDED
@@ -0,0 +1,85 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {%- if tools %}
2
+ {{- '<|im_start|>system\n' }}
3
+ {%- if messages[0].role == 'system' %}
4
+ {{- messages[0].content + '\n\n' }}
5
+ {%- endif %}
6
+ {{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
7
+ {%- for tool in tools %}
8
+ {{- "\n" }}
9
+ {{- tool | tojson }}
10
+ {%- endfor %}
11
+ {{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
12
+ {%- else %}
13
+ {%- if messages[0].role == 'system' %}
14
+ {{- '<|im_start|>system\n' + messages[0].content + '<|im_end|>\n' }}
15
+ {%- endif %}
16
+ {%- endif %}
17
+ {%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
18
+ {%- for message in messages[::-1] %}
19
+ {%- set index = (messages|length - 1) - loop.index0 %}
20
+ {%- if ns.multi_step_tool and message.role == "user" and not(message.content.startswith('<tool_response>') and message.content.endswith('</tool_response>')) %}
21
+ {%- set ns.multi_step_tool = false %}
22
+ {%- set ns.last_query_index = index %}
23
+ {%- endif %}
24
+ {%- endfor %}
25
+ {%- for message in messages %}
26
+ {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
27
+ {{- '<|im_start|>' + message.role + '\n' + message.content + '<|im_end|>' + '\n' }}
28
+ {%- elif message.role == "assistant" %}
29
+ {%- set content = message.content %}
30
+ {%- set reasoning_content = '' %}
31
+ {%- if message.reasoning_content is defined and message.reasoning_content is not none %}
32
+ {%- set reasoning_content = message.reasoning_content %}
33
+ {%- else %}
34
+ {%- if '</think>' in message.content %}
35
+ {%- set content = message.content.split('</think>')[-1].lstrip('\n') %}
36
+ {%- set reasoning_content = message.content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
37
+ {%- endif %}
38
+ {%- endif %}
39
+ {%- if loop.index0 > ns.last_query_index %}
40
+ {%- if loop.last or (not loop.last and reasoning_content) %}
41
+ {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
42
+ {%- else %}
43
+ {{- '<|im_start|>' + message.role + '\n' + content }}
44
+ {%- endif %}
45
+ {%- else %}
46
+ {{- '<|im_start|>' + message.role + '\n' + content }}
47
+ {%- endif %}
48
+ {%- if message.tool_calls %}
49
+ {%- for tool_call in message.tool_calls %}
50
+ {%- if (loop.first and content) or (not loop.first) %}
51
+ {{- '\n' }}
52
+ {%- endif %}
53
+ {%- if tool_call.function %}
54
+ {%- set tool_call = tool_call.function %}
55
+ {%- endif %}
56
+ {{- '<tool_call>\n{"name": "' }}
57
+ {{- tool_call.name }}
58
+ {{- '", "arguments": ' }}
59
+ {%- if tool_call.arguments is string %}
60
+ {{- tool_call.arguments }}
61
+ {%- else %}
62
+ {{- tool_call.arguments | tojson }}
63
+ {%- endif %}
64
+ {{- '}\n</tool_call>' }}
65
+ {%- endfor %}
66
+ {%- endif %}
67
+ {{- '<|im_end|>\n' }}
68
+ {%- elif message.role == "tool" %}
69
+ {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
70
+ {{- '<|im_start|>user' }}
71
+ {%- endif %}
72
+ {{- '\n<tool_response>\n' }}
73
+ {{- message.content }}
74
+ {{- '\n</tool_response>' }}
75
+ {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
76
+ {{- '<|im_end|>\n' }}
77
+ {%- endif %}
78
+ {%- endif %}
79
+ {%- endfor %}
80
+ {%- if add_generation_prompt %}
81
+ {{- '<|im_start|>assistant\n' }}
82
+ {%- if enable_thinking is defined and enable_thinking is false %}
83
+ {{- '<think>\n\n</think>\n\n' }}
84
+ {%- endif %}
85
+ {%- endif %}
runs/t2a/t2a_150M_C0_baseline/checkpoint-2232/config.json ADDED
@@ -0,0 +1,32 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "LlamaForCausalLM"
4
+ ],
5
+ "attention_bias": false,
6
+ "attention_dropout": 0.0,
7
+ "bos_token_id": null,
8
+ "dtype": "float32",
9
+ "eos_token_id": 151645,
10
+ "head_dim": 128,
11
+ "hidden_act": "silu",
12
+ "hidden_size": 512,
13
+ "initializer_range": 0.02,
14
+ "intermediate_size": 1536,
15
+ "max_position_embeddings": 2048,
16
+ "mlp_bias": false,
17
+ "model_type": "llama",
18
+ "num_attention_heads": 4,
19
+ "num_hidden_layers": 20,
20
+ "num_key_value_heads": 4,
21
+ "pad_token_id": 151645,
22
+ "pretraining_tp": 1,
23
+ "rms_norm_eps": 1e-06,
24
+ "rope_parameters": {
25
+ "rope_theta": 10000.0,
26
+ "rope_type": "default"
27
+ },
28
+ "tie_word_embeddings": true,
29
+ "transformers_version": "5.15.0",
30
+ "use_cache": false,
31
+ "vocab_size": 151671
32
+ }
runs/t2a/t2a_150M_C0_baseline/checkpoint-2232/generation_config.json ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "eos_token_id": [
4
+ 151645
5
+ ],
6
+ "output_attentions": false,
7
+ "output_hidden_states": false,
8
+ "pad_token_id": 151645,
9
+ "transformers_version": "5.15.0",
10
+ "use_cache": true
11
+ }
runs/t2a/t2a_150M_C0_baseline/checkpoint-2232/tokenizer_config.json ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "backend": "tokenizers",
4
+ "bos_token": null,
5
+ "clean_up_tokenization_spaces": false,
6
+ "eos_token": "<|im_end|>",
7
+ "errors": "replace",
8
+ "extra_special_tokens": [
9
+ "<|l2r_pred|>",
10
+ "<|r2l_pred|>"
11
+ ],
12
+ "is_local": false,
13
+ "local_files_only": false,
14
+ "model_max_length": 131072,
15
+ "pad_token": "<|im_end|>",
16
+ "split_special_tokens": false,
17
+ "tokenizer_class": "Qwen2Tokenizer",
18
+ "unk_token": null
19
+ }
runs/t2a/t2a_150M_C0_baseline/checkpoint-2232/trainer_state.json ADDED
@@ -0,0 +1,1850 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 15.609457092819614,
6
+ "eval_steps": 72,
7
+ "global_step": 2232,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "epoch": 0.0070052539404553416,
14
+ "grad_norm": 1.7456934452056885,
15
+ "learning_rate": 0.0,
16
+ "loss": 12.037310600280762,
17
+ "step": 1
18
+ },
19
+ {
20
+ "epoch": 0.07005253940455342,
21
+ "grad_norm": 1.7848070859909058,
22
+ "learning_rate": 5.399999999999999e-05,
23
+ "loss": 11.75309583875868,
24
+ "step": 10
25
+ },
26
+ {
27
+ "epoch": 0.14010507880910683,
28
+ "grad_norm": 1.6792213916778564,
29
+ "learning_rate": 0.00011399999999999999,
30
+ "loss": 11.042081451416015,
31
+ "step": 20
32
+ },
33
+ {
34
+ "epoch": 0.21015761821366025,
35
+ "grad_norm": 1.6428947448730469,
36
+ "learning_rate": 0.00017399999999999997,
37
+ "loss": 10.273934173583985,
38
+ "step": 30
39
+ },
40
+ {
41
+ "epoch": 0.28021015761821366,
42
+ "grad_norm": 1.5020601749420166,
43
+ "learning_rate": 0.000234,
44
+ "loss": 9.222610473632812,
45
+ "step": 40
46
+ },
47
+ {
48
+ "epoch": 0.3502626970227671,
49
+ "grad_norm": 0.9085643887519836,
50
+ "learning_rate": 0.000294,
51
+ "loss": 8.245561218261718,
52
+ "step": 50
53
+ },
54
+ {
55
+ "epoch": 0.4203152364273205,
56
+ "grad_norm": 0.4325217604637146,
57
+ "learning_rate": 0.00035399999999999993,
58
+ "loss": 7.676107788085938,
59
+ "step": 60
60
+ },
61
+ {
62
+ "epoch": 0.4903677758318739,
63
+ "grad_norm": 0.4937305450439453,
64
+ "learning_rate": 0.0004139999999999999,
65
+ "loss": 7.524372100830078,
66
+ "step": 70
67
+ },
68
+ {
69
+ "epoch": 0.5043782837127846,
70
+ "eval_loss": 7.435024261474609,
71
+ "eval_runtime": 27.6967,
72
+ "eval_samples_per_second": 88.133,
73
+ "eval_steps_per_second": 1.408,
74
+ "step": 72
75
+ },
76
+ {
77
+ "epoch": 0.5604203152364273,
78
+ "grad_norm": 0.42533794045448303,
79
+ "learning_rate": 0.000474,
80
+ "loss": 7.372554016113281,
81
+ "step": 80
82
+ },
83
+ {
84
+ "epoch": 0.6304728546409807,
85
+ "grad_norm": 0.8951543569564819,
86
+ "learning_rate": 0.000534,
87
+ "loss": 7.21534423828125,
88
+ "step": 90
89
+ },
90
+ {
91
+ "epoch": 0.7005253940455342,
92
+ "grad_norm": 0.5405721664428711,
93
+ "learning_rate": 0.0005939999999999999,
94
+ "loss": 7.055476379394531,
95
+ "step": 100
96
+ },
97
+ {
98
+ "epoch": 0.7705779334500875,
99
+ "grad_norm": 0.4531368315219879,
100
+ "learning_rate": 0.0006,
101
+ "loss": 6.896892547607422,
102
+ "step": 110
103
+ },
104
+ {
105
+ "epoch": 0.840630472854641,
106
+ "grad_norm": 0.3535282611846924,
107
+ "learning_rate": 0.0006,
108
+ "loss": 6.763951110839844,
109
+ "step": 120
110
+ },
111
+ {
112
+ "epoch": 0.9106830122591943,
113
+ "grad_norm": 0.772462785243988,
114
+ "learning_rate": 0.0006,
115
+ "loss": 6.64959487915039,
116
+ "step": 130
117
+ },
118
+ {
119
+ "epoch": 0.9807355516637478,
120
+ "grad_norm": 0.6359318494796753,
121
+ "learning_rate": 0.0006,
122
+ "loss": 6.573255157470703,
123
+ "step": 140
124
+ },
125
+ {
126
+ "epoch": 1.0070052539404553,
127
+ "eval_loss": 6.504467487335205,
128
+ "eval_runtime": 3.2635,
129
+ "eval_samples_per_second": 747.973,
130
+ "eval_steps_per_second": 11.95,
131
+ "step": 144
132
+ },
133
+ {
134
+ "epoch": 1.0490367775831875,
135
+ "grad_norm": 0.6233965754508972,
136
+ "learning_rate": 0.0006,
137
+ "loss": 6.474125671386719,
138
+ "step": 150
139
+ },
140
+ {
141
+ "epoch": 1.1190893169877407,
142
+ "grad_norm": 1.1296820640563965,
143
+ "learning_rate": 0.0006,
144
+ "loss": 6.41492919921875,
145
+ "step": 160
146
+ },
147
+ {
148
+ "epoch": 1.1891418563922942,
149
+ "grad_norm": 0.3820130527019501,
150
+ "learning_rate": 0.0006,
151
+ "loss": 6.331824493408203,
152
+ "step": 170
153
+ },
154
+ {
155
+ "epoch": 1.2591943957968477,
156
+ "grad_norm": 1.0479296445846558,
157
+ "learning_rate": 0.0006,
158
+ "loss": 6.272692489624023,
159
+ "step": 180
160
+ },
161
+ {
162
+ "epoch": 1.3292469352014011,
163
+ "grad_norm": 0.6026280522346497,
164
+ "learning_rate": 0.0006,
165
+ "loss": 6.26782455444336,
166
+ "step": 190
167
+ },
168
+ {
169
+ "epoch": 1.3992994746059544,
170
+ "grad_norm": 1.3044335842132568,
171
+ "learning_rate": 0.0006,
172
+ "loss": 6.164206314086914,
173
+ "step": 200
174
+ },
175
+ {
176
+ "epoch": 1.4693520140105079,
177
+ "grad_norm": 0.414527952671051,
178
+ "learning_rate": 0.0006,
179
+ "loss": 6.096501159667969,
180
+ "step": 210
181
+ },
182
+ {
183
+ "epoch": 1.51138353765324,
184
+ "eval_loss": 6.077937126159668,
185
+ "eval_runtime": 3.2373,
186
+ "eval_samples_per_second": 754.014,
187
+ "eval_steps_per_second": 12.047,
188
+ "step": 216
189
+ },
190
+ {
191
+ "epoch": 1.5394045534150613,
192
+ "grad_norm": 1.2787466049194336,
193
+ "learning_rate": 0.0006,
194
+ "loss": 6.050628662109375,
195
+ "step": 220
196
+ },
197
+ {
198
+ "epoch": 1.6094570928196146,
199
+ "grad_norm": 0.6056044101715088,
200
+ "learning_rate": 0.0006,
201
+ "loss": 5.993916702270508,
202
+ "step": 230
203
+ },
204
+ {
205
+ "epoch": 1.679509632224168,
206
+ "grad_norm": 0.5148183107376099,
207
+ "learning_rate": 0.0006,
208
+ "loss": 5.94757080078125,
209
+ "step": 240
210
+ },
211
+ {
212
+ "epoch": 1.7495621716287215,
213
+ "grad_norm": 0.31190764904022217,
214
+ "learning_rate": 0.0006,
215
+ "loss": 5.883531188964843,
216
+ "step": 250
217
+ },
218
+ {
219
+ "epoch": 1.819614711033275,
220
+ "grad_norm": 0.36509600281715393,
221
+ "learning_rate": 0.0006,
222
+ "loss": 5.829424667358398,
223
+ "step": 260
224
+ },
225
+ {
226
+ "epoch": 1.8896672504378285,
227
+ "grad_norm": 1.0162628889083862,
228
+ "learning_rate": 0.0006,
229
+ "loss": 5.784934997558594,
230
+ "step": 270
231
+ },
232
+ {
233
+ "epoch": 1.959719789842382,
234
+ "grad_norm": 0.39998742938041687,
235
+ "learning_rate": 0.0006,
236
+ "loss": 5.737149810791015,
237
+ "step": 280
238
+ },
239
+ {
240
+ "epoch": 2.0140105078809105,
241
+ "eval_loss": 5.6931023597717285,
242
+ "eval_runtime": 3.7798,
243
+ "eval_samples_per_second": 645.803,
244
+ "eval_steps_per_second": 10.318,
245
+ "step": 288
246
+ },
247
+ {
248
+ "epoch": 2.0280210157618215,
249
+ "grad_norm": 0.4628411531448364,
250
+ "learning_rate": 0.0006,
251
+ "loss": 5.674631500244141,
252
+ "step": 290
253
+ },
254
+ {
255
+ "epoch": 2.098073555166375,
256
+ "grad_norm": 0.48857519030570984,
257
+ "learning_rate": 0.0006,
258
+ "loss": 5.590349197387695,
259
+ "step": 300
260
+ },
261
+ {
262
+ "epoch": 2.168126094570928,
263
+ "grad_norm": 0.55524742603302,
264
+ "learning_rate": 0.0006,
265
+ "loss": 5.591045761108399,
266
+ "step": 310
267
+ },
268
+ {
269
+ "epoch": 2.2381786339754814,
270
+ "grad_norm": 0.3205523192882538,
271
+ "learning_rate": 0.0006,
272
+ "loss": 5.515919494628906,
273
+ "step": 320
274
+ },
275
+ {
276
+ "epoch": 2.308231173380035,
277
+ "grad_norm": 0.6817094087600708,
278
+ "learning_rate": 0.0006,
279
+ "loss": 5.462646865844727,
280
+ "step": 330
281
+ },
282
+ {
283
+ "epoch": 2.3782837127845884,
284
+ "grad_norm": 0.782426655292511,
285
+ "learning_rate": 0.0006,
286
+ "loss": 5.457918167114258,
287
+ "step": 340
288
+ },
289
+ {
290
+ "epoch": 2.448336252189142,
291
+ "grad_norm": 0.30008888244628906,
292
+ "learning_rate": 0.0006,
293
+ "loss": 5.380028533935547,
294
+ "step": 350
295
+ },
296
+ {
297
+ "epoch": 2.5183887915936953,
298
+ "grad_norm": 0.49766749143600464,
299
+ "learning_rate": 0.0006,
300
+ "loss": 5.30742301940918,
301
+ "step": 360
302
+ },
303
+ {
304
+ "epoch": 2.5183887915936953,
305
+ "eval_loss": 5.35524320602417,
306
+ "eval_runtime": 3.985,
307
+ "eval_samples_per_second": 612.553,
308
+ "eval_steps_per_second": 9.787,
309
+ "step": 360
310
+ },
311
+ {
312
+ "epoch": 2.588441330998249,
313
+ "grad_norm": 1.1236608028411865,
314
+ "learning_rate": 0.0006,
315
+ "loss": 5.256686401367188,
316
+ "step": 370
317
+ },
318
+ {
319
+ "epoch": 2.6584938704028023,
320
+ "grad_norm": 0.9686716794967651,
321
+ "learning_rate": 0.0006,
322
+ "loss": 5.27575569152832,
323
+ "step": 380
324
+ },
325
+ {
326
+ "epoch": 2.7285464098073557,
327
+ "grad_norm": 0.29338839650154114,
328
+ "learning_rate": 0.0006,
329
+ "loss": 5.202019119262696,
330
+ "step": 390
331
+ },
332
+ {
333
+ "epoch": 2.7985989492119088,
334
+ "grad_norm": 0.34476038813591003,
335
+ "learning_rate": 0.0006,
336
+ "loss": 5.1141807556152346,
337
+ "step": 400
338
+ },
339
+ {
340
+ "epoch": 2.8686514886164622,
341
+ "grad_norm": 0.8026102185249329,
342
+ "learning_rate": 0.0006,
343
+ "loss": 5.106670379638672,
344
+ "step": 410
345
+ },
346
+ {
347
+ "epoch": 2.9387040280210157,
348
+ "grad_norm": 0.41981688141822815,
349
+ "learning_rate": 0.0006,
350
+ "loss": 5.072551727294922,
351
+ "step": 420
352
+ },
353
+ {
354
+ "epoch": 3.0070052539404553,
355
+ "grad_norm": 0.2799515426158905,
356
+ "learning_rate": 0.0006,
357
+ "loss": 4.986613464355469,
358
+ "step": 430
359
+ },
360
+ {
361
+ "epoch": 3.021015761821366,
362
+ "eval_loss": 5.031318664550781,
363
+ "eval_runtime": 3.2607,
364
+ "eval_samples_per_second": 748.624,
365
+ "eval_steps_per_second": 11.961,
366
+ "step": 432
367
+ },
368
+ {
369
+ "epoch": 3.0770577933450087,
370
+ "grad_norm": 0.7299258708953857,
371
+ "learning_rate": 0.0006,
372
+ "loss": 4.936912536621094,
373
+ "step": 440
374
+ },
375
+ {
376
+ "epoch": 3.147110332749562,
377
+ "grad_norm": 0.7612623572349548,
378
+ "learning_rate": 0.0006,
379
+ "loss": 4.938148117065429,
380
+ "step": 450
381
+ },
382
+ {
383
+ "epoch": 3.2171628721541157,
384
+ "grad_norm": 0.30043086409568787,
385
+ "learning_rate": 0.0006,
386
+ "loss": 4.860009002685547,
387
+ "step": 460
388
+ },
389
+ {
390
+ "epoch": 3.287215411558669,
391
+ "grad_norm": 0.34863752126693726,
392
+ "learning_rate": 0.0006,
393
+ "loss": 4.803009796142578,
394
+ "step": 470
395
+ },
396
+ {
397
+ "epoch": 3.357267950963222,
398
+ "grad_norm": 0.6028016805648804,
399
+ "learning_rate": 0.0006,
400
+ "loss": 4.771705245971679,
401
+ "step": 480
402
+ },
403
+ {
404
+ "epoch": 3.4273204903677756,
405
+ "grad_norm": 0.38058093190193176,
406
+ "learning_rate": 0.0006,
407
+ "loss": 4.7452739715576175,
408
+ "step": 490
409
+ },
410
+ {
411
+ "epoch": 3.497373029772329,
412
+ "grad_norm": 0.6069619059562683,
413
+ "learning_rate": 0.0006,
414
+ "loss": 4.701860809326172,
415
+ "step": 500
416
+ },
417
+ {
418
+ "epoch": 3.5253940455341506,
419
+ "eval_loss": 4.769557476043701,
420
+ "eval_runtime": 4.0934,
421
+ "eval_samples_per_second": 596.332,
422
+ "eval_steps_per_second": 9.528,
423
+ "step": 504
424
+ },
425
+ {
426
+ "epoch": 3.5674255691768826,
427
+ "grad_norm": 0.3816678524017334,
428
+ "learning_rate": 0.0006,
429
+ "loss": 4.671141815185547,
430
+ "step": 510
431
+ },
432
+ {
433
+ "epoch": 3.637478108581436,
434
+ "grad_norm": 0.34364593029022217,
435
+ "learning_rate": 0.0006,
436
+ "loss": 4.621066665649414,
437
+ "step": 520
438
+ },
439
+ {
440
+ "epoch": 3.7075306479859895,
441
+ "grad_norm": 0.7322856187820435,
442
+ "learning_rate": 0.0006,
443
+ "loss": 4.612473678588867,
444
+ "step": 530
445
+ },
446
+ {
447
+ "epoch": 3.777583187390543,
448
+ "grad_norm": 0.42280662059783936,
449
+ "learning_rate": 0.0006,
450
+ "loss": 4.61462173461914,
451
+ "step": 540
452
+ },
453
+ {
454
+ "epoch": 3.8476357267950965,
455
+ "grad_norm": 0.2599335312843323,
456
+ "learning_rate": 0.0006,
457
+ "loss": 4.5415092468261715,
458
+ "step": 550
459
+ },
460
+ {
461
+ "epoch": 3.91768826619965,
462
+ "grad_norm": 0.403529554605484,
463
+ "learning_rate": 0.0006,
464
+ "loss": 4.4888755798339846,
465
+ "step": 560
466
+ },
467
+ {
468
+ "epoch": 3.9877408056042034,
469
+ "grad_norm": 0.2667478621006012,
470
+ "learning_rate": 0.0006,
471
+ "loss": 4.468472290039062,
472
+ "step": 570
473
+ },
474
+ {
475
+ "epoch": 4.028021015761821,
476
+ "eval_loss": 4.5557074546813965,
477
+ "eval_runtime": 3.4966,
478
+ "eval_samples_per_second": 698.116,
479
+ "eval_steps_per_second": 11.154,
480
+ "step": 576
481
+ },
482
+ {
483
+ "epoch": 4.056042031523643,
484
+ "grad_norm": 0.4902321994304657,
485
+ "learning_rate": 0.0006,
486
+ "loss": 4.418440628051758,
487
+ "step": 580
488
+ },
489
+ {
490
+ "epoch": 4.126094570928196,
491
+ "grad_norm": 0.6080674529075623,
492
+ "learning_rate": 0.0006,
493
+ "loss": 4.395100402832031,
494
+ "step": 590
495
+ },
496
+ {
497
+ "epoch": 4.19614711033275,
498
+ "grad_norm": 0.3149462938308716,
499
+ "learning_rate": 0.0006,
500
+ "loss": 4.390229034423828,
501
+ "step": 600
502
+ },
503
+ {
504
+ "epoch": 4.266199649737303,
505
+ "grad_norm": 0.3042010962963104,
506
+ "learning_rate": 0.0006,
507
+ "loss": 4.345519638061523,
508
+ "step": 610
509
+ },
510
+ {
511
+ "epoch": 4.336252189141856,
512
+ "grad_norm": 0.24239438772201538,
513
+ "learning_rate": 0.0006,
514
+ "loss": 4.321981811523438,
515
+ "step": 620
516
+ },
517
+ {
518
+ "epoch": 4.406304728546409,
519
+ "grad_norm": 0.38201919198036194,
520
+ "learning_rate": 0.0006,
521
+ "loss": 4.2978263854980465,
522
+ "step": 630
523
+ },
524
+ {
525
+ "epoch": 4.476357267950963,
526
+ "grad_norm": 0.3829366862773895,
527
+ "learning_rate": 0.0006,
528
+ "loss": 4.280644226074219,
529
+ "step": 640
530
+ },
531
+ {
532
+ "epoch": 4.532399299474606,
533
+ "eval_loss": 4.381662845611572,
534
+ "eval_runtime": 3.691,
535
+ "eval_samples_per_second": 661.341,
536
+ "eval_steps_per_second": 10.566,
537
+ "step": 648
538
+ },
539
+ {
540
+ "epoch": 4.546409807355516,
541
+ "grad_norm": 0.32211166620254517,
542
+ "learning_rate": 0.0006,
543
+ "loss": 4.259928131103516,
544
+ "step": 650
545
+ },
546
+ {
547
+ "epoch": 4.61646234676007,
548
+ "grad_norm": 0.4348321259021759,
549
+ "learning_rate": 0.0006,
550
+ "loss": 4.251454925537109,
551
+ "step": 660
552
+ },
553
+ {
554
+ "epoch": 4.686514886164623,
555
+ "grad_norm": 0.3812483549118042,
556
+ "learning_rate": 0.0006,
557
+ "loss": 4.236824035644531,
558
+ "step": 670
559
+ },
560
+ {
561
+ "epoch": 4.756567425569177,
562
+ "grad_norm": 0.29503557085990906,
563
+ "learning_rate": 0.0006,
564
+ "loss": 4.216230010986328,
565
+ "step": 680
566
+ },
567
+ {
568
+ "epoch": 4.82661996497373,
569
+ "grad_norm": 0.2546517848968506,
570
+ "learning_rate": 0.0006,
571
+ "loss": 4.186763000488281,
572
+ "step": 690
573
+ },
574
+ {
575
+ "epoch": 4.896672504378284,
576
+ "grad_norm": 0.3478173613548279,
577
+ "learning_rate": 0.0006,
578
+ "loss": 4.161574172973633,
579
+ "step": 700
580
+ },
581
+ {
582
+ "epoch": 4.966725043782837,
583
+ "grad_norm": 0.21839696168899536,
584
+ "learning_rate": 0.0006,
585
+ "loss": 4.142048645019531,
586
+ "step": 710
587
+ },
588
+ {
589
+ "epoch": 5.035026269702277,
590
+ "grad_norm": 0.5530551671981812,
591
+ "learning_rate": 0.0006,
592
+ "loss": 4.121431732177735,
593
+ "step": 720
594
+ },
595
+ {
596
+ "epoch": 5.035026269702277,
597
+ "eval_loss": 4.278331756591797,
598
+ "eval_runtime": 3.908,
599
+ "eval_samples_per_second": 624.624,
600
+ "eval_steps_per_second": 9.98,
601
+ "step": 720
602
+ },
603
+ {
604
+ "epoch": 5.10507880910683,
605
+ "grad_norm": 0.4000762701034546,
606
+ "learning_rate": 0.0006,
607
+ "loss": 4.094532012939453,
608
+ "step": 730
609
+ },
610
+ {
611
+ "epoch": 5.175131348511384,
612
+ "grad_norm": 0.30992069840431213,
613
+ "learning_rate": 0.0006,
614
+ "loss": 4.089750671386719,
615
+ "step": 740
616
+ },
617
+ {
618
+ "epoch": 5.245183887915937,
619
+ "grad_norm": 0.23498931527137756,
620
+ "learning_rate": 0.0006,
621
+ "loss": 4.051603698730469,
622
+ "step": 750
623
+ },
624
+ {
625
+ "epoch": 5.315236427320491,
626
+ "grad_norm": 0.24305176734924316,
627
+ "learning_rate": 0.0006,
628
+ "loss": 4.043153381347656,
629
+ "step": 760
630
+ },
631
+ {
632
+ "epoch": 5.385288966725044,
633
+ "grad_norm": 0.24310357868671417,
634
+ "learning_rate": 0.0006,
635
+ "loss": 4.023551940917969,
636
+ "step": 770
637
+ },
638
+ {
639
+ "epoch": 5.455341506129598,
640
+ "grad_norm": 0.41956567764282227,
641
+ "learning_rate": 0.0006,
642
+ "loss": 4.035160064697266,
643
+ "step": 780
644
+ },
645
+ {
646
+ "epoch": 5.525394045534151,
647
+ "grad_norm": 0.32964956760406494,
648
+ "learning_rate": 0.0006,
649
+ "loss": 4.013400268554688,
650
+ "step": 790
651
+ },
652
+ {
653
+ "epoch": 5.5394045534150615,
654
+ "eval_loss": 4.1707587242126465,
655
+ "eval_runtime": 3.4448,
656
+ "eval_samples_per_second": 708.602,
657
+ "eval_steps_per_second": 11.321,
658
+ "step": 792
659
+ },
660
+ {
661
+ "epoch": 5.5954465849387045,
662
+ "grad_norm": 0.23922166228294373,
663
+ "learning_rate": 0.0006,
664
+ "loss": 4.003241729736328,
665
+ "step": 800
666
+ },
667
+ {
668
+ "epoch": 5.665499124343257,
669
+ "grad_norm": 0.22956278920173645,
670
+ "learning_rate": 0.0006,
671
+ "loss": 3.984221649169922,
672
+ "step": 810
673
+ },
674
+ {
675
+ "epoch": 5.735551663747811,
676
+ "grad_norm": 0.2851473391056061,
677
+ "learning_rate": 0.0006,
678
+ "loss": 3.9737266540527343,
679
+ "step": 820
680
+ },
681
+ {
682
+ "epoch": 5.805604203152364,
683
+ "grad_norm": 0.24042366445064545,
684
+ "learning_rate": 0.0006,
685
+ "loss": 3.972766876220703,
686
+ "step": 830
687
+ },
688
+ {
689
+ "epoch": 5.8756567425569175,
690
+ "grad_norm": 0.28479474782943726,
691
+ "learning_rate": 0.0006,
692
+ "loss": 3.972808837890625,
693
+ "step": 840
694
+ },
695
+ {
696
+ "epoch": 5.945709281961471,
697
+ "grad_norm": 0.38838624954223633,
698
+ "learning_rate": 0.0006,
699
+ "loss": 3.948868179321289,
700
+ "step": 850
701
+ },
702
+ {
703
+ "epoch": 6.0140105078809105,
704
+ "grad_norm": 0.3256464898586273,
705
+ "learning_rate": 0.0006,
706
+ "loss": 3.9456851959228514,
707
+ "step": 860
708
+ },
709
+ {
710
+ "epoch": 6.042031523642732,
711
+ "eval_loss": 4.092311859130859,
712
+ "eval_runtime": 3.3951,
713
+ "eval_samples_per_second": 718.982,
714
+ "eval_steps_per_second": 11.487,
715
+ "step": 864
716
+ },
717
+ {
718
+ "epoch": 6.084063047285464,
719
+ "grad_norm": 0.40261223912239075,
720
+ "learning_rate": 0.0006,
721
+ "loss": 3.878314208984375,
722
+ "step": 870
723
+ },
724
+ {
725
+ "epoch": 6.1541155866900175,
726
+ "grad_norm": 0.24834436178207397,
727
+ "learning_rate": 0.0006,
728
+ "loss": 3.8830192565917967,
729
+ "step": 880
730
+ },
731
+ {
732
+ "epoch": 6.224168126094571,
733
+ "grad_norm": 0.29639580845832825,
734
+ "learning_rate": 0.0006,
735
+ "loss": 3.8620361328125,
736
+ "step": 890
737
+ },
738
+ {
739
+ "epoch": 6.294220665499124,
740
+ "grad_norm": 0.24953483045101166,
741
+ "learning_rate": 0.0006,
742
+ "loss": 3.857699966430664,
743
+ "step": 900
744
+ },
745
+ {
746
+ "epoch": 6.364273204903678,
747
+ "grad_norm": 0.32599544525146484,
748
+ "learning_rate": 0.0006,
749
+ "loss": 3.8431934356689452,
750
+ "step": 910
751
+ },
752
+ {
753
+ "epoch": 6.434325744308231,
754
+ "grad_norm": 0.2872227132320404,
755
+ "learning_rate": 0.0006,
756
+ "loss": 3.8442756652832033,
757
+ "step": 920
758
+ },
759
+ {
760
+ "epoch": 6.504378283712785,
761
+ "grad_norm": 0.23899482190608978,
762
+ "learning_rate": 0.0006,
763
+ "loss": 3.8262901306152344,
764
+ "step": 930
765
+ },
766
+ {
767
+ "epoch": 6.546409807355516,
768
+ "eval_loss": 4.0181989669799805,
769
+ "eval_runtime": 3.4934,
770
+ "eval_samples_per_second": 698.739,
771
+ "eval_steps_per_second": 11.164,
772
+ "step": 936
773
+ },
774
+ {
775
+ "epoch": 6.574430823117338,
776
+ "grad_norm": 0.27845045924186707,
777
+ "learning_rate": 0.0006,
778
+ "loss": 3.8415252685546877,
779
+ "step": 940
780
+ },
781
+ {
782
+ "epoch": 6.644483362521892,
783
+ "grad_norm": 0.27635708451271057,
784
+ "learning_rate": 0.0006,
785
+ "loss": 3.828055572509766,
786
+ "step": 950
787
+ },
788
+ {
789
+ "epoch": 6.714535901926444,
790
+ "grad_norm": 0.27892908453941345,
791
+ "learning_rate": 0.0006,
792
+ "loss": 3.8211326599121094,
793
+ "step": 960
794
+ },
795
+ {
796
+ "epoch": 6.784588441330998,
797
+ "grad_norm": 0.26431673765182495,
798
+ "learning_rate": 0.0006,
799
+ "loss": 3.809326171875,
800
+ "step": 970
801
+ },
802
+ {
803
+ "epoch": 6.854640980735551,
804
+ "grad_norm": 0.2798311412334442,
805
+ "learning_rate": 0.0006,
806
+ "loss": 3.8122833251953123,
807
+ "step": 980
808
+ },
809
+ {
810
+ "epoch": 6.924693520140105,
811
+ "grad_norm": 0.33427661657333374,
812
+ "learning_rate": 0.0006,
813
+ "loss": 3.8152488708496093,
814
+ "step": 990
815
+ },
816
+ {
817
+ "epoch": 6.994746059544658,
818
+ "grad_norm": 0.2408551573753357,
819
+ "learning_rate": 0.0006,
820
+ "loss": 3.7974853515625,
821
+ "step": 1000
822
+ },
823
+ {
824
+ "epoch": 7.049036777583187,
825
+ "eval_loss": 3.9776611328125,
826
+ "eval_runtime": 3.5281,
827
+ "eval_samples_per_second": 691.883,
828
+ "eval_steps_per_second": 11.054,
829
+ "step": 1008
830
+ },
831
+ {
832
+ "epoch": 7.063047285464098,
833
+ "grad_norm": 0.4962931275367737,
834
+ "learning_rate": 0.0006,
835
+ "loss": 3.739080810546875,
836
+ "step": 1010
837
+ },
838
+ {
839
+ "epoch": 7.133099824868651,
840
+ "grad_norm": 0.3556423783302307,
841
+ "learning_rate": 0.0006,
842
+ "loss": 3.7520721435546873,
843
+ "step": 1020
844
+ },
845
+ {
846
+ "epoch": 7.203152364273205,
847
+ "grad_norm": 0.29888731241226196,
848
+ "learning_rate": 0.0006,
849
+ "loss": 3.739801788330078,
850
+ "step": 1030
851
+ },
852
+ {
853
+ "epoch": 7.273204903677758,
854
+ "grad_norm": 0.22400957345962524,
855
+ "learning_rate": 0.0006,
856
+ "loss": 3.7231216430664062,
857
+ "step": 1040
858
+ },
859
+ {
860
+ "epoch": 7.343257443082312,
861
+ "grad_norm": 0.22138261795043945,
862
+ "learning_rate": 0.0006,
863
+ "loss": 3.723231887817383,
864
+ "step": 1050
865
+ },
866
+ {
867
+ "epoch": 7.413309982486865,
868
+ "grad_norm": 0.24107135832309723,
869
+ "learning_rate": 0.0006,
870
+ "loss": 3.713814544677734,
871
+ "step": 1060
872
+ },
873
+ {
874
+ "epoch": 7.483362521891419,
875
+ "grad_norm": 0.2450643926858902,
876
+ "learning_rate": 0.0006,
877
+ "loss": 3.7051460266113283,
878
+ "step": 1070
879
+ },
880
+ {
881
+ "epoch": 7.553415061295972,
882
+ "grad_norm": 0.2829040586948395,
883
+ "learning_rate": 0.0006,
884
+ "loss": 3.710027313232422,
885
+ "step": 1080
886
+ },
887
+ {
888
+ "epoch": 7.553415061295972,
889
+ "eval_loss": 3.9317781925201416,
890
+ "eval_runtime": 3.9557,
891
+ "eval_samples_per_second": 617.084,
892
+ "eval_steps_per_second": 9.859,
893
+ "step": 1080
894
+ },
895
+ {
896
+ "epoch": 7.6234676007005255,
897
+ "grad_norm": 0.34311461448669434,
898
+ "learning_rate": 0.0006,
899
+ "loss": 3.7191070556640624,
900
+ "step": 1090
901
+ },
902
+ {
903
+ "epoch": 7.693520140105079,
904
+ "grad_norm": 0.2540404796600342,
905
+ "learning_rate": 0.0006,
906
+ "loss": 3.709515380859375,
907
+ "step": 1100
908
+ },
909
+ {
910
+ "epoch": 7.7635726795096325,
911
+ "grad_norm": 0.2629285454750061,
912
+ "learning_rate": 0.0006,
913
+ "loss": 3.6871673583984377,
914
+ "step": 1110
915
+ },
916
+ {
917
+ "epoch": 7.833625218914186,
918
+ "grad_norm": 0.23696254193782806,
919
+ "learning_rate": 0.0006,
920
+ "loss": 3.6875625610351563,
921
+ "step": 1120
922
+ },
923
+ {
924
+ "epoch": 7.903677758318739,
925
+ "grad_norm": 0.3052332401275635,
926
+ "learning_rate": 0.0006,
927
+ "loss": 3.695769500732422,
928
+ "step": 1130
929
+ },
930
+ {
931
+ "epoch": 7.973730297723293,
932
+ "grad_norm": 0.21110214293003082,
933
+ "learning_rate": 0.0006,
934
+ "loss": 3.6882614135742187,
935
+ "step": 1140
936
+ },
937
+ {
938
+ "epoch": 8.042031523642732,
939
+ "grad_norm": 0.422978937625885,
940
+ "learning_rate": 0.0006,
941
+ "loss": 3.6368148803710936,
942
+ "step": 1150
943
+ },
944
+ {
945
+ "epoch": 8.056042031523642,
946
+ "eval_loss": 3.90358304977417,
947
+ "eval_runtime": 3.3844,
948
+ "eval_samples_per_second": 721.261,
949
+ "eval_steps_per_second": 11.524,
950
+ "step": 1152
951
+ },
952
+ {
953
+ "epoch": 8.112084063047286,
954
+ "grad_norm": 0.34560948610305786,
955
+ "learning_rate": 0.0006,
956
+ "loss": 3.6172374725341796,
957
+ "step": 1160
958
+ },
959
+ {
960
+ "epoch": 8.18213660245184,
961
+ "grad_norm": 0.33484628796577454,
962
+ "learning_rate": 0.0006,
963
+ "loss": 3.613117218017578,
964
+ "step": 1170
965
+ },
966
+ {
967
+ "epoch": 8.252189141856393,
968
+ "grad_norm": 0.3299332559108734,
969
+ "learning_rate": 0.0006,
970
+ "loss": 3.6176902770996096,
971
+ "step": 1180
972
+ },
973
+ {
974
+ "epoch": 8.322241681260946,
975
+ "grad_norm": 0.25558748841285706,
976
+ "learning_rate": 0.0006,
977
+ "loss": 3.6145286560058594,
978
+ "step": 1190
979
+ },
980
+ {
981
+ "epoch": 8.3922942206655,
982
+ "grad_norm": 0.23964306712150574,
983
+ "learning_rate": 0.0006,
984
+ "loss": 3.6121849060058593,
985
+ "step": 1200
986
+ },
987
+ {
988
+ "epoch": 8.462346760070053,
989
+ "grad_norm": 0.23925594985485077,
990
+ "learning_rate": 0.0006,
991
+ "loss": 3.615177536010742,
992
+ "step": 1210
993
+ },
994
+ {
995
+ "epoch": 8.532399299474607,
996
+ "grad_norm": 0.23387187719345093,
997
+ "learning_rate": 0.0006,
998
+ "loss": 3.6193164825439452,
999
+ "step": 1220
1000
+ },
1001
+ {
1002
+ "epoch": 8.560420315236428,
1003
+ "eval_loss": 3.8646886348724365,
1004
+ "eval_runtime": 3.4045,
1005
+ "eval_samples_per_second": 716.998,
1006
+ "eval_steps_per_second": 11.456,
1007
+ "step": 1224
1008
+ },
1009
+ {
1010
+ "epoch": 8.602451838879158,
1011
+ "grad_norm": 0.24263112246990204,
1012
+ "learning_rate": 0.0006,
1013
+ "loss": 3.6070907592773436,
1014
+ "step": 1230
1015
+ },
1016
+ {
1017
+ "epoch": 8.672504378283712,
1018
+ "grad_norm": 0.24607565999031067,
1019
+ "learning_rate": 0.0006,
1020
+ "loss": 3.6143115997314452,
1021
+ "step": 1240
1022
+ },
1023
+ {
1024
+ "epoch": 8.742556917688265,
1025
+ "grad_norm": 0.273237943649292,
1026
+ "learning_rate": 0.0006,
1027
+ "loss": 3.607615661621094,
1028
+ "step": 1250
1029
+ },
1030
+ {
1031
+ "epoch": 8.812609457092819,
1032
+ "grad_norm": 0.22331637144088745,
1033
+ "learning_rate": 0.0006,
1034
+ "loss": 3.6233566284179686,
1035
+ "step": 1260
1036
+ },
1037
+ {
1038
+ "epoch": 8.882661996497372,
1039
+ "grad_norm": 0.3021933436393738,
1040
+ "learning_rate": 0.0006,
1041
+ "loss": 3.601115417480469,
1042
+ "step": 1270
1043
+ },
1044
+ {
1045
+ "epoch": 8.952714535901926,
1046
+ "grad_norm": 0.249200239777565,
1047
+ "learning_rate": 0.0006,
1048
+ "loss": 3.597786712646484,
1049
+ "step": 1280
1050
+ },
1051
+ {
1052
+ "epoch": 9.021015761821365,
1053
+ "grad_norm": 0.25300464034080505,
1054
+ "learning_rate": 0.0006,
1055
+ "loss": 3.591754913330078,
1056
+ "step": 1290
1057
+ },
1058
+ {
1059
+ "epoch": 9.063047285464098,
1060
+ "eval_loss": 3.844236373901367,
1061
+ "eval_runtime": 3.4005,
1062
+ "eval_samples_per_second": 717.835,
1063
+ "eval_steps_per_second": 11.469,
1064
+ "step": 1296
1065
+ },
1066
+ {
1067
+ "epoch": 9.091068301225919,
1068
+ "grad_norm": 0.3442411720752716,
1069
+ "learning_rate": 0.0006,
1070
+ "loss": 3.5079620361328123,
1071
+ "step": 1300
1072
+ },
1073
+ {
1074
+ "epoch": 9.161120840630472,
1075
+ "grad_norm": 0.35806402564048767,
1076
+ "learning_rate": 0.0006,
1077
+ "loss": 3.5409027099609376,
1078
+ "step": 1310
1079
+ },
1080
+ {
1081
+ "epoch": 9.231173380035026,
1082
+ "grad_norm": 0.35811489820480347,
1083
+ "learning_rate": 0.0006,
1084
+ "loss": 3.54935302734375,
1085
+ "step": 1320
1086
+ },
1087
+ {
1088
+ "epoch": 9.30122591943958,
1089
+ "grad_norm": 0.2234598696231842,
1090
+ "learning_rate": 0.0006,
1091
+ "loss": 3.5450748443603515,
1092
+ "step": 1330
1093
+ },
1094
+ {
1095
+ "epoch": 9.371278458844133,
1096
+ "grad_norm": 0.21218106150627136,
1097
+ "learning_rate": 0.0006,
1098
+ "loss": 3.524667739868164,
1099
+ "step": 1340
1100
+ },
1101
+ {
1102
+ "epoch": 9.441330998248686,
1103
+ "grad_norm": 0.2407359778881073,
1104
+ "learning_rate": 0.0006,
1105
+ "loss": 3.54589729309082,
1106
+ "step": 1350
1107
+ },
1108
+ {
1109
+ "epoch": 9.51138353765324,
1110
+ "grad_norm": 0.29132887721061707,
1111
+ "learning_rate": 0.0006,
1112
+ "loss": 3.532334899902344,
1113
+ "step": 1360
1114
+ },
1115
+ {
1116
+ "epoch": 9.567425569176883,
1117
+ "eval_loss": 3.823045015335083,
1118
+ "eval_runtime": 3.5809,
1119
+ "eval_samples_per_second": 681.681,
1120
+ "eval_steps_per_second": 10.891,
1121
+ "step": 1368
1122
+ },
1123
+ {
1124
+ "epoch": 9.581436077057793,
1125
+ "grad_norm": 0.27332165837287903,
1126
+ "learning_rate": 0.0006,
1127
+ "loss": 3.5407684326171873,
1128
+ "step": 1370
1129
+ },
1130
+ {
1131
+ "epoch": 9.651488616462347,
1132
+ "grad_norm": 0.22323289513587952,
1133
+ "learning_rate": 0.0006,
1134
+ "loss": 3.5315406799316404,
1135
+ "step": 1380
1136
+ },
1137
+ {
1138
+ "epoch": 9.7215411558669,
1139
+ "grad_norm": 0.23029769957065582,
1140
+ "learning_rate": 0.0006,
1141
+ "loss": 3.5354686737060548,
1142
+ "step": 1390
1143
+ },
1144
+ {
1145
+ "epoch": 9.791593695271454,
1146
+ "grad_norm": 0.239872008562088,
1147
+ "learning_rate": 0.0006,
1148
+ "loss": 3.5226299285888674,
1149
+ "step": 1400
1150
+ },
1151
+ {
1152
+ "epoch": 9.861646234676007,
1153
+ "grad_norm": 0.2355087697505951,
1154
+ "learning_rate": 0.0006,
1155
+ "loss": 3.5222801208496093,
1156
+ "step": 1410
1157
+ },
1158
+ {
1159
+ "epoch": 9.93169877408056,
1160
+ "grad_norm": 0.22256873548030853,
1161
+ "learning_rate": 0.0006,
1162
+ "loss": 3.5412212371826173,
1163
+ "step": 1420
1164
+ },
1165
+ {
1166
+ "epoch": 10.0,
1167
+ "grad_norm": 0.25026267766952515,
1168
+ "learning_rate": 0.0006,
1169
+ "loss": 3.519300079345703,
1170
+ "step": 1430
1171
+ },
1172
+ {
1173
+ "epoch": 10.070052539404553,
1174
+ "grad_norm": 0.418517142534256,
1175
+ "learning_rate": 0.0006,
1176
+ "loss": 3.4538883209228515,
1177
+ "step": 1440
1178
+ },
1179
+ {
1180
+ "epoch": 10.070052539404553,
1181
+ "eval_loss": 3.8263518810272217,
1182
+ "eval_runtime": 3.5579,
1183
+ "eval_samples_per_second": 686.082,
1184
+ "eval_steps_per_second": 10.962,
1185
+ "step": 1440
1186
+ },
1187
+ {
1188
+ "epoch": 10.140105078809107,
1189
+ "grad_norm": 0.4168876111507416,
1190
+ "learning_rate": 0.0006,
1191
+ "loss": 3.474140930175781,
1192
+ "step": 1450
1193
+ },
1194
+ {
1195
+ "epoch": 10.21015761821366,
1196
+ "grad_norm": 0.3406018316745758,
1197
+ "learning_rate": 0.0006,
1198
+ "loss": 3.473613739013672,
1199
+ "step": 1460
1200
+ },
1201
+ {
1202
+ "epoch": 10.280210157618214,
1203
+ "grad_norm": 0.21272197365760803,
1204
+ "learning_rate": 0.0006,
1205
+ "loss": 3.461037445068359,
1206
+ "step": 1470
1207
+ },
1208
+ {
1209
+ "epoch": 10.350262697022767,
1210
+ "grad_norm": 0.2185119390487671,
1211
+ "learning_rate": 0.0006,
1212
+ "loss": 3.467035675048828,
1213
+ "step": 1480
1214
+ },
1215
+ {
1216
+ "epoch": 10.42031523642732,
1217
+ "grad_norm": 0.2328590750694275,
1218
+ "learning_rate": 0.0006,
1219
+ "loss": 3.461333465576172,
1220
+ "step": 1490
1221
+ },
1222
+ {
1223
+ "epoch": 10.490367775831874,
1224
+ "grad_norm": 0.2336580902338028,
1225
+ "learning_rate": 0.0006,
1226
+ "loss": 3.4777130126953124,
1227
+ "step": 1500
1228
+ },
1229
+ {
1230
+ "epoch": 10.560420315236428,
1231
+ "grad_norm": 0.28968092799186707,
1232
+ "learning_rate": 0.0006,
1233
+ "loss": 3.466571807861328,
1234
+ "step": 1510
1235
+ },
1236
+ {
1237
+ "epoch": 10.574430823117337,
1238
+ "eval_loss": 3.788224697113037,
1239
+ "eval_runtime": 3.7388,
1240
+ "eval_samples_per_second": 652.875,
1241
+ "eval_steps_per_second": 10.431,
1242
+ "step": 1512
1243
+ },
1244
+ {
1245
+ "epoch": 10.630472854640981,
1246
+ "grad_norm": 0.22140789031982422,
1247
+ "learning_rate": 0.0006,
1248
+ "loss": 3.46275634765625,
1249
+ "step": 1520
1250
+ },
1251
+ {
1252
+ "epoch": 10.700525394045535,
1253
+ "grad_norm": 0.21839596331119537,
1254
+ "learning_rate": 0.0006,
1255
+ "loss": 3.4691925048828125,
1256
+ "step": 1530
1257
+ },
1258
+ {
1259
+ "epoch": 10.770577933450088,
1260
+ "grad_norm": 0.26760339736938477,
1261
+ "learning_rate": 0.0006,
1262
+ "loss": 3.4725528717041017,
1263
+ "step": 1540
1264
+ },
1265
+ {
1266
+ "epoch": 10.840630472854642,
1267
+ "grad_norm": 0.2242419421672821,
1268
+ "learning_rate": 0.0006,
1269
+ "loss": 3.475904846191406,
1270
+ "step": 1550
1271
+ },
1272
+ {
1273
+ "epoch": 10.910683012259195,
1274
+ "grad_norm": 0.22520668804645538,
1275
+ "learning_rate": 0.0006,
1276
+ "loss": 3.4760223388671876,
1277
+ "step": 1560
1278
+ },
1279
+ {
1280
+ "epoch": 10.980735551663749,
1281
+ "grad_norm": 0.23904766142368317,
1282
+ "learning_rate": 0.0006,
1283
+ "loss": 3.4671859741210938,
1284
+ "step": 1570
1285
+ },
1286
+ {
1287
+ "epoch": 11.049036777583188,
1288
+ "grad_norm": 0.3444315791130066,
1289
+ "learning_rate": 0.0006,
1290
+ "loss": 3.4159122467041017,
1291
+ "step": 1580
1292
+ },
1293
+ {
1294
+ "epoch": 11.07705779334501,
1295
+ "eval_loss": 3.791930913925171,
1296
+ "eval_runtime": 3.4685,
1297
+ "eval_samples_per_second": 703.773,
1298
+ "eval_steps_per_second": 11.244,
1299
+ "step": 1584
1300
+ },
1301
+ {
1302
+ "epoch": 11.119089316987742,
1303
+ "grad_norm": 0.30959776043891907,
1304
+ "learning_rate": 0.0006,
1305
+ "loss": 3.38648681640625,
1306
+ "step": 1590
1307
+ },
1308
+ {
1309
+ "epoch": 11.189141856392295,
1310
+ "grad_norm": 0.2670822739601135,
1311
+ "learning_rate": 0.0006,
1312
+ "loss": 3.387676239013672,
1313
+ "step": 1600
1314
+ },
1315
+ {
1316
+ "epoch": 11.259194395796847,
1317
+ "grad_norm": 0.2430114895105362,
1318
+ "learning_rate": 0.0006,
1319
+ "loss": 3.3922351837158202,
1320
+ "step": 1610
1321
+ },
1322
+ {
1323
+ "epoch": 11.3292469352014,
1324
+ "grad_norm": 0.259003609418869,
1325
+ "learning_rate": 0.0006,
1326
+ "loss": 3.401170349121094,
1327
+ "step": 1620
1328
+ },
1329
+ {
1330
+ "epoch": 11.399299474605954,
1331
+ "grad_norm": 0.28309041261672974,
1332
+ "learning_rate": 0.0006,
1333
+ "loss": 3.405831146240234,
1334
+ "step": 1630
1335
+ },
1336
+ {
1337
+ "epoch": 11.469352014010507,
1338
+ "grad_norm": 0.2640569508075714,
1339
+ "learning_rate": 0.0006,
1340
+ "loss": 3.4174739837646486,
1341
+ "step": 1640
1342
+ },
1343
+ {
1344
+ "epoch": 11.53940455341506,
1345
+ "grad_norm": 0.24435488879680634,
1346
+ "learning_rate": 0.0006,
1347
+ "loss": 3.40150146484375,
1348
+ "step": 1650
1349
+ },
1350
+ {
1351
+ "epoch": 11.581436077057793,
1352
+ "eval_loss": 3.7666003704071045,
1353
+ "eval_runtime": 3.5185,
1354
+ "eval_samples_per_second": 693.764,
1355
+ "eval_steps_per_second": 11.084,
1356
+ "step": 1656
1357
+ },
1358
+ {
1359
+ "epoch": 11.609457092819614,
1360
+ "grad_norm": 0.23578906059265137,
1361
+ "learning_rate": 0.0006,
1362
+ "loss": 3.4165611267089844,
1363
+ "step": 1660
1364
+ },
1365
+ {
1366
+ "epoch": 11.679509632224168,
1367
+ "grad_norm": 0.23872224986553192,
1368
+ "learning_rate": 0.0006,
1369
+ "loss": 3.4270652770996093,
1370
+ "step": 1670
1371
+ },
1372
+ {
1373
+ "epoch": 11.749562171628721,
1374
+ "grad_norm": 0.2594001293182373,
1375
+ "learning_rate": 0.0006,
1376
+ "loss": 3.4211357116699217,
1377
+ "step": 1680
1378
+ },
1379
+ {
1380
+ "epoch": 11.819614711033275,
1381
+ "grad_norm": 0.24929523468017578,
1382
+ "learning_rate": 0.0006,
1383
+ "loss": 3.423053741455078,
1384
+ "step": 1690
1385
+ },
1386
+ {
1387
+ "epoch": 11.889667250437828,
1388
+ "grad_norm": 0.2450554370880127,
1389
+ "learning_rate": 0.0006,
1390
+ "loss": 3.4230796813964846,
1391
+ "step": 1700
1392
+ },
1393
+ {
1394
+ "epoch": 11.959719789842381,
1395
+ "grad_norm": 0.2280256599187851,
1396
+ "learning_rate": 0.0006,
1397
+ "loss": 3.423610305786133,
1398
+ "step": 1710
1399
+ },
1400
+ {
1401
+ "epoch": 12.028021015761821,
1402
+ "grad_norm": 0.40455758571624756,
1403
+ "learning_rate": 0.0006,
1404
+ "loss": 3.389678192138672,
1405
+ "step": 1720
1406
+ },
1407
+ {
1408
+ "epoch": 12.084063047285465,
1409
+ "eval_loss": 3.7763803005218506,
1410
+ "eval_runtime": 3.6455,
1411
+ "eval_samples_per_second": 669.594,
1412
+ "eval_steps_per_second": 10.698,
1413
+ "step": 1728
1414
+ },
1415
+ {
1416
+ "epoch": 12.098073555166375,
1417
+ "grad_norm": 0.3537187874317169,
1418
+ "learning_rate": 0.0006,
1419
+ "loss": 3.333454132080078,
1420
+ "step": 1730
1421
+ },
1422
+ {
1423
+ "epoch": 12.168126094570928,
1424
+ "grad_norm": 0.29340294003486633,
1425
+ "learning_rate": 0.0006,
1426
+ "loss": 3.344945526123047,
1427
+ "step": 1740
1428
+ },
1429
+ {
1430
+ "epoch": 12.238178633975481,
1431
+ "grad_norm": 0.30197280645370483,
1432
+ "learning_rate": 0.0006,
1433
+ "loss": 3.360484313964844,
1434
+ "step": 1750
1435
+ },
1436
+ {
1437
+ "epoch": 12.308231173380035,
1438
+ "grad_norm": 0.22601750493049622,
1439
+ "learning_rate": 0.0006,
1440
+ "loss": 3.350424957275391,
1441
+ "step": 1760
1442
+ },
1443
+ {
1444
+ "epoch": 12.378283712784588,
1445
+ "grad_norm": 0.23398347198963165,
1446
+ "learning_rate": 0.0006,
1447
+ "loss": 3.3569656372070313,
1448
+ "step": 1770
1449
+ },
1450
+ {
1451
+ "epoch": 12.448336252189142,
1452
+ "grad_norm": 0.2278864085674286,
1453
+ "learning_rate": 0.0006,
1454
+ "loss": 3.3536346435546873,
1455
+ "step": 1780
1456
+ },
1457
+ {
1458
+ "epoch": 12.518388791593695,
1459
+ "grad_norm": 0.23192206025123596,
1460
+ "learning_rate": 0.0006,
1461
+ "loss": 3.361003875732422,
1462
+ "step": 1790
1463
+ },
1464
+ {
1465
+ "epoch": 12.588441330998249,
1466
+ "grad_norm": 0.2884629964828491,
1467
+ "learning_rate": 0.0006,
1468
+ "loss": 3.3694381713867188,
1469
+ "step": 1800
1470
+ },
1471
+ {
1472
+ "epoch": 12.588441330998249,
1473
+ "eval_loss": 3.753028154373169,
1474
+ "eval_runtime": 3.5308,
1475
+ "eval_samples_per_second": 691.353,
1476
+ "eval_steps_per_second": 11.046,
1477
+ "step": 1800
1478
+ },
1479
+ {
1480
+ "epoch": 12.658493870402802,
1481
+ "grad_norm": 0.22837434709072113,
1482
+ "learning_rate": 0.0006,
1483
+ "loss": 3.3743656158447264,
1484
+ "step": 1810
1485
+ },
1486
+ {
1487
+ "epoch": 12.728546409807356,
1488
+ "grad_norm": 0.23902975022792816,
1489
+ "learning_rate": 0.0006,
1490
+ "loss": 3.3757350921630858,
1491
+ "step": 1820
1492
+ },
1493
+ {
1494
+ "epoch": 12.79859894921191,
1495
+ "grad_norm": 0.24309372901916504,
1496
+ "learning_rate": 0.0006,
1497
+ "loss": 3.373929977416992,
1498
+ "step": 1830
1499
+ },
1500
+ {
1501
+ "epoch": 12.868651488616463,
1502
+ "grad_norm": 0.265935480594635,
1503
+ "learning_rate": 0.0006,
1504
+ "loss": 3.3781661987304688,
1505
+ "step": 1840
1506
+ },
1507
+ {
1508
+ "epoch": 12.938704028021016,
1509
+ "grad_norm": 0.2791769802570343,
1510
+ "learning_rate": 0.0006,
1511
+ "loss": 3.3764923095703123,
1512
+ "step": 1850
1513
+ },
1514
+ {
1515
+ "epoch": 13.007005253940456,
1516
+ "grad_norm": 0.259432852268219,
1517
+ "learning_rate": 0.0006,
1518
+ "loss": 3.360120391845703,
1519
+ "step": 1860
1520
+ },
1521
+ {
1522
+ "epoch": 13.07705779334501,
1523
+ "grad_norm": 0.29833364486694336,
1524
+ "learning_rate": 0.0006,
1525
+ "loss": 3.2793312072753906,
1526
+ "step": 1870
1527
+ },
1528
+ {
1529
+ "epoch": 13.091068301225919,
1530
+ "eval_loss": 3.760260820388794,
1531
+ "eval_runtime": 3.4544,
1532
+ "eval_samples_per_second": 706.638,
1533
+ "eval_steps_per_second": 11.29,
1534
+ "step": 1872
1535
+ },
1536
+ {
1537
+ "epoch": 13.147110332749563,
1538
+ "grad_norm": 0.34551334381103516,
1539
+ "learning_rate": 0.0006,
1540
+ "loss": 3.28541259765625,
1541
+ "step": 1880
1542
+ },
1543
+ {
1544
+ "epoch": 13.217162872154116,
1545
+ "grad_norm": 0.30898651480674744,
1546
+ "learning_rate": 0.0006,
1547
+ "loss": 3.2950508117675783,
1548
+ "step": 1890
1549
+ },
1550
+ {
1551
+ "epoch": 13.28721541155867,
1552
+ "grad_norm": 0.2684871256351471,
1553
+ "learning_rate": 0.0006,
1554
+ "loss": 3.3074264526367188,
1555
+ "step": 1900
1556
+ },
1557
+ {
1558
+ "epoch": 13.357267950963223,
1559
+ "grad_norm": 0.22169773280620575,
1560
+ "learning_rate": 0.0006,
1561
+ "loss": 3.3030105590820313,
1562
+ "step": 1910
1563
+ },
1564
+ {
1565
+ "epoch": 13.427320490367777,
1566
+ "grad_norm": 0.27856582403182983,
1567
+ "learning_rate": 0.0006,
1568
+ "loss": 3.3179840087890624,
1569
+ "step": 1920
1570
+ },
1571
+ {
1572
+ "epoch": 13.49737302977233,
1573
+ "grad_norm": 0.24769793450832367,
1574
+ "learning_rate": 0.0006,
1575
+ "loss": 3.3265071868896485,
1576
+ "step": 1930
1577
+ },
1578
+ {
1579
+ "epoch": 13.567425569176883,
1580
+ "grad_norm": 0.224507674574852,
1581
+ "learning_rate": 0.0006,
1582
+ "loss": 3.3263214111328123,
1583
+ "step": 1940
1584
+ },
1585
+ {
1586
+ "epoch": 13.595446584938705,
1587
+ "eval_loss": 3.7428011894226074,
1588
+ "eval_runtime": 3.3783,
1589
+ "eval_samples_per_second": 722.558,
1590
+ "eval_steps_per_second": 11.544,
1591
+ "step": 1944
1592
+ },
1593
+ {
1594
+ "epoch": 13.637478108581437,
1595
+ "grad_norm": 0.25914138555526733,
1596
+ "learning_rate": 0.0006,
1597
+ "loss": 3.332944869995117,
1598
+ "step": 1950
1599
+ },
1600
+ {
1601
+ "epoch": 13.70753064798599,
1602
+ "grad_norm": 0.2343108355998993,
1603
+ "learning_rate": 0.0006,
1604
+ "loss": 3.3334304809570314,
1605
+ "step": 1960
1606
+ },
1607
+ {
1608
+ "epoch": 13.777583187390542,
1609
+ "grad_norm": 0.23315395414829254,
1610
+ "learning_rate": 0.0006,
1611
+ "loss": 3.3350761413574217,
1612
+ "step": 1970
1613
+ },
1614
+ {
1615
+ "epoch": 13.847635726795096,
1616
+ "grad_norm": 0.2602457106113434,
1617
+ "learning_rate": 0.0006,
1618
+ "loss": 3.3352535247802733,
1619
+ "step": 1980
1620
+ },
1621
+ {
1622
+ "epoch": 13.917688266199649,
1623
+ "grad_norm": 0.24317029118537903,
1624
+ "learning_rate": 0.0006,
1625
+ "loss": 3.331440734863281,
1626
+ "step": 1990
1627
+ },
1628
+ {
1629
+ "epoch": 13.987740805604203,
1630
+ "grad_norm": 0.23694735765457153,
1631
+ "learning_rate": 0.0006,
1632
+ "loss": 3.3404006958007812,
1633
+ "step": 2000
1634
+ },
1635
+ {
1636
+ "epoch": 14.056042031523642,
1637
+ "grad_norm": 0.32518285512924194,
1638
+ "learning_rate": 0.0006,
1639
+ "loss": 3.254779052734375,
1640
+ "step": 2010
1641
+ },
1642
+ {
1643
+ "epoch": 14.098073555166375,
1644
+ "eval_loss": 3.7570059299468994,
1645
+ "eval_runtime": 3.4079,
1646
+ "eval_samples_per_second": 716.283,
1647
+ "eval_steps_per_second": 11.444,
1648
+ "step": 2016
1649
+ },
1650
+ {
1651
+ "epoch": 14.126094570928196,
1652
+ "grad_norm": 0.3998780846595764,
1653
+ "learning_rate": 0.0006,
1654
+ "loss": 3.2511558532714844,
1655
+ "step": 2020
1656
+ },
1657
+ {
1658
+ "epoch": 14.196147110332749,
1659
+ "grad_norm": 0.2794221043586731,
1660
+ "learning_rate": 0.0006,
1661
+ "loss": 3.2535163879394533,
1662
+ "step": 2030
1663
+ },
1664
+ {
1665
+ "epoch": 14.266199649737302,
1666
+ "grad_norm": 0.2672958970069885,
1667
+ "learning_rate": 0.0006,
1668
+ "loss": 3.2557903289794923,
1669
+ "step": 2040
1670
+ },
1671
+ {
1672
+ "epoch": 14.336252189141856,
1673
+ "grad_norm": 0.3607710599899292,
1674
+ "learning_rate": 0.0006,
1675
+ "loss": 3.2687118530273436,
1676
+ "step": 2050
1677
+ },
1678
+ {
1679
+ "epoch": 14.40630472854641,
1680
+ "grad_norm": 0.26717787981033325,
1681
+ "learning_rate": 0.0006,
1682
+ "loss": 3.283890151977539,
1683
+ "step": 2060
1684
+ },
1685
+ {
1686
+ "epoch": 14.476357267950963,
1687
+ "grad_norm": 0.22727353870868683,
1688
+ "learning_rate": 0.0006,
1689
+ "loss": 3.2842716217041015,
1690
+ "step": 2070
1691
+ },
1692
+ {
1693
+ "epoch": 14.546409807355516,
1694
+ "grad_norm": 0.22838088870048523,
1695
+ "learning_rate": 0.0006,
1696
+ "loss": 3.278900909423828,
1697
+ "step": 2080
1698
+ },
1699
+ {
1700
+ "epoch": 14.602451838879158,
1701
+ "eval_loss": 3.73828125,
1702
+ "eval_runtime": 3.4709,
1703
+ "eval_samples_per_second": 703.278,
1704
+ "eval_steps_per_second": 11.236,
1705
+ "step": 2088
1706
+ },
1707
+ {
1708
+ "epoch": 14.61646234676007,
1709
+ "grad_norm": 0.22517885267734528,
1710
+ "learning_rate": 0.0006,
1711
+ "loss": 3.290675735473633,
1712
+ "step": 2090
1713
+ },
1714
+ {
1715
+ "epoch": 14.686514886164623,
1716
+ "grad_norm": 0.24148719012737274,
1717
+ "learning_rate": 0.0006,
1718
+ "loss": 3.2893081665039063,
1719
+ "step": 2100
1720
+ },
1721
+ {
1722
+ "epoch": 14.756567425569177,
1723
+ "grad_norm": 0.22470487654209137,
1724
+ "learning_rate": 0.0006,
1725
+ "loss": 3.29949836730957,
1726
+ "step": 2110
1727
+ },
1728
+ {
1729
+ "epoch": 14.82661996497373,
1730
+ "grad_norm": 0.2813766300678253,
1731
+ "learning_rate": 0.0006,
1732
+ "loss": 3.3061370849609375,
1733
+ "step": 2120
1734
+ },
1735
+ {
1736
+ "epoch": 14.896672504378284,
1737
+ "grad_norm": 0.2263512760400772,
1738
+ "learning_rate": 0.0006,
1739
+ "loss": 3.3054107666015624,
1740
+ "step": 2130
1741
+ },
1742
+ {
1743
+ "epoch": 14.966725043782837,
1744
+ "grad_norm": 0.24720951914787292,
1745
+ "learning_rate": 0.0006,
1746
+ "loss": 3.290102005004883,
1747
+ "step": 2140
1748
+ },
1749
+ {
1750
+ "epoch": 15.035026269702277,
1751
+ "grad_norm": 0.3907608687877655,
1752
+ "learning_rate": 0.0006,
1753
+ "loss": 3.239105224609375,
1754
+ "step": 2150
1755
+ },
1756
+ {
1757
+ "epoch": 15.10507880910683,
1758
+ "grad_norm": 0.4218903183937073,
1759
+ "learning_rate": 0.0006,
1760
+ "loss": 3.2074462890625,
1761
+ "step": 2160
1762
+ },
1763
+ {
1764
+ "epoch": 15.10507880910683,
1765
+ "eval_loss": 3.769059896469116,
1766
+ "eval_runtime": 4.3518,
1767
+ "eval_samples_per_second": 560.917,
1768
+ "eval_steps_per_second": 8.962,
1769
+ "step": 2160
1770
+ },
1771
+ {
1772
+ "epoch": 15.175131348511384,
1773
+ "grad_norm": 0.3093777000904083,
1774
+ "learning_rate": 0.0006,
1775
+ "loss": 3.230516815185547,
1776
+ "step": 2170
1777
+ },
1778
+ {
1779
+ "epoch": 15.245183887915937,
1780
+ "grad_norm": 0.24631434679031372,
1781
+ "learning_rate": 0.0006,
1782
+ "loss": 3.235198211669922,
1783
+ "step": 2180
1784
+ },
1785
+ {
1786
+ "epoch": 15.31523642732049,
1787
+ "grad_norm": 0.22635217010974884,
1788
+ "learning_rate": 0.0006,
1789
+ "loss": 3.2339107513427736,
1790
+ "step": 2190
1791
+ },
1792
+ {
1793
+ "epoch": 15.385288966725044,
1794
+ "grad_norm": 0.24024364352226257,
1795
+ "learning_rate": 0.0006,
1796
+ "loss": 3.2306838989257813,
1797
+ "step": 2200
1798
+ },
1799
+ {
1800
+ "epoch": 15.455341506129598,
1801
+ "grad_norm": 0.25333160161972046,
1802
+ "learning_rate": 0.0006,
1803
+ "loss": 3.2406005859375,
1804
+ "step": 2210
1805
+ },
1806
+ {
1807
+ "epoch": 15.525394045534151,
1808
+ "grad_norm": 0.2500869035720825,
1809
+ "learning_rate": 0.0006,
1810
+ "loss": 3.242127227783203,
1811
+ "step": 2220
1812
+ },
1813
+ {
1814
+ "epoch": 15.595446584938705,
1815
+ "grad_norm": 0.2822413444519043,
1816
+ "learning_rate": 0.0006,
1817
+ "loss": 3.2631065368652346,
1818
+ "step": 2230
1819
+ },
1820
+ {
1821
+ "epoch": 15.609457092819614,
1822
+ "eval_loss": 3.7376556396484375,
1823
+ "eval_runtime": 3.2511,
1824
+ "eval_samples_per_second": 750.83,
1825
+ "eval_steps_per_second": 11.996,
1826
+ "step": 2232
1827
+ }
1828
+ ],
1829
+ "logging_steps": 10,
1830
+ "max_steps": 7150,
1831
+ "num_input_tokens_seen": 0,
1832
+ "num_train_epochs": 50,
1833
+ "save_steps": 500,
1834
+ "stateful_callbacks": {
1835
+ "TrainerControl": {
1836
+ "args": {
1837
+ "should_epoch_stop": false,
1838
+ "should_evaluate": false,
1839
+ "should_log": false,
1840
+ "should_save": false,
1841
+ "should_training_stop": false
1842
+ },
1843
+ "attributes": {}
1844
+ }
1845
+ },
1846
+ "total_flos": 9.557891315128074e+17,
1847
+ "train_batch_size": 8,
1848
+ "trial_name": null,
1849
+ "trial_params": null
1850
+ }
runs/t2a/t2a_150M_C0_baseline/train.log ADDED
The diff for this file is too large to render. See raw diff
 
runs/t2a/t2a_19M_C0_baseline/checkpoint-576/chat_template.jinja ADDED
@@ -0,0 +1,85 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {%- if tools %}
2
+ {{- '<|im_start|>system\n' }}
3
+ {%- if messages[0].role == 'system' %}
4
+ {{- messages[0].content + '\n\n' }}
5
+ {%- endif %}
6
+ {{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
7
+ {%- for tool in tools %}
8
+ {{- "\n" }}
9
+ {{- tool | tojson }}
10
+ {%- endfor %}
11
+ {{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
12
+ {%- else %}
13
+ {%- if messages[0].role == 'system' %}
14
+ {{- '<|im_start|>system\n' + messages[0].content + '<|im_end|>\n' }}
15
+ {%- endif %}
16
+ {%- endif %}
17
+ {%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
18
+ {%- for message in messages[::-1] %}
19
+ {%- set index = (messages|length - 1) - loop.index0 %}
20
+ {%- if ns.multi_step_tool and message.role == "user" and not(message.content.startswith('<tool_response>') and message.content.endswith('</tool_response>')) %}
21
+ {%- set ns.multi_step_tool = false %}
22
+ {%- set ns.last_query_index = index %}
23
+ {%- endif %}
24
+ {%- endfor %}
25
+ {%- for message in messages %}
26
+ {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
27
+ {{- '<|im_start|>' + message.role + '\n' + message.content + '<|im_end|>' + '\n' }}
28
+ {%- elif message.role == "assistant" %}
29
+ {%- set content = message.content %}
30
+ {%- set reasoning_content = '' %}
31
+ {%- if message.reasoning_content is defined and message.reasoning_content is not none %}
32
+ {%- set reasoning_content = message.reasoning_content %}
33
+ {%- else %}
34
+ {%- if '</think>' in message.content %}
35
+ {%- set content = message.content.split('</think>')[-1].lstrip('\n') %}
36
+ {%- set reasoning_content = message.content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
37
+ {%- endif %}
38
+ {%- endif %}
39
+ {%- if loop.index0 > ns.last_query_index %}
40
+ {%- if loop.last or (not loop.last and reasoning_content) %}
41
+ {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
42
+ {%- else %}
43
+ {{- '<|im_start|>' + message.role + '\n' + content }}
44
+ {%- endif %}
45
+ {%- else %}
46
+ {{- '<|im_start|>' + message.role + '\n' + content }}
47
+ {%- endif %}
48
+ {%- if message.tool_calls %}
49
+ {%- for tool_call in message.tool_calls %}
50
+ {%- if (loop.first and content) or (not loop.first) %}
51
+ {{- '\n' }}
52
+ {%- endif %}
53
+ {%- if tool_call.function %}
54
+ {%- set tool_call = tool_call.function %}
55
+ {%- endif %}
56
+ {{- '<tool_call>\n{"name": "' }}
57
+ {{- tool_call.name }}
58
+ {{- '", "arguments": ' }}
59
+ {%- if tool_call.arguments is string %}
60
+ {{- tool_call.arguments }}
61
+ {%- else %}
62
+ {{- tool_call.arguments | tojson }}
63
+ {%- endif %}
64
+ {{- '}\n</tool_call>' }}
65
+ {%- endfor %}
66
+ {%- endif %}
67
+ {{- '<|im_end|>\n' }}
68
+ {%- elif message.role == "tool" %}
69
+ {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
70
+ {{- '<|im_start|>user' }}
71
+ {%- endif %}
72
+ {{- '\n<tool_response>\n' }}
73
+ {{- message.content }}
74
+ {{- '\n</tool_response>' }}
75
+ {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
76
+ {{- '<|im_end|>\n' }}
77
+ {%- endif %}
78
+ {%- endif %}
79
+ {%- endfor %}
80
+ {%- if add_generation_prompt %}
81
+ {{- '<|im_start|>assistant\n' }}
82
+ {%- if enable_thinking is defined and enable_thinking is false %}
83
+ {{- '<think>\n\n</think>\n\n' }}
84
+ {%- endif %}
85
+ {%- endif %}
runs/t2a/t2a_19M_C0_baseline/checkpoint-576/config.json ADDED
@@ -0,0 +1,32 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "LlamaForCausalLM"
4
+ ],
5
+ "attention_bias": false,
6
+ "attention_dropout": 0.0,
7
+ "bos_token_id": null,
8
+ "dtype": "float32",
9
+ "eos_token_id": 151645,
10
+ "head_dim": 128,
11
+ "hidden_act": "silu",
12
+ "hidden_size": 512,
13
+ "initializer_range": 0.02,
14
+ "intermediate_size": 1536,
15
+ "max_position_embeddings": 2048,
16
+ "mlp_bias": false,
17
+ "model_type": "llama",
18
+ "num_attention_heads": 4,
19
+ "num_hidden_layers": 20,
20
+ "num_key_value_heads": 4,
21
+ "pad_token_id": 151645,
22
+ "pretraining_tp": 1,
23
+ "rms_norm_eps": 1e-06,
24
+ "rope_parameters": {
25
+ "rope_theta": 10000.0,
26
+ "rope_type": "default"
27
+ },
28
+ "tie_word_embeddings": true,
29
+ "transformers_version": "5.15.0",
30
+ "use_cache": false,
31
+ "vocab_size": 151671
32
+ }
runs/t2a/t2a_19M_C0_baseline/checkpoint-576/generation_config.json ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "eos_token_id": [
4
+ 151645
5
+ ],
6
+ "output_attentions": false,
7
+ "output_hidden_states": false,
8
+ "pad_token_id": 151645,
9
+ "transformers_version": "5.15.0",
10
+ "use_cache": true
11
+ }
runs/t2a/t2a_19M_C0_baseline/checkpoint-576/tokenizer_config.json ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "backend": "tokenizers",
4
+ "bos_token": null,
5
+ "clean_up_tokenization_spaces": false,
6
+ "eos_token": "<|im_end|>",
7
+ "errors": "replace",
8
+ "extra_special_tokens": [
9
+ "<|l2r_pred|>",
10
+ "<|r2l_pred|>"
11
+ ],
12
+ "is_local": false,
13
+ "local_files_only": false,
14
+ "model_max_length": 131072,
15
+ "pad_token": "<|im_end|>",
16
+ "split_special_tokens": false,
17
+ "tokenizer_class": "Qwen2Tokenizer",
18
+ "unk_token": null
19
+ }
runs/t2a/t2a_19M_C0_baseline/checkpoint-576/trainer_state.json ADDED
@@ -0,0 +1,504 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 30.322147651006713,
6
+ "eval_steps": 72,
7
+ "global_step": 576,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "epoch": 0.053691275167785234,
14
+ "grad_norm": 1.8142167329788208,
15
+ "learning_rate": 0.0,
16
+ "loss": 12.011314392089844,
17
+ "step": 1
18
+ },
19
+ {
20
+ "epoch": 0.5369127516778524,
21
+ "grad_norm": 1.7725874185562134,
22
+ "learning_rate": 5.399999999999999e-05,
23
+ "loss": 11.70957522922092,
24
+ "step": 10
25
+ },
26
+ {
27
+ "epoch": 1.0536912751677852,
28
+ "grad_norm": 1.685107707977295,
29
+ "learning_rate": 0.00011399999999999999,
30
+ "loss": 11.0172119140625,
31
+ "step": 20
32
+ },
33
+ {
34
+ "epoch": 1.5906040268456376,
35
+ "grad_norm": 1.6527771949768066,
36
+ "learning_rate": 0.00017399999999999997,
37
+ "loss": 10.258322143554688,
38
+ "step": 30
39
+ },
40
+ {
41
+ "epoch": 2.1073825503355703,
42
+ "grad_norm": 1.4742095470428467,
43
+ "learning_rate": 0.000234,
44
+ "loss": 9.209019470214844,
45
+ "step": 40
46
+ },
47
+ {
48
+ "epoch": 2.6442953020134228,
49
+ "grad_norm": 0.8990502953529358,
50
+ "learning_rate": 0.000294,
51
+ "loss": 8.233860778808594,
52
+ "step": 50
53
+ },
54
+ {
55
+ "epoch": 3.1610738255033555,
56
+ "grad_norm": 0.32120978832244873,
57
+ "learning_rate": 0.00035399999999999993,
58
+ "loss": 7.701448822021485,
59
+ "step": 60
60
+ },
61
+ {
62
+ "epoch": 3.697986577181208,
63
+ "grad_norm": 1.2037309408187866,
64
+ "learning_rate": 0.0004139999999999999,
65
+ "loss": 7.5403800964355465,
66
+ "step": 70
67
+ },
68
+ {
69
+ "epoch": 3.8053691275167782,
70
+ "eval_loss": 7.4664626121521,
71
+ "eval_runtime": 27.3161,
72
+ "eval_samples_per_second": 89.361,
73
+ "eval_steps_per_second": 1.428,
74
+ "step": 72
75
+ },
76
+ {
77
+ "epoch": 4.214765100671141,
78
+ "grad_norm": 0.43744996190071106,
79
+ "learning_rate": 0.000474,
80
+ "loss": 7.394523620605469,
81
+ "step": 80
82
+ },
83
+ {
84
+ "epoch": 4.751677852348993,
85
+ "grad_norm": 1.3217568397521973,
86
+ "learning_rate": 0.000534,
87
+ "loss": 7.174855804443359,
88
+ "step": 90
89
+ },
90
+ {
91
+ "epoch": 5.268456375838926,
92
+ "grad_norm": 0.5023097991943359,
93
+ "learning_rate": 0.0005939999999999999,
94
+ "loss": 7.081169891357422,
95
+ "step": 100
96
+ },
97
+ {
98
+ "epoch": 5.805369127516778,
99
+ "grad_norm": 0.30793970823287964,
100
+ "learning_rate": 0.0006,
101
+ "loss": 6.914817810058594,
102
+ "step": 110
103
+ },
104
+ {
105
+ "epoch": 6.322147651006711,
106
+ "grad_norm": 0.2508816123008728,
107
+ "learning_rate": 0.0006,
108
+ "loss": 6.747209167480468,
109
+ "step": 120
110
+ },
111
+ {
112
+ "epoch": 6.859060402684563,
113
+ "grad_norm": 0.6191902756690979,
114
+ "learning_rate": 0.0006,
115
+ "loss": 6.628437042236328,
116
+ "step": 130
117
+ },
118
+ {
119
+ "epoch": 7.375838926174497,
120
+ "grad_norm": 0.45684921741485596,
121
+ "learning_rate": 0.0006,
122
+ "loss": 6.524767303466797,
123
+ "step": 140
124
+ },
125
+ {
126
+ "epoch": 7.590604026845638,
127
+ "eval_loss": 6.43440580368042,
128
+ "eval_runtime": 3.2557,
129
+ "eval_samples_per_second": 749.764,
130
+ "eval_steps_per_second": 11.979,
131
+ "step": 144
132
+ },
133
+ {
134
+ "epoch": 7.912751677852349,
135
+ "grad_norm": 0.32593798637390137,
136
+ "learning_rate": 0.0006,
137
+ "loss": 6.398002624511719,
138
+ "step": 150
139
+ },
140
+ {
141
+ "epoch": 8.429530201342281,
142
+ "grad_norm": 0.49603816866874695,
143
+ "learning_rate": 0.0006,
144
+ "loss": 6.30065803527832,
145
+ "step": 160
146
+ },
147
+ {
148
+ "epoch": 8.966442953020135,
149
+ "grad_norm": 0.5765122771263123,
150
+ "learning_rate": 0.0006,
151
+ "loss": 6.225212097167969,
152
+ "step": 170
153
+ },
154
+ {
155
+ "epoch": 9.483221476510067,
156
+ "grad_norm": 0.35036981105804443,
157
+ "learning_rate": 0.0006,
158
+ "loss": 6.138700485229492,
159
+ "step": 180
160
+ },
161
+ {
162
+ "epoch": 10.0,
163
+ "grad_norm": 0.864263653755188,
164
+ "learning_rate": 0.0006,
165
+ "loss": 6.088552856445313,
166
+ "step": 190
167
+ },
168
+ {
169
+ "epoch": 10.536912751677852,
170
+ "grad_norm": 0.3657694160938263,
171
+ "learning_rate": 0.0006,
172
+ "loss": 6.047029876708985,
173
+ "step": 200
174
+ },
175
+ {
176
+ "epoch": 11.053691275167786,
177
+ "grad_norm": 0.32556647062301636,
178
+ "learning_rate": 0.0006,
179
+ "loss": 5.944820404052734,
180
+ "step": 210
181
+ },
182
+ {
183
+ "epoch": 11.375838926174497,
184
+ "eval_loss": 5.944965362548828,
185
+ "eval_runtime": 3.251,
186
+ "eval_samples_per_second": 750.841,
187
+ "eval_steps_per_second": 11.996,
188
+ "step": 216
189
+ },
190
+ {
191
+ "epoch": 11.590604026845638,
192
+ "grad_norm": 0.8797701001167297,
193
+ "learning_rate": 0.0006,
194
+ "loss": 5.853334426879883,
195
+ "step": 220
196
+ },
197
+ {
198
+ "epoch": 12.10738255033557,
199
+ "grad_norm": 0.7158633470535278,
200
+ "learning_rate": 0.0006,
201
+ "loss": 5.845497894287109,
202
+ "step": 230
203
+ },
204
+ {
205
+ "epoch": 12.644295302013422,
206
+ "grad_norm": 0.3370794355869293,
207
+ "learning_rate": 0.0006,
208
+ "loss": 5.740810775756836,
209
+ "step": 240
210
+ },
211
+ {
212
+ "epoch": 13.161073825503356,
213
+ "grad_norm": 1.0377658605575562,
214
+ "learning_rate": 0.0006,
215
+ "loss": 5.6702430725097654,
216
+ "step": 250
217
+ },
218
+ {
219
+ "epoch": 13.697986577181208,
220
+ "grad_norm": 0.6437246799468994,
221
+ "learning_rate": 0.0006,
222
+ "loss": 5.641211700439453,
223
+ "step": 260
224
+ },
225
+ {
226
+ "epoch": 14.21476510067114,
227
+ "grad_norm": 0.45554235577583313,
228
+ "learning_rate": 0.0006,
229
+ "loss": 5.558935928344726,
230
+ "step": 270
231
+ },
232
+ {
233
+ "epoch": 14.751677852348994,
234
+ "grad_norm": 0.7522556781768799,
235
+ "learning_rate": 0.0006,
236
+ "loss": 5.494405364990234,
237
+ "step": 280
238
+ },
239
+ {
240
+ "epoch": 15.161073825503356,
241
+ "eval_loss": 5.567083835601807,
242
+ "eval_runtime": 3.2645,
243
+ "eval_samples_per_second": 747.75,
244
+ "eval_steps_per_second": 11.947,
245
+ "step": 288
246
+ },
247
+ {
248
+ "epoch": 15.268456375838927,
249
+ "grad_norm": 0.8695115447044373,
250
+ "learning_rate": 0.0006,
251
+ "loss": 5.467164993286133,
252
+ "step": 290
253
+ },
254
+ {
255
+ "epoch": 15.805369127516778,
256
+ "grad_norm": 0.297993540763855,
257
+ "learning_rate": 0.0006,
258
+ "loss": 5.397716522216797,
259
+ "step": 300
260
+ },
261
+ {
262
+ "epoch": 16.322147651006713,
263
+ "grad_norm": 0.4352909326553345,
264
+ "learning_rate": 0.0006,
265
+ "loss": 5.3047119140625,
266
+ "step": 310
267
+ },
268
+ {
269
+ "epoch": 16.859060402684563,
270
+ "grad_norm": 0.6090894937515259,
271
+ "learning_rate": 0.0006,
272
+ "loss": 5.284146881103515,
273
+ "step": 320
274
+ },
275
+ {
276
+ "epoch": 17.375838926174495,
277
+ "grad_norm": 0.38252320885658264,
278
+ "learning_rate": 0.0006,
279
+ "loss": 5.229152679443359,
280
+ "step": 330
281
+ },
282
+ {
283
+ "epoch": 17.91275167785235,
284
+ "grad_norm": 0.5717241168022156,
285
+ "learning_rate": 0.0006,
286
+ "loss": 5.158631896972656,
287
+ "step": 340
288
+ },
289
+ {
290
+ "epoch": 18.42953020134228,
291
+ "grad_norm": 0.8796677589416504,
292
+ "learning_rate": 0.0006,
293
+ "loss": 5.094086837768555,
294
+ "step": 350
295
+ },
296
+ {
297
+ "epoch": 18.966442953020135,
298
+ "grad_norm": 0.5505955815315247,
299
+ "learning_rate": 0.0006,
300
+ "loss": 5.084417343139648,
301
+ "step": 360
302
+ },
303
+ {
304
+ "epoch": 18.966442953020135,
305
+ "eval_loss": 5.213481903076172,
306
+ "eval_runtime": 3.2663,
307
+ "eval_samples_per_second": 747.319,
308
+ "eval_steps_per_second": 11.94,
309
+ "step": 360
310
+ },
311
+ {
312
+ "epoch": 19.483221476510067,
313
+ "grad_norm": 0.3969738781452179,
314
+ "learning_rate": 0.0006,
315
+ "loss": 4.99476318359375,
316
+ "step": 370
317
+ },
318
+ {
319
+ "epoch": 20.0,
320
+ "grad_norm": 0.23706182837486267,
321
+ "learning_rate": 0.0006,
322
+ "loss": 4.9330909729003904,
323
+ "step": 380
324
+ },
325
+ {
326
+ "epoch": 20.536912751677853,
327
+ "grad_norm": 0.8291894197463989,
328
+ "learning_rate": 0.0006,
329
+ "loss": 4.859904861450195,
330
+ "step": 390
331
+ },
332
+ {
333
+ "epoch": 21.053691275167786,
334
+ "grad_norm": 0.483011931180954,
335
+ "learning_rate": 0.0006,
336
+ "loss": 4.925698089599609,
337
+ "step": 400
338
+ },
339
+ {
340
+ "epoch": 21.59060402684564,
341
+ "grad_norm": 0.4645936191082001,
342
+ "learning_rate": 0.0006,
343
+ "loss": 4.812581634521484,
344
+ "step": 410
345
+ },
346
+ {
347
+ "epoch": 22.107382550335572,
348
+ "grad_norm": 0.4653514623641968,
349
+ "learning_rate": 0.0006,
350
+ "loss": 4.752062225341797,
351
+ "step": 420
352
+ },
353
+ {
354
+ "epoch": 22.644295302013422,
355
+ "grad_norm": 0.45866936445236206,
356
+ "learning_rate": 0.0006,
357
+ "loss": 4.681341171264648,
358
+ "step": 430
359
+ },
360
+ {
361
+ "epoch": 22.751677852348994,
362
+ "eval_loss": 4.928530216217041,
363
+ "eval_runtime": 3.243,
364
+ "eval_samples_per_second": 752.706,
365
+ "eval_steps_per_second": 12.026,
366
+ "step": 432
367
+ },
368
+ {
369
+ "epoch": 23.161073825503355,
370
+ "grad_norm": 0.4152670204639435,
371
+ "learning_rate": 0.0006,
372
+ "loss": 4.620679473876953,
373
+ "step": 440
374
+ },
375
+ {
376
+ "epoch": 23.697986577181208,
377
+ "grad_norm": 0.8872674703598022,
378
+ "learning_rate": 0.0006,
379
+ "loss": 4.583133697509766,
380
+ "step": 450
381
+ },
382
+ {
383
+ "epoch": 24.21476510067114,
384
+ "grad_norm": 0.7488914132118225,
385
+ "learning_rate": 0.0006,
386
+ "loss": 4.62310905456543,
387
+ "step": 460
388
+ },
389
+ {
390
+ "epoch": 24.751677852348994,
391
+ "grad_norm": 0.404926061630249,
392
+ "learning_rate": 0.0006,
393
+ "loss": 4.546733856201172,
394
+ "step": 470
395
+ },
396
+ {
397
+ "epoch": 25.268456375838927,
398
+ "grad_norm": 0.31850552558898926,
399
+ "learning_rate": 0.0006,
400
+ "loss": 4.459835052490234,
401
+ "step": 480
402
+ },
403
+ {
404
+ "epoch": 25.80536912751678,
405
+ "grad_norm": 0.2806507647037506,
406
+ "learning_rate": 0.0006,
407
+ "loss": 4.37512435913086,
408
+ "step": 490
409
+ },
410
+ {
411
+ "epoch": 26.322147651006713,
412
+ "grad_norm": 0.5341393351554871,
413
+ "learning_rate": 0.0006,
414
+ "loss": 4.31119384765625,
415
+ "step": 500
416
+ },
417
+ {
418
+ "epoch": 26.536912751677853,
419
+ "eval_loss": 4.76123571395874,
420
+ "eval_runtime": 3.2635,
421
+ "eval_samples_per_second": 747.966,
422
+ "eval_steps_per_second": 11.95,
423
+ "step": 504
424
+ },
425
+ {
426
+ "epoch": 26.859060402684563,
427
+ "grad_norm": 1.043302297592163,
428
+ "learning_rate": 0.0006,
429
+ "loss": 4.311373138427735,
430
+ "step": 510
431
+ },
432
+ {
433
+ "epoch": 27.375838926174495,
434
+ "grad_norm": 0.9856887459754944,
435
+ "learning_rate": 0.0006,
436
+ "loss": 4.348341369628907,
437
+ "step": 520
438
+ },
439
+ {
440
+ "epoch": 27.91275167785235,
441
+ "grad_norm": 0.6892461180686951,
442
+ "learning_rate": 0.0006,
443
+ "loss": 4.312998962402344,
444
+ "step": 530
445
+ },
446
+ {
447
+ "epoch": 28.42953020134228,
448
+ "grad_norm": 0.3091641962528229,
449
+ "learning_rate": 0.0006,
450
+ "loss": 4.186044692993164,
451
+ "step": 540
452
+ },
453
+ {
454
+ "epoch": 28.966442953020135,
455
+ "grad_norm": 0.2504050135612488,
456
+ "learning_rate": 0.0006,
457
+ "loss": 4.133544540405273,
458
+ "step": 550
459
+ },
460
+ {
461
+ "epoch": 29.483221476510067,
462
+ "grad_norm": 0.5265411734580994,
463
+ "learning_rate": 0.0006,
464
+ "loss": 4.017610931396485,
465
+ "step": 560
466
+ },
467
+ {
468
+ "epoch": 30.0,
469
+ "grad_norm": 0.5696446299552917,
470
+ "learning_rate": 0.0006,
471
+ "loss": 4.023490905761719,
472
+ "step": 570
473
+ },
474
+ {
475
+ "epoch": 30.322147651006713,
476
+ "eval_loss": 4.713936805725098,
477
+ "eval_runtime": 3.2534,
478
+ "eval_samples_per_second": 750.303,
479
+ "eval_steps_per_second": 11.988,
480
+ "step": 576
481
+ }
482
+ ],
483
+ "logging_steps": 10,
484
+ "max_steps": 7505,
485
+ "num_input_tokens_seen": 0,
486
+ "num_train_epochs": 395,
487
+ "save_steps": 500,
488
+ "stateful_callbacks": {
489
+ "TrainerControl": {
490
+ "args": {
491
+ "should_epoch_stop": false,
492
+ "should_evaluate": false,
493
+ "should_log": false,
494
+ "should_save": false,
495
+ "should_training_stop": false
496
+ },
497
+ "attributes": {}
498
+ }
499
+ },
500
+ "total_flos": 2.4224477549376307e+17,
501
+ "train_batch_size": 8,
502
+ "trial_name": null,
503
+ "trial_params": null
504
+ }
runs/t2a/t2a_19M_C0_baseline/train.log ADDED
The diff for this file is too large to render. See raw diff