Hugging Face's logo

SJTUFreshman
/
collaborativelearning

Text Generation
PEFT
Safetensors
English
lora
multi-agent
collaborative-learning
jca
qwen3
Model card Files
xet
 Community
SJTUFreshman commited on
Commit
a86e230
·
verified ·
1 Parent(s): 368a263

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. Conifer/PD/Qwen3-1.7B/checkpoint-400/adapter_config.json +45 -0
  2. Conifer/PD/Qwen3-1.7B/checkpoint-400/chat_template.jinja +89 -0
  3. Conifer/PD/Qwen3-1.7B/checkpoint-400/tokenizer_config.json +30 -0
  4. Conifer/PD/Qwen3-1.7B/checkpoint-400/trainer_state.json +355 -0
  5. Conifer/PD/Qwen3-1.7B/checkpoint-500/adapter_config.json +45 -0
  6. Conifer/PD/Qwen3-1.7B/checkpoint-500/chat_template.jinja +89 -0
  7. Conifer/PD/Qwen3-1.7B/checkpoint-500/tokenizer_config.json +30 -0
  8. Conifer/PD/Qwen3-1.7B/checkpoint-500/trainer_state.json +433 -0
  9. Conifer/PD/Qwen3-1.7B/checkpoint-558/adapter_config.json +45 -0
  10. Conifer/PD/Qwen3-1.7B/checkpoint-558/chat_template.jinja +89 -0
  11. Conifer/PD/Qwen3-1.7B/checkpoint-558/tokenizer_config.json +30 -0
  12. Conifer/PD/Qwen3-1.7B/checkpoint-558/trainer_state.json +476 -0
  13. Conifer/PD/Qwen3-1.7B/final/adapter_config.json +45 -0
  14. Conifer/PD/Qwen3-1.7B/final/chat_template.jinja +89 -0
  15. Conifer/PD/Qwen3-1.7B/final/final_eval_metrics.json +7 -0
  16. Conifer/PD/Qwen3-1.7B/final/tokenizer_config.json +30 -0
  17. Conifer/PD/Qwen3-1.7B/final/train_summary.json +12 -0
  18. Conifer/PD/Qwen3-4B/checkpoint-500/adapter_config.json +45 -0
  19. Conifer/PD/Qwen3-4B/checkpoint-500/chat_template.jinja +89 -0
  20. Conifer/PD/Qwen3-4B/checkpoint-500/tokenizer_config.json +30 -0
  21. Conifer/PD/Qwen3-4B/checkpoint-500/trainer_state.json +433 -0
  22. Conifer/PD/Qwen3-4B/checkpoint-600/adapter_config.json +45 -0
  23. Conifer/PD/Qwen3-4B/checkpoint-600/chat_template.jinja +89 -0
  24. Conifer/PD/Qwen3-4B/checkpoint-600/tokenizer_config.json +30 -0
  25. Conifer/PD/Qwen3-4B/checkpoint-600/trainer_state.json +511 -0
  26. Conifer/PD/Qwen3-4B/checkpoint-621/adapter_config.json +45 -0
  27. Conifer/PD/Qwen3-4B/checkpoint-621/chat_template.jinja +89 -0
  28. Conifer/PD/Qwen3-4B/checkpoint-621/tokenizer_config.json +30 -0
  29. Conifer/PD/Qwen3-4B/checkpoint-621/trainer_state.json +533 -0
  30. Conifer/PD/Qwen3-4B/final/adapter_config.json +45 -0
  31. Conifer/PD/Qwen3-4B/final/chat_template.jinja +89 -0
  32. Conifer/PD/Qwen3-4B/final/final_eval_metrics.json +7 -0
  33. Conifer/PD/Qwen3-4B/final/tokenizer_config.json +30 -0
  34. Conifer/PD/Qwen3-4B/final/train_summary.json +12 -0
  35. Conifer/PD/Qwen3-8B/checkpoint-500/adapter_config.json +45 -0
  36. Conifer/PD/Qwen3-8B/checkpoint-500/tokenizer_config.json +30 -0
  37. Conifer/PD/Qwen3-8B/checkpoint-500/trainer_state.json +433 -0
  38. Conifer/PD/Qwen3-8B/checkpoint-600/adapter_config.json +45 -0
  39. Conifer/PD/Qwen3-8B/checkpoint-600/chat_template.jinja +89 -0
  40. Conifer/PD/Qwen3-8B/checkpoint-600/tokenizer_config.json +30 -0
  41. Conifer/PD/Qwen3-8B/checkpoint-600/trainer_state.json +511 -0
  42. Conifer/PD/Qwen3-8B/checkpoint-618/adapter_config.json +45 -0
  43. Conifer/PD/Qwen3-8B/checkpoint-618/chat_template.jinja +89 -0
  44. Conifer/PD/Qwen3-8B/checkpoint-618/tokenizer_config.json +30 -0
  45. Conifer/PD/Qwen3-8B/checkpoint-618/trainer_state.json +526 -0
  46. Conifer/PD/Qwen3-8B/final/adapter_config.json +45 -0
  47. Conifer/PD/Qwen3-8B/final/chat_template.jinja +89 -0
  48. Conifer/PD/Qwen3-8B/final/final_eval_metrics.json +7 -0
  49. Conifer/PD/Qwen3-8B/final/tokenizer_config.json +30 -0
  50. Conifer/PD/Qwen3-8B/final/train_summary.json +12 -0
Conifer/PD/Qwen3-1.7B/checkpoint-400/adapter_config.json ADDED
@@ -0,0 +1,45 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alora_invocation_tokens": null,
3
+ "alpha_pattern": {},
4
+ "arrow_config": null,
5
+ "auto_mapping": null,
6
+ "base_model_name_or_path": "Qwen/Qwen3-1.7B",
7
+ "bias": "none",
8
+ "corda_config": null,
9
+ "ensure_weight_tying": false,
10
+ "eva_config": null,
11
+ "exclude_modules": null,
12
+ "fan_in_fan_out": false,
13
+ "inference_mode": true,
14
+ "init_lora_weights": true,
15
+ "layer_replication": null,
16
+ "layers_pattern": null,
17
+ "layers_to_transform": null,
18
+ "loftq_config": {},
19
+ "lora_alpha": 64,
20
+ "lora_bias": false,
21
+ "lora_dropout": 0.05,
22
+ "lora_ga_config": null,
23
+ "megatron_config": null,
24
+ "megatron_core": "megatron.core",
25
+ "modules_to_save": null,
26
+ "peft_type": "LORA",
27
+ "peft_version": "0.19.1",
28
+ "qalora_group_size": 16,
29
+ "r": 64,
30
+ "rank_pattern": {},
31
+ "revision": null,
32
+ "target_modules": [
33
+ "k_proj",
34
+ "v_proj",
35
+ "q_proj",
36
+ "o_proj"
37
+ ],
38
+ "target_parameters": null,
39
+ "task_type": "CAUSAL_LM",
40
+ "trainable_token_indices": null,
41
+ "use_bdlora": null,
42
+ "use_dora": false,
43
+ "use_qalora": false,
44
+ "use_rslora": false
45
+ }
Conifer/PD/Qwen3-1.7B/checkpoint-400/chat_template.jinja ADDED
@@ -0,0 +1,89 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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 message.content is string 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.content is string %}
27
+ {%- set content = message.content %}
28
+ {%- else %}
29
+ {%- set content = '' %}
30
+ {%- endif %}
31
+ {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
32
+ {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
33
+ {%- elif message.role == "assistant" %}
34
+ {%- set reasoning_content = '' %}
35
+ {%- if message.reasoning_content is string %}
36
+ {%- set reasoning_content = message.reasoning_content %}
37
+ {%- else %}
38
+ {%- if '</think>' in content %}
39
+ {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
40
+ {%- set content = content.split('</think>')[-1].lstrip('\n') %}
41
+ {%- endif %}
42
+ {%- endif %}
43
+ {%- if loop.index0 > ns.last_query_index %}
44
+ {%- if loop.last or (not loop.last and reasoning_content) %}
45
+ {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
46
+ {%- else %}
47
+ {{- '<|im_start|>' + message.role + '\n' + content }}
48
+ {%- endif %}
49
+ {%- else %}
50
+ {{- '<|im_start|>' + message.role + '\n' + content }}
51
+ {%- endif %}
52
+ {%- if message.tool_calls %}
53
+ {%- for tool_call in message.tool_calls %}
54
+ {%- if (loop.first and content) or (not loop.first) %}
55
+ {{- '\n' }}
56
+ {%- endif %}
57
+ {%- if tool_call.function %}
58
+ {%- set tool_call = tool_call.function %}
59
+ {%- endif %}
60
+ {{- '<tool_call>\n{"name": "' }}
61
+ {{- tool_call.name }}
62
+ {{- '", "arguments": ' }}
63
+ {%- if tool_call.arguments is string %}
64
+ {{- tool_call.arguments }}
65
+ {%- else %}
66
+ {{- tool_call.arguments | tojson }}
67
+ {%- endif %}
68
+ {{- '}\n</tool_call>' }}
69
+ {%- endfor %}
70
+ {%- endif %}
71
+ {{- '<|im_end|>\n' }}
72
+ {%- elif message.role == "tool" %}
73
+ {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
74
+ {{- '<|im_start|>user' }}
75
+ {%- endif %}
76
+ {{- '\n<tool_response>\n' }}
77
+ {{- content }}
78
+ {{- '\n</tool_response>' }}
79
+ {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
80
+ {{- '<|im_end|>\n' }}
81
+ {%- endif %}
82
+ {%- endif %}
83
+ {%- endfor %}
84
+ {%- if add_generation_prompt %}
85
+ {{- '<|im_start|>assistant\n' }}
86
+ {%- if enable_thinking is defined and enable_thinking is false %}
87
+ {{- '<think>\n\n</think>\n\n' }}
88
+ {%- endif %}
89
+ {%- endif %}
Conifer/PD/Qwen3-1.7B/checkpoint-400/tokenizer_config.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
+ "<|im_start|>",
10
+ "<|im_end|>",
11
+ "<|object_ref_start|>",
12
+ "<|object_ref_end|>",
13
+ "<|box_start|>",
14
+ "<|box_end|>",
15
+ "<|quad_start|>",
16
+ "<|quad_end|>",
17
+ "<|vision_start|>",
18
+ "<|vision_end|>",
19
+ "<|vision_pad|>",
20
+ "<|image_pad|>",
21
+ "<|video_pad|>"
22
+ ],
23
+ "is_local": true,
24
+ "local_files_only": false,
25
+ "model_max_length": 131072,
26
+ "pad_token": "<|endoftext|>",
27
+ "split_special_tokens": false,
28
+ "tokenizer_class": "Qwen2Tokenizer",
29
+ "unk_token": null
30
+ }
Conifer/PD/Qwen3-1.7B/checkpoint-400/trainer_state.json ADDED
@@ -0,0 +1,355 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": 400,
3
+ "best_metric": 0.1748906821012497,
4
+ "best_model_checkpoint": "source_project/conifer_training_hub/11_runs/sft/conifer_stratified_sft_full_rl_20260902/A1/checkpoint-400",
5
+ "epoch": 2.1507402422611035,
6
+ "eval_steps": 100,
7
+ "global_step": 400,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "epoch": 0.05383580080753701,
14
+ "grad_norm": 0.9229851961135864,
15
+ "learning_rate": 6.4285714285714295e-06,
16
+ "loss": 0.3553433418273926,
17
+ "step": 10
18
+ },
19
+ {
20
+ "epoch": 0.10767160161507403,
21
+ "grad_norm": 0.3959938883781433,
22
+ "learning_rate": 1.3571428571428574e-05,
23
+ "loss": 0.32763218879699707,
24
+ "step": 20
25
+ },
26
+ {
27
+ "epoch": 0.16150740242261102,
28
+ "grad_norm": 0.13564534485340118,
29
+ "learning_rate": 1.9999824322258754e-05,
30
+ "loss": 0.25324430465698244,
31
+ "step": 30
32
+ },
33
+ {
34
+ "epoch": 0.21534320323014805,
35
+ "grad_norm": 0.10541348159313202,
36
+ "learning_rate": 1.9978750461051698e-05,
37
+ "loss": 0.2291196346282959,
38
+ "step": 40
39
+ },
40
+ {
41
+ "epoch": 0.2691790040376851,
42
+ "grad_norm": 0.08480330556631088,
43
+ "learning_rate": 1.9922625874911624e-05,
44
+ "loss": 0.22366776466369628,
45
+ "step": 50
46
+ },
47
+ {
48
+ "epoch": 0.32301480484522205,
49
+ "grad_norm": 0.09931497275829315,
50
+ "learning_rate": 1.9831647703494287e-05,
51
+ "loss": 0.220938777923584,
52
+ "step": 60
53
+ },
54
+ {
55
+ "epoch": 0.3768506056527591,
56
+ "grad_norm": 0.08036182820796967,
57
+ "learning_rate": 1.9706135510944894e-05,
58
+ "loss": 0.20970940589904785,
59
+ "step": 70
60
+ },
61
+ {
62
+ "epoch": 0.4306864064602961,
63
+ "grad_norm": 0.08407500386238098,
64
+ "learning_rate": 1.9546530163417374e-05,
65
+ "loss": 0.2213752269744873,
66
+ "step": 80
67
+ },
68
+ {
69
+ "epoch": 0.4845222072678331,
70
+ "grad_norm": 0.08968329429626465,
71
+ "learning_rate": 1.9353392280515938e-05,
72
+ "loss": 0.19955954551696778,
73
+ "step": 90
74
+ },
75
+ {
76
+ "epoch": 0.5383580080753702,
77
+ "grad_norm": 0.09176259487867355,
78
+ "learning_rate": 1.912740026609828e-05,
79
+ "loss": 0.19643588066101075,
80
+ "step": 100
81
+ },
82
+ {
83
+ "epoch": 0.5383580080753702,
84
+ "eval_loss": 0.19940482079982758,
85
+ "eval_runtime": 59.3187,
86
+ "eval_samples_per_second": 22.253,
87
+ "eval_steps_per_second": 1.399,
88
+ "step": 100
89
+ },
90
+ {
91
+ "epoch": 0.5921938088829072,
92
+ "grad_norm": 0.08072628080844879,
93
+ "learning_rate": 1.8869347925357275e-05,
94
+ "loss": 0.18773868083953857,
95
+ "step": 110
96
+ },
97
+ {
98
+ "epoch": 0.6460296096904441,
99
+ "grad_norm": 0.090627022087574,
100
+ "learning_rate": 1.8580141676551298e-05,
101
+ "loss": 0.19360114336013795,
102
+ "step": 120
103
+ },
104
+ {
105
+ "epoch": 0.6998654104979811,
106
+ "grad_norm": 0.076926589012146,
107
+ "learning_rate": 1.826079736717702e-05,
108
+ "loss": 0.1843669056892395,
109
+ "step": 130
110
+ },
111
+ {
112
+ "epoch": 0.7537012113055181,
113
+ "grad_norm": 0.07070426642894745,
114
+ "learning_rate": 1.7912436705768045e-05,
115
+ "loss": 0.1653825283050537,
116
+ "step": 140
117
+ },
118
+ {
119
+ "epoch": 0.8075370121130552,
120
+ "grad_norm": 0.07505597174167633,
121
+ "learning_rate": 1.753628332185275e-05,
122
+ "loss": 0.17851850986480713,
123
+ "step": 150
124
+ },
125
+ {
126
+ "epoch": 0.8613728129205922,
127
+ "grad_norm": 0.0769675076007843,
128
+ "learning_rate": 1.7133658467910968e-05,
129
+ "loss": 0.20060482025146484,
130
+ "step": 160
131
+ },
132
+ {
133
+ "epoch": 0.9152086137281292,
134
+ "grad_norm": 0.08414696902036667,
135
+ "learning_rate": 1.6705976378426447e-05,
136
+ "loss": 0.1934957981109619,
137
+ "step": 170
138
+ },
139
+ {
140
+ "epoch": 0.9690444145356663,
141
+ "grad_norm": 0.07856564223766327,
142
+ "learning_rate": 1.625473930233664e-05,
143
+ "loss": 0.18108274936676025,
144
+ "step": 180
145
+ },
146
+ {
147
+ "epoch": 1.0215343203230147,
148
+ "grad_norm": 0.09861781448125839,
149
+ "learning_rate": 1.5781532226328544e-05,
150
+ "loss": 0.2005953311920166,
151
+ "step": 190
152
+ },
153
+ {
154
+ "epoch": 1.0753701211305517,
155
+ "grad_norm": 0.09661930799484253,
156
+ "learning_rate": 1.5288017307515142e-05,
157
+ "loss": 0.18249776363372802,
158
+ "step": 200
159
+ },
160
+ {
161
+ "epoch": 1.0753701211305517,
162
+ "eval_loss": 0.1815323382616043,
163
+ "eval_runtime": 60.9448,
164
+ "eval_samples_per_second": 21.659,
165
+ "eval_steps_per_second": 1.362,
166
+ "step": 200
167
+ },
168
+ {
169
+ "epoch": 1.1292059219380888,
170
+ "grad_norm": 0.08623727411031723,
171
+ "learning_rate": 1.4775928035047928e-05,
172
+ "loss": 0.17418408393859863,
173
+ "step": 210
174
+ },
175
+ {
176
+ "epoch": 1.1830417227456258,
177
+ "grad_norm": 0.09250917285680771,
178
+ "learning_rate": 1.424706314117303e-05,
179
+ "loss": 0.19033379554748536,
180
+ "step": 220
181
+ },
182
+ {
183
+ "epoch": 1.2368775235531628,
184
+ "grad_norm": 0.0896449089050293,
185
+ "learning_rate": 1.3703280283118601e-05,
186
+ "loss": 0.17169111967086792,
187
+ "step": 230
188
+ },
189
+ {
190
+ "epoch": 1.2907133243606999,
191
+ "grad_norm": 0.07717910408973694,
192
+ "learning_rate": 1.3146489518006065e-05,
193
+ "loss": 0.1807547926902771,
194
+ "step": 240
195
+ },
196
+ {
197
+ "epoch": 1.3445491251682369,
198
+ "grad_norm": 0.0989438146352768,
199
+ "learning_rate": 1.2578646593704786e-05,
200
+ "loss": 0.16931524276733398,
201
+ "step": 250
202
+ },
203
+ {
204
+ "epoch": 1.398384925975774,
205
+ "grad_norm": 0.08172167092561722,
206
+ "learning_rate": 1.2001746079196381e-05,
207
+ "loss": 0.17504738569259642,
208
+ "step": 260
209
+ },
210
+ {
211
+ "epoch": 1.452220726783311,
212
+ "grad_norm": 0.08974647521972656,
213
+ "learning_rate": 1.14178143585784e-05,
214
+ "loss": 0.1663546919822693,
215
+ "step": 270
216
+ },
217
+ {
218
+ "epoch": 1.506056527590848,
219
+ "grad_norm": 0.0908568724989891,
220
+ "learning_rate": 1.0828902513316299e-05,
221
+ "loss": 0.1784202814102173,
222
+ "step": 280
223
+ },
224
+ {
225
+ "epoch": 1.559892328398385,
226
+ "grad_norm": 0.08752070367336273,
227
+ "learning_rate": 1.0237079117744977e-05,
228
+ "loss": 0.17578048706054689,
229
+ "step": 290
230
+ },
231
+ {
232
+ "epoch": 1.613728129205922,
233
+ "grad_norm": 0.08162529021501541,
234
+ "learning_rate": 9.644422973125977e-06,
235
+ "loss": 0.1648276448249817,
236
+ "step": 300
237
+ },
238
+ {
239
+ "epoch": 1.613728129205922,
240
+ "eval_loss": 0.17673912644386292,
241
+ "eval_runtime": 62.7878,
242
+ "eval_samples_per_second": 21.023,
243
+ "eval_steps_per_second": 1.322,
244
+ "step": 300
245
+ },
246
+ {
247
+ "epoch": 1.667563930013459,
248
+ "grad_norm": 0.09499260038137436,
249
+ "learning_rate": 9.053015805782249e-06,
250
+ "loss": 0.17204223871231078,
251
+ "step": 310
252
+ },
253
+ {
254
+ "epoch": 1.721399730820996,
255
+ "grad_norm": 0.0870763510465622,
256
+ "learning_rate": 8.46493495495851e-06,
257
+ "loss": 0.17899261713027953,
258
+ "step": 320
259
+ },
260
+ {
261
+ "epoch": 1.775235531628533,
262
+ "grad_norm": 0.0877470001578331,
263
+ "learning_rate": 7.882246076091385e-06,
264
+ "loss": 0.17512272596359252,
265
+ "step": 330
266
+ },
267
+ {
268
+ "epoch": 1.8290713324360701,
269
+ "grad_norm": 0.10601353645324707,
270
+ "learning_rate": 7.306995885119349e-06,
271
+ "loss": 0.1890477180480957,
272
+ "step": 340
273
+ },
274
+ {
275
+ "epoch": 1.8829071332436071,
276
+ "grad_norm": 0.08982375264167786,
277
+ "learning_rate": 6.741204969318343e-06,
278
+ "loss": 0.16031824350357055,
279
+ "step": 350
280
+ },
281
+ {
282
+ "epoch": 1.9367429340511442,
283
+ "grad_norm": 0.07413674145936966,
284
+ "learning_rate": 6.186860689915286e-06,
285
+ "loss": 0.17462934255599977,
286
+ "step": 360
287
+ },
288
+ {
289
+ "epoch": 1.990578734858681,
290
+ "grad_norm": 0.07907533645629883,
291
+ "learning_rate": 5.645910201409268e-06,
292
+ "loss": 0.1853417158126831,
293
+ "step": 370
294
+ },
295
+ {
296
+ "epoch": 2.0430686406460294,
297
+ "grad_norm": 0.07409507036209106,
298
+ "learning_rate": 5.120253612120363e-06,
299
+ "loss": 0.1760026216506958,
300
+ "step": 380
301
+ },
302
+ {
303
+ "epoch": 2.0969044414535665,
304
+ "grad_norm": 0.09365784376859665,
305
+ "learning_rate": 4.611737309989861e-06,
306
+ "loss": 0.17717018127441406,
307
+ "step": 390
308
+ },
309
+ {
310
+ "epoch": 2.1507402422611035,
311
+ "grad_norm": 0.08592981100082397,
312
+ "learning_rate": 4.12214747707527e-06,
313
+ "loss": 0.1779325485229492,
314
+ "step": 400
315
+ },
316
+ {
317
+ "epoch": 2.1507402422611035,
318
+ "eval_loss": 0.1748906821012497,
319
+ "eval_runtime": 61.9646,
320
+ "eval_samples_per_second": 21.302,
321
+ "eval_steps_per_second": 1.339,
322
+ "step": 400
323
+ }
324
+ ],
325
+ "logging_steps": 10,
326
+ "max_steps": 558,
327
+ "num_input_tokens_seen": 0,
328
+ "num_train_epochs": 3,
329
+ "save_steps": 100,
330
+ "stateful_callbacks": {
331
+ "EarlyStoppingCallback": {
332
+ "args": {
333
+ "early_stopping_patience": 3,
334
+ "early_stopping_threshold": 0.0
335
+ },
336
+ "attributes": {
337
+ "early_stopping_patience_counter": 0
338
+ }
339
+ },
340
+ "TrainerControl": {
341
+ "args": {
342
+ "should_epoch_stop": false,
343
+ "should_evaluate": false,
344
+ "should_log": false,
345
+ "should_save": true,
346
+ "should_training_stop": false
347
+ },
348
+ "attributes": {}
349
+ }
350
+ },
351
+ "total_flos": 4.077453831254835e+17,
352
+ "train_batch_size": 2,
353
+ "trial_name": null,
354
+ "trial_params": null
355
+ }
Conifer/PD/Qwen3-1.7B/checkpoint-500/adapter_config.json ADDED
@@ -0,0 +1,45 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alora_invocation_tokens": null,
3
+ "alpha_pattern": {},
4
+ "arrow_config": null,
5
+ "auto_mapping": null,
6
+ "base_model_name_or_path": "Qwen/Qwen3-1.7B",
7
+ "bias": "none",
8
+ "corda_config": null,
9
+ "ensure_weight_tying": false,
10
+ "eva_config": null,
11
+ "exclude_modules": null,
12
+ "fan_in_fan_out": false,
13
+ "inference_mode": true,
14
+ "init_lora_weights": true,
15
+ "layer_replication": null,
16
+ "layers_pattern": null,
17
+ "layers_to_transform": null,
18
+ "loftq_config": {},
19
+ "lora_alpha": 64,
20
+ "lora_bias": false,
21
+ "lora_dropout": 0.05,
22
+ "lora_ga_config": null,
23
+ "megatron_config": null,
24
+ "megatron_core": "megatron.core",
25
+ "modules_to_save": null,
26
+ "peft_type": "LORA",
27
+ "peft_version": "0.19.1",
28
+ "qalora_group_size": 16,
29
+ "r": 64,
30
+ "rank_pattern": {},
31
+ "revision": null,
32
+ "target_modules": [
33
+ "k_proj",
34
+ "v_proj",
35
+ "q_proj",
36
+ "o_proj"
37
+ ],
38
+ "target_parameters": null,
39
+ "task_type": "CAUSAL_LM",
40
+ "trainable_token_indices": null,
41
+ "use_bdlora": null,
42
+ "use_dora": false,
43
+ "use_qalora": false,
44
+ "use_rslora": false
45
+ }
Conifer/PD/Qwen3-1.7B/checkpoint-500/chat_template.jinja ADDED
@@ -0,0 +1,89 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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 message.content is string 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.content is string %}
27
+ {%- set content = message.content %}
28
+ {%- else %}
29
+ {%- set content = '' %}
30
+ {%- endif %}
31
+ {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
32
+ {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
33
+ {%- elif message.role == "assistant" %}
34
+ {%- set reasoning_content = '' %}
35
+ {%- if message.reasoning_content is string %}
36
+ {%- set reasoning_content = message.reasoning_content %}
37
+ {%- else %}
38
+ {%- if '</think>' in content %}
39
+ {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
40
+ {%- set content = content.split('</think>')[-1].lstrip('\n') %}
41
+ {%- endif %}
42
+ {%- endif %}
43
+ {%- if loop.index0 > ns.last_query_index %}
44
+ {%- if loop.last or (not loop.last and reasoning_content) %}
45
+ {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
46
+ {%- else %}
47
+ {{- '<|im_start|>' + message.role + '\n' + content }}
48
+ {%- endif %}
49
+ {%- else %}
50
+ {{- '<|im_start|>' + message.role + '\n' + content }}
51
+ {%- endif %}
52
+ {%- if message.tool_calls %}
53
+ {%- for tool_call in message.tool_calls %}
54
+ {%- if (loop.first and content) or (not loop.first) %}
55
+ {{- '\n' }}
56
+ {%- endif %}
57
+ {%- if tool_call.function %}
58
+ {%- set tool_call = tool_call.function %}
59
+ {%- endif %}
60
+ {{- '<tool_call>\n{"name": "' }}
61
+ {{- tool_call.name }}
62
+ {{- '", "arguments": ' }}
63
+ {%- if tool_call.arguments is string %}
64
+ {{- tool_call.arguments }}
65
+ {%- else %}
66
+ {{- tool_call.arguments | tojson }}
67
+ {%- endif %}
68
+ {{- '}\n</tool_call>' }}
69
+ {%- endfor %}
70
+ {%- endif %}
71
+ {{- '<|im_end|>\n' }}
72
+ {%- elif message.role == "tool" %}
73
+ {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
74
+ {{- '<|im_start|>user' }}
75
+ {%- endif %}
76
+ {{- '\n<tool_response>\n' }}
77
+ {{- content }}
78
+ {{- '\n</tool_response>' }}
79
+ {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
80
+ {{- '<|im_end|>\n' }}
81
+ {%- endif %}
82
+ {%- endif %}
83
+ {%- endfor %}
84
+ {%- if add_generation_prompt %}
85
+ {{- '<|im_start|>assistant\n' }}
86
+ {%- if enable_thinking is defined and enable_thinking is false %}
87
+ {{- '<think>\n\n</think>\n\n' }}
88
+ {%- endif %}
89
+ {%- endif %}
Conifer/PD/Qwen3-1.7B/checkpoint-500/tokenizer_config.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
+ "<|im_start|>",
10
+ "<|im_end|>",
11
+ "<|object_ref_start|>",
12
+ "<|object_ref_end|>",
13
+ "<|box_start|>",
14
+ "<|box_end|>",
15
+ "<|quad_start|>",
16
+ "<|quad_end|>",
17
+ "<|vision_start|>",
18
+ "<|vision_end|>",
19
+ "<|vision_pad|>",
20
+ "<|image_pad|>",
21
+ "<|video_pad|>"
22
+ ],
23
+ "is_local": true,
24
+ "local_files_only": false,
25
+ "model_max_length": 131072,
26
+ "pad_token": "<|endoftext|>",
27
+ "split_special_tokens": false,
28
+ "tokenizer_class": "Qwen2Tokenizer",
29
+ "unk_token": null
30
+ }
Conifer/PD/Qwen3-1.7B/checkpoint-500/trainer_state.json ADDED
@@ -0,0 +1,433 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": 500,
3
+ "best_metric": 0.17438176274299622,
4
+ "best_model_checkpoint": "source_project/conifer_training_hub/11_runs/sft/conifer_stratified_sft_full_rl_20260902/A1/checkpoint-500",
5
+ "epoch": 2.6890982503364738,
6
+ "eval_steps": 100,
7
+ "global_step": 500,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "epoch": 0.05383580080753701,
14
+ "grad_norm": 0.9229851961135864,
15
+ "learning_rate": 6.4285714285714295e-06,
16
+ "loss": 0.3553433418273926,
17
+ "step": 10
18
+ },
19
+ {
20
+ "epoch": 0.10767160161507403,
21
+ "grad_norm": 0.3959938883781433,
22
+ "learning_rate": 1.3571428571428574e-05,
23
+ "loss": 0.32763218879699707,
24
+ "step": 20
25
+ },
26
+ {
27
+ "epoch": 0.16150740242261102,
28
+ "grad_norm": 0.13564534485340118,
29
+ "learning_rate": 1.9999824322258754e-05,
30
+ "loss": 0.25324430465698244,
31
+ "step": 30
32
+ },
33
+ {
34
+ "epoch": 0.21534320323014805,
35
+ "grad_norm": 0.10541348159313202,
36
+ "learning_rate": 1.9978750461051698e-05,
37
+ "loss": 0.2291196346282959,
38
+ "step": 40
39
+ },
40
+ {
41
+ "epoch": 0.2691790040376851,
42
+ "grad_norm": 0.08480330556631088,
43
+ "learning_rate": 1.9922625874911624e-05,
44
+ "loss": 0.22366776466369628,
45
+ "step": 50
46
+ },
47
+ {
48
+ "epoch": 0.32301480484522205,
49
+ "grad_norm": 0.09931497275829315,
50
+ "learning_rate": 1.9831647703494287e-05,
51
+ "loss": 0.220938777923584,
52
+ "step": 60
53
+ },
54
+ {
55
+ "epoch": 0.3768506056527591,
56
+ "grad_norm": 0.08036182820796967,
57
+ "learning_rate": 1.9706135510944894e-05,
58
+ "loss": 0.20970940589904785,
59
+ "step": 70
60
+ },
61
+ {
62
+ "epoch": 0.4306864064602961,
63
+ "grad_norm": 0.08407500386238098,
64
+ "learning_rate": 1.9546530163417374e-05,
65
+ "loss": 0.2213752269744873,
66
+ "step": 80
67
+ },
68
+ {
69
+ "epoch": 0.4845222072678331,
70
+ "grad_norm": 0.08968329429626465,
71
+ "learning_rate": 1.9353392280515938e-05,
72
+ "loss": 0.19955954551696778,
73
+ "step": 90
74
+ },
75
+ {
76
+ "epoch": 0.5383580080753702,
77
+ "grad_norm": 0.09176259487867355,
78
+ "learning_rate": 1.912740026609828e-05,
79
+ "loss": 0.19643588066101075,
80
+ "step": 100
81
+ },
82
+ {
83
+ "epoch": 0.5383580080753702,
84
+ "eval_loss": 0.19940482079982758,
85
+ "eval_runtime": 59.3187,
86
+ "eval_samples_per_second": 22.253,
87
+ "eval_steps_per_second": 1.399,
88
+ "step": 100
89
+ },
90
+ {
91
+ "epoch": 0.5921938088829072,
92
+ "grad_norm": 0.08072628080844879,
93
+ "learning_rate": 1.8869347925357275e-05,
94
+ "loss": 0.18773868083953857,
95
+ "step": 110
96
+ },
97
+ {
98
+ "epoch": 0.6460296096904441,
99
+ "grad_norm": 0.090627022087574,
100
+ "learning_rate": 1.8580141676551298e-05,
101
+ "loss": 0.19360114336013795,
102
+ "step": 120
103
+ },
104
+ {
105
+ "epoch": 0.6998654104979811,
106
+ "grad_norm": 0.076926589012146,
107
+ "learning_rate": 1.826079736717702e-05,
108
+ "loss": 0.1843669056892395,
109
+ "step": 130
110
+ },
111
+ {
112
+ "epoch": 0.7537012113055181,
113
+ "grad_norm": 0.07070426642894745,
114
+ "learning_rate": 1.7912436705768045e-05,
115
+ "loss": 0.1653825283050537,
116
+ "step": 140
117
+ },
118
+ {
119
+ "epoch": 0.8075370121130552,
120
+ "grad_norm": 0.07505597174167633,
121
+ "learning_rate": 1.753628332185275e-05,
122
+ "loss": 0.17851850986480713,
123
+ "step": 150
124
+ },
125
+ {
126
+ "epoch": 0.8613728129205922,
127
+ "grad_norm": 0.0769675076007843,
128
+ "learning_rate": 1.7133658467910968e-05,
129
+ "loss": 0.20060482025146484,
130
+ "step": 160
131
+ },
132
+ {
133
+ "epoch": 0.9152086137281292,
134
+ "grad_norm": 0.08414696902036667,
135
+ "learning_rate": 1.6705976378426447e-05,
136
+ "loss": 0.1934957981109619,
137
+ "step": 170
138
+ },
139
+ {
140
+ "epoch": 0.9690444145356663,
141
+ "grad_norm": 0.07856564223766327,
142
+ "learning_rate": 1.625473930233664e-05,
143
+ "loss": 0.18108274936676025,
144
+ "step": 180
145
+ },
146
+ {
147
+ "epoch": 1.0215343203230147,
148
+ "grad_norm": 0.09861781448125839,
149
+ "learning_rate": 1.5781532226328544e-05,
150
+ "loss": 0.2005953311920166,
151
+ "step": 190
152
+ },
153
+ {
154
+ "epoch": 1.0753701211305517,
155
+ "grad_norm": 0.09661930799484253,
156
+ "learning_rate": 1.5288017307515142e-05,
157
+ "loss": 0.18249776363372802,
158
+ "step": 200
159
+ },
160
+ {
161
+ "epoch": 1.0753701211305517,
162
+ "eval_loss": 0.1815323382616043,
163
+ "eval_runtime": 60.9448,
164
+ "eval_samples_per_second": 21.659,
165
+ "eval_steps_per_second": 1.362,
166
+ "step": 200
167
+ },
168
+ {
169
+ "epoch": 1.1292059219380888,
170
+ "grad_norm": 0.08623727411031723,
171
+ "learning_rate": 1.4775928035047928e-05,
172
+ "loss": 0.17418408393859863,
173
+ "step": 210
174
+ },
175
+ {
176
+ "epoch": 1.1830417227456258,
177
+ "grad_norm": 0.09250917285680771,
178
+ "learning_rate": 1.424706314117303e-05,
179
+ "loss": 0.19033379554748536,
180
+ "step": 220
181
+ },
182
+ {
183
+ "epoch": 1.2368775235531628,
184
+ "grad_norm": 0.0896449089050293,
185
+ "learning_rate": 1.3703280283118601e-05,
186
+ "loss": 0.17169111967086792,
187
+ "step": 230
188
+ },
189
+ {
190
+ "epoch": 1.2907133243606999,
191
+ "grad_norm": 0.07717910408973694,
192
+ "learning_rate": 1.3146489518006065e-05,
193
+ "loss": 0.1807547926902771,
194
+ "step": 240
195
+ },
196
+ {
197
+ "epoch": 1.3445491251682369,
198
+ "grad_norm": 0.0989438146352768,
199
+ "learning_rate": 1.2578646593704786e-05,
200
+ "loss": 0.16931524276733398,
201
+ "step": 250
202
+ },
203
+ {
204
+ "epoch": 1.398384925975774,
205
+ "grad_norm": 0.08172167092561722,
206
+ "learning_rate": 1.2001746079196381e-05,
207
+ "loss": 0.17504738569259642,
208
+ "step": 260
209
+ },
210
+ {
211
+ "epoch": 1.452220726783311,
212
+ "grad_norm": 0.08974647521972656,
213
+ "learning_rate": 1.14178143585784e-05,
214
+ "loss": 0.1663546919822693,
215
+ "step": 270
216
+ },
217
+ {
218
+ "epoch": 1.506056527590848,
219
+ "grad_norm": 0.0908568724989891,
220
+ "learning_rate": 1.0828902513316299e-05,
221
+ "loss": 0.1784202814102173,
222
+ "step": 280
223
+ },
224
+ {
225
+ "epoch": 1.559892328398385,
226
+ "grad_norm": 0.08752070367336273,
227
+ "learning_rate": 1.0237079117744977e-05,
228
+ "loss": 0.17578048706054689,
229
+ "step": 290
230
+ },
231
+ {
232
+ "epoch": 1.613728129205922,
233
+ "grad_norm": 0.08162529021501541,
234
+ "learning_rate": 9.644422973125977e-06,
235
+ "loss": 0.1648276448249817,
236
+ "step": 300
237
+ },
238
+ {
239
+ "epoch": 1.613728129205922,
240
+ "eval_loss": 0.17673912644386292,
241
+ "eval_runtime": 62.7878,
242
+ "eval_samples_per_second": 21.023,
243
+ "eval_steps_per_second": 1.322,
244
+ "step": 300
245
+ },
246
+ {
247
+ "epoch": 1.667563930013459,
248
+ "grad_norm": 0.09499260038137436,
249
+ "learning_rate": 9.053015805782249e-06,
250
+ "loss": 0.17204223871231078,
251
+ "step": 310
252
+ },
253
+ {
254
+ "epoch": 1.721399730820996,
255
+ "grad_norm": 0.0870763510465622,
256
+ "learning_rate": 8.46493495495851e-06,
257
+ "loss": 0.17899261713027953,
258
+ "step": 320
259
+ },
260
+ {
261
+ "epoch": 1.775235531628533,
262
+ "grad_norm": 0.0877470001578331,
263
+ "learning_rate": 7.882246076091385e-06,
264
+ "loss": 0.17512272596359252,
265
+ "step": 330
266
+ },
267
+ {
268
+ "epoch": 1.8290713324360701,
269
+ "grad_norm": 0.10601353645324707,
270
+ "learning_rate": 7.306995885119349e-06,
271
+ "loss": 0.1890477180480957,
272
+ "step": 340
273
+ },
274
+ {
275
+ "epoch": 1.8829071332436071,
276
+ "grad_norm": 0.08982375264167786,
277
+ "learning_rate": 6.741204969318343e-06,
278
+ "loss": 0.16031824350357055,
279
+ "step": 350
280
+ },
281
+ {
282
+ "epoch": 1.9367429340511442,
283
+ "grad_norm": 0.07413674145936966,
284
+ "learning_rate": 6.186860689915286e-06,
285
+ "loss": 0.17462934255599977,
286
+ "step": 360
287
+ },
288
+ {
289
+ "epoch": 1.990578734858681,
290
+ "grad_norm": 0.07907533645629883,
291
+ "learning_rate": 5.645910201409268e-06,
292
+ "loss": 0.1853417158126831,
293
+ "step": 370
294
+ },
295
+ {
296
+ "epoch": 2.0430686406460294,
297
+ "grad_norm": 0.07409507036209106,
298
+ "learning_rate": 5.120253612120363e-06,
299
+ "loss": 0.1760026216506958,
300
+ "step": 380
301
+ },
302
+ {
303
+ "epoch": 2.0969044414535665,
304
+ "grad_norm": 0.09365784376859665,
305
+ "learning_rate": 4.611737309989861e-06,
306
+ "loss": 0.17717018127441406,
307
+ "step": 390
308
+ },
309
+ {
310
+ "epoch": 2.1507402422611035,
311
+ "grad_norm": 0.08592981100082397,
312
+ "learning_rate": 4.12214747707527e-06,
313
+ "loss": 0.1779325485229492,
314
+ "step": 400
315
+ },
316
+ {
317
+ "epoch": 2.1507402422611035,
318
+ "eval_loss": 0.1748906821012497,
319
+ "eval_runtime": 61.9646,
320
+ "eval_samples_per_second": 21.302,
321
+ "eval_steps_per_second": 1.339,
322
+ "step": 400
323
+ },
324
+ {
325
+ "epoch": 2.2045760430686405,
326
+ "grad_norm": 0.09960120916366577,
327
+ "learning_rate": 3.653203815520714e-06,
328
+ "loss": 0.17972402572631835,
329
+ "step": 410
330
+ },
331
+ {
332
+ "epoch": 2.2584118438761775,
333
+ "grad_norm": 0.10758821666240692,
334
+ "learning_rate": 3.2065535070404085e-06,
335
+ "loss": 0.1703733205795288,
336
+ "step": 420
337
+ },
338
+ {
339
+ "epoch": 2.3122476446837146,
340
+ "grad_norm": 0.0882757231593132,
341
+ "learning_rate": 2.783765427132837e-06,
342
+ "loss": 0.1815846562385559,
343
+ "step": 430
344
+ },
345
+ {
346
+ "epoch": 2.3660834454912516,
347
+ "grad_norm": 0.08758968114852905,
348
+ "learning_rate": 2.3863246343483306e-06,
349
+ "loss": 0.15952233076095582,
350
+ "step": 440
351
+ },
352
+ {
353
+ "epoch": 2.4199192462987886,
354
+ "grad_norm": 0.10103869438171387,
355
+ "learning_rate": 2.0156271539667517e-06,
356
+ "loss": 0.17301445007324218,
357
+ "step": 450
358
+ },
359
+ {
360
+ "epoch": 2.4737550471063257,
361
+ "grad_norm": 0.09046496450901031,
362
+ "learning_rate": 1.6729750744077755e-06,
363
+ "loss": 0.16349974870681763,
364
+ "step": 460
365
+ },
366
+ {
367
+ "epoch": 2.5275908479138627,
368
+ "grad_norm": 0.08671458810567856,
369
+ "learning_rate": 1.3595719735978451e-06,
370
+ "loss": 0.1653430461883545,
371
+ "step": 470
372
+ },
373
+ {
374
+ "epoch": 2.5814266487213997,
375
+ "grad_norm": 0.10391299426555634,
376
+ "learning_rate": 1.0765186913587866e-06,
377
+ "loss": 0.17563060522079468,
378
+ "step": 480
379
+ },
380
+ {
381
+ "epoch": 2.6352624495289367,
382
+ "grad_norm": 0.08818709850311279,
383
+ "learning_rate": 8.24809462667805e-07,
384
+ "loss": 0.17396645545959472,
385
+ "step": 490
386
+ },
387
+ {
388
+ "epoch": 2.6890982503364738,
389
+ "grad_norm": 0.11322963237762451,
390
+ "learning_rate": 6.053284253708547e-07,
391
+ "loss": 0.16821794509887694,
392
+ "step": 500
393
+ },
394
+ {
395
+ "epoch": 2.6890982503364738,
396
+ "eval_loss": 0.17438176274299622,
397
+ "eval_runtime": 61.1022,
398
+ "eval_samples_per_second": 21.603,
399
+ "eval_steps_per_second": 1.358,
400
+ "step": 500
401
+ }
402
+ ],
403
+ "logging_steps": 10,
404
+ "max_steps": 558,
405
+ "num_input_tokens_seen": 0,
406
+ "num_train_epochs": 3,
407
+ "save_steps": 100,
408
+ "stateful_callbacks": {
409
+ "EarlyStoppingCallback": {
410
+ "args": {
411
+ "early_stopping_patience": 3,
412
+ "early_stopping_threshold": 0.0
413
+ },
414
+ "attributes": {
415
+ "early_stopping_patience_counter": 0
416
+ }
417
+ },
418
+ "TrainerControl": {
419
+ "args": {
420
+ "should_epoch_stop": false,
421
+ "should_evaluate": false,
422
+ "should_log": false,
423
+ "should_save": true,
424
+ "should_training_stop": false
425
+ },
426
+ "attributes": {}
427
+ }
428
+ },
429
+ "total_flos": 5.099886225737646e+17,
430
+ "train_batch_size": 2,
431
+ "trial_name": null,
432
+ "trial_params": null
433
+ }
Conifer/PD/Qwen3-1.7B/checkpoint-558/adapter_config.json ADDED
@@ -0,0 +1,45 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alora_invocation_tokens": null,
3
+ "alpha_pattern": {},
4
+ "arrow_config": null,
5
+ "auto_mapping": null,
6
+ "base_model_name_or_path": "Qwen/Qwen3-1.7B",
7
+ "bias": "none",
8
+ "corda_config": null,
9
+ "ensure_weight_tying": false,
10
+ "eva_config": null,
11
+ "exclude_modules": null,
12
+ "fan_in_fan_out": false,
13
+ "inference_mode": true,
14
+ "init_lora_weights": true,
15
+ "layer_replication": null,
16
+ "layers_pattern": null,
17
+ "layers_to_transform": null,
18
+ "loftq_config": {},
19
+ "lora_alpha": 64,
20
+ "lora_bias": false,
21
+ "lora_dropout": 0.05,
22
+ "lora_ga_config": null,
23
+ "megatron_config": null,
24
+ "megatron_core": "megatron.core",
25
+ "modules_to_save": null,
26
+ "peft_type": "LORA",
27
+ "peft_version": "0.19.1",
28
+ "qalora_group_size": 16,
29
+ "r": 64,
30
+ "rank_pattern": {},
31
+ "revision": null,
32
+ "target_modules": [
33
+ "k_proj",
34
+ "v_proj",
35
+ "q_proj",
36
+ "o_proj"
37
+ ],
38
+ "target_parameters": null,
39
+ "task_type": "CAUSAL_LM",
40
+ "trainable_token_indices": null,
41
+ "use_bdlora": null,
42
+ "use_dora": false,
43
+ "use_qalora": false,
44
+ "use_rslora": false
45
+ }
Conifer/PD/Qwen3-1.7B/checkpoint-558/chat_template.jinja ADDED
@@ -0,0 +1,89 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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 message.content is string 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.content is string %}
27
+ {%- set content = message.content %}
28
+ {%- else %}
29
+ {%- set content = '' %}
30
+ {%- endif %}
31
+ {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
32
+ {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
33
+ {%- elif message.role == "assistant" %}
34
+ {%- set reasoning_content = '' %}
35
+ {%- if message.reasoning_content is string %}
36
+ {%- set reasoning_content = message.reasoning_content %}
37
+ {%- else %}
38
+ {%- if '</think>' in content %}
39
+ {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
40
+ {%- set content = content.split('</think>')[-1].lstrip('\n') %}
41
+ {%- endif %}
42
+ {%- endif %}
43
+ {%- if loop.index0 > ns.last_query_index %}
44
+ {%- if loop.last or (not loop.last and reasoning_content) %}
45
+ {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
46
+ {%- else %}
47
+ {{- '<|im_start|>' + message.role + '\n' + content }}
48
+ {%- endif %}
49
+ {%- else %}
50
+ {{- '<|im_start|>' + message.role + '\n' + content }}
51
+ {%- endif %}
52
+ {%- if message.tool_calls %}
53
+ {%- for tool_call in message.tool_calls %}
54
+ {%- if (loop.first and content) or (not loop.first) %}
55
+ {{- '\n' }}
56
+ {%- endif %}
57
+ {%- if tool_call.function %}
58
+ {%- set tool_call = tool_call.function %}
59
+ {%- endif %}
60
+ {{- '<tool_call>\n{"name": "' }}
61
+ {{- tool_call.name }}
62
+ {{- '", "arguments": ' }}
63
+ {%- if tool_call.arguments is string %}
64
+ {{- tool_call.arguments }}
65
+ {%- else %}
66
+ {{- tool_call.arguments | tojson }}
67
+ {%- endif %}
68
+ {{- '}\n</tool_call>' }}
69
+ {%- endfor %}
70
+ {%- endif %}
71
+ {{- '<|im_end|>\n' }}
72
+ {%- elif message.role == "tool" %}
73
+ {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
74
+ {{- '<|im_start|>user' }}
75
+ {%- endif %}
76
+ {{- '\n<tool_response>\n' }}
77
+ {{- content }}
78
+ {{- '\n</tool_response>' }}
79
+ {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
80
+ {{- '<|im_end|>\n' }}
81
+ {%- endif %}
82
+ {%- endif %}
83
+ {%- endfor %}
84
+ {%- if add_generation_prompt %}
85
+ {{- '<|im_start|>assistant\n' }}
86
+ {%- if enable_thinking is defined and enable_thinking is false %}
87
+ {{- '<think>\n\n</think>\n\n' }}
88
+ {%- endif %}
89
+ {%- endif %}
Conifer/PD/Qwen3-1.7B/checkpoint-558/tokenizer_config.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
+ "<|im_start|>",
10
+ "<|im_end|>",
11
+ "<|object_ref_start|>",
12
+ "<|object_ref_end|>",
13
+ "<|box_start|>",
14
+ "<|box_end|>",
15
+ "<|quad_start|>",
16
+ "<|quad_end|>",
17
+ "<|vision_start|>",
18
+ "<|vision_end|>",
19
+ "<|vision_pad|>",
20
+ "<|image_pad|>",
21
+ "<|video_pad|>"
22
+ ],
23
+ "is_local": true,
24
+ "local_files_only": false,
25
+ "model_max_length": 131072,
26
+ "pad_token": "<|endoftext|>",
27
+ "split_special_tokens": false,
28
+ "tokenizer_class": "Qwen2Tokenizer",
29
+ "unk_token": null
30
+ }
Conifer/PD/Qwen3-1.7B/checkpoint-558/trainer_state.json ADDED
@@ -0,0 +1,476 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": 500,
3
+ "best_metric": 0.17438176274299622,
4
+ "best_model_checkpoint": "source_project/conifer_training_hub/11_runs/sft/conifer_stratified_sft_full_rl_20260902/A1/checkpoint-500",
5
+ "epoch": 3.0,
6
+ "eval_steps": 100,
7
+ "global_step": 558,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "epoch": 0.05383580080753701,
14
+ "grad_norm": 0.9229851961135864,
15
+ "learning_rate": 6.4285714285714295e-06,
16
+ "loss": 0.3553433418273926,
17
+ "step": 10
18
+ },
19
+ {
20
+ "epoch": 0.10767160161507403,
21
+ "grad_norm": 0.3959938883781433,
22
+ "learning_rate": 1.3571428571428574e-05,
23
+ "loss": 0.32763218879699707,
24
+ "step": 20
25
+ },
26
+ {
27
+ "epoch": 0.16150740242261102,
28
+ "grad_norm": 0.13564534485340118,
29
+ "learning_rate": 1.9999824322258754e-05,
30
+ "loss": 0.25324430465698244,
31
+ "step": 30
32
+ },
33
+ {
34
+ "epoch": 0.21534320323014805,
35
+ "grad_norm": 0.10541348159313202,
36
+ "learning_rate": 1.9978750461051698e-05,
37
+ "loss": 0.2291196346282959,
38
+ "step": 40
39
+ },
40
+ {
41
+ "epoch": 0.2691790040376851,
42
+ "grad_norm": 0.08480330556631088,
43
+ "learning_rate": 1.9922625874911624e-05,
44
+ "loss": 0.22366776466369628,
45
+ "step": 50
46
+ },
47
+ {
48
+ "epoch": 0.32301480484522205,
49
+ "grad_norm": 0.09931497275829315,
50
+ "learning_rate": 1.9831647703494287e-05,
51
+ "loss": 0.220938777923584,
52
+ "step": 60
53
+ },
54
+ {
55
+ "epoch": 0.3768506056527591,
56
+ "grad_norm": 0.08036182820796967,
57
+ "learning_rate": 1.9706135510944894e-05,
58
+ "loss": 0.20970940589904785,
59
+ "step": 70
60
+ },
61
+ {
62
+ "epoch": 0.4306864064602961,
63
+ "grad_norm": 0.08407500386238098,
64
+ "learning_rate": 1.9546530163417374e-05,
65
+ "loss": 0.2213752269744873,
66
+ "step": 80
67
+ },
68
+ {
69
+ "epoch": 0.4845222072678331,
70
+ "grad_norm": 0.08968329429626465,
71
+ "learning_rate": 1.9353392280515938e-05,
72
+ "loss": 0.19955954551696778,
73
+ "step": 90
74
+ },
75
+ {
76
+ "epoch": 0.5383580080753702,
77
+ "grad_norm": 0.09176259487867355,
78
+ "learning_rate": 1.912740026609828e-05,
79
+ "loss": 0.19643588066101075,
80
+ "step": 100
81
+ },
82
+ {
83
+ "epoch": 0.5383580080753702,
84
+ "eval_loss": 0.19940482079982758,
85
+ "eval_runtime": 59.3187,
86
+ "eval_samples_per_second": 22.253,
87
+ "eval_steps_per_second": 1.399,
88
+ "step": 100
89
+ },
90
+ {
91
+ "epoch": 0.5921938088829072,
92
+ "grad_norm": 0.08072628080844879,
93
+ "learning_rate": 1.8869347925357275e-05,
94
+ "loss": 0.18773868083953857,
95
+ "step": 110
96
+ },
97
+ {
98
+ "epoch": 0.6460296096904441,
99
+ "grad_norm": 0.090627022087574,
100
+ "learning_rate": 1.8580141676551298e-05,
101
+ "loss": 0.19360114336013795,
102
+ "step": 120
103
+ },
104
+ {
105
+ "epoch": 0.6998654104979811,
106
+ "grad_norm": 0.076926589012146,
107
+ "learning_rate": 1.826079736717702e-05,
108
+ "loss": 0.1843669056892395,
109
+ "step": 130
110
+ },
111
+ {
112
+ "epoch": 0.7537012113055181,
113
+ "grad_norm": 0.07070426642894745,
114
+ "learning_rate": 1.7912436705768045e-05,
115
+ "loss": 0.1653825283050537,
116
+ "step": 140
117
+ },
118
+ {
119
+ "epoch": 0.8075370121130552,
120
+ "grad_norm": 0.07505597174167633,
121
+ "learning_rate": 1.753628332185275e-05,
122
+ "loss": 0.17851850986480713,
123
+ "step": 150
124
+ },
125
+ {
126
+ "epoch": 0.8613728129205922,
127
+ "grad_norm": 0.0769675076007843,
128
+ "learning_rate": 1.7133658467910968e-05,
129
+ "loss": 0.20060482025146484,
130
+ "step": 160
131
+ },
132
+ {
133
+ "epoch": 0.9152086137281292,
134
+ "grad_norm": 0.08414696902036667,
135
+ "learning_rate": 1.6705976378426447e-05,
136
+ "loss": 0.1934957981109619,
137
+ "step": 170
138
+ },
139
+ {
140
+ "epoch": 0.9690444145356663,
141
+ "grad_norm": 0.07856564223766327,
142
+ "learning_rate": 1.625473930233664e-05,
143
+ "loss": 0.18108274936676025,
144
+ "step": 180
145
+ },
146
+ {
147
+ "epoch": 1.0215343203230147,
148
+ "grad_norm": 0.09861781448125839,
149
+ "learning_rate": 1.5781532226328544e-05,
150
+ "loss": 0.2005953311920166,
151
+ "step": 190
152
+ },
153
+ {
154
+ "epoch": 1.0753701211305517,
155
+ "grad_norm": 0.09661930799484253,
156
+ "learning_rate": 1.5288017307515142e-05,
157
+ "loss": 0.18249776363372802,
158
+ "step": 200
159
+ },
160
+ {
161
+ "epoch": 1.0753701211305517,
162
+ "eval_loss": 0.1815323382616043,
163
+ "eval_runtime": 60.9448,
164
+ "eval_samples_per_second": 21.659,
165
+ "eval_steps_per_second": 1.362,
166
+ "step": 200
167
+ },
168
+ {
169
+ "epoch": 1.1292059219380888,
170
+ "grad_norm": 0.08623727411031723,
171
+ "learning_rate": 1.4775928035047928e-05,
172
+ "loss": 0.17418408393859863,
173
+ "step": 210
174
+ },
175
+ {
176
+ "epoch": 1.1830417227456258,
177
+ "grad_norm": 0.09250917285680771,
178
+ "learning_rate": 1.424706314117303e-05,
179
+ "loss": 0.19033379554748536,
180
+ "step": 220
181
+ },
182
+ {
183
+ "epoch": 1.2368775235531628,
184
+ "grad_norm": 0.0896449089050293,
185
+ "learning_rate": 1.3703280283118601e-05,
186
+ "loss": 0.17169111967086792,
187
+ "step": 230
188
+ },
189
+ {
190
+ "epoch": 1.2907133243606999,
191
+ "grad_norm": 0.07717910408973694,
192
+ "learning_rate": 1.3146489518006065e-05,
193
+ "loss": 0.1807547926902771,
194
+ "step": 240
195
+ },
196
+ {
197
+ "epoch": 1.3445491251682369,
198
+ "grad_norm": 0.0989438146352768,
199
+ "learning_rate": 1.2578646593704786e-05,
200
+ "loss": 0.16931524276733398,
201
+ "step": 250
202
+ },
203
+ {
204
+ "epoch": 1.398384925975774,
205
+ "grad_norm": 0.08172167092561722,
206
+ "learning_rate": 1.2001746079196381e-05,
207
+ "loss": 0.17504738569259642,
208
+ "step": 260
209
+ },
210
+ {
211
+ "epoch": 1.452220726783311,
212
+ "grad_norm": 0.08974647521972656,
213
+ "learning_rate": 1.14178143585784e-05,
214
+ "loss": 0.1663546919822693,
215
+ "step": 270
216
+ },
217
+ {
218
+ "epoch": 1.506056527590848,
219
+ "grad_norm": 0.0908568724989891,
220
+ "learning_rate": 1.0828902513316299e-05,
221
+ "loss": 0.1784202814102173,
222
+ "step": 280
223
+ },
224
+ {
225
+ "epoch": 1.559892328398385,
226
+ "grad_norm": 0.08752070367336273,
227
+ "learning_rate": 1.0237079117744977e-05,
228
+ "loss": 0.17578048706054689,
229
+ "step": 290
230
+ },
231
+ {
232
+ "epoch": 1.613728129205922,
233
+ "grad_norm": 0.08162529021501541,
234
+ "learning_rate": 9.644422973125977e-06,
235
+ "loss": 0.1648276448249817,
236
+ "step": 300
237
+ },
238
+ {
239
+ "epoch": 1.613728129205922,
240
+ "eval_loss": 0.17673912644386292,
241
+ "eval_runtime": 62.7878,
242
+ "eval_samples_per_second": 21.023,
243
+ "eval_steps_per_second": 1.322,
244
+ "step": 300
245
+ },
246
+ {
247
+ "epoch": 1.667563930013459,
248
+ "grad_norm": 0.09499260038137436,
249
+ "learning_rate": 9.053015805782249e-06,
250
+ "loss": 0.17204223871231078,
251
+ "step": 310
252
+ },
253
+ {
254
+ "epoch": 1.721399730820996,
255
+ "grad_norm": 0.0870763510465622,
256
+ "learning_rate": 8.46493495495851e-06,
257
+ "loss": 0.17899261713027953,
258
+ "step": 320
259
+ },
260
+ {
261
+ "epoch": 1.775235531628533,
262
+ "grad_norm": 0.0877470001578331,
263
+ "learning_rate": 7.882246076091385e-06,
264
+ "loss": 0.17512272596359252,
265
+ "step": 330
266
+ },
267
+ {
268
+ "epoch": 1.8290713324360701,
269
+ "grad_norm": 0.10601353645324707,
270
+ "learning_rate": 7.306995885119349e-06,
271
+ "loss": 0.1890477180480957,
272
+ "step": 340
273
+ },
274
+ {
275
+ "epoch": 1.8829071332436071,
276
+ "grad_norm": 0.08982375264167786,
277
+ "learning_rate": 6.741204969318343e-06,
278
+ "loss": 0.16031824350357055,
279
+ "step": 350
280
+ },
281
+ {
282
+ "epoch": 1.9367429340511442,
283
+ "grad_norm": 0.07413674145936966,
284
+ "learning_rate": 6.186860689915286e-06,
285
+ "loss": 0.17462934255599977,
286
+ "step": 360
287
+ },
288
+ {
289
+ "epoch": 1.990578734858681,
290
+ "grad_norm": 0.07907533645629883,
291
+ "learning_rate": 5.645910201409268e-06,
292
+ "loss": 0.1853417158126831,
293
+ "step": 370
294
+ },
295
+ {
296
+ "epoch": 2.0430686406460294,
297
+ "grad_norm": 0.07409507036209106,
298
+ "learning_rate": 5.120253612120363e-06,
299
+ "loss": 0.1760026216506958,
300
+ "step": 380
301
+ },
302
+ {
303
+ "epoch": 2.0969044414535665,
304
+ "grad_norm": 0.09365784376859665,
305
+ "learning_rate": 4.611737309989861e-06,
306
+ "loss": 0.17717018127441406,
307
+ "step": 390
308
+ },
309
+ {
310
+ "epoch": 2.1507402422611035,
311
+ "grad_norm": 0.08592981100082397,
312
+ "learning_rate": 4.12214747707527e-06,
313
+ "loss": 0.1779325485229492,
314
+ "step": 400
315
+ },
316
+ {
317
+ "epoch": 2.1507402422611035,
318
+ "eval_loss": 0.1748906821012497,
319
+ "eval_runtime": 61.9646,
320
+ "eval_samples_per_second": 21.302,
321
+ "eval_steps_per_second": 1.339,
322
+ "step": 400
323
+ },
324
+ {
325
+ "epoch": 2.2045760430686405,
326
+ "grad_norm": 0.09960120916366577,
327
+ "learning_rate": 3.653203815520714e-06,
328
+ "loss": 0.17972402572631835,
329
+ "step": 410
330
+ },
331
+ {
332
+ "epoch": 2.2584118438761775,
333
+ "grad_norm": 0.10758821666240692,
334
+ "learning_rate": 3.2065535070404085e-06,
335
+ "loss": 0.1703733205795288,
336
+ "step": 420
337
+ },
338
+ {
339
+ "epoch": 2.3122476446837146,
340
+ "grad_norm": 0.0882757231593132,
341
+ "learning_rate": 2.783765427132837e-06,
342
+ "loss": 0.1815846562385559,
343
+ "step": 430
344
+ },
345
+ {
346
+ "epoch": 2.3660834454912516,
347
+ "grad_norm": 0.08758968114852905,
348
+ "learning_rate": 2.3863246343483306e-06,
349
+ "loss": 0.15952233076095582,
350
+ "step": 440
351
+ },
352
+ {
353
+ "epoch": 2.4199192462987886,
354
+ "grad_norm": 0.10103869438171387,
355
+ "learning_rate": 2.0156271539667517e-06,
356
+ "loss": 0.17301445007324218,
357
+ "step": 450
358
+ },
359
+ {
360
+ "epoch": 2.4737550471063257,
361
+ "grad_norm": 0.09046496450901031,
362
+ "learning_rate": 1.6729750744077755e-06,
363
+ "loss": 0.16349974870681763,
364
+ "step": 460
365
+ },
366
+ {
367
+ "epoch": 2.5275908479138627,
368
+ "grad_norm": 0.08671458810567856,
369
+ "learning_rate": 1.3595719735978451e-06,
370
+ "loss": 0.1653430461883545,
371
+ "step": 470
372
+ },
373
+ {
374
+ "epoch": 2.5814266487213997,
375
+ "grad_norm": 0.10391299426555634,
376
+ "learning_rate": 1.0765186913587866e-06,
377
+ "loss": 0.17563060522079468,
378
+ "step": 480
379
+ },
380
+ {
381
+ "epoch": 2.6352624495289367,
382
+ "grad_norm": 0.08818709850311279,
383
+ "learning_rate": 8.24809462667805e-07,
384
+ "loss": 0.17396645545959472,
385
+ "step": 490
386
+ },
387
+ {
388
+ "epoch": 2.6890982503364738,
389
+ "grad_norm": 0.11322963237762451,
390
+ "learning_rate": 6.053284253708547e-07,
391
+ "loss": 0.16821794509887694,
392
+ "step": 500
393
+ },
394
+ {
395
+ "epoch": 2.6890982503364738,
396
+ "eval_loss": 0.17438176274299622,
397
+ "eval_runtime": 61.1022,
398
+ "eval_samples_per_second": 21.603,
399
+ "eval_steps_per_second": 1.358,
400
+ "step": 500
401
+ },
402
+ {
403
+ "epoch": 2.742934051144011,
404
+ "grad_norm": 0.09662002325057983,
405
+ "learning_rate": 4.188465146162113e-07,
406
+ "loss": 0.173370361328125,
407
+ "step": 510
408
+ },
409
+ {
410
+ "epoch": 2.796769851951548,
411
+ "grad_norm": 0.08389712870121002,
412
+ "learning_rate": 2.660187549166504e-07,
413
+ "loss": 0.16504586935043336,
414
+ "step": 520
415
+ },
416
+ {
417
+ "epoch": 2.850605652759085,
418
+ "grad_norm": 0.09145800769329071,
419
+ "learning_rate": 1.4738195935194987e-07,
420
+ "loss": 0.16487127542495728,
421
+ "step": 530
422
+ },
423
+ {
424
+ "epoch": 2.904441453566622,
425
+ "grad_norm": 0.10042046755552292,
426
+ "learning_rate": 6.335284399337726e-08,
427
+ "loss": 0.18167952299118043,
428
+ "step": 540
429
+ },
430
+ {
431
+ "epoch": 2.958277254374159,
432
+ "grad_norm": 0.09333372861146927,
433
+ "learning_rate": 1.4226564173286383e-08,
434
+ "loss": 0.16996867656707765,
435
+ "step": 550
436
+ },
437
+ {
438
+ "epoch": 3.0,
439
+ "eval_loss": 0.17438960075378418,
440
+ "eval_runtime": 59.9311,
441
+ "eval_samples_per_second": 22.025,
442
+ "eval_steps_per_second": 1.385,
443
+ "step": 558
444
+ }
445
+ ],
446
+ "logging_steps": 10,
447
+ "max_steps": 558,
448
+ "num_input_tokens_seen": 0,
449
+ "num_train_epochs": 3,
450
+ "save_steps": 100,
451
+ "stateful_callbacks": {
452
+ "EarlyStoppingCallback": {
453
+ "args": {
454
+ "early_stopping_patience": 3,
455
+ "early_stopping_threshold": 0.0
456
+ },
457
+ "attributes": {
458
+ "early_stopping_patience_counter": 1
459
+ }
460
+ },
461
+ "TrainerControl": {
462
+ "args": {
463
+ "should_epoch_stop": false,
464
+ "should_evaluate": false,
465
+ "should_log": false,
466
+ "should_save": true,
467
+ "should_training_stop": true
468
+ },
469
+ "attributes": {}
470
+ }
471
+ },
472
+ "total_flos": 5.694249489431265e+17,
473
+ "train_batch_size": 2,
474
+ "trial_name": null,
475
+ "trial_params": null
476
+ }
Conifer/PD/Qwen3-1.7B/final/adapter_config.json ADDED
@@ -0,0 +1,45 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alora_invocation_tokens": null,
3
+ "alpha_pattern": {},
4
+ "arrow_config": null,
5
+ "auto_mapping": null,
6
+ "base_model_name_or_path": "Qwen/Qwen3-1.7B",
7
+ "bias": "none",
8
+ "corda_config": null,
9
+ "ensure_weight_tying": false,
10
+ "eva_config": null,
11
+ "exclude_modules": null,
12
+ "fan_in_fan_out": false,
13
+ "inference_mode": true,
14
+ "init_lora_weights": true,
15
+ "layer_replication": null,
16
+ "layers_pattern": null,
17
+ "layers_to_transform": null,
18
+ "loftq_config": {},
19
+ "lora_alpha": 64,
20
+ "lora_bias": false,
21
+ "lora_dropout": 0.05,
22
+ "lora_ga_config": null,
23
+ "megatron_config": null,
24
+ "megatron_core": "megatron.core",
25
+ "modules_to_save": null,
26
+ "peft_type": "LORA",
27
+ "peft_version": "0.19.1",
28
+ "qalora_group_size": 16,
29
+ "r": 64,
30
+ "rank_pattern": {},
31
+ "revision": null,
32
+ "target_modules": [
33
+ "k_proj",
34
+ "v_proj",
35
+ "q_proj",
36
+ "o_proj"
37
+ ],
38
+ "target_parameters": null,
39
+ "task_type": "CAUSAL_LM",
40
+ "trainable_token_indices": null,
41
+ "use_bdlora": null,
42
+ "use_dora": false,
43
+ "use_qalora": false,
44
+ "use_rslora": false
45
+ }
Conifer/PD/Qwen3-1.7B/final/chat_template.jinja ADDED
@@ -0,0 +1,89 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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 message.content is string 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.content is string %}
27
+ {%- set content = message.content %}
28
+ {%- else %}
29
+ {%- set content = '' %}
30
+ {%- endif %}
31
+ {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
32
+ {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
33
+ {%- elif message.role == "assistant" %}
34
+ {%- set reasoning_content = '' %}
35
+ {%- if message.reasoning_content is string %}
36
+ {%- set reasoning_content = message.reasoning_content %}
37
+ {%- else %}
38
+ {%- if '</think>' in content %}
39
+ {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
40
+ {%- set content = content.split('</think>')[-1].lstrip('\n') %}
41
+ {%- endif %}
42
+ {%- endif %}
43
+ {%- if loop.index0 > ns.last_query_index %}
44
+ {%- if loop.last or (not loop.last and reasoning_content) %}
45
+ {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
46
+ {%- else %}
47
+ {{- '<|im_start|>' + message.role + '\n' + content }}
48
+ {%- endif %}
49
+ {%- else %}
50
+ {{- '<|im_start|>' + message.role + '\n' + content }}
51
+ {%- endif %}
52
+ {%- if message.tool_calls %}
53
+ {%- for tool_call in message.tool_calls %}
54
+ {%- if (loop.first and content) or (not loop.first) %}
55
+ {{- '\n' }}
56
+ {%- endif %}
57
+ {%- if tool_call.function %}
58
+ {%- set tool_call = tool_call.function %}
59
+ {%- endif %}
60
+ {{- '<tool_call>\n{"name": "' }}
61
+ {{- tool_call.name }}
62
+ {{- '", "arguments": ' }}
63
+ {%- if tool_call.arguments is string %}
64
+ {{- tool_call.arguments }}
65
+ {%- else %}
66
+ {{- tool_call.arguments | tojson }}
67
+ {%- endif %}
68
+ {{- '}\n</tool_call>' }}
69
+ {%- endfor %}
70
+ {%- endif %}
71
+ {{- '<|im_end|>\n' }}
72
+ {%- elif message.role == "tool" %}
73
+ {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
74
+ {{- '<|im_start|>user' }}
75
+ {%- endif %}
76
+ {{- '\n<tool_response>\n' }}
77
+ {{- content }}
78
+ {{- '\n</tool_response>' }}
79
+ {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
80
+ {{- '<|im_end|>\n' }}
81
+ {%- endif %}
82
+ {%- endif %}
83
+ {%- endfor %}
84
+ {%- if add_generation_prompt %}
85
+ {{- '<|im_start|>assistant\n' }}
86
+ {%- if enable_thinking is defined and enable_thinking is false %}
87
+ {{- '<think>\n\n</think>\n\n' }}
88
+ {%- endif %}
89
+ {%- endif %}
Conifer/PD/Qwen3-1.7B/final/final_eval_metrics.json ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ {
2
+ "eval_loss": 0.17438960075378418,
3
+ "eval_runtime": 60.0034,
4
+ "eval_samples_per_second": 21.999,
5
+ "eval_steps_per_second": 1.383,
6
+ "epoch": 3.0
7
+ }
Conifer/PD/Qwen3-1.7B/final/tokenizer_config.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
+ "<|im_start|>",
10
+ "<|im_end|>",
11
+ "<|object_ref_start|>",
12
+ "<|object_ref_end|>",
13
+ "<|box_start|>",
14
+ "<|box_end|>",
15
+ "<|quad_start|>",
16
+ "<|quad_end|>",
17
+ "<|vision_start|>",
18
+ "<|vision_end|>",
19
+ "<|vision_pad|>",
20
+ "<|image_pad|>",
21
+ "<|video_pad|>"
22
+ ],
23
+ "is_local": true,
24
+ "local_files_only": false,
25
+ "model_max_length": 131072,
26
+ "pad_token": "<|endoftext|>",
27
+ "split_special_tokens": false,
28
+ "tokenizer_class": "Qwen2Tokenizer",
29
+ "unk_token": null
30
+ }
Conifer/PD/Qwen3-1.7B/final/train_summary.json ADDED
@@ -0,0 +1,12 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "global_step": 558,
3
+ "training_loss": 0.1885697089643034,
4
+ "metrics": {
5
+ "train_runtime": 1937.8949,
6
+ "train_samples_per_second": 18.385,
7
+ "train_steps_per_second": 0.288,
8
+ "total_flos": 5.694249489431265e+17,
9
+ "train_loss": 0.1885697089643034,
10
+ "epoch": 3.0
11
+ }
12
+ }
Conifer/PD/Qwen3-4B/checkpoint-500/adapter_config.json ADDED
@@ -0,0 +1,45 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alora_invocation_tokens": null,
3
+ "alpha_pattern": {},
4
+ "arrow_config": null,
5
+ "auto_mapping": null,
6
+ "base_model_name_or_path": "Qwen/Qwen3-4B",
7
+ "bias": "none",
8
+ "corda_config": null,
9
+ "ensure_weight_tying": false,
10
+ "eva_config": null,
11
+ "exclude_modules": null,
12
+ "fan_in_fan_out": false,
13
+ "inference_mode": true,
14
+ "init_lora_weights": true,
15
+ "layer_replication": null,
16
+ "layers_pattern": null,
17
+ "layers_to_transform": null,
18
+ "loftq_config": {},
19
+ "lora_alpha": 64,
20
+ "lora_bias": false,
21
+ "lora_dropout": 0.05,
22
+ "lora_ga_config": null,
23
+ "megatron_config": null,
24
+ "megatron_core": "megatron.core",
25
+ "modules_to_save": null,
26
+ "peft_type": "LORA",
27
+ "peft_version": "0.19.1",
28
+ "qalora_group_size": 16,
29
+ "r": 64,
30
+ "rank_pattern": {},
31
+ "revision": null,
32
+ "target_modules": [
33
+ "o_proj",
34
+ "q_proj",
35
+ "v_proj",
36
+ "k_proj"
37
+ ],
38
+ "target_parameters": null,
39
+ "task_type": "CAUSAL_LM",
40
+ "trainable_token_indices": null,
41
+ "use_bdlora": null,
42
+ "use_dora": false,
43
+ "use_qalora": false,
44
+ "use_rslora": false
45
+ }
Conifer/PD/Qwen3-4B/checkpoint-500/chat_template.jinja ADDED
@@ -0,0 +1,89 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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 message.content is string 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.content is string %}
27
+ {%- set content = message.content %}
28
+ {%- else %}
29
+ {%- set content = '' %}
30
+ {%- endif %}
31
+ {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
32
+ {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
33
+ {%- elif message.role == "assistant" %}
34
+ {%- set reasoning_content = '' %}
35
+ {%- if message.reasoning_content is string %}
36
+ {%- set reasoning_content = message.reasoning_content %}
37
+ {%- else %}
38
+ {%- if '</think>' in content %}
39
+ {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
40
+ {%- set content = content.split('</think>')[-1].lstrip('\n') %}
41
+ {%- endif %}
42
+ {%- endif %}
43
+ {%- if loop.index0 > ns.last_query_index %}
44
+ {%- if loop.last or (not loop.last and reasoning_content) %}
45
+ {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
46
+ {%- else %}
47
+ {{- '<|im_start|>' + message.role + '\n' + content }}
48
+ {%- endif %}
49
+ {%- else %}
50
+ {{- '<|im_start|>' + message.role + '\n' + content }}
51
+ {%- endif %}
52
+ {%- if message.tool_calls %}
53
+ {%- for tool_call in message.tool_calls %}
54
+ {%- if (loop.first and content) or (not loop.first) %}
55
+ {{- '\n' }}
56
+ {%- endif %}
57
+ {%- if tool_call.function %}
58
+ {%- set tool_call = tool_call.function %}
59
+ {%- endif %}
60
+ {{- '<tool_call>\n{"name": "' }}
61
+ {{- tool_call.name }}
62
+ {{- '", "arguments": ' }}
63
+ {%- if tool_call.arguments is string %}
64
+ {{- tool_call.arguments }}
65
+ {%- else %}
66
+ {{- tool_call.arguments | tojson }}
67
+ {%- endif %}
68
+ {{- '}\n</tool_call>' }}
69
+ {%- endfor %}
70
+ {%- endif %}
71
+ {{- '<|im_end|>\n' }}
72
+ {%- elif message.role == "tool" %}
73
+ {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
74
+ {{- '<|im_start|>user' }}
75
+ {%- endif %}
76
+ {{- '\n<tool_response>\n' }}
77
+ {{- content }}
78
+ {{- '\n</tool_response>' }}
79
+ {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
80
+ {{- '<|im_end|>\n' }}
81
+ {%- endif %}
82
+ {%- endif %}
83
+ {%- endfor %}
84
+ {%- if add_generation_prompt %}
85
+ {{- '<|im_start|>assistant\n' }}
86
+ {%- if enable_thinking is defined and enable_thinking is false %}
87
+ {{- '<think>\n\n</think>\n\n' }}
88
+ {%- endif %}
89
+ {%- endif %}
Conifer/PD/Qwen3-4B/checkpoint-500/tokenizer_config.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
+ "<|im_start|>",
10
+ "<|im_end|>",
11
+ "<|object_ref_start|>",
12
+ "<|object_ref_end|>",
13
+ "<|box_start|>",
14
+ "<|box_end|>",
15
+ "<|quad_start|>",
16
+ "<|quad_end|>",
17
+ "<|vision_start|>",
18
+ "<|vision_end|>",
19
+ "<|vision_pad|>",
20
+ "<|image_pad|>",
21
+ "<|video_pad|>"
22
+ ],
23
+ "is_local": true,
24
+ "local_files_only": false,
25
+ "model_max_length": 131072,
26
+ "pad_token": "<|endoftext|>",
27
+ "split_special_tokens": false,
28
+ "tokenizer_class": "Qwen2Tokenizer",
29
+ "unk_token": null
30
+ }
Conifer/PD/Qwen3-4B/checkpoint-500/trainer_state.json ADDED
@@ -0,0 +1,433 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": 500,
3
+ "best_metric": 0.13559849560260773,
4
+ "best_model_checkpoint": "source_project/conifer_training_hub/11_runs/sft/conifer_stratified_sft_full_rl_20260902/A2/checkpoint-500",
5
+ "epoch": 2.4154589371980677,
6
+ "eval_steps": 100,
7
+ "global_step": 500,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "epoch": 0.04830917874396135,
14
+ "grad_norm": 0.6730383038520813,
15
+ "learning_rate": 5.625e-06,
16
+ "loss": 0.34637789726257323,
17
+ "step": 10
18
+ },
19
+ {
20
+ "epoch": 0.0966183574879227,
21
+ "grad_norm": 0.404668927192688,
22
+ "learning_rate": 1.1875e-05,
23
+ "loss": 0.294273829460144,
24
+ "step": 20
25
+ },
26
+ {
27
+ "epoch": 0.14492753623188406,
28
+ "grad_norm": 0.12716145813465118,
29
+ "learning_rate": 1.8125e-05,
30
+ "loss": 0.23454363346099855,
31
+ "step": 30
32
+ },
33
+ {
34
+ "epoch": 0.1932367149758454,
35
+ "grad_norm": 0.10648500174283981,
36
+ "learning_rate": 1.999303077013914e-05,
37
+ "loss": 0.20498347282409668,
38
+ "step": 40
39
+ },
40
+ {
41
+ "epoch": 0.24154589371980675,
42
+ "grad_norm": 0.08959633857011795,
43
+ "learning_rate": 1.9958919149562405e-05,
44
+ "loss": 0.1963196039199829,
45
+ "step": 50
46
+ },
47
+ {
48
+ "epoch": 0.2898550724637681,
49
+ "grad_norm": 0.09468325227499008,
50
+ "learning_rate": 1.9896481975871063e-05,
51
+ "loss": 0.19790533781051636,
52
+ "step": 60
53
+ },
54
+ {
55
+ "epoch": 0.33816425120772947,
56
+ "grad_norm": 0.09737477451562881,
57
+ "learning_rate": 1.980589683535266e-05,
58
+ "loss": 0.15979270935058593,
59
+ "step": 70
60
+ },
61
+ {
62
+ "epoch": 0.3864734299516908,
63
+ "grad_norm": 0.09095712006092072,
64
+ "learning_rate": 1.9687421373859175e-05,
65
+ "loss": 0.16852133274078368,
66
+ "step": 80
67
+ },
68
+ {
69
+ "epoch": 0.43478260869565216,
70
+ "grad_norm": 0.08449332416057587,
71
+ "learning_rate": 1.954139256400049e-05,
72
+ "loss": 0.16047848463058473,
73
+ "step": 90
74
+ },
75
+ {
76
+ "epoch": 0.4830917874396135,
77
+ "grad_norm": 0.07700371742248535,
78
+ "learning_rate": 1.936822574671348e-05,
79
+ "loss": 0.14818867444992065,
80
+ "step": 100
81
+ },
82
+ {
83
+ "epoch": 0.4830917874396135,
84
+ "eval_loss": 0.15427540242671967,
85
+ "eval_runtime": 93.9405,
86
+ "eval_samples_per_second": 15.669,
87
+ "eval_steps_per_second": 0.979,
88
+ "step": 100
89
+ },
90
+ {
91
+ "epoch": 0.5314009661835749,
92
+ "grad_norm": 0.07393220067024231,
93
+ "learning_rate": 1.9168413449932856e-05,
94
+ "loss": 0.1506107807159424,
95
+ "step": 110
96
+ },
97
+ {
98
+ "epoch": 0.5797101449275363,
99
+ "grad_norm": 0.0800199806690216,
100
+ "learning_rate": 1.894252398772368e-05,
101
+ "loss": 0.14703266620635985,
102
+ "step": 120
103
+ },
104
+ {
105
+ "epoch": 0.6280193236714976,
106
+ "grad_norm": 0.07923530787229538,
107
+ "learning_rate": 1.8691199843859917e-05,
108
+ "loss": 0.14741261005401612,
109
+ "step": 130
110
+ },
111
+ {
112
+ "epoch": 0.6763285024154589,
113
+ "grad_norm": 0.09360925853252411,
114
+ "learning_rate": 1.8415155844446593e-05,
115
+ "loss": 0.13760238885879517,
116
+ "step": 140
117
+ },
118
+ {
119
+ "epoch": 0.7246376811594203,
120
+ "grad_norm": 0.09434773027896881,
121
+ "learning_rate": 1.8115177124783026e-05,
122
+ "loss": 0.15733144283294678,
123
+ "step": 150
124
+ },
125
+ {
126
+ "epoch": 0.7729468599033816,
127
+ "grad_norm": 0.08686012029647827,
128
+ "learning_rate": 1.7792116896249833e-05,
129
+ "loss": 0.1565406560897827,
130
+ "step": 160
131
+ },
132
+ {
133
+ "epoch": 0.821256038647343,
134
+ "grad_norm": 0.09214519709348679,
135
+ "learning_rate": 1.7446894019571273e-05,
136
+ "loss": 0.14771753549575806,
137
+ "step": 170
138
+ },
139
+ {
140
+ "epoch": 0.8695652173913043,
141
+ "grad_norm": 0.11540012806653976,
142
+ "learning_rate": 1.7080490391355083e-05,
143
+ "loss": 0.15460551977157594,
144
+ "step": 180
145
+ },
146
+ {
147
+ "epoch": 0.9178743961352657,
148
+ "grad_norm": 0.0938999280333519,
149
+ "learning_rate": 1.6693948151343193e-05,
150
+ "loss": 0.1382729411125183,
151
+ "step": 190
152
+ },
153
+ {
154
+ "epoch": 0.966183574879227,
155
+ "grad_norm": 0.08514069765806198,
156
+ "learning_rate": 1.6288366718316504e-05,
157
+ "loss": 0.15198016166687012,
158
+ "step": 200
159
+ },
160
+ {
161
+ "epoch": 0.966183574879227,
162
+ "eval_loss": 0.14136852324008942,
163
+ "eval_runtime": 92.172,
164
+ "eval_samples_per_second": 15.97,
165
+ "eval_steps_per_second": 0.998,
166
+ "step": 200
167
+ },
168
+ {
169
+ "epoch": 1.0144927536231885,
170
+ "grad_norm": 0.0952739343047142,
171
+ "learning_rate": 1.5864899663084356e-05,
172
+ "loss": 0.13684881925582887,
173
+ "step": 210
174
+ },
175
+ {
176
+ "epoch": 1.0628019323671498,
177
+ "grad_norm": 0.08769699186086655,
178
+ "learning_rate": 1.5424751427452644e-05,
179
+ "loss": 0.1316512107849121,
180
+ "step": 220
181
+ },
182
+ {
183
+ "epoch": 1.1111111111111112,
184
+ "grad_norm": 0.10097978264093399,
185
+ "learning_rate": 1.4969173898502625e-05,
186
+ "loss": 0.12527782917022706,
187
+ "step": 230
188
+ },
189
+ {
190
+ "epoch": 1.1594202898550725,
191
+ "grad_norm": 0.08655855059623718,
192
+ "learning_rate": 1.4499462847924016e-05,
193
+ "loss": 0.13935062885284424,
194
+ "step": 240
195
+ },
196
+ {
197
+ "epoch": 1.2077294685990339,
198
+ "grad_norm": 0.1034281924366951,
199
+ "learning_rate": 1.4016954246529697e-05,
200
+ "loss": 0.13535609245300292,
201
+ "step": 250
202
+ },
203
+ {
204
+ "epoch": 1.2560386473429952,
205
+ "grad_norm": 0.08799777925014496,
206
+ "learning_rate": 1.3523020464434515e-05,
207
+ "loss": 0.14618408679962158,
208
+ "step": 260
209
+ },
210
+ {
211
+ "epoch": 1.3043478260869565,
212
+ "grad_norm": 0.11097506433725357,
213
+ "learning_rate": 1.3019066367705734e-05,
214
+ "loss": 0.14520273208618165,
215
+ "step": 270
216
+ },
217
+ {
218
+ "epoch": 1.3526570048309179,
219
+ "grad_norm": 0.10301961749792099,
220
+ "learning_rate": 1.2506525322587207e-05,
221
+ "loss": 0.14716745615005494,
222
+ "step": 280
223
+ },
224
+ {
225
+ "epoch": 1.4009661835748792,
226
+ "grad_norm": 0.10164006799459457,
227
+ "learning_rate": 1.1986855118662207e-05,
228
+ "loss": 0.14633605480194092,
229
+ "step": 290
230
+ },
231
+ {
232
+ "epoch": 1.4492753623188406,
233
+ "grad_norm": 0.08774349838495255,
234
+ "learning_rate": 1.1461533822550443e-05,
235
+ "loss": 0.13926935195922852,
236
+ "step": 300
237
+ },
238
+ {
239
+ "epoch": 1.4492753623188406,
240
+ "eval_loss": 0.13798561692237854,
241
+ "eval_runtime": 90.8254,
242
+ "eval_samples_per_second": 16.207,
243
+ "eval_steps_per_second": 1.013,
244
+ "step": 300
245
+ },
246
+ {
247
+ "epoch": 1.497584541062802,
248
+ "grad_norm": 0.10154459625482559,
249
+ "learning_rate": 1.0932055573932317e-05,
250
+ "loss": 0.13777120113372804,
251
+ "step": 310
252
+ },
253
+ {
254
+ "epoch": 1.5458937198067633,
255
+ "grad_norm": 0.09672509878873825,
256
+ "learning_rate": 1.0399926335857532e-05,
257
+ "loss": 0.14109740257263184,
258
+ "step": 320
259
+ },
260
+ {
261
+ "epoch": 1.5942028985507246,
262
+ "grad_norm": 0.08732195943593979,
263
+ "learning_rate": 9.866659611425227e-06,
264
+ "loss": 0.13123960494995118,
265
+ "step": 330
266
+ },
267
+ {
268
+ "epoch": 1.642512077294686,
269
+ "grad_norm": 0.0913168340921402,
270
+ "learning_rate": 9.333772139018388e-06,
271
+ "loss": 0.13618662357330322,
272
+ "step": 340
273
+ },
274
+ {
275
+ "epoch": 1.6908212560386473,
276
+ "grad_norm": 0.08482403308153152,
277
+ "learning_rate": 8.80277957833633e-06,
278
+ "loss": 0.12878143787384033,
279
+ "step": 350
280
+ },
281
+ {
282
+ "epoch": 1.7391304347826086,
283
+ "grad_norm": 0.10442593693733215,
284
+ "learning_rate": 8.275192199495237e-06,
285
+ "loss": 0.1377718925476074,
286
+ "step": 360
287
+ },
288
+ {
289
+ "epoch": 1.78743961352657,
290
+ "grad_norm": 0.10647054761648178,
291
+ "learning_rate": 7.75251058745795e-06,
292
+ "loss": 0.14234890937805175,
293
+ "step": 370
294
+ },
295
+ {
296
+ "epoch": 1.8357487922705316,
297
+ "grad_norm": 0.10604608058929443,
298
+ "learning_rate": 7.236221374010648e-06,
299
+ "loss": 0.14028689861297608,
300
+ "step": 380
301
+ },
302
+ {
303
+ "epoch": 1.8840579710144927,
304
+ "grad_norm": 0.10070566833019257,
305
+ "learning_rate": 6.7277930094257405e-06,
306
+ "loss": 0.12577431201934813,
307
+ "step": 390
308
+ },
309
+ {
310
+ "epoch": 1.9323671497584543,
311
+ "grad_norm": 0.11515983194112778,
312
+ "learning_rate": 6.2286715858372894e-06,
313
+ "loss": 0.15147969722747803,
314
+ "step": 400
315
+ },
316
+ {
317
+ "epoch": 1.9323671497584543,
318
+ "eval_loss": 0.13627424836158752,
319
+ "eval_runtime": 92.4651,
320
+ "eval_samples_per_second": 15.92,
321
+ "eval_steps_per_second": 0.995,
322
+ "step": 400
323
+ },
324
+ {
325
+ "epoch": 1.9806763285024154,
326
+ "grad_norm": 0.10686878859996796,
327
+ "learning_rate": 5.740276724208397e-06,
328
+ "loss": 0.13868520259857178,
329
+ "step": 410
330
+ },
331
+ {
332
+ "epoch": 2.028985507246377,
333
+ "grad_norm": 0.09904751926660538,
334
+ "learning_rate": 5.2639975365888915e-06,
335
+ "loss": 0.13737874031066893,
336
+ "step": 420
337
+ },
338
+ {
339
+ "epoch": 2.077294685990338,
340
+ "grad_norm": 0.0951017439365387,
341
+ "learning_rate": 4.8011886751477196e-06,
342
+ "loss": 0.14747107028961182,
343
+ "step": 430
344
+ },
345
+ {
346
+ "epoch": 2.1256038647342996,
347
+ "grad_norm": 0.0876847431063652,
348
+ "learning_rate": 4.3531664792174776e-06,
349
+ "loss": 0.12845938205718993,
350
+ "step": 440
351
+ },
352
+ {
353
+ "epoch": 2.1739130434782608,
354
+ "grad_norm": 0.12756787240505219,
355
+ "learning_rate": 3.921205231309818e-06,
356
+ "loss": 0.12847700119018554,
357
+ "step": 450
358
+ },
359
+ {
360
+ "epoch": 2.2222222222222223,
361
+ "grad_norm": 0.10848432034254074,
362
+ "learning_rate": 3.5065335327505867e-06,
363
+ "loss": 0.13653106689453126,
364
+ "step": 460
365
+ },
366
+ {
367
+ "epoch": 2.2705314009661834,
368
+ "grad_norm": 0.10308367758989334,
369
+ "learning_rate": 3.110330809243134e-06,
370
+ "loss": 0.135123872756958,
371
+ "step": 470
372
+ },
373
+ {
374
+ "epoch": 2.318840579710145,
375
+ "grad_norm": 0.11499480158090591,
376
+ "learning_rate": 2.7337239562989517e-06,
377
+ "loss": 0.1404101610183716,
378
+ "step": 480
379
+ },
380
+ {
381
+ "epoch": 2.367149758454106,
382
+ "grad_norm": 0.09369100630283356,
383
+ "learning_rate": 2.377784134076698e-06,
384
+ "loss": 0.1466900110244751,
385
+ "step": 490
386
+ },
387
+ {
388
+ "epoch": 2.4154589371980677,
389
+ "grad_norm": 0.09752967953681946,
390
+ "learning_rate": 2.04352372074599e-06,
391
+ "loss": 0.13244171142578126,
392
+ "step": 500
393
+ },
394
+ {
395
+ "epoch": 2.4154589371980677,
396
+ "eval_loss": 0.13559849560260773,
397
+ "eval_runtime": 96.1099,
398
+ "eval_samples_per_second": 15.316,
399
+ "eval_steps_per_second": 0.957,
400
+ "step": 500
401
+ }
402
+ ],
403
+ "logging_steps": 10,
404
+ "max_steps": 621,
405
+ "num_input_tokens_seen": 0,
406
+ "num_train_epochs": 3,
407
+ "save_steps": 100,
408
+ "stateful_callbacks": {
409
+ "EarlyStoppingCallback": {
410
+ "args": {
411
+ "early_stopping_patience": 3,
412
+ "early_stopping_threshold": 0.0
413
+ },
414
+ "attributes": {
415
+ "early_stopping_patience_counter": 0
416
+ }
417
+ },
418
+ "TrainerControl": {
419
+ "args": {
420
+ "should_epoch_stop": false,
421
+ "should_evaluate": false,
422
+ "should_log": false,
423
+ "should_save": true,
424
+ "should_training_stop": false
425
+ },
426
+ "attributes": {}
427
+ }
428
+ },
429
+ "total_flos": 1.1255263778318582e+18,
430
+ "train_batch_size": 2,
431
+ "trial_name": null,
432
+ "trial_params": null
433
+ }
Conifer/PD/Qwen3-4B/checkpoint-600/adapter_config.json ADDED
@@ -0,0 +1,45 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alora_invocation_tokens": null,
3
+ "alpha_pattern": {},
4
+ "arrow_config": null,
5
+ "auto_mapping": null,
6
+ "base_model_name_or_path": "Qwen/Qwen3-4B",
7
+ "bias": "none",
8
+ "corda_config": null,
9
+ "ensure_weight_tying": false,
10
+ "eva_config": null,
11
+ "exclude_modules": null,
12
+ "fan_in_fan_out": false,
13
+ "inference_mode": true,
14
+ "init_lora_weights": true,
15
+ "layer_replication": null,
16
+ "layers_pattern": null,
17
+ "layers_to_transform": null,
18
+ "loftq_config": {},
19
+ "lora_alpha": 64,
20
+ "lora_bias": false,
21
+ "lora_dropout": 0.05,
22
+ "lora_ga_config": null,
23
+ "megatron_config": null,
24
+ "megatron_core": "megatron.core",
25
+ "modules_to_save": null,
26
+ "peft_type": "LORA",
27
+ "peft_version": "0.19.1",
28
+ "qalora_group_size": 16,
29
+ "r": 64,
30
+ "rank_pattern": {},
31
+ "revision": null,
32
+ "target_modules": [
33
+ "o_proj",
34
+ "q_proj",
35
+ "v_proj",
36
+ "k_proj"
37
+ ],
38
+ "target_parameters": null,
39
+ "task_type": "CAUSAL_LM",
40
+ "trainable_token_indices": null,
41
+ "use_bdlora": null,
42
+ "use_dora": false,
43
+ "use_qalora": false,
44
+ "use_rslora": false
45
+ }
Conifer/PD/Qwen3-4B/checkpoint-600/chat_template.jinja ADDED
@@ -0,0 +1,89 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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 message.content is string 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.content is string %}
27
+ {%- set content = message.content %}
28
+ {%- else %}
29
+ {%- set content = '' %}
30
+ {%- endif %}
31
+ {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
32
+ {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
33
+ {%- elif message.role == "assistant" %}
34
+ {%- set reasoning_content = '' %}
35
+ {%- if message.reasoning_content is string %}
36
+ {%- set reasoning_content = message.reasoning_content %}
37
+ {%- else %}
38
+ {%- if '</think>' in content %}
39
+ {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
40
+ {%- set content = content.split('</think>')[-1].lstrip('\n') %}
41
+ {%- endif %}
42
+ {%- endif %}
43
+ {%- if loop.index0 > ns.last_query_index %}
44
+ {%- if loop.last or (not loop.last and reasoning_content) %}
45
+ {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
46
+ {%- else %}
47
+ {{- '<|im_start|>' + message.role + '\n' + content }}
48
+ {%- endif %}
49
+ {%- else %}
50
+ {{- '<|im_start|>' + message.role + '\n' + content }}
51
+ {%- endif %}
52
+ {%- if message.tool_calls %}
53
+ {%- for tool_call in message.tool_calls %}
54
+ {%- if (loop.first and content) or (not loop.first) %}
55
+ {{- '\n' }}
56
+ {%- endif %}
57
+ {%- if tool_call.function %}
58
+ {%- set tool_call = tool_call.function %}
59
+ {%- endif %}
60
+ {{- '<tool_call>\n{"name": "' }}
61
+ {{- tool_call.name }}
62
+ {{- '", "arguments": ' }}
63
+ {%- if tool_call.arguments is string %}
64
+ {{- tool_call.arguments }}
65
+ {%- else %}
66
+ {{- tool_call.arguments | tojson }}
67
+ {%- endif %}
68
+ {{- '}\n</tool_call>' }}
69
+ {%- endfor %}
70
+ {%- endif %}
71
+ {{- '<|im_end|>\n' }}
72
+ {%- elif message.role == "tool" %}
73
+ {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
74
+ {{- '<|im_start|>user' }}
75
+ {%- endif %}
76
+ {{- '\n<tool_response>\n' }}
77
+ {{- content }}
78
+ {{- '\n</tool_response>' }}
79
+ {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
80
+ {{- '<|im_end|>\n' }}
81
+ {%- endif %}
82
+ {%- endif %}
83
+ {%- endfor %}
84
+ {%- if add_generation_prompt %}
85
+ {{- '<|im_start|>assistant\n' }}
86
+ {%- if enable_thinking is defined and enable_thinking is false %}
87
+ {{- '<think>\n\n</think>\n\n' }}
88
+ {%- endif %}
89
+ {%- endif %}
Conifer/PD/Qwen3-4B/checkpoint-600/tokenizer_config.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
+ "<|im_start|>",
10
+ "<|im_end|>",
11
+ "<|object_ref_start|>",
12
+ "<|object_ref_end|>",
13
+ "<|box_start|>",
14
+ "<|box_end|>",
15
+ "<|quad_start|>",
16
+ "<|quad_end|>",
17
+ "<|vision_start|>",
18
+ "<|vision_end|>",
19
+ "<|vision_pad|>",
20
+ "<|image_pad|>",
21
+ "<|video_pad|>"
22
+ ],
23
+ "is_local": true,
24
+ "local_files_only": false,
25
+ "model_max_length": 131072,
26
+ "pad_token": "<|endoftext|>",
27
+ "split_special_tokens": false,
28
+ "tokenizer_class": "Qwen2Tokenizer",
29
+ "unk_token": null
30
+ }
Conifer/PD/Qwen3-4B/checkpoint-600/trainer_state.json ADDED
@@ -0,0 +1,511 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": 600,
3
+ "best_metric": 0.13550187647342682,
4
+ "best_model_checkpoint": "source_project/conifer_training_hub/11_runs/sft/conifer_stratified_sft_full_rl_20260902/A2/checkpoint-600",
5
+ "epoch": 2.898550724637681,
6
+ "eval_steps": 100,
7
+ "global_step": 600,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "epoch": 0.04830917874396135,
14
+ "grad_norm": 0.6730383038520813,
15
+ "learning_rate": 5.625e-06,
16
+ "loss": 0.34637789726257323,
17
+ "step": 10
18
+ },
19
+ {
20
+ "epoch": 0.0966183574879227,
21
+ "grad_norm": 0.404668927192688,
22
+ "learning_rate": 1.1875e-05,
23
+ "loss": 0.294273829460144,
24
+ "step": 20
25
+ },
26
+ {
27
+ "epoch": 0.14492753623188406,
28
+ "grad_norm": 0.12716145813465118,
29
+ "learning_rate": 1.8125e-05,
30
+ "loss": 0.23454363346099855,
31
+ "step": 30
32
+ },
33
+ {
34
+ "epoch": 0.1932367149758454,
35
+ "grad_norm": 0.10648500174283981,
36
+ "learning_rate": 1.999303077013914e-05,
37
+ "loss": 0.20498347282409668,
38
+ "step": 40
39
+ },
40
+ {
41
+ "epoch": 0.24154589371980675,
42
+ "grad_norm": 0.08959633857011795,
43
+ "learning_rate": 1.9958919149562405e-05,
44
+ "loss": 0.1963196039199829,
45
+ "step": 50
46
+ },
47
+ {
48
+ "epoch": 0.2898550724637681,
49
+ "grad_norm": 0.09468325227499008,
50
+ "learning_rate": 1.9896481975871063e-05,
51
+ "loss": 0.19790533781051636,
52
+ "step": 60
53
+ },
54
+ {
55
+ "epoch": 0.33816425120772947,
56
+ "grad_norm": 0.09737477451562881,
57
+ "learning_rate": 1.980589683535266e-05,
58
+ "loss": 0.15979270935058593,
59
+ "step": 70
60
+ },
61
+ {
62
+ "epoch": 0.3864734299516908,
63
+ "grad_norm": 0.09095712006092072,
64
+ "learning_rate": 1.9687421373859175e-05,
65
+ "loss": 0.16852133274078368,
66
+ "step": 80
67
+ },
68
+ {
69
+ "epoch": 0.43478260869565216,
70
+ "grad_norm": 0.08449332416057587,
71
+ "learning_rate": 1.954139256400049e-05,
72
+ "loss": 0.16047848463058473,
73
+ "step": 90
74
+ },
75
+ {
76
+ "epoch": 0.4830917874396135,
77
+ "grad_norm": 0.07700371742248535,
78
+ "learning_rate": 1.936822574671348e-05,
79
+ "loss": 0.14818867444992065,
80
+ "step": 100
81
+ },
82
+ {
83
+ "epoch": 0.4830917874396135,
84
+ "eval_loss": 0.15427540242671967,
85
+ "eval_runtime": 93.9405,
86
+ "eval_samples_per_second": 15.669,
87
+ "eval_steps_per_second": 0.979,
88
+ "step": 100
89
+ },
90
+ {
91
+ "epoch": 0.5314009661835749,
92
+ "grad_norm": 0.07393220067024231,
93
+ "learning_rate": 1.9168413449932856e-05,
94
+ "loss": 0.1506107807159424,
95
+ "step": 110
96
+ },
97
+ {
98
+ "epoch": 0.5797101449275363,
99
+ "grad_norm": 0.0800199806690216,
100
+ "learning_rate": 1.894252398772368e-05,
101
+ "loss": 0.14703266620635985,
102
+ "step": 120
103
+ },
104
+ {
105
+ "epoch": 0.6280193236714976,
106
+ "grad_norm": 0.07923530787229538,
107
+ "learning_rate": 1.8691199843859917e-05,
108
+ "loss": 0.14741261005401612,
109
+ "step": 130
110
+ },
111
+ {
112
+ "epoch": 0.6763285024154589,
113
+ "grad_norm": 0.09360925853252411,
114
+ "learning_rate": 1.8415155844446593e-05,
115
+ "loss": 0.13760238885879517,
116
+ "step": 140
117
+ },
118
+ {
119
+ "epoch": 0.7246376811594203,
120
+ "grad_norm": 0.09434773027896881,
121
+ "learning_rate": 1.8115177124783026e-05,
122
+ "loss": 0.15733144283294678,
123
+ "step": 150
124
+ },
125
+ {
126
+ "epoch": 0.7729468599033816,
127
+ "grad_norm": 0.08686012029647827,
128
+ "learning_rate": 1.7792116896249833e-05,
129
+ "loss": 0.1565406560897827,
130
+ "step": 160
131
+ },
132
+ {
133
+ "epoch": 0.821256038647343,
134
+ "grad_norm": 0.09214519709348679,
135
+ "learning_rate": 1.7446894019571273e-05,
136
+ "loss": 0.14771753549575806,
137
+ "step": 170
138
+ },
139
+ {
140
+ "epoch": 0.8695652173913043,
141
+ "grad_norm": 0.11540012806653976,
142
+ "learning_rate": 1.7080490391355083e-05,
143
+ "loss": 0.15460551977157594,
144
+ "step": 180
145
+ },
146
+ {
147
+ "epoch": 0.9178743961352657,
148
+ "grad_norm": 0.0938999280333519,
149
+ "learning_rate": 1.6693948151343193e-05,
150
+ "loss": 0.1382729411125183,
151
+ "step": 190
152
+ },
153
+ {
154
+ "epoch": 0.966183574879227,
155
+ "grad_norm": 0.08514069765806198,
156
+ "learning_rate": 1.6288366718316504e-05,
157
+ "loss": 0.15198016166687012,
158
+ "step": 200
159
+ },
160
+ {
161
+ "epoch": 0.966183574879227,
162
+ "eval_loss": 0.14136852324008942,
163
+ "eval_runtime": 92.172,
164
+ "eval_samples_per_second": 15.97,
165
+ "eval_steps_per_second": 0.998,
166
+ "step": 200
167
+ },
168
+ {
169
+ "epoch": 1.0144927536231885,
170
+ "grad_norm": 0.0952739343047142,
171
+ "learning_rate": 1.5864899663084356e-05,
172
+ "loss": 0.13684881925582887,
173
+ "step": 210
174
+ },
175
+ {
176
+ "epoch": 1.0628019323671498,
177
+ "grad_norm": 0.08769699186086655,
178
+ "learning_rate": 1.5424751427452644e-05,
179
+ "loss": 0.1316512107849121,
180
+ "step": 220
181
+ },
182
+ {
183
+ "epoch": 1.1111111111111112,
184
+ "grad_norm": 0.10097978264093399,
185
+ "learning_rate": 1.4969173898502625e-05,
186
+ "loss": 0.12527782917022706,
187
+ "step": 230
188
+ },
189
+ {
190
+ "epoch": 1.1594202898550725,
191
+ "grad_norm": 0.08655855059623718,
192
+ "learning_rate": 1.4499462847924016e-05,
193
+ "loss": 0.13935062885284424,
194
+ "step": 240
195
+ },
196
+ {
197
+ "epoch": 1.2077294685990339,
198
+ "grad_norm": 0.1034281924366951,
199
+ "learning_rate": 1.4016954246529697e-05,
200
+ "loss": 0.13535609245300292,
201
+ "step": 250
202
+ },
203
+ {
204
+ "epoch": 1.2560386473429952,
205
+ "grad_norm": 0.08799777925014496,
206
+ "learning_rate": 1.3523020464434515e-05,
207
+ "loss": 0.14618408679962158,
208
+ "step": 260
209
+ },
210
+ {
211
+ "epoch": 1.3043478260869565,
212
+ "grad_norm": 0.11097506433725357,
213
+ "learning_rate": 1.3019066367705734e-05,
214
+ "loss": 0.14520273208618165,
215
+ "step": 270
216
+ },
217
+ {
218
+ "epoch": 1.3526570048309179,
219
+ "grad_norm": 0.10301961749792099,
220
+ "learning_rate": 1.2506525322587207e-05,
221
+ "loss": 0.14716745615005494,
222
+ "step": 280
223
+ },
224
+ {
225
+ "epoch": 1.4009661835748792,
226
+ "grad_norm": 0.10164006799459457,
227
+ "learning_rate": 1.1986855118662207e-05,
228
+ "loss": 0.14633605480194092,
229
+ "step": 290
230
+ },
231
+ {
232
+ "epoch": 1.4492753623188406,
233
+ "grad_norm": 0.08774349838495255,
234
+ "learning_rate": 1.1461533822550443e-05,
235
+ "loss": 0.13926935195922852,
236
+ "step": 300
237
+ },
238
+ {
239
+ "epoch": 1.4492753623188406,
240
+ "eval_loss": 0.13798561692237854,
241
+ "eval_runtime": 90.8254,
242
+ "eval_samples_per_second": 16.207,
243
+ "eval_steps_per_second": 1.013,
244
+ "step": 300
245
+ },
246
+ {
247
+ "epoch": 1.497584541062802,
248
+ "grad_norm": 0.10154459625482559,
249
+ "learning_rate": 1.0932055573932317e-05,
250
+ "loss": 0.13777120113372804,
251
+ "step": 310
252
+ },
253
+ {
254
+ "epoch": 1.5458937198067633,
255
+ "grad_norm": 0.09672509878873825,
256
+ "learning_rate": 1.0399926335857532e-05,
257
+ "loss": 0.14109740257263184,
258
+ "step": 320
259
+ },
260
+ {
261
+ "epoch": 1.5942028985507246,
262
+ "grad_norm": 0.08732195943593979,
263
+ "learning_rate": 9.866659611425227e-06,
264
+ "loss": 0.13123960494995118,
265
+ "step": 330
266
+ },
267
+ {
268
+ "epoch": 1.642512077294686,
269
+ "grad_norm": 0.0913168340921402,
270
+ "learning_rate": 9.333772139018388e-06,
271
+ "loss": 0.13618662357330322,
272
+ "step": 340
273
+ },
274
+ {
275
+ "epoch": 1.6908212560386473,
276
+ "grad_norm": 0.08482403308153152,
277
+ "learning_rate": 8.80277957833633e-06,
278
+ "loss": 0.12878143787384033,
279
+ "step": 350
280
+ },
281
+ {
282
+ "epoch": 1.7391304347826086,
283
+ "grad_norm": 0.10442593693733215,
284
+ "learning_rate": 8.275192199495237e-06,
285
+ "loss": 0.1377718925476074,
286
+ "step": 360
287
+ },
288
+ {
289
+ "epoch": 1.78743961352657,
290
+ "grad_norm": 0.10647054761648178,
291
+ "learning_rate": 7.75251058745795e-06,
292
+ "loss": 0.14234890937805175,
293
+ "step": 370
294
+ },
295
+ {
296
+ "epoch": 1.8357487922705316,
297
+ "grad_norm": 0.10604608058929443,
298
+ "learning_rate": 7.236221374010648e-06,
299
+ "loss": 0.14028689861297608,
300
+ "step": 380
301
+ },
302
+ {
303
+ "epoch": 1.8840579710144927,
304
+ "grad_norm": 0.10070566833019257,
305
+ "learning_rate": 6.7277930094257405e-06,
306
+ "loss": 0.12577431201934813,
307
+ "step": 390
308
+ },
309
+ {
310
+ "epoch": 1.9323671497584543,
311
+ "grad_norm": 0.11515983194112778,
312
+ "learning_rate": 6.2286715858372894e-06,
313
+ "loss": 0.15147969722747803,
314
+ "step": 400
315
+ },
316
+ {
317
+ "epoch": 1.9323671497584543,
318
+ "eval_loss": 0.13627424836158752,
319
+ "eval_runtime": 92.4651,
320
+ "eval_samples_per_second": 15.92,
321
+ "eval_steps_per_second": 0.995,
322
+ "step": 400
323
+ },
324
+ {
325
+ "epoch": 1.9806763285024154,
326
+ "grad_norm": 0.10686878859996796,
327
+ "learning_rate": 5.740276724208397e-06,
328
+ "loss": 0.13868520259857178,
329
+ "step": 410
330
+ },
331
+ {
332
+ "epoch": 2.028985507246377,
333
+ "grad_norm": 0.09904751926660538,
334
+ "learning_rate": 5.2639975365888915e-06,
335
+ "loss": 0.13737874031066893,
336
+ "step": 420
337
+ },
338
+ {
339
+ "epoch": 2.077294685990338,
340
+ "grad_norm": 0.0951017439365387,
341
+ "learning_rate": 4.8011886751477196e-06,
342
+ "loss": 0.14747107028961182,
343
+ "step": 430
344
+ },
345
+ {
346
+ "epoch": 2.1256038647342996,
347
+ "grad_norm": 0.0876847431063652,
348
+ "learning_rate": 4.3531664792174776e-06,
349
+ "loss": 0.12845938205718993,
350
+ "step": 440
351
+ },
352
+ {
353
+ "epoch": 2.1739130434782608,
354
+ "grad_norm": 0.12756787240505219,
355
+ "learning_rate": 3.921205231309818e-06,
356
+ "loss": 0.12847700119018554,
357
+ "step": 450
358
+ },
359
+ {
360
+ "epoch": 2.2222222222222223,
361
+ "grad_norm": 0.10848432034254074,
362
+ "learning_rate": 3.5065335327505867e-06,
363
+ "loss": 0.13653106689453126,
364
+ "step": 460
365
+ },
366
+ {
367
+ "epoch": 2.2705314009661834,
368
+ "grad_norm": 0.10308367758989334,
369
+ "learning_rate": 3.110330809243134e-06,
370
+ "loss": 0.135123872756958,
371
+ "step": 470
372
+ },
373
+ {
374
+ "epoch": 2.318840579710145,
375
+ "grad_norm": 0.11499480158090591,
376
+ "learning_rate": 2.7337239562989517e-06,
377
+ "loss": 0.1404101610183716,
378
+ "step": 480
379
+ },
380
+ {
381
+ "epoch": 2.367149758454106,
382
+ "grad_norm": 0.09369100630283356,
383
+ "learning_rate": 2.377784134076698e-06,
384
+ "loss": 0.1466900110244751,
385
+ "step": 490
386
+ },
387
+ {
388
+ "epoch": 2.4154589371980677,
389
+ "grad_norm": 0.09752967953681946,
390
+ "learning_rate": 2.04352372074599e-06,
391
+ "loss": 0.13244171142578126,
392
+ "step": 500
393
+ },
394
+ {
395
+ "epoch": 2.4154589371980677,
396
+ "eval_loss": 0.13559849560260773,
397
+ "eval_runtime": 96.1099,
398
+ "eval_samples_per_second": 15.316,
399
+ "eval_steps_per_second": 0.957,
400
+ "step": 500
401
+ },
402
+ {
403
+ "epoch": 2.463768115942029,
404
+ "grad_norm": 0.11568186432123184,
405
+ "learning_rate": 1.7318934330412197e-06,
406
+ "loss": 0.1342388391494751,
407
+ "step": 510
408
+ },
409
+ {
410
+ "epoch": 2.5120772946859904,
411
+ "grad_norm": 0.08561587333679199,
412
+ "learning_rate": 1.4437796221952748e-06,
413
+ "loss": 0.11313736438751221,
414
+ "step": 520
415
+ },
416
+ {
417
+ "epoch": 2.5603864734299515,
418
+ "grad_norm": 0.10464368015527725,
419
+ "learning_rate": 1.180001752944202e-06,
420
+ "loss": 0.13915711641311646,
421
+ "step": 530
422
+ },
423
+ {
424
+ "epoch": 2.608695652173913,
425
+ "grad_norm": 0.09740391373634338,
426
+ "learning_rate": 9.413100727730628e-07,
427
+ "loss": 0.14361197948455812,
428
+ "step": 540
429
+ },
430
+ {
431
+ "epoch": 2.6570048309178746,
432
+ "grad_norm": 0.24509626626968384,
433
+ "learning_rate": 7.283834780322763e-07,
434
+ "loss": 0.14554858207702637,
435
+ "step": 550
436
+ },
437
+ {
438
+ "epoch": 2.7053140096618358,
439
+ "grad_norm": 0.09961432218551636,
440
+ "learning_rate": 5.418275829936537e-07,
441
+ "loss": 0.1291545271873474,
442
+ "step": 560
443
+ },
444
+ {
445
+ "epoch": 2.753623188405797,
446
+ "grad_norm": 0.11762341111898422,
447
+ "learning_rate": 3.8217299733826817e-07,
448
+ "loss": 0.14881887435913085,
449
+ "step": 570
450
+ },
451
+ {
452
+ "epoch": 2.8019323671497585,
453
+ "grad_norm": 0.10903816670179367,
454
+ "learning_rate": 2.498738169753323e-07,
455
+ "loss": 0.13594473600387574,
456
+ "step": 580
457
+ },
458
+ {
459
+ "epoch": 2.85024154589372,
460
+ "grad_norm": 0.11383962631225586,
461
+ "learning_rate": 1.453063324845827e-07,
462
+ "loss": 0.12993814945220947,
463
+ "step": 590
464
+ },
465
+ {
466
+ "epoch": 2.898550724637681,
467
+ "grad_norm": 0.09717977046966553,
468
+ "learning_rate": 6.876795885569554e-08,
469
+ "loss": 0.1270850658416748,
470
+ "step": 600
471
+ },
472
+ {
473
+ "epoch": 2.898550724637681,
474
+ "eval_loss": 0.13550187647342682,
475
+ "eval_runtime": 93.2558,
476
+ "eval_samples_per_second": 15.785,
477
+ "eval_steps_per_second": 0.987,
478
+ "step": 600
479
+ }
480
+ ],
481
+ "logging_steps": 10,
482
+ "max_steps": 621,
483
+ "num_input_tokens_seen": 0,
484
+ "num_train_epochs": 3,
485
+ "save_steps": 100,
486
+ "stateful_callbacks": {
487
+ "EarlyStoppingCallback": {
488
+ "args": {
489
+ "early_stopping_patience": 3,
490
+ "early_stopping_threshold": 0.0
491
+ },
492
+ "attributes": {
493
+ "early_stopping_patience_counter": 0
494
+ }
495
+ },
496
+ "TrainerControl": {
497
+ "args": {
498
+ "should_epoch_stop": false,
499
+ "should_evaluate": false,
500
+ "should_log": false,
501
+ "should_save": true,
502
+ "should_training_stop": false
503
+ },
504
+ "attributes": {}
505
+ }
506
+ },
507
+ "total_flos": 1.3479697477732925e+18,
508
+ "train_batch_size": 2,
509
+ "trial_name": null,
510
+ "trial_params": null
511
+ }
Conifer/PD/Qwen3-4B/checkpoint-621/adapter_config.json ADDED
@@ -0,0 +1,45 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alora_invocation_tokens": null,
3
+ "alpha_pattern": {},
4
+ "arrow_config": null,
5
+ "auto_mapping": null,
6
+ "base_model_name_or_path": "Qwen/Qwen3-4B",
7
+ "bias": "none",
8
+ "corda_config": null,
9
+ "ensure_weight_tying": false,
10
+ "eva_config": null,
11
+ "exclude_modules": null,
12
+ "fan_in_fan_out": false,
13
+ "inference_mode": true,
14
+ "init_lora_weights": true,
15
+ "layer_replication": null,
16
+ "layers_pattern": null,
17
+ "layers_to_transform": null,
18
+ "loftq_config": {},
19
+ "lora_alpha": 64,
20
+ "lora_bias": false,
21
+ "lora_dropout": 0.05,
22
+ "lora_ga_config": null,
23
+ "megatron_config": null,
24
+ "megatron_core": "megatron.core",
25
+ "modules_to_save": null,
26
+ "peft_type": "LORA",
27
+ "peft_version": "0.19.1",
28
+ "qalora_group_size": 16,
29
+ "r": 64,
30
+ "rank_pattern": {},
31
+ "revision": null,
32
+ "target_modules": [
33
+ "o_proj",
34
+ "q_proj",
35
+ "v_proj",
36
+ "k_proj"
37
+ ],
38
+ "target_parameters": null,
39
+ "task_type": "CAUSAL_LM",
40
+ "trainable_token_indices": null,
41
+ "use_bdlora": null,
42
+ "use_dora": false,
43
+ "use_qalora": false,
44
+ "use_rslora": false
45
+ }
Conifer/PD/Qwen3-4B/checkpoint-621/chat_template.jinja ADDED
@@ -0,0 +1,89 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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 message.content is string 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.content is string %}
27
+ {%- set content = message.content %}
28
+ {%- else %}
29
+ {%- set content = '' %}
30
+ {%- endif %}
31
+ {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
32
+ {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
33
+ {%- elif message.role == "assistant" %}
34
+ {%- set reasoning_content = '' %}
35
+ {%- if message.reasoning_content is string %}
36
+ {%- set reasoning_content = message.reasoning_content %}
37
+ {%- else %}
38
+ {%- if '</think>' in content %}
39
+ {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
40
+ {%- set content = content.split('</think>')[-1].lstrip('\n') %}
41
+ {%- endif %}
42
+ {%- endif %}
43
+ {%- if loop.index0 > ns.last_query_index %}
44
+ {%- if loop.last or (not loop.last and reasoning_content) %}
45
+ {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
46
+ {%- else %}
47
+ {{- '<|im_start|>' + message.role + '\n' + content }}
48
+ {%- endif %}
49
+ {%- else %}
50
+ {{- '<|im_start|>' + message.role + '\n' + content }}
51
+ {%- endif %}
52
+ {%- if message.tool_calls %}
53
+ {%- for tool_call in message.tool_calls %}
54
+ {%- if (loop.first and content) or (not loop.first) %}
55
+ {{- '\n' }}
56
+ {%- endif %}
57
+ {%- if tool_call.function %}
58
+ {%- set tool_call = tool_call.function %}
59
+ {%- endif %}
60
+ {{- '<tool_call>\n{"name": "' }}
61
+ {{- tool_call.name }}
62
+ {{- '", "arguments": ' }}
63
+ {%- if tool_call.arguments is string %}
64
+ {{- tool_call.arguments }}
65
+ {%- else %}
66
+ {{- tool_call.arguments | tojson }}
67
+ {%- endif %}
68
+ {{- '}\n</tool_call>' }}
69
+ {%- endfor %}
70
+ {%- endif %}
71
+ {{- '<|im_end|>\n' }}
72
+ {%- elif message.role == "tool" %}
73
+ {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
74
+ {{- '<|im_start|>user' }}
75
+ {%- endif %}
76
+ {{- '\n<tool_response>\n' }}
77
+ {{- content }}
78
+ {{- '\n</tool_response>' }}
79
+ {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
80
+ {{- '<|im_end|>\n' }}
81
+ {%- endif %}
82
+ {%- endif %}
83
+ {%- endfor %}
84
+ {%- if add_generation_prompt %}
85
+ {{- '<|im_start|>assistant\n' }}
86
+ {%- if enable_thinking is defined and enable_thinking is false %}
87
+ {{- '<think>\n\n</think>\n\n' }}
88
+ {%- endif %}
89
+ {%- endif %}
Conifer/PD/Qwen3-4B/checkpoint-621/tokenizer_config.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
+ "<|im_start|>",
10
+ "<|im_end|>",
11
+ "<|object_ref_start|>",
12
+ "<|object_ref_end|>",
13
+ "<|box_start|>",
14
+ "<|box_end|>",
15
+ "<|quad_start|>",
16
+ "<|quad_end|>",
17
+ "<|vision_start|>",
18
+ "<|vision_end|>",
19
+ "<|vision_pad|>",
20
+ "<|image_pad|>",
21
+ "<|video_pad|>"
22
+ ],
23
+ "is_local": true,
24
+ "local_files_only": false,
25
+ "model_max_length": 131072,
26
+ "pad_token": "<|endoftext|>",
27
+ "split_special_tokens": false,
28
+ "tokenizer_class": "Qwen2Tokenizer",
29
+ "unk_token": null
30
+ }
Conifer/PD/Qwen3-4B/checkpoint-621/trainer_state.json ADDED
@@ -0,0 +1,533 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": 621,
3
+ "best_metric": 0.13547885417938232,
4
+ "best_model_checkpoint": "source_project/conifer_training_hub/11_runs/sft/conifer_stratified_sft_full_rl_20260902/A2/checkpoint-621",
5
+ "epoch": 3.0,
6
+ "eval_steps": 100,
7
+ "global_step": 621,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "epoch": 0.04830917874396135,
14
+ "grad_norm": 0.6730383038520813,
15
+ "learning_rate": 5.625e-06,
16
+ "loss": 0.34637789726257323,
17
+ "step": 10
18
+ },
19
+ {
20
+ "epoch": 0.0966183574879227,
21
+ "grad_norm": 0.404668927192688,
22
+ "learning_rate": 1.1875e-05,
23
+ "loss": 0.294273829460144,
24
+ "step": 20
25
+ },
26
+ {
27
+ "epoch": 0.14492753623188406,
28
+ "grad_norm": 0.12716145813465118,
29
+ "learning_rate": 1.8125e-05,
30
+ "loss": 0.23454363346099855,
31
+ "step": 30
32
+ },
33
+ {
34
+ "epoch": 0.1932367149758454,
35
+ "grad_norm": 0.10648500174283981,
36
+ "learning_rate": 1.999303077013914e-05,
37
+ "loss": 0.20498347282409668,
38
+ "step": 40
39
+ },
40
+ {
41
+ "epoch": 0.24154589371980675,
42
+ "grad_norm": 0.08959633857011795,
43
+ "learning_rate": 1.9958919149562405e-05,
44
+ "loss": 0.1963196039199829,
45
+ "step": 50
46
+ },
47
+ {
48
+ "epoch": 0.2898550724637681,
49
+ "grad_norm": 0.09468325227499008,
50
+ "learning_rate": 1.9896481975871063e-05,
51
+ "loss": 0.19790533781051636,
52
+ "step": 60
53
+ },
54
+ {
55
+ "epoch": 0.33816425120772947,
56
+ "grad_norm": 0.09737477451562881,
57
+ "learning_rate": 1.980589683535266e-05,
58
+ "loss": 0.15979270935058593,
59
+ "step": 70
60
+ },
61
+ {
62
+ "epoch": 0.3864734299516908,
63
+ "grad_norm": 0.09095712006092072,
64
+ "learning_rate": 1.9687421373859175e-05,
65
+ "loss": 0.16852133274078368,
66
+ "step": 80
67
+ },
68
+ {
69
+ "epoch": 0.43478260869565216,
70
+ "grad_norm": 0.08449332416057587,
71
+ "learning_rate": 1.954139256400049e-05,
72
+ "loss": 0.16047848463058473,
73
+ "step": 90
74
+ },
75
+ {
76
+ "epoch": 0.4830917874396135,
77
+ "grad_norm": 0.07700371742248535,
78
+ "learning_rate": 1.936822574671348e-05,
79
+ "loss": 0.14818867444992065,
80
+ "step": 100
81
+ },
82
+ {
83
+ "epoch": 0.4830917874396135,
84
+ "eval_loss": 0.15427540242671967,
85
+ "eval_runtime": 93.9405,
86
+ "eval_samples_per_second": 15.669,
87
+ "eval_steps_per_second": 0.979,
88
+ "step": 100
89
+ },
90
+ {
91
+ "epoch": 0.5314009661835749,
92
+ "grad_norm": 0.07393220067024231,
93
+ "learning_rate": 1.9168413449932856e-05,
94
+ "loss": 0.1506107807159424,
95
+ "step": 110
96
+ },
97
+ {
98
+ "epoch": 0.5797101449275363,
99
+ "grad_norm": 0.0800199806690216,
100
+ "learning_rate": 1.894252398772368e-05,
101
+ "loss": 0.14703266620635985,
102
+ "step": 120
103
+ },
104
+ {
105
+ "epoch": 0.6280193236714976,
106
+ "grad_norm": 0.07923530787229538,
107
+ "learning_rate": 1.8691199843859917e-05,
108
+ "loss": 0.14741261005401612,
109
+ "step": 130
110
+ },
111
+ {
112
+ "epoch": 0.6763285024154589,
113
+ "grad_norm": 0.09360925853252411,
114
+ "learning_rate": 1.8415155844446593e-05,
115
+ "loss": 0.13760238885879517,
116
+ "step": 140
117
+ },
118
+ {
119
+ "epoch": 0.7246376811594203,
120
+ "grad_norm": 0.09434773027896881,
121
+ "learning_rate": 1.8115177124783026e-05,
122
+ "loss": 0.15733144283294678,
123
+ "step": 150
124
+ },
125
+ {
126
+ "epoch": 0.7729468599033816,
127
+ "grad_norm": 0.08686012029647827,
128
+ "learning_rate": 1.7792116896249833e-05,
129
+ "loss": 0.1565406560897827,
130
+ "step": 160
131
+ },
132
+ {
133
+ "epoch": 0.821256038647343,
134
+ "grad_norm": 0.09214519709348679,
135
+ "learning_rate": 1.7446894019571273e-05,
136
+ "loss": 0.14771753549575806,
137
+ "step": 170
138
+ },
139
+ {
140
+ "epoch": 0.8695652173913043,
141
+ "grad_norm": 0.11540012806653976,
142
+ "learning_rate": 1.7080490391355083e-05,
143
+ "loss": 0.15460551977157594,
144
+ "step": 180
145
+ },
146
+ {
147
+ "epoch": 0.9178743961352657,
148
+ "grad_norm": 0.0938999280333519,
149
+ "learning_rate": 1.6693948151343193e-05,
150
+ "loss": 0.1382729411125183,
151
+ "step": 190
152
+ },
153
+ {
154
+ "epoch": 0.966183574879227,
155
+ "grad_norm": 0.08514069765806198,
156
+ "learning_rate": 1.6288366718316504e-05,
157
+ "loss": 0.15198016166687012,
158
+ "step": 200
159
+ },
160
+ {
161
+ "epoch": 0.966183574879227,
162
+ "eval_loss": 0.14136852324008942,
163
+ "eval_runtime": 92.172,
164
+ "eval_samples_per_second": 15.97,
165
+ "eval_steps_per_second": 0.998,
166
+ "step": 200
167
+ },
168
+ {
169
+ "epoch": 1.0144927536231885,
170
+ "grad_norm": 0.0952739343047142,
171
+ "learning_rate": 1.5864899663084356e-05,
172
+ "loss": 0.13684881925582887,
173
+ "step": 210
174
+ },
175
+ {
176
+ "epoch": 1.0628019323671498,
177
+ "grad_norm": 0.08769699186086655,
178
+ "learning_rate": 1.5424751427452644e-05,
179
+ "loss": 0.1316512107849121,
180
+ "step": 220
181
+ },
182
+ {
183
+ "epoch": 1.1111111111111112,
184
+ "grad_norm": 0.10097978264093399,
185
+ "learning_rate": 1.4969173898502625e-05,
186
+ "loss": 0.12527782917022706,
187
+ "step": 230
188
+ },
189
+ {
190
+ "epoch": 1.1594202898550725,
191
+ "grad_norm": 0.08655855059623718,
192
+ "learning_rate": 1.4499462847924016e-05,
193
+ "loss": 0.13935062885284424,
194
+ "step": 240
195
+ },
196
+ {
197
+ "epoch": 1.2077294685990339,
198
+ "grad_norm": 0.1034281924366951,
199
+ "learning_rate": 1.4016954246529697e-05,
200
+ "loss": 0.13535609245300292,
201
+ "step": 250
202
+ },
203
+ {
204
+ "epoch": 1.2560386473429952,
205
+ "grad_norm": 0.08799777925014496,
206
+ "learning_rate": 1.3523020464434515e-05,
207
+ "loss": 0.14618408679962158,
208
+ "step": 260
209
+ },
210
+ {
211
+ "epoch": 1.3043478260869565,
212
+ "grad_norm": 0.11097506433725357,
213
+ "learning_rate": 1.3019066367705734e-05,
214
+ "loss": 0.14520273208618165,
215
+ "step": 270
216
+ },
217
+ {
218
+ "epoch": 1.3526570048309179,
219
+ "grad_norm": 0.10301961749792099,
220
+ "learning_rate": 1.2506525322587207e-05,
221
+ "loss": 0.14716745615005494,
222
+ "step": 280
223
+ },
224
+ {
225
+ "epoch": 1.4009661835748792,
226
+ "grad_norm": 0.10164006799459457,
227
+ "learning_rate": 1.1986855118662207e-05,
228
+ "loss": 0.14633605480194092,
229
+ "step": 290
230
+ },
231
+ {
232
+ "epoch": 1.4492753623188406,
233
+ "grad_norm": 0.08774349838495255,
234
+ "learning_rate": 1.1461533822550443e-05,
235
+ "loss": 0.13926935195922852,
236
+ "step": 300
237
+ },
238
+ {
239
+ "epoch": 1.4492753623188406,
240
+ "eval_loss": 0.13798561692237854,
241
+ "eval_runtime": 90.8254,
242
+ "eval_samples_per_second": 16.207,
243
+ "eval_steps_per_second": 1.013,
244
+ "step": 300
245
+ },
246
+ {
247
+ "epoch": 1.497584541062802,
248
+ "grad_norm": 0.10154459625482559,
249
+ "learning_rate": 1.0932055573932317e-05,
250
+ "loss": 0.13777120113372804,
251
+ "step": 310
252
+ },
253
+ {
254
+ "epoch": 1.5458937198067633,
255
+ "grad_norm": 0.09672509878873825,
256
+ "learning_rate": 1.0399926335857532e-05,
257
+ "loss": 0.14109740257263184,
258
+ "step": 320
259
+ },
260
+ {
261
+ "epoch": 1.5942028985507246,
262
+ "grad_norm": 0.08732195943593979,
263
+ "learning_rate": 9.866659611425227e-06,
264
+ "loss": 0.13123960494995118,
265
+ "step": 330
266
+ },
267
+ {
268
+ "epoch": 1.642512077294686,
269
+ "grad_norm": 0.0913168340921402,
270
+ "learning_rate": 9.333772139018388e-06,
271
+ "loss": 0.13618662357330322,
272
+ "step": 340
273
+ },
274
+ {
275
+ "epoch": 1.6908212560386473,
276
+ "grad_norm": 0.08482403308153152,
277
+ "learning_rate": 8.80277957833633e-06,
278
+ "loss": 0.12878143787384033,
279
+ "step": 350
280
+ },
281
+ {
282
+ "epoch": 1.7391304347826086,
283
+ "grad_norm": 0.10442593693733215,
284
+ "learning_rate": 8.275192199495237e-06,
285
+ "loss": 0.1377718925476074,
286
+ "step": 360
287
+ },
288
+ {
289
+ "epoch": 1.78743961352657,
290
+ "grad_norm": 0.10647054761648178,
291
+ "learning_rate": 7.75251058745795e-06,
292
+ "loss": 0.14234890937805175,
293
+ "step": 370
294
+ },
295
+ {
296
+ "epoch": 1.8357487922705316,
297
+ "grad_norm": 0.10604608058929443,
298
+ "learning_rate": 7.236221374010648e-06,
299
+ "loss": 0.14028689861297608,
300
+ "step": 380
301
+ },
302
+ {
303
+ "epoch": 1.8840579710144927,
304
+ "grad_norm": 0.10070566833019257,
305
+ "learning_rate": 6.7277930094257405e-06,
306
+ "loss": 0.12577431201934813,
307
+ "step": 390
308
+ },
309
+ {
310
+ "epoch": 1.9323671497584543,
311
+ "grad_norm": 0.11515983194112778,
312
+ "learning_rate": 6.2286715858372894e-06,
313
+ "loss": 0.15147969722747803,
314
+ "step": 400
315
+ },
316
+ {
317
+ "epoch": 1.9323671497584543,
318
+ "eval_loss": 0.13627424836158752,
319
+ "eval_runtime": 92.4651,
320
+ "eval_samples_per_second": 15.92,
321
+ "eval_steps_per_second": 0.995,
322
+ "step": 400
323
+ },
324
+ {
325
+ "epoch": 1.9806763285024154,
326
+ "grad_norm": 0.10686878859996796,
327
+ "learning_rate": 5.740276724208397e-06,
328
+ "loss": 0.13868520259857178,
329
+ "step": 410
330
+ },
331
+ {
332
+ "epoch": 2.028985507246377,
333
+ "grad_norm": 0.09904751926660538,
334
+ "learning_rate": 5.2639975365888915e-06,
335
+ "loss": 0.13737874031066893,
336
+ "step": 420
337
+ },
338
+ {
339
+ "epoch": 2.077294685990338,
340
+ "grad_norm": 0.0951017439365387,
341
+ "learning_rate": 4.8011886751477196e-06,
342
+ "loss": 0.14747107028961182,
343
+ "step": 430
344
+ },
345
+ {
346
+ "epoch": 2.1256038647342996,
347
+ "grad_norm": 0.0876847431063652,
348
+ "learning_rate": 4.3531664792174776e-06,
349
+ "loss": 0.12845938205718993,
350
+ "step": 440
351
+ },
352
+ {
353
+ "epoch": 2.1739130434782608,
354
+ "grad_norm": 0.12756787240505219,
355
+ "learning_rate": 3.921205231309818e-06,
356
+ "loss": 0.12847700119018554,
357
+ "step": 450
358
+ },
359
+ {
360
+ "epoch": 2.2222222222222223,
361
+ "grad_norm": 0.10848432034254074,
362
+ "learning_rate": 3.5065335327505867e-06,
363
+ "loss": 0.13653106689453126,
364
+ "step": 460
365
+ },
366
+ {
367
+ "epoch": 2.2705314009661834,
368
+ "grad_norm": 0.10308367758989334,
369
+ "learning_rate": 3.110330809243134e-06,
370
+ "loss": 0.135123872756958,
371
+ "step": 470
372
+ },
373
+ {
374
+ "epoch": 2.318840579710145,
375
+ "grad_norm": 0.11499480158090591,
376
+ "learning_rate": 2.7337239562989517e-06,
377
+ "loss": 0.1404101610183716,
378
+ "step": 480
379
+ },
380
+ {
381
+ "epoch": 2.367149758454106,
382
+ "grad_norm": 0.09369100630283356,
383
+ "learning_rate": 2.377784134076698e-06,
384
+ "loss": 0.1466900110244751,
385
+ "step": 490
386
+ },
387
+ {
388
+ "epoch": 2.4154589371980677,
389
+ "grad_norm": 0.09752967953681946,
390
+ "learning_rate": 2.04352372074599e-06,
391
+ "loss": 0.13244171142578126,
392
+ "step": 500
393
+ },
394
+ {
395
+ "epoch": 2.4154589371980677,
396
+ "eval_loss": 0.13559849560260773,
397
+ "eval_runtime": 96.1099,
398
+ "eval_samples_per_second": 15.316,
399
+ "eval_steps_per_second": 0.957,
400
+ "step": 500
401
+ },
402
+ {
403
+ "epoch": 2.463768115942029,
404
+ "grad_norm": 0.11568186432123184,
405
+ "learning_rate": 1.7318934330412197e-06,
406
+ "loss": 0.1342388391494751,
407
+ "step": 510
408
+ },
409
+ {
410
+ "epoch": 2.5120772946859904,
411
+ "grad_norm": 0.08561587333679199,
412
+ "learning_rate": 1.4437796221952748e-06,
413
+ "loss": 0.11313736438751221,
414
+ "step": 520
415
+ },
416
+ {
417
+ "epoch": 2.5603864734299515,
418
+ "grad_norm": 0.10464368015527725,
419
+ "learning_rate": 1.180001752944202e-06,
420
+ "loss": 0.13915711641311646,
421
+ "step": 530
422
+ },
423
+ {
424
+ "epoch": 2.608695652173913,
425
+ "grad_norm": 0.09740391373634338,
426
+ "learning_rate": 9.413100727730628e-07,
427
+ "loss": 0.14361197948455812,
428
+ "step": 540
429
+ },
430
+ {
431
+ "epoch": 2.6570048309178746,
432
+ "grad_norm": 0.24509626626968384,
433
+ "learning_rate": 7.283834780322763e-07,
434
+ "loss": 0.14554858207702637,
435
+ "step": 550
436
+ },
437
+ {
438
+ "epoch": 2.7053140096618358,
439
+ "grad_norm": 0.09961432218551636,
440
+ "learning_rate": 5.418275829936537e-07,
441
+ "loss": 0.1291545271873474,
442
+ "step": 560
443
+ },
444
+ {
445
+ "epoch": 2.753623188405797,
446
+ "grad_norm": 0.11762341111898422,
447
+ "learning_rate": 3.8217299733826817e-07,
448
+ "loss": 0.14881887435913085,
449
+ "step": 570
450
+ },
451
+ {
452
+ "epoch": 2.8019323671497585,
453
+ "grad_norm": 0.10903816670179367,
454
+ "learning_rate": 2.498738169753323e-07,
455
+ "loss": 0.13594473600387574,
456
+ "step": 580
457
+ },
458
+ {
459
+ "epoch": 2.85024154589372,
460
+ "grad_norm": 0.11383962631225586,
461
+ "learning_rate": 1.453063324845827e-07,
462
+ "loss": 0.12993814945220947,
463
+ "step": 590
464
+ },
465
+ {
466
+ "epoch": 2.898550724637681,
467
+ "grad_norm": 0.09717977046966553,
468
+ "learning_rate": 6.876795885569554e-08,
469
+ "loss": 0.1270850658416748,
470
+ "step": 600
471
+ },
472
+ {
473
+ "epoch": 2.898550724637681,
474
+ "eval_loss": 0.13550187647342682,
475
+ "eval_runtime": 93.2558,
476
+ "eval_samples_per_second": 15.785,
477
+ "eval_steps_per_second": 0.987,
478
+ "step": 600
479
+ },
480
+ {
481
+ "epoch": 2.9468599033816423,
482
+ "grad_norm": 0.09067437052726746,
483
+ "learning_rate": 2.0476389568749867e-08,
484
+ "loss": 0.1349723219871521,
485
+ "step": 610
486
+ },
487
+ {
488
+ "epoch": 2.995169082125604,
489
+ "grad_norm": 0.08941265940666199,
490
+ "learning_rate": 5.689774218065047e-10,
491
+ "loss": 0.12579793930053712,
492
+ "step": 620
493
+ },
494
+ {
495
+ "epoch": 3.0,
496
+ "eval_loss": 0.13547885417938232,
497
+ "eval_runtime": 93.4291,
498
+ "eval_samples_per_second": 15.755,
499
+ "eval_steps_per_second": 0.985,
500
+ "step": 621
501
+ }
502
+ ],
503
+ "logging_steps": 10,
504
+ "max_steps": 621,
505
+ "num_input_tokens_seen": 0,
506
+ "num_train_epochs": 3,
507
+ "save_steps": 100,
508
+ "stateful_callbacks": {
509
+ "EarlyStoppingCallback": {
510
+ "args": {
511
+ "early_stopping_patience": 3,
512
+ "early_stopping_threshold": 0.0
513
+ },
514
+ "attributes": {
515
+ "early_stopping_patience_counter": 0
516
+ }
517
+ },
518
+ "TrainerControl": {
519
+ "args": {
520
+ "should_epoch_stop": false,
521
+ "should_evaluate": false,
522
+ "should_log": false,
523
+ "should_save": true,
524
+ "should_training_stop": true
525
+ },
526
+ "attributes": {}
527
+ }
528
+ },
529
+ "total_flos": 1.3966547804739338e+18,
530
+ "train_batch_size": 2,
531
+ "trial_name": null,
532
+ "trial_params": null
533
+ }
Conifer/PD/Qwen3-4B/final/adapter_config.json ADDED
@@ -0,0 +1,45 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alora_invocation_tokens": null,
3
+ "alpha_pattern": {},
4
+ "arrow_config": null,
5
+ "auto_mapping": null,
6
+ "base_model_name_or_path": "Qwen/Qwen3-4B",
7
+ "bias": "none",
8
+ "corda_config": null,
9
+ "ensure_weight_tying": false,
10
+ "eva_config": null,
11
+ "exclude_modules": null,
12
+ "fan_in_fan_out": false,
13
+ "inference_mode": true,
14
+ "init_lora_weights": true,
15
+ "layer_replication": null,
16
+ "layers_pattern": null,
17
+ "layers_to_transform": null,
18
+ "loftq_config": {},
19
+ "lora_alpha": 64,
20
+ "lora_bias": false,
21
+ "lora_dropout": 0.05,
22
+ "lora_ga_config": null,
23
+ "megatron_config": null,
24
+ "megatron_core": "megatron.core",
25
+ "modules_to_save": null,
26
+ "peft_type": "LORA",
27
+ "peft_version": "0.19.1",
28
+ "qalora_group_size": 16,
29
+ "r": 64,
30
+ "rank_pattern": {},
31
+ "revision": null,
32
+ "target_modules": [
33
+ "o_proj",
34
+ "q_proj",
35
+ "v_proj",
36
+ "k_proj"
37
+ ],
38
+ "target_parameters": null,
39
+ "task_type": "CAUSAL_LM",
40
+ "trainable_token_indices": null,
41
+ "use_bdlora": null,
42
+ "use_dora": false,
43
+ "use_qalora": false,
44
+ "use_rslora": false
45
+ }
Conifer/PD/Qwen3-4B/final/chat_template.jinja ADDED
@@ -0,0 +1,89 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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 message.content is string 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.content is string %}
27
+ {%- set content = message.content %}
28
+ {%- else %}
29
+ {%- set content = '' %}
30
+ {%- endif %}
31
+ {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
32
+ {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
33
+ {%- elif message.role == "assistant" %}
34
+ {%- set reasoning_content = '' %}
35
+ {%- if message.reasoning_content is string %}
36
+ {%- set reasoning_content = message.reasoning_content %}
37
+ {%- else %}
38
+ {%- if '</think>' in content %}
39
+ {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
40
+ {%- set content = content.split('</think>')[-1].lstrip('\n') %}
41
+ {%- endif %}
42
+ {%- endif %}
43
+ {%- if loop.index0 > ns.last_query_index %}
44
+ {%- if loop.last or (not loop.last and reasoning_content) %}
45
+ {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
46
+ {%- else %}
47
+ {{- '<|im_start|>' + message.role + '\n' + content }}
48
+ {%- endif %}
49
+ {%- else %}
50
+ {{- '<|im_start|>' + message.role + '\n' + content }}
51
+ {%- endif %}
52
+ {%- if message.tool_calls %}
53
+ {%- for tool_call in message.tool_calls %}
54
+ {%- if (loop.first and content) or (not loop.first) %}
55
+ {{- '\n' }}
56
+ {%- endif %}
57
+ {%- if tool_call.function %}
58
+ {%- set tool_call = tool_call.function %}
59
+ {%- endif %}
60
+ {{- '<tool_call>\n{"name": "' }}
61
+ {{- tool_call.name }}
62
+ {{- '", "arguments": ' }}
63
+ {%- if tool_call.arguments is string %}
64
+ {{- tool_call.arguments }}
65
+ {%- else %}
66
+ {{- tool_call.arguments | tojson }}
67
+ {%- endif %}
68
+ {{- '}\n</tool_call>' }}
69
+ {%- endfor %}
70
+ {%- endif %}
71
+ {{- '<|im_end|>\n' }}
72
+ {%- elif message.role == "tool" %}
73
+ {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
74
+ {{- '<|im_start|>user' }}
75
+ {%- endif %}
76
+ {{- '\n<tool_response>\n' }}
77
+ {{- content }}
78
+ {{- '\n</tool_response>' }}
79
+ {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
80
+ {{- '<|im_end|>\n' }}
81
+ {%- endif %}
82
+ {%- endif %}
83
+ {%- endfor %}
84
+ {%- if add_generation_prompt %}
85
+ {{- '<|im_start|>assistant\n' }}
86
+ {%- if enable_thinking is defined and enable_thinking is false %}
87
+ {{- '<think>\n\n</think>\n\n' }}
88
+ {%- endif %}
89
+ {%- endif %}
Conifer/PD/Qwen3-4B/final/final_eval_metrics.json ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ {
2
+ "eval_loss": 0.13547885417938232,
3
+ "eval_runtime": 91.4187,
4
+ "eval_samples_per_second": 16.102,
5
+ "eval_steps_per_second": 1.006,
6
+ "epoch": 3.0
7
+ }
Conifer/PD/Qwen3-4B/final/tokenizer_config.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
+ "<|im_start|>",
10
+ "<|im_end|>",
11
+ "<|object_ref_start|>",
12
+ "<|object_ref_end|>",
13
+ "<|box_start|>",
14
+ "<|box_end|>",
15
+ "<|quad_start|>",
16
+ "<|quad_end|>",
17
+ "<|vision_start|>",
18
+ "<|vision_end|>",
19
+ "<|vision_pad|>",
20
+ "<|image_pad|>",
21
+ "<|video_pad|>"
22
+ ],
23
+ "is_local": true,
24
+ "local_files_only": false,
25
+ "model_max_length": 131072,
26
+ "pad_token": "<|endoftext|>",
27
+ "split_special_tokens": false,
28
+ "tokenizer_class": "Qwen2Tokenizer",
29
+ "unk_token": null
30
+ }
Conifer/PD/Qwen3-4B/final/train_summary.json ADDED
@@ -0,0 +1,12 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "global_step": 621,
3
+ "training_loss": 0.15068361007169057,
4
+ "metrics": {
5
+ "train_runtime": 3686.3991,
6
+ "train_samples_per_second": 10.77,
7
+ "train_steps_per_second": 0.168,
8
+ "total_flos": 1.3966547804739338e+18,
9
+ "train_loss": 0.15068361007169057,
10
+ "epoch": 3.0
11
+ }
12
+ }
Conifer/PD/Qwen3-8B/checkpoint-500/adapter_config.json ADDED
@@ -0,0 +1,45 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alora_invocation_tokens": null,
3
+ "alpha_pattern": {},
4
+ "arrow_config": null,
5
+ "auto_mapping": null,
6
+ "base_model_name_or_path": "Qwen/Qwen3-8B",
7
+ "bias": "none",
8
+ "corda_config": null,
9
+ "ensure_weight_tying": false,
10
+ "eva_config": null,
11
+ "exclude_modules": null,
12
+ "fan_in_fan_out": false,
13
+ "inference_mode": true,
14
+ "init_lora_weights": true,
15
+ "layer_replication": null,
16
+ "layers_pattern": null,
17
+ "layers_to_transform": null,
18
+ "loftq_config": {},
19
+ "lora_alpha": 64,
20
+ "lora_bias": false,
21
+ "lora_dropout": 0.05,
22
+ "lora_ga_config": null,
23
+ "megatron_config": null,
24
+ "megatron_core": "megatron.core",
25
+ "modules_to_save": null,
26
+ "peft_type": "LORA",
27
+ "peft_version": "0.19.1",
28
+ "qalora_group_size": 16,
29
+ "r": 64,
30
+ "rank_pattern": {},
31
+ "revision": null,
32
+ "target_modules": [
33
+ "o_proj",
34
+ "v_proj",
35
+ "k_proj",
36
+ "q_proj"
37
+ ],
38
+ "target_parameters": null,
39
+ "task_type": "CAUSAL_LM",
40
+ "trainable_token_indices": null,
41
+ "use_bdlora": null,
42
+ "use_dora": false,
43
+ "use_qalora": false,
44
+ "use_rslora": false
45
+ }
Conifer/PD/Qwen3-8B/checkpoint-500/tokenizer_config.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
+ "<|im_start|>",
10
+ "<|im_end|>",
11
+ "<|object_ref_start|>",
12
+ "<|object_ref_end|>",
13
+ "<|box_start|>",
14
+ "<|box_end|>",
15
+ "<|quad_start|>",
16
+ "<|quad_end|>",
17
+ "<|vision_start|>",
18
+ "<|vision_end|>",
19
+ "<|vision_pad|>",
20
+ "<|image_pad|>",
21
+ "<|video_pad|>"
22
+ ],
23
+ "is_local": true,
24
+ "local_files_only": false,
25
+ "model_max_length": 131072,
26
+ "pad_token": "<|endoftext|>",
27
+ "split_special_tokens": false,
28
+ "tokenizer_class": "Qwen2Tokenizer",
29
+ "unk_token": null
30
+ }
Conifer/PD/Qwen3-8B/checkpoint-500/trainer_state.json ADDED
@@ -0,0 +1,433 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": 500,
3
+ "best_metric": 0.035489436239004135,
4
+ "best_model_checkpoint": "source_project/conifer_training_hub/11_runs/sft/conifer_stratified_sft_full_rl_20260902/A3/checkpoint-500",
5
+ "epoch": 2.4282238442822384,
6
+ "eval_steps": 100,
7
+ "global_step": 500,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "epoch": 0.04866180048661801,
14
+ "grad_norm": 0.15639817714691162,
15
+ "learning_rate": 5.806451612903226e-06,
16
+ "loss": 0.1349215030670166,
17
+ "step": 10
18
+ },
19
+ {
20
+ "epoch": 0.09732360097323602,
21
+ "grad_norm": 0.13050897419452667,
22
+ "learning_rate": 1.2258064516129034e-05,
23
+ "loss": 0.1392993688583374,
24
+ "step": 20
25
+ },
26
+ {
27
+ "epoch": 0.145985401459854,
28
+ "grad_norm": 0.0693754181265831,
29
+ "learning_rate": 1.870967741935484e-05,
30
+ "loss": 0.10144758224487305,
31
+ "step": 30
32
+ },
33
+ {
34
+ "epoch": 0.19464720194647203,
35
+ "grad_norm": 0.0953354462981224,
36
+ "learning_rate": 1.999083553376548e-05,
37
+ "loss": 0.08977606892585754,
38
+ "step": 40
39
+ },
40
+ {
41
+ "epoch": 0.24330900243309003,
42
+ "grad_norm": 0.0660717561841011,
43
+ "learning_rate": 1.995363367648267e-05,
44
+ "loss": 0.07811610102653503,
45
+ "step": 50
46
+ },
47
+ {
48
+ "epoch": 0.291970802919708,
49
+ "grad_norm": 0.06015173718333244,
50
+ "learning_rate": 1.988792810005708e-05,
51
+ "loss": 0.06340082287788391,
52
+ "step": 60
53
+ },
54
+ {
55
+ "epoch": 0.340632603406326,
56
+ "grad_norm": 0.053529202938079834,
57
+ "learning_rate": 1.9793906962236682e-05,
58
+ "loss": 0.05219377875328064,
59
+ "step": 70
60
+ },
61
+ {
62
+ "epoch": 0.38929440389294406,
63
+ "grad_norm": 0.05288362875580788,
64
+ "learning_rate": 1.9671839506615617e-05,
65
+ "loss": 0.04788380265235901,
66
+ "step": 80
67
+ },
68
+ {
69
+ "epoch": 0.43795620437956206,
70
+ "grad_norm": 0.05314455181360245,
71
+ "learning_rate": 1.952207529161492e-05,
72
+ "loss": 0.04456350803375244,
73
+ "step": 90
74
+ },
75
+ {
76
+ "epoch": 0.48661800486618007,
77
+ "grad_norm": 0.05196147412061691,
78
+ "learning_rate": 1.9345043189469676e-05,
79
+ "loss": 0.043999499082565306,
80
+ "step": 100
81
+ },
82
+ {
83
+ "epoch": 0.48661800486618007,
84
+ "eval_loss": 0.042818356305360794,
85
+ "eval_runtime": 110.4984,
86
+ "eval_samples_per_second": 13.249,
87
+ "eval_steps_per_second": 0.833,
88
+ "step": 100
89
+ },
90
+ {
91
+ "epoch": 0.5352798053527981,
92
+ "grad_norm": 0.05217790976166725,
93
+ "learning_rate": 1.9141250158089202e-05,
94
+ "loss": 0.04414570331573486,
95
+ "step": 110
96
+ },
97
+ {
98
+ "epoch": 0.583941605839416,
99
+ "grad_norm": 0.04231223836541176,
100
+ "learning_rate": 1.8911279789307254e-05,
101
+ "loss": 0.040081775188446044,
102
+ "step": 120
103
+ },
104
+ {
105
+ "epoch": 0.6326034063260341,
106
+ "grad_norm": 0.0464063324034214,
107
+ "learning_rate": 1.865579063767948e-05,
108
+ "loss": 0.038861709833145144,
109
+ "step": 130
110
+ },
111
+ {
112
+ "epoch": 0.681265206812652,
113
+ "grad_norm": 0.04705721512436867,
114
+ "learning_rate": 1.8375514334613934e-05,
115
+ "loss": 0.03908354640007019,
116
+ "step": 140
117
+ },
118
+ {
119
+ "epoch": 0.7299270072992701,
120
+ "grad_norm": 0.050299860537052155,
121
+ "learning_rate": 1.8071253493235094e-05,
122
+ "loss": 0.03936109244823456,
123
+ "step": 150
124
+ },
125
+ {
126
+ "epoch": 0.7785888077858881,
127
+ "grad_norm": 0.048315174877643585,
128
+ "learning_rate": 1.7743879409981094e-05,
129
+ "loss": 0.03659357130527496,
130
+ "step": 160
131
+ },
132
+ {
133
+ "epoch": 0.8272506082725061,
134
+ "grad_norm": 0.054264940321445465,
135
+ "learning_rate": 1.739432956951603e-05,
136
+ "loss": 0.037407022714614865,
137
+ "step": 170
138
+ },
139
+ {
140
+ "epoch": 0.8759124087591241,
141
+ "grad_norm": 0.04875651001930237,
142
+ "learning_rate": 1.7023604960102357e-05,
143
+ "loss": 0.04030710756778717,
144
+ "step": 180
145
+ },
146
+ {
147
+ "epoch": 0.9245742092457421,
148
+ "grad_norm": 0.05363527685403824,
149
+ "learning_rate": 1.6632767207121268e-05,
150
+ "loss": 0.036510169506073,
151
+ "step": 190
152
+ },
153
+ {
154
+ "epoch": 0.9732360097323601,
155
+ "grad_norm": 0.058934297412633896,
156
+ "learning_rate": 1.6222935532949586e-05,
157
+ "loss": 0.03854315280914307,
158
+ "step": 200
159
+ },
160
+ {
161
+ "epoch": 0.9732360097323601,
162
+ "eval_loss": 0.03737252205610275,
163
+ "eval_runtime": 111.4043,
164
+ "eval_samples_per_second": 13.141,
165
+ "eval_steps_per_second": 0.826,
166
+ "step": 200
167
+ },
168
+ {
169
+ "epoch": 1.0194647201946472,
170
+ "grad_norm": 0.05371764674782753,
171
+ "learning_rate": 1.5795283551899036e-05,
172
+ "loss": 0.0347901463508606,
173
+ "step": 210
174
+ },
175
+ {
176
+ "epoch": 1.0681265206812651,
177
+ "grad_norm": 0.05175907164812088,
178
+ "learning_rate": 1.535103590939609e-05,
179
+ "loss": 0.036166781187057497,
180
+ "step": 220
181
+ },
182
+ {
183
+ "epoch": 1.1167883211678833,
184
+ "grad_norm": 0.05254443362355232,
185
+ "learning_rate": 1.4891464775026584e-05,
186
+ "loss": 0.03544372022151947,
187
+ "step": 230
188
+ },
189
+ {
190
+ "epoch": 1.1654501216545012,
191
+ "grad_norm": 0.045467473566532135,
192
+ "learning_rate": 1.4417886199487777e-05,
193
+ "loss": 0.03554186224937439,
194
+ "step": 240
195
+ },
196
+ {
197
+ "epoch": 1.2141119221411192,
198
+ "grad_norm": 0.04005361348390579,
199
+ "learning_rate": 1.3931656345880289e-05,
200
+ "loss": 0.0348221093416214,
201
+ "step": 250
202
+ },
203
+ {
204
+ "epoch": 1.2627737226277373,
205
+ "grad_norm": 0.05931864306330681,
206
+ "learning_rate": 1.3434167606132194e-05,
207
+ "loss": 0.03547696173191071,
208
+ "step": 260
209
+ },
210
+ {
211
+ "epoch": 1.3114355231143553,
212
+ "grad_norm": 0.06791075319051743,
213
+ "learning_rate": 1.2926844613676399e-05,
214
+ "loss": 0.038187041878700256,
215
+ "step": 270
216
+ },
217
+ {
218
+ "epoch": 1.3600973236009732,
219
+ "grad_norm": 0.0669725090265274,
220
+ "learning_rate": 1.2411140163799694e-05,
221
+ "loss": 0.03823740482330322,
222
+ "step": 280
223
+ },
224
+ {
225
+ "epoch": 1.4087591240875912,
226
+ "grad_norm": 0.04331658408045769,
227
+ "learning_rate": 1.1888531053346102e-05,
228
+ "loss": 0.03448362648487091,
229
+ "step": 290
230
+ },
231
+ {
232
+ "epoch": 1.4574209245742091,
233
+ "grad_norm": 0.044026345014572144,
234
+ "learning_rate": 1.136051385168817e-05,
235
+ "loss": 0.034500351548194884,
236
+ "step": 300
237
+ },
238
+ {
239
+ "epoch": 1.4574209245742091,
240
+ "eval_loss": 0.03616208955645561,
241
+ "eval_runtime": 109.8142,
242
+ "eval_samples_per_second": 13.332,
243
+ "eval_steps_per_second": 0.838,
244
+ "step": 300
245
+ },
246
+ {
247
+ "epoch": 1.5060827250608273,
248
+ "grad_norm": 0.0620030015707016,
249
+ "learning_rate": 1.0828600615076695e-05,
250
+ "loss": 0.03327265977859497,
251
+ "step": 310
252
+ },
253
+ {
254
+ "epoch": 1.5547445255474452,
255
+ "grad_norm": 0.059572961181402206,
256
+ "learning_rate": 1.0294314556641415e-05,
257
+ "loss": 0.03631613552570343,
258
+ "step": 320
259
+ },
260
+ {
261
+ "epoch": 1.6034063260340634,
262
+ "grad_norm": 0.05534299090504646,
263
+ "learning_rate": 9.759185684442364e-06,
264
+ "loss": 0.03574211597442627,
265
+ "step": 330
266
+ },
267
+ {
268
+ "epoch": 1.6520681265206814,
269
+ "grad_norm": 0.054647643119096756,
270
+ "learning_rate": 9.224746420062857e-06,
271
+ "loss": 0.034722316265106204,
272
+ "step": 340
273
+ },
274
+ {
275
+ "epoch": 1.7007299270072993,
276
+ "grad_norm": 0.04994694143533707,
277
+ "learning_rate": 8.692527210291061e-06,
278
+ "loss": 0.03766582608222961,
279
+ "step": 350
280
+ },
281
+ {
282
+ "epoch": 1.7493917274939172,
283
+ "grad_norm": 0.04726022481918335,
284
+ "learning_rate": 8.164052144456645e-06,
285
+ "loss": 0.037096422910690305,
286
+ "step": 360
287
+ },
288
+ {
289
+ "epoch": 1.7980535279805352,
290
+ "grad_norm": 0.045089419931173325,
291
+ "learning_rate": 7.640834589972964e-06,
292
+ "loss": 0.03496133089065552,
293
+ "step": 370
294
+ },
295
+ {
296
+ "epoch": 1.8467153284671531,
297
+ "grad_norm": 0.045662347227334976,
298
+ "learning_rate": 7.124372858583077e-06,
299
+ "loss": 0.03283970355987549,
300
+ "step": 380
301
+ },
302
+ {
303
+ "epoch": 1.8953771289537713,
304
+ "grad_norm": 0.05277340114116669,
305
+ "learning_rate": 6.616145915719912e-06,
306
+ "loss": 0.03219903707504272,
307
+ "step": 390
308
+ },
309
+ {
310
+ "epoch": 1.9440389294403893,
311
+ "grad_norm": 0.05256184935569763,
312
+ "learning_rate": 6.117609145267484e-06,
313
+ "loss": 0.03142733573913574,
314
+ "step": 400
315
+ },
316
+ {
317
+ "epoch": 1.9440389294403893,
318
+ "eval_loss": 0.035637106746435165,
319
+ "eval_runtime": 110.3846,
320
+ "eval_samples_per_second": 13.263,
321
+ "eval_steps_per_second": 0.833,
322
+ "step": 400
323
+ },
324
+ {
325
+ "epoch": 1.9927007299270074,
326
+ "grad_norm": 0.054978180676698685,
327
+ "learning_rate": 5.63019018185142e-06,
328
+ "loss": 0.03756006956100464,
329
+ "step": 410
330
+ },
331
+ {
332
+ "epoch": 2.0389294403892944,
333
+ "grad_norm": 0.05100775137543678,
334
+ "learning_rate": 5.155284822593612e-06,
335
+ "loss": 0.03562599122524261,
336
+ "step": 420
337
+ },
338
+ {
339
+ "epoch": 2.0875912408759123,
340
+ "grad_norm": 0.04862402379512787,
341
+ "learning_rate": 4.694253030038364e-06,
342
+ "loss": 0.03520449101924896,
343
+ "step": 430
344
+ },
345
+ {
346
+ "epoch": 2.1362530413625302,
347
+ "grad_norm": 0.05179072543978691,
348
+ "learning_rate": 4.248415037696172e-06,
349
+ "loss": 0.03458556532859802,
350
+ "step": 440
351
+ },
352
+ {
353
+ "epoch": 2.1849148418491486,
354
+ "grad_norm": 0.04908761382102966,
355
+ "learning_rate": 3.819047569357552e-06,
356
+ "loss": 0.03751436769962311,
357
+ "step": 450
358
+ },
359
+ {
360
+ "epoch": 2.2335766423357666,
361
+ "grad_norm": 0.048701491206884384,
362
+ "learning_rate": 3.4073801830034758e-06,
363
+ "loss": 0.03350481688976288,
364
+ "step": 460
365
+ },
366
+ {
367
+ "epoch": 2.2822384428223845,
368
+ "grad_norm": 0.04559648036956787,
369
+ "learning_rate": 3.0145917497821353e-06,
370
+ "loss": 0.031913352012634275,
371
+ "step": 470
372
+ },
373
+ {
374
+ "epoch": 2.3309002433090025,
375
+ "grad_norm": 0.06077806279063225,
376
+ "learning_rate": 2.641807078135048e-06,
377
+ "loss": 0.034614771604537964,
378
+ "step": 480
379
+ },
380
+ {
381
+ "epoch": 2.3795620437956204,
382
+ "grad_norm": 0.05987059697508812,
383
+ "learning_rate": 2.2900936927398065e-06,
384
+ "loss": 0.034870317578315733,
385
+ "step": 490
386
+ },
387
+ {
388
+ "epoch": 2.4282238442822384,
389
+ "grad_norm": 0.05105074867606163,
390
+ "learning_rate": 1.960458777493458e-06,
391
+ "loss": 0.03284066021442413,
392
+ "step": 500
393
+ },
394
+ {
395
+ "epoch": 2.4282238442822384,
396
+ "eval_loss": 0.035489436239004135,
397
+ "eval_runtime": 109.5355,
398
+ "eval_samples_per_second": 13.366,
399
+ "eval_steps_per_second": 0.84,
400
+ "step": 500
401
+ }
402
+ ],
403
+ "logging_steps": 10,
404
+ "max_steps": 618,
405
+ "num_input_tokens_seen": 0,
406
+ "num_train_epochs": 3,
407
+ "save_steps": 100,
408
+ "stateful_callbacks": {
409
+ "EarlyStoppingCallback": {
410
+ "args": {
411
+ "early_stopping_patience": 3,
412
+ "early_stopping_threshold": 0.0
413
+ },
414
+ "attributes": {
415
+ "early_stopping_patience_counter": 0
416
+ }
417
+ },
418
+ "TrainerControl": {
419
+ "args": {
420
+ "should_epoch_stop": false,
421
+ "should_evaluate": false,
422
+ "should_log": false,
423
+ "should_save": true,
424
+ "should_training_stop": false
425
+ },
426
+ "attributes": {}
427
+ }
428
+ },
429
+ "total_flos": 2.710903711361663e+18,
430
+ "train_batch_size": 2,
431
+ "trial_name": null,
432
+ "trial_params": null
433
+ }
Conifer/PD/Qwen3-8B/checkpoint-600/adapter_config.json ADDED
@@ -0,0 +1,45 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alora_invocation_tokens": null,
3
+ "alpha_pattern": {},
4
+ "arrow_config": null,
5
+ "auto_mapping": null,
6
+ "base_model_name_or_path": "Qwen/Qwen3-8B",
7
+ "bias": "none",
8
+ "corda_config": null,
9
+ "ensure_weight_tying": false,
10
+ "eva_config": null,
11
+ "exclude_modules": null,
12
+ "fan_in_fan_out": false,
13
+ "inference_mode": true,
14
+ "init_lora_weights": true,
15
+ "layer_replication": null,
16
+ "layers_pattern": null,
17
+ "layers_to_transform": null,
18
+ "loftq_config": {},
19
+ "lora_alpha": 64,
20
+ "lora_bias": false,
21
+ "lora_dropout": 0.05,
22
+ "lora_ga_config": null,
23
+ "megatron_config": null,
24
+ "megatron_core": "megatron.core",
25
+ "modules_to_save": null,
26
+ "peft_type": "LORA",
27
+ "peft_version": "0.19.1",
28
+ "qalora_group_size": 16,
29
+ "r": 64,
30
+ "rank_pattern": {},
31
+ "revision": null,
32
+ "target_modules": [
33
+ "o_proj",
34
+ "v_proj",
35
+ "k_proj",
36
+ "q_proj"
37
+ ],
38
+ "target_parameters": null,
39
+ "task_type": "CAUSAL_LM",
40
+ "trainable_token_indices": null,
41
+ "use_bdlora": null,
42
+ "use_dora": false,
43
+ "use_qalora": false,
44
+ "use_rslora": false
45
+ }
Conifer/PD/Qwen3-8B/checkpoint-600/chat_template.jinja ADDED
@@ -0,0 +1,89 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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 message.content is string 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.content is string %}
27
+ {%- set content = message.content %}
28
+ {%- else %}
29
+ {%- set content = '' %}
30
+ {%- endif %}
31
+ {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
32
+ {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
33
+ {%- elif message.role == "assistant" %}
34
+ {%- set reasoning_content = '' %}
35
+ {%- if message.reasoning_content is string %}
36
+ {%- set reasoning_content = message.reasoning_content %}
37
+ {%- else %}
38
+ {%- if '</think>' in content %}
39
+ {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
40
+ {%- set content = content.split('</think>')[-1].lstrip('\n') %}
41
+ {%- endif %}
42
+ {%- endif %}
43
+ {%- if loop.index0 > ns.last_query_index %}
44
+ {%- if loop.last or (not loop.last and reasoning_content) %}
45
+ {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
46
+ {%- else %}
47
+ {{- '<|im_start|>' + message.role + '\n' + content }}
48
+ {%- endif %}
49
+ {%- else %}
50
+ {{- '<|im_start|>' + message.role + '\n' + content }}
51
+ {%- endif %}
52
+ {%- if message.tool_calls %}
53
+ {%- for tool_call in message.tool_calls %}
54
+ {%- if (loop.first and content) or (not loop.first) %}
55
+ {{- '\n' }}
56
+ {%- endif %}
57
+ {%- if tool_call.function %}
58
+ {%- set tool_call = tool_call.function %}
59
+ {%- endif %}
60
+ {{- '<tool_call>\n{"name": "' }}
61
+ {{- tool_call.name }}
62
+ {{- '", "arguments": ' }}
63
+ {%- if tool_call.arguments is string %}
64
+ {{- tool_call.arguments }}
65
+ {%- else %}
66
+ {{- tool_call.arguments | tojson }}
67
+ {%- endif %}
68
+ {{- '}\n</tool_call>' }}
69
+ {%- endfor %}
70
+ {%- endif %}
71
+ {{- '<|im_end|>\n' }}
72
+ {%- elif message.role == "tool" %}
73
+ {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
74
+ {{- '<|im_start|>user' }}
75
+ {%- endif %}
76
+ {{- '\n<tool_response>\n' }}
77
+ {{- content }}
78
+ {{- '\n</tool_response>' }}
79
+ {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
80
+ {{- '<|im_end|>\n' }}
81
+ {%- endif %}
82
+ {%- endif %}
83
+ {%- endfor %}
84
+ {%- if add_generation_prompt %}
85
+ {{- '<|im_start|>assistant\n' }}
86
+ {%- if enable_thinking is defined and enable_thinking is false %}
87
+ {{- '<think>\n\n</think>\n\n' }}
88
+ {%- endif %}
89
+ {%- endif %}
Conifer/PD/Qwen3-8B/checkpoint-600/tokenizer_config.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
+ "<|im_start|>",
10
+ "<|im_end|>",
11
+ "<|object_ref_start|>",
12
+ "<|object_ref_end|>",
13
+ "<|box_start|>",
14
+ "<|box_end|>",
15
+ "<|quad_start|>",
16
+ "<|quad_end|>",
17
+ "<|vision_start|>",
18
+ "<|vision_end|>",
19
+ "<|vision_pad|>",
20
+ "<|image_pad|>",
21
+ "<|video_pad|>"
22
+ ],
23
+ "is_local": true,
24
+ "local_files_only": false,
25
+ "model_max_length": 131072,
26
+ "pad_token": "<|endoftext|>",
27
+ "split_special_tokens": false,
28
+ "tokenizer_class": "Qwen2Tokenizer",
29
+ "unk_token": null
30
+ }
Conifer/PD/Qwen3-8B/checkpoint-600/trainer_state.json ADDED
@@ -0,0 +1,511 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": 600,
3
+ "best_metric": 0.03542821854352951,
4
+ "best_model_checkpoint": "source_project/conifer_training_hub/11_runs/sft/conifer_stratified_sft_full_rl_20260902/A3/checkpoint-600",
5
+ "epoch": 2.9148418491484183,
6
+ "eval_steps": 100,
7
+ "global_step": 600,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "epoch": 0.04866180048661801,
14
+ "grad_norm": 0.15639817714691162,
15
+ "learning_rate": 5.806451612903226e-06,
16
+ "loss": 0.1349215030670166,
17
+ "step": 10
18
+ },
19
+ {
20
+ "epoch": 0.09732360097323602,
21
+ "grad_norm": 0.13050897419452667,
22
+ "learning_rate": 1.2258064516129034e-05,
23
+ "loss": 0.1392993688583374,
24
+ "step": 20
25
+ },
26
+ {
27
+ "epoch": 0.145985401459854,
28
+ "grad_norm": 0.0693754181265831,
29
+ "learning_rate": 1.870967741935484e-05,
30
+ "loss": 0.10144758224487305,
31
+ "step": 30
32
+ },
33
+ {
34
+ "epoch": 0.19464720194647203,
35
+ "grad_norm": 0.0953354462981224,
36
+ "learning_rate": 1.999083553376548e-05,
37
+ "loss": 0.08977606892585754,
38
+ "step": 40
39
+ },
40
+ {
41
+ "epoch": 0.24330900243309003,
42
+ "grad_norm": 0.0660717561841011,
43
+ "learning_rate": 1.995363367648267e-05,
44
+ "loss": 0.07811610102653503,
45
+ "step": 50
46
+ },
47
+ {
48
+ "epoch": 0.291970802919708,
49
+ "grad_norm": 0.06015173718333244,
50
+ "learning_rate": 1.988792810005708e-05,
51
+ "loss": 0.06340082287788391,
52
+ "step": 60
53
+ },
54
+ {
55
+ "epoch": 0.340632603406326,
56
+ "grad_norm": 0.053529202938079834,
57
+ "learning_rate": 1.9793906962236682e-05,
58
+ "loss": 0.05219377875328064,
59
+ "step": 70
60
+ },
61
+ {
62
+ "epoch": 0.38929440389294406,
63
+ "grad_norm": 0.05288362875580788,
64
+ "learning_rate": 1.9671839506615617e-05,
65
+ "loss": 0.04788380265235901,
66
+ "step": 80
67
+ },
68
+ {
69
+ "epoch": 0.43795620437956206,
70
+ "grad_norm": 0.05314455181360245,
71
+ "learning_rate": 1.952207529161492e-05,
72
+ "loss": 0.04456350803375244,
73
+ "step": 90
74
+ },
75
+ {
76
+ "epoch": 0.48661800486618007,
77
+ "grad_norm": 0.05196147412061691,
78
+ "learning_rate": 1.9345043189469676e-05,
79
+ "loss": 0.043999499082565306,
80
+ "step": 100
81
+ },
82
+ {
83
+ "epoch": 0.48661800486618007,
84
+ "eval_loss": 0.042818356305360794,
85
+ "eval_runtime": 110.4984,
86
+ "eval_samples_per_second": 13.249,
87
+ "eval_steps_per_second": 0.833,
88
+ "step": 100
89
+ },
90
+ {
91
+ "epoch": 0.5352798053527981,
92
+ "grad_norm": 0.05217790976166725,
93
+ "learning_rate": 1.9141250158089202e-05,
94
+ "loss": 0.04414570331573486,
95
+ "step": 110
96
+ },
97
+ {
98
+ "epoch": 0.583941605839416,
99
+ "grad_norm": 0.04231223836541176,
100
+ "learning_rate": 1.8911279789307254e-05,
101
+ "loss": 0.040081775188446044,
102
+ "step": 120
103
+ },
104
+ {
105
+ "epoch": 0.6326034063260341,
106
+ "grad_norm": 0.0464063324034214,
107
+ "learning_rate": 1.865579063767948e-05,
108
+ "loss": 0.038861709833145144,
109
+ "step": 130
110
+ },
111
+ {
112
+ "epoch": 0.681265206812652,
113
+ "grad_norm": 0.04705721512436867,
114
+ "learning_rate": 1.8375514334613934e-05,
115
+ "loss": 0.03908354640007019,
116
+ "step": 140
117
+ },
118
+ {
119
+ "epoch": 0.7299270072992701,
120
+ "grad_norm": 0.050299860537052155,
121
+ "learning_rate": 1.8071253493235094e-05,
122
+ "loss": 0.03936109244823456,
123
+ "step": 150
124
+ },
125
+ {
126
+ "epoch": 0.7785888077858881,
127
+ "grad_norm": 0.048315174877643585,
128
+ "learning_rate": 1.7743879409981094e-05,
129
+ "loss": 0.03659357130527496,
130
+ "step": 160
131
+ },
132
+ {
133
+ "epoch": 0.8272506082725061,
134
+ "grad_norm": 0.054264940321445465,
135
+ "learning_rate": 1.739432956951603e-05,
136
+ "loss": 0.037407022714614865,
137
+ "step": 170
138
+ },
139
+ {
140
+ "epoch": 0.8759124087591241,
141
+ "grad_norm": 0.04875651001930237,
142
+ "learning_rate": 1.7023604960102357e-05,
143
+ "loss": 0.04030710756778717,
144
+ "step": 180
145
+ },
146
+ {
147
+ "epoch": 0.9245742092457421,
148
+ "grad_norm": 0.05363527685403824,
149
+ "learning_rate": 1.6632767207121268e-05,
150
+ "loss": 0.036510169506073,
151
+ "step": 190
152
+ },
153
+ {
154
+ "epoch": 0.9732360097323601,
155
+ "grad_norm": 0.058934297412633896,
156
+ "learning_rate": 1.6222935532949586e-05,
157
+ "loss": 0.03854315280914307,
158
+ "step": 200
159
+ },
160
+ {
161
+ "epoch": 0.9732360097323601,
162
+ "eval_loss": 0.03737252205610275,
163
+ "eval_runtime": 111.4043,
164
+ "eval_samples_per_second": 13.141,
165
+ "eval_steps_per_second": 0.826,
166
+ "step": 200
167
+ },
168
+ {
169
+ "epoch": 1.0194647201946472,
170
+ "grad_norm": 0.05371764674782753,
171
+ "learning_rate": 1.5795283551899036e-05,
172
+ "loss": 0.0347901463508606,
173
+ "step": 210
174
+ },
175
+ {
176
+ "epoch": 1.0681265206812651,
177
+ "grad_norm": 0.05175907164812088,
178
+ "learning_rate": 1.535103590939609e-05,
179
+ "loss": 0.036166781187057497,
180
+ "step": 220
181
+ },
182
+ {
183
+ "epoch": 1.1167883211678833,
184
+ "grad_norm": 0.05254443362355232,
185
+ "learning_rate": 1.4891464775026584e-05,
186
+ "loss": 0.03544372022151947,
187
+ "step": 230
188
+ },
189
+ {
190
+ "epoch": 1.1654501216545012,
191
+ "grad_norm": 0.045467473566532135,
192
+ "learning_rate": 1.4417886199487777e-05,
193
+ "loss": 0.03554186224937439,
194
+ "step": 240
195
+ },
196
+ {
197
+ "epoch": 1.2141119221411192,
198
+ "grad_norm": 0.04005361348390579,
199
+ "learning_rate": 1.3931656345880289e-05,
200
+ "loss": 0.0348221093416214,
201
+ "step": 250
202
+ },
203
+ {
204
+ "epoch": 1.2627737226277373,
205
+ "grad_norm": 0.05931864306330681,
206
+ "learning_rate": 1.3434167606132194e-05,
207
+ "loss": 0.03547696173191071,
208
+ "step": 260
209
+ },
210
+ {
211
+ "epoch": 1.3114355231143553,
212
+ "grad_norm": 0.06791075319051743,
213
+ "learning_rate": 1.2926844613676399e-05,
214
+ "loss": 0.038187041878700256,
215
+ "step": 270
216
+ },
217
+ {
218
+ "epoch": 1.3600973236009732,
219
+ "grad_norm": 0.0669725090265274,
220
+ "learning_rate": 1.2411140163799694e-05,
221
+ "loss": 0.03823740482330322,
222
+ "step": 280
223
+ },
224
+ {
225
+ "epoch": 1.4087591240875912,
226
+ "grad_norm": 0.04331658408045769,
227
+ "learning_rate": 1.1888531053346102e-05,
228
+ "loss": 0.03448362648487091,
229
+ "step": 290
230
+ },
231
+ {
232
+ "epoch": 1.4574209245742091,
233
+ "grad_norm": 0.044026345014572144,
234
+ "learning_rate": 1.136051385168817e-05,
235
+ "loss": 0.034500351548194884,
236
+ "step": 300
237
+ },
238
+ {
239
+ "epoch": 1.4574209245742091,
240
+ "eval_loss": 0.03616208955645561,
241
+ "eval_runtime": 109.8142,
242
+ "eval_samples_per_second": 13.332,
243
+ "eval_steps_per_second": 0.838,
244
+ "step": 300
245
+ },
246
+ {
247
+ "epoch": 1.5060827250608273,
248
+ "grad_norm": 0.0620030015707016,
249
+ "learning_rate": 1.0828600615076695e-05,
250
+ "loss": 0.03327265977859497,
251
+ "step": 310
252
+ },
253
+ {
254
+ "epoch": 1.5547445255474452,
255
+ "grad_norm": 0.059572961181402206,
256
+ "learning_rate": 1.0294314556641415e-05,
257
+ "loss": 0.03631613552570343,
258
+ "step": 320
259
+ },
260
+ {
261
+ "epoch": 1.6034063260340634,
262
+ "grad_norm": 0.05534299090504646,
263
+ "learning_rate": 9.759185684442364e-06,
264
+ "loss": 0.03574211597442627,
265
+ "step": 330
266
+ },
267
+ {
268
+ "epoch": 1.6520681265206814,
269
+ "grad_norm": 0.054647643119096756,
270
+ "learning_rate": 9.224746420062857e-06,
271
+ "loss": 0.034722316265106204,
272
+ "step": 340
273
+ },
274
+ {
275
+ "epoch": 1.7007299270072993,
276
+ "grad_norm": 0.04994694143533707,
277
+ "learning_rate": 8.692527210291061e-06,
278
+ "loss": 0.03766582608222961,
279
+ "step": 350
280
+ },
281
+ {
282
+ "epoch": 1.7493917274939172,
283
+ "grad_norm": 0.04726022481918335,
284
+ "learning_rate": 8.164052144456645e-06,
285
+ "loss": 0.037096422910690305,
286
+ "step": 360
287
+ },
288
+ {
289
+ "epoch": 1.7980535279805352,
290
+ "grad_norm": 0.045089419931173325,
291
+ "learning_rate": 7.640834589972964e-06,
292
+ "loss": 0.03496133089065552,
293
+ "step": 370
294
+ },
295
+ {
296
+ "epoch": 1.8467153284671531,
297
+ "grad_norm": 0.045662347227334976,
298
+ "learning_rate": 7.124372858583077e-06,
299
+ "loss": 0.03283970355987549,
300
+ "step": 380
301
+ },
302
+ {
303
+ "epoch": 1.8953771289537713,
304
+ "grad_norm": 0.05277340114116669,
305
+ "learning_rate": 6.616145915719912e-06,
306
+ "loss": 0.03219903707504272,
307
+ "step": 390
308
+ },
309
+ {
310
+ "epoch": 1.9440389294403893,
311
+ "grad_norm": 0.05256184935569763,
312
+ "learning_rate": 6.117609145267484e-06,
313
+ "loss": 0.03142733573913574,
314
+ "step": 400
315
+ },
316
+ {
317
+ "epoch": 1.9440389294403893,
318
+ "eval_loss": 0.035637106746435165,
319
+ "eval_runtime": 110.3846,
320
+ "eval_samples_per_second": 13.263,
321
+ "eval_steps_per_second": 0.833,
322
+ "step": 400
323
+ },
324
+ {
325
+ "epoch": 1.9927007299270074,
326
+ "grad_norm": 0.054978180676698685,
327
+ "learning_rate": 5.63019018185142e-06,
328
+ "loss": 0.03756006956100464,
329
+ "step": 410
330
+ },
331
+ {
332
+ "epoch": 2.0389294403892944,
333
+ "grad_norm": 0.05100775137543678,
334
+ "learning_rate": 5.155284822593612e-06,
335
+ "loss": 0.03562599122524261,
336
+ "step": 420
337
+ },
338
+ {
339
+ "epoch": 2.0875912408759123,
340
+ "grad_norm": 0.04862402379512787,
341
+ "learning_rate": 4.694253030038364e-06,
342
+ "loss": 0.03520449101924896,
343
+ "step": 430
344
+ },
345
+ {
346
+ "epoch": 2.1362530413625302,
347
+ "grad_norm": 0.05179072543978691,
348
+ "learning_rate": 4.248415037696172e-06,
349
+ "loss": 0.03458556532859802,
350
+ "step": 440
351
+ },
352
+ {
353
+ "epoch": 2.1849148418491486,
354
+ "grad_norm": 0.04908761382102966,
355
+ "learning_rate": 3.819047569357552e-06,
356
+ "loss": 0.03751436769962311,
357
+ "step": 450
358
+ },
359
+ {
360
+ "epoch": 2.2335766423357666,
361
+ "grad_norm": 0.048701491206884384,
362
+ "learning_rate": 3.4073801830034758e-06,
363
+ "loss": 0.03350481688976288,
364
+ "step": 460
365
+ },
366
+ {
367
+ "epoch": 2.2822384428223845,
368
+ "grad_norm": 0.04559648036956787,
369
+ "learning_rate": 3.0145917497821353e-06,
370
+ "loss": 0.031913352012634275,
371
+ "step": 470
372
+ },
373
+ {
374
+ "epoch": 2.3309002433090025,
375
+ "grad_norm": 0.06077806279063225,
376
+ "learning_rate": 2.641807078135048e-06,
377
+ "loss": 0.034614771604537964,
378
+ "step": 480
379
+ },
380
+ {
381
+ "epoch": 2.3795620437956204,
382
+ "grad_norm": 0.05987059697508812,
383
+ "learning_rate": 2.2900936927398065e-06,
384
+ "loss": 0.034870317578315733,
385
+ "step": 490
386
+ },
387
+ {
388
+ "epoch": 2.4282238442822384,
389
+ "grad_norm": 0.05105074867606163,
390
+ "learning_rate": 1.960458777493458e-06,
391
+ "loss": 0.03284066021442413,
392
+ "step": 500
393
+ },
394
+ {
395
+ "epoch": 2.4282238442822384,
396
+ "eval_loss": 0.035489436239004135,
397
+ "eval_runtime": 109.5355,
398
+ "eval_samples_per_second": 13.366,
399
+ "eval_steps_per_second": 0.84,
400
+ "step": 500
401
+ },
402
+ {
403
+ "epoch": 2.4768856447688563,
404
+ "grad_norm": 0.0426454097032547,
405
+ "learning_rate": 1.653846291290736e-06,
406
+ "loss": 0.03034111261367798,
407
+ "step": 510
408
+ },
409
+ {
410
+ "epoch": 2.5255474452554747,
411
+ "grad_norm": 0.05250120162963867,
412
+ "learning_rate": 1.3711342648565395e-06,
413
+ "loss": 0.033456772565841675,
414
+ "step": 520
415
+ },
416
+ {
417
+ "epoch": 2.574209245742092,
418
+ "grad_norm": 0.04732771962881088,
419
+ "learning_rate": 1.1131322863736093e-06,
420
+ "loss": 0.032463949918746945,
421
+ "step": 530
422
+ },
423
+ {
424
+ "epoch": 2.6228710462287106,
425
+ "grad_norm": 0.04851313680410385,
426
+ "learning_rate": 8.805791831056321e-07,
427
+ "loss": 0.034820270538330075,
428
+ "step": 540
429
+ },
430
+ {
431
+ "epoch": 2.6715328467153285,
432
+ "grad_norm": 0.04829060658812523,
433
+ "learning_rate": 6.741409056548498e-07,
434
+ "loss": 0.03143848776817322,
435
+ "step": 550
436
+ },
437
+ {
438
+ "epoch": 2.7201946472019465,
439
+ "grad_norm": 0.05005922541022301,
440
+ "learning_rate": 4.944086209128762e-07,
441
+ "loss": 0.03460255265235901,
442
+ "step": 560
443
+ },
444
+ {
445
+ "epoch": 2.7688564476885644,
446
+ "grad_norm": 0.05089850351214409,
447
+ "learning_rate": 3.418970191658566e-07,
448
+ "loss": 0.036028966307640076,
449
+ "step": 570
450
+ },
451
+ {
452
+ "epoch": 2.8175182481751824,
453
+ "grad_norm": 0.052207909524440765,
454
+ "learning_rate": 2.170428402018454e-07,
455
+ "loss": 0.03278985619544983,
456
+ "step": 580
457
+ },
458
+ {
459
+ "epoch": 2.8661800486618008,
460
+ "grad_norm": 0.047765858471393585,
461
+ "learning_rate": 1.2020362264107298e-07,
462
+ "loss": 0.036867809295654294,
463
+ "step": 590
464
+ },
465
+ {
466
+ "epoch": 2.9148418491484183,
467
+ "grad_norm": 0.05039682239294052,
468
+ "learning_rate": 5.165668007062641e-08,
469
+ "loss": 0.03270910978317261,
470
+ "step": 600
471
+ },
472
+ {
473
+ "epoch": 2.9148418491484183,
474
+ "eval_loss": 0.03542821854352951,
475
+ "eval_runtime": 110.8026,
476
+ "eval_samples_per_second": 13.213,
477
+ "eval_steps_per_second": 0.83,
478
+ "step": 600
479
+ }
480
+ ],
481
+ "logging_steps": 10,
482
+ "max_steps": 618,
483
+ "num_input_tokens_seen": 0,
484
+ "num_train_epochs": 3,
485
+ "save_steps": 100,
486
+ "stateful_callbacks": {
487
+ "EarlyStoppingCallback": {
488
+ "args": {
489
+ "early_stopping_patience": 3,
490
+ "early_stopping_threshold": 0.0
491
+ },
492
+ "attributes": {
493
+ "early_stopping_patience_counter": 0
494
+ }
495
+ },
496
+ "TrainerControl": {
497
+ "args": {
498
+ "should_epoch_stop": false,
499
+ "should_evaluate": false,
500
+ "should_log": false,
501
+ "should_save": true,
502
+ "should_training_stop": false
503
+ },
504
+ "attributes": {}
505
+ }
506
+ },
507
+ "total_flos": 3.254857660087927e+18,
508
+ "train_batch_size": 2,
509
+ "trial_name": null,
510
+ "trial_params": null
511
+ }
Conifer/PD/Qwen3-8B/checkpoint-618/adapter_config.json ADDED
@@ -0,0 +1,45 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alora_invocation_tokens": null,
3
+ "alpha_pattern": {},
4
+ "arrow_config": null,
5
+ "auto_mapping": null,
6
+ "base_model_name_or_path": "Qwen/Qwen3-8B",
7
+ "bias": "none",
8
+ "corda_config": null,
9
+ "ensure_weight_tying": false,
10
+ "eva_config": null,
11
+ "exclude_modules": null,
12
+ "fan_in_fan_out": false,
13
+ "inference_mode": true,
14
+ "init_lora_weights": true,
15
+ "layer_replication": null,
16
+ "layers_pattern": null,
17
+ "layers_to_transform": null,
18
+ "loftq_config": {},
19
+ "lora_alpha": 64,
20
+ "lora_bias": false,
21
+ "lora_dropout": 0.05,
22
+ "lora_ga_config": null,
23
+ "megatron_config": null,
24
+ "megatron_core": "megatron.core",
25
+ "modules_to_save": null,
26
+ "peft_type": "LORA",
27
+ "peft_version": "0.19.1",
28
+ "qalora_group_size": 16,
29
+ "r": 64,
30
+ "rank_pattern": {},
31
+ "revision": null,
32
+ "target_modules": [
33
+ "o_proj",
34
+ "v_proj",
35
+ "k_proj",
36
+ "q_proj"
37
+ ],
38
+ "target_parameters": null,
39
+ "task_type": "CAUSAL_LM",
40
+ "trainable_token_indices": null,
41
+ "use_bdlora": null,
42
+ "use_dora": false,
43
+ "use_qalora": false,
44
+ "use_rslora": false
45
+ }
Conifer/PD/Qwen3-8B/checkpoint-618/chat_template.jinja ADDED
@@ -0,0 +1,89 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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 message.content is string 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.content is string %}
27
+ {%- set content = message.content %}
28
+ {%- else %}
29
+ {%- set content = '' %}
30
+ {%- endif %}
31
+ {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
32
+ {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
33
+ {%- elif message.role == "assistant" %}
34
+ {%- set reasoning_content = '' %}
35
+ {%- if message.reasoning_content is string %}
36
+ {%- set reasoning_content = message.reasoning_content %}
37
+ {%- else %}
38
+ {%- if '</think>' in content %}
39
+ {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
40
+ {%- set content = content.split('</think>')[-1].lstrip('\n') %}
41
+ {%- endif %}
42
+ {%- endif %}
43
+ {%- if loop.index0 > ns.last_query_index %}
44
+ {%- if loop.last or (not loop.last and reasoning_content) %}
45
+ {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
46
+ {%- else %}
47
+ {{- '<|im_start|>' + message.role + '\n' + content }}
48
+ {%- endif %}
49
+ {%- else %}
50
+ {{- '<|im_start|>' + message.role + '\n' + content }}
51
+ {%- endif %}
52
+ {%- if message.tool_calls %}
53
+ {%- for tool_call in message.tool_calls %}
54
+ {%- if (loop.first and content) or (not loop.first) %}
55
+ {{- '\n' }}
56
+ {%- endif %}
57
+ {%- if tool_call.function %}
58
+ {%- set tool_call = tool_call.function %}
59
+ {%- endif %}
60
+ {{- '<tool_call>\n{"name": "' }}
61
+ {{- tool_call.name }}
62
+ {{- '", "arguments": ' }}
63
+ {%- if tool_call.arguments is string %}
64
+ {{- tool_call.arguments }}
65
+ {%- else %}
66
+ {{- tool_call.arguments | tojson }}
67
+ {%- endif %}
68
+ {{- '}\n</tool_call>' }}
69
+ {%- endfor %}
70
+ {%- endif %}
71
+ {{- '<|im_end|>\n' }}
72
+ {%- elif message.role == "tool" %}
73
+ {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
74
+ {{- '<|im_start|>user' }}
75
+ {%- endif %}
76
+ {{- '\n<tool_response>\n' }}
77
+ {{- content }}
78
+ {{- '\n</tool_response>' }}
79
+ {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
80
+ {{- '<|im_end|>\n' }}
81
+ {%- endif %}
82
+ {%- endif %}
83
+ {%- endfor %}
84
+ {%- if add_generation_prompt %}
85
+ {{- '<|im_start|>assistant\n' }}
86
+ {%- if enable_thinking is defined and enable_thinking is false %}
87
+ {{- '<think>\n\n</think>\n\n' }}
88
+ {%- endif %}
89
+ {%- endif %}
Conifer/PD/Qwen3-8B/checkpoint-618/tokenizer_config.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
+ "<|im_start|>",
10
+ "<|im_end|>",
11
+ "<|object_ref_start|>",
12
+ "<|object_ref_end|>",
13
+ "<|box_start|>",
14
+ "<|box_end|>",
15
+ "<|quad_start|>",
16
+ "<|quad_end|>",
17
+ "<|vision_start|>",
18
+ "<|vision_end|>",
19
+ "<|vision_pad|>",
20
+ "<|image_pad|>",
21
+ "<|video_pad|>"
22
+ ],
23
+ "is_local": true,
24
+ "local_files_only": false,
25
+ "model_max_length": 131072,
26
+ "pad_token": "<|endoftext|>",
27
+ "split_special_tokens": false,
28
+ "tokenizer_class": "Qwen2Tokenizer",
29
+ "unk_token": null
30
+ }
Conifer/PD/Qwen3-8B/checkpoint-618/trainer_state.json ADDED
@@ -0,0 +1,526 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": 618,
3
+ "best_metric": 0.035394057631492615,
4
+ "best_model_checkpoint": "source_project/conifer_training_hub/11_runs/sft/conifer_stratified_sft_full_rl_20260902/A3/checkpoint-618",
5
+ "epoch": 3.0,
6
+ "eval_steps": 100,
7
+ "global_step": 618,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "epoch": 0.04866180048661801,
14
+ "grad_norm": 0.15639817714691162,
15
+ "learning_rate": 5.806451612903226e-06,
16
+ "loss": 0.1349215030670166,
17
+ "step": 10
18
+ },
19
+ {
20
+ "epoch": 0.09732360097323602,
21
+ "grad_norm": 0.13050897419452667,
22
+ "learning_rate": 1.2258064516129034e-05,
23
+ "loss": 0.1392993688583374,
24
+ "step": 20
25
+ },
26
+ {
27
+ "epoch": 0.145985401459854,
28
+ "grad_norm": 0.0693754181265831,
29
+ "learning_rate": 1.870967741935484e-05,
30
+ "loss": 0.10144758224487305,
31
+ "step": 30
32
+ },
33
+ {
34
+ "epoch": 0.19464720194647203,
35
+ "grad_norm": 0.0953354462981224,
36
+ "learning_rate": 1.999083553376548e-05,
37
+ "loss": 0.08977606892585754,
38
+ "step": 40
39
+ },
40
+ {
41
+ "epoch": 0.24330900243309003,
42
+ "grad_norm": 0.0660717561841011,
43
+ "learning_rate": 1.995363367648267e-05,
44
+ "loss": 0.07811610102653503,
45
+ "step": 50
46
+ },
47
+ {
48
+ "epoch": 0.291970802919708,
49
+ "grad_norm": 0.06015173718333244,
50
+ "learning_rate": 1.988792810005708e-05,
51
+ "loss": 0.06340082287788391,
52
+ "step": 60
53
+ },
54
+ {
55
+ "epoch": 0.340632603406326,
56
+ "grad_norm": 0.053529202938079834,
57
+ "learning_rate": 1.9793906962236682e-05,
58
+ "loss": 0.05219377875328064,
59
+ "step": 70
60
+ },
61
+ {
62
+ "epoch": 0.38929440389294406,
63
+ "grad_norm": 0.05288362875580788,
64
+ "learning_rate": 1.9671839506615617e-05,
65
+ "loss": 0.04788380265235901,
66
+ "step": 80
67
+ },
68
+ {
69
+ "epoch": 0.43795620437956206,
70
+ "grad_norm": 0.05314455181360245,
71
+ "learning_rate": 1.952207529161492e-05,
72
+ "loss": 0.04456350803375244,
73
+ "step": 90
74
+ },
75
+ {
76
+ "epoch": 0.48661800486618007,
77
+ "grad_norm": 0.05196147412061691,
78
+ "learning_rate": 1.9345043189469676e-05,
79
+ "loss": 0.043999499082565306,
80
+ "step": 100
81
+ },
82
+ {
83
+ "epoch": 0.48661800486618007,
84
+ "eval_loss": 0.042818356305360794,
85
+ "eval_runtime": 110.4984,
86
+ "eval_samples_per_second": 13.249,
87
+ "eval_steps_per_second": 0.833,
88
+ "step": 100
89
+ },
90
+ {
91
+ "epoch": 0.5352798053527981,
92
+ "grad_norm": 0.05217790976166725,
93
+ "learning_rate": 1.9141250158089202e-05,
94
+ "loss": 0.04414570331573486,
95
+ "step": 110
96
+ },
97
+ {
98
+ "epoch": 0.583941605839416,
99
+ "grad_norm": 0.04231223836541176,
100
+ "learning_rate": 1.8911279789307254e-05,
101
+ "loss": 0.040081775188446044,
102
+ "step": 120
103
+ },
104
+ {
105
+ "epoch": 0.6326034063260341,
106
+ "grad_norm": 0.0464063324034214,
107
+ "learning_rate": 1.865579063767948e-05,
108
+ "loss": 0.038861709833145144,
109
+ "step": 130
110
+ },
111
+ {
112
+ "epoch": 0.681265206812652,
113
+ "grad_norm": 0.04705721512436867,
114
+ "learning_rate": 1.8375514334613934e-05,
115
+ "loss": 0.03908354640007019,
116
+ "step": 140
117
+ },
118
+ {
119
+ "epoch": 0.7299270072992701,
120
+ "grad_norm": 0.050299860537052155,
121
+ "learning_rate": 1.8071253493235094e-05,
122
+ "loss": 0.03936109244823456,
123
+ "step": 150
124
+ },
125
+ {
126
+ "epoch": 0.7785888077858881,
127
+ "grad_norm": 0.048315174877643585,
128
+ "learning_rate": 1.7743879409981094e-05,
129
+ "loss": 0.03659357130527496,
130
+ "step": 160
131
+ },
132
+ {
133
+ "epoch": 0.8272506082725061,
134
+ "grad_norm": 0.054264940321445465,
135
+ "learning_rate": 1.739432956951603e-05,
136
+ "loss": 0.037407022714614865,
137
+ "step": 170
138
+ },
139
+ {
140
+ "epoch": 0.8759124087591241,
141
+ "grad_norm": 0.04875651001930237,
142
+ "learning_rate": 1.7023604960102357e-05,
143
+ "loss": 0.04030710756778717,
144
+ "step": 180
145
+ },
146
+ {
147
+ "epoch": 0.9245742092457421,
148
+ "grad_norm": 0.05363527685403824,
149
+ "learning_rate": 1.6632767207121268e-05,
150
+ "loss": 0.036510169506073,
151
+ "step": 190
152
+ },
153
+ {
154
+ "epoch": 0.9732360097323601,
155
+ "grad_norm": 0.058934297412633896,
156
+ "learning_rate": 1.6222935532949586e-05,
157
+ "loss": 0.03854315280914307,
158
+ "step": 200
159
+ },
160
+ {
161
+ "epoch": 0.9732360097323601,
162
+ "eval_loss": 0.03737252205610275,
163
+ "eval_runtime": 111.4043,
164
+ "eval_samples_per_second": 13.141,
165
+ "eval_steps_per_second": 0.826,
166
+ "step": 200
167
+ },
168
+ {
169
+ "epoch": 1.0194647201946472,
170
+ "grad_norm": 0.05371764674782753,
171
+ "learning_rate": 1.5795283551899036e-05,
172
+ "loss": 0.0347901463508606,
173
+ "step": 210
174
+ },
175
+ {
176
+ "epoch": 1.0681265206812651,
177
+ "grad_norm": 0.05175907164812088,
178
+ "learning_rate": 1.535103590939609e-05,
179
+ "loss": 0.036166781187057497,
180
+ "step": 220
181
+ },
182
+ {
183
+ "epoch": 1.1167883211678833,
184
+ "grad_norm": 0.05254443362355232,
185
+ "learning_rate": 1.4891464775026584e-05,
186
+ "loss": 0.03544372022151947,
187
+ "step": 230
188
+ },
189
+ {
190
+ "epoch": 1.1654501216545012,
191
+ "grad_norm": 0.045467473566532135,
192
+ "learning_rate": 1.4417886199487777e-05,
193
+ "loss": 0.03554186224937439,
194
+ "step": 240
195
+ },
196
+ {
197
+ "epoch": 1.2141119221411192,
198
+ "grad_norm": 0.04005361348390579,
199
+ "learning_rate": 1.3931656345880289e-05,
200
+ "loss": 0.0348221093416214,
201
+ "step": 250
202
+ },
203
+ {
204
+ "epoch": 1.2627737226277373,
205
+ "grad_norm": 0.05931864306330681,
206
+ "learning_rate": 1.3434167606132194e-05,
207
+ "loss": 0.03547696173191071,
208
+ "step": 260
209
+ },
210
+ {
211
+ "epoch": 1.3114355231143553,
212
+ "grad_norm": 0.06791075319051743,
213
+ "learning_rate": 1.2926844613676399e-05,
214
+ "loss": 0.038187041878700256,
215
+ "step": 270
216
+ },
217
+ {
218
+ "epoch": 1.3600973236009732,
219
+ "grad_norm": 0.0669725090265274,
220
+ "learning_rate": 1.2411140163799694e-05,
221
+ "loss": 0.03823740482330322,
222
+ "step": 280
223
+ },
224
+ {
225
+ "epoch": 1.4087591240875912,
226
+ "grad_norm": 0.04331658408045769,
227
+ "learning_rate": 1.1888531053346102e-05,
228
+ "loss": 0.03448362648487091,
229
+ "step": 290
230
+ },
231
+ {
232
+ "epoch": 1.4574209245742091,
233
+ "grad_norm": 0.044026345014572144,
234
+ "learning_rate": 1.136051385168817e-05,
235
+ "loss": 0.034500351548194884,
236
+ "step": 300
237
+ },
238
+ {
239
+ "epoch": 1.4574209245742091,
240
+ "eval_loss": 0.03616208955645561,
241
+ "eval_runtime": 109.8142,
242
+ "eval_samples_per_second": 13.332,
243
+ "eval_steps_per_second": 0.838,
244
+ "step": 300
245
+ },
246
+ {
247
+ "epoch": 1.5060827250608273,
248
+ "grad_norm": 0.0620030015707016,
249
+ "learning_rate": 1.0828600615076695e-05,
250
+ "loss": 0.03327265977859497,
251
+ "step": 310
252
+ },
253
+ {
254
+ "epoch": 1.5547445255474452,
255
+ "grad_norm": 0.059572961181402206,
256
+ "learning_rate": 1.0294314556641415e-05,
257
+ "loss": 0.03631613552570343,
258
+ "step": 320
259
+ },
260
+ {
261
+ "epoch": 1.6034063260340634,
262
+ "grad_norm": 0.05534299090504646,
263
+ "learning_rate": 9.759185684442364e-06,
264
+ "loss": 0.03574211597442627,
265
+ "step": 330
266
+ },
267
+ {
268
+ "epoch": 1.6520681265206814,
269
+ "grad_norm": 0.054647643119096756,
270
+ "learning_rate": 9.224746420062857e-06,
271
+ "loss": 0.034722316265106204,
272
+ "step": 340
273
+ },
274
+ {
275
+ "epoch": 1.7007299270072993,
276
+ "grad_norm": 0.04994694143533707,
277
+ "learning_rate": 8.692527210291061e-06,
278
+ "loss": 0.03766582608222961,
279
+ "step": 350
280
+ },
281
+ {
282
+ "epoch": 1.7493917274939172,
283
+ "grad_norm": 0.04726022481918335,
284
+ "learning_rate": 8.164052144456645e-06,
285
+ "loss": 0.037096422910690305,
286
+ "step": 360
287
+ },
288
+ {
289
+ "epoch": 1.7980535279805352,
290
+ "grad_norm": 0.045089419931173325,
291
+ "learning_rate": 7.640834589972964e-06,
292
+ "loss": 0.03496133089065552,
293
+ "step": 370
294
+ },
295
+ {
296
+ "epoch": 1.8467153284671531,
297
+ "grad_norm": 0.045662347227334976,
298
+ "learning_rate": 7.124372858583077e-06,
299
+ "loss": 0.03283970355987549,
300
+ "step": 380
301
+ },
302
+ {
303
+ "epoch": 1.8953771289537713,
304
+ "grad_norm": 0.05277340114116669,
305
+ "learning_rate": 6.616145915719912e-06,
306
+ "loss": 0.03219903707504272,
307
+ "step": 390
308
+ },
309
+ {
310
+ "epoch": 1.9440389294403893,
311
+ "grad_norm": 0.05256184935569763,
312
+ "learning_rate": 6.117609145267484e-06,
313
+ "loss": 0.03142733573913574,
314
+ "step": 400
315
+ },
316
+ {
317
+ "epoch": 1.9440389294403893,
318
+ "eval_loss": 0.035637106746435165,
319
+ "eval_runtime": 110.3846,
320
+ "eval_samples_per_second": 13.263,
321
+ "eval_steps_per_second": 0.833,
322
+ "step": 400
323
+ },
324
+ {
325
+ "epoch": 1.9927007299270074,
326
+ "grad_norm": 0.054978180676698685,
327
+ "learning_rate": 5.63019018185142e-06,
328
+ "loss": 0.03756006956100464,
329
+ "step": 410
330
+ },
331
+ {
332
+ "epoch": 2.0389294403892944,
333
+ "grad_norm": 0.05100775137543678,
334
+ "learning_rate": 5.155284822593612e-06,
335
+ "loss": 0.03562599122524261,
336
+ "step": 420
337
+ },
338
+ {
339
+ "epoch": 2.0875912408759123,
340
+ "grad_norm": 0.04862402379512787,
341
+ "learning_rate": 4.694253030038364e-06,
342
+ "loss": 0.03520449101924896,
343
+ "step": 430
344
+ },
345
+ {
346
+ "epoch": 2.1362530413625302,
347
+ "grad_norm": 0.05179072543978691,
348
+ "learning_rate": 4.248415037696172e-06,
349
+ "loss": 0.03458556532859802,
350
+ "step": 440
351
+ },
352
+ {
353
+ "epoch": 2.1849148418491486,
354
+ "grad_norm": 0.04908761382102966,
355
+ "learning_rate": 3.819047569357552e-06,
356
+ "loss": 0.03751436769962311,
357
+ "step": 450
358
+ },
359
+ {
360
+ "epoch": 2.2335766423357666,
361
+ "grad_norm": 0.048701491206884384,
362
+ "learning_rate": 3.4073801830034758e-06,
363
+ "loss": 0.03350481688976288,
364
+ "step": 460
365
+ },
366
+ {
367
+ "epoch": 2.2822384428223845,
368
+ "grad_norm": 0.04559648036956787,
369
+ "learning_rate": 3.0145917497821353e-06,
370
+ "loss": 0.031913352012634275,
371
+ "step": 470
372
+ },
373
+ {
374
+ "epoch": 2.3309002433090025,
375
+ "grad_norm": 0.06077806279063225,
376
+ "learning_rate": 2.641807078135048e-06,
377
+ "loss": 0.034614771604537964,
378
+ "step": 480
379
+ },
380
+ {
381
+ "epoch": 2.3795620437956204,
382
+ "grad_norm": 0.05987059697508812,
383
+ "learning_rate": 2.2900936927398065e-06,
384
+ "loss": 0.034870317578315733,
385
+ "step": 490
386
+ },
387
+ {
388
+ "epoch": 2.4282238442822384,
389
+ "grad_norm": 0.05105074867606163,
390
+ "learning_rate": 1.960458777493458e-06,
391
+ "loss": 0.03284066021442413,
392
+ "step": 500
393
+ },
394
+ {
395
+ "epoch": 2.4282238442822384,
396
+ "eval_loss": 0.035489436239004135,
397
+ "eval_runtime": 109.5355,
398
+ "eval_samples_per_second": 13.366,
399
+ "eval_steps_per_second": 0.84,
400
+ "step": 500
401
+ },
402
+ {
403
+ "epoch": 2.4768856447688563,
404
+ "grad_norm": 0.0426454097032547,
405
+ "learning_rate": 1.653846291290736e-06,
406
+ "loss": 0.03034111261367798,
407
+ "step": 510
408
+ },
409
+ {
410
+ "epoch": 2.5255474452554747,
411
+ "grad_norm": 0.05250120162963867,
412
+ "learning_rate": 1.3711342648565395e-06,
413
+ "loss": 0.033456772565841675,
414
+ "step": 520
415
+ },
416
+ {
417
+ "epoch": 2.574209245742092,
418
+ "grad_norm": 0.04732771962881088,
419
+ "learning_rate": 1.1131322863736093e-06,
420
+ "loss": 0.032463949918746945,
421
+ "step": 530
422
+ },
423
+ {
424
+ "epoch": 2.6228710462287106,
425
+ "grad_norm": 0.04851313680410385,
426
+ "learning_rate": 8.805791831056321e-07,
427
+ "loss": 0.034820270538330075,
428
+ "step": 540
429
+ },
430
+ {
431
+ "epoch": 2.6715328467153285,
432
+ "grad_norm": 0.04829060658812523,
433
+ "learning_rate": 6.741409056548498e-07,
434
+ "loss": 0.03143848776817322,
435
+ "step": 550
436
+ },
437
+ {
438
+ "epoch": 2.7201946472019465,
439
+ "grad_norm": 0.05005922541022301,
440
+ "learning_rate": 4.944086209128762e-07,
441
+ "loss": 0.03460255265235901,
442
+ "step": 560
443
+ },
444
+ {
445
+ "epoch": 2.7688564476885644,
446
+ "grad_norm": 0.05089850351214409,
447
+ "learning_rate": 3.418970191658566e-07,
448
+ "loss": 0.036028966307640076,
449
+ "step": 570
450
+ },
451
+ {
452
+ "epoch": 2.8175182481751824,
453
+ "grad_norm": 0.052207909524440765,
454
+ "learning_rate": 2.170428402018454e-07,
455
+ "loss": 0.03278985619544983,
456
+ "step": 580
457
+ },
458
+ {
459
+ "epoch": 2.8661800486618008,
460
+ "grad_norm": 0.047765858471393585,
461
+ "learning_rate": 1.2020362264107298e-07,
462
+ "loss": 0.036867809295654294,
463
+ "step": 590
464
+ },
465
+ {
466
+ "epoch": 2.9148418491484183,
467
+ "grad_norm": 0.05039682239294052,
468
+ "learning_rate": 5.165668007062641e-08,
469
+ "loss": 0.03270910978317261,
470
+ "step": 600
471
+ },
472
+ {
473
+ "epoch": 2.9148418491484183,
474
+ "eval_loss": 0.03542821854352951,
475
+ "eval_runtime": 110.8026,
476
+ "eval_samples_per_second": 13.213,
477
+ "eval_steps_per_second": 0.83,
478
+ "step": 600
479
+ },
480
+ {
481
+ "epoch": 2.9635036496350367,
482
+ "grad_norm": 0.04082901030778885,
483
+ "learning_rate": 1.1598306915499013e-08,
484
+ "loss": 0.03378779888153076,
485
+ "step": 610
486
+ },
487
+ {
488
+ "epoch": 3.0,
489
+ "eval_loss": 0.035394057631492615,
490
+ "eval_runtime": 111.6352,
491
+ "eval_samples_per_second": 13.114,
492
+ "eval_steps_per_second": 0.824,
493
+ "step": 618
494
+ }
495
+ ],
496
+ "logging_steps": 10,
497
+ "max_steps": 618,
498
+ "num_input_tokens_seen": 0,
499
+ "num_train_epochs": 3,
500
+ "save_steps": 100,
501
+ "stateful_callbacks": {
502
+ "EarlyStoppingCallback": {
503
+ "args": {
504
+ "early_stopping_patience": 3,
505
+ "early_stopping_threshold": 0.0
506
+ },
507
+ "attributes": {
508
+ "early_stopping_patience_counter": 0
509
+ }
510
+ },
511
+ "TrainerControl": {
512
+ "args": {
513
+ "should_epoch_stop": false,
514
+ "should_evaluate": false,
515
+ "should_log": false,
516
+ "should_save": true,
517
+ "should_training_stop": true
518
+ },
519
+ "attributes": {}
520
+ }
521
+ },
522
+ "total_flos": 3.3504636406868214e+18,
523
+ "train_batch_size": 2,
524
+ "trial_name": null,
525
+ "trial_params": null
526
+ }
Conifer/PD/Qwen3-8B/final/adapter_config.json ADDED
@@ -0,0 +1,45 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alora_invocation_tokens": null,
3
+ "alpha_pattern": {},
4
+ "arrow_config": null,
5
+ "auto_mapping": null,
6
+ "base_model_name_or_path": "Qwen/Qwen3-8B",
7
+ "bias": "none",
8
+ "corda_config": null,
9
+ "ensure_weight_tying": false,
10
+ "eva_config": null,
11
+ "exclude_modules": null,
12
+ "fan_in_fan_out": false,
13
+ "inference_mode": true,
14
+ "init_lora_weights": true,
15
+ "layer_replication": null,
16
+ "layers_pattern": null,
17
+ "layers_to_transform": null,
18
+ "loftq_config": {},
19
+ "lora_alpha": 64,
20
+ "lora_bias": false,
21
+ "lora_dropout": 0.05,
22
+ "lora_ga_config": null,
23
+ "megatron_config": null,
24
+ "megatron_core": "megatron.core",
25
+ "modules_to_save": null,
26
+ "peft_type": "LORA",
27
+ "peft_version": "0.19.1",
28
+ "qalora_group_size": 16,
29
+ "r": 64,
30
+ "rank_pattern": {},
31
+ "revision": null,
32
+ "target_modules": [
33
+ "o_proj",
34
+ "v_proj",
35
+ "k_proj",
36
+ "q_proj"
37
+ ],
38
+ "target_parameters": null,
39
+ "task_type": "CAUSAL_LM",
40
+ "trainable_token_indices": null,
41
+ "use_bdlora": null,
42
+ "use_dora": false,
43
+ "use_qalora": false,
44
+ "use_rslora": false
45
+ }
Conifer/PD/Qwen3-8B/final/chat_template.jinja ADDED
@@ -0,0 +1,89 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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 message.content is string 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.content is string %}
27
+ {%- set content = message.content %}
28
+ {%- else %}
29
+ {%- set content = '' %}
30
+ {%- endif %}
31
+ {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
32
+ {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
33
+ {%- elif message.role == "assistant" %}
34
+ {%- set reasoning_content = '' %}
35
+ {%- if message.reasoning_content is string %}
36
+ {%- set reasoning_content = message.reasoning_content %}
37
+ {%- else %}
38
+ {%- if '</think>' in content %}
39
+ {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
40
+ {%- set content = content.split('</think>')[-1].lstrip('\n') %}
41
+ {%- endif %}
42
+ {%- endif %}
43
+ {%- if loop.index0 > ns.last_query_index %}
44
+ {%- if loop.last or (not loop.last and reasoning_content) %}
45
+ {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
46
+ {%- else %}
47
+ {{- '<|im_start|>' + message.role + '\n' + content }}
48
+ {%- endif %}
49
+ {%- else %}
50
+ {{- '<|im_start|>' + message.role + '\n' + content }}
51
+ {%- endif %}
52
+ {%- if message.tool_calls %}
53
+ {%- for tool_call in message.tool_calls %}
54
+ {%- if (loop.first and content) or (not loop.first) %}
55
+ {{- '\n' }}
56
+ {%- endif %}
57
+ {%- if tool_call.function %}
58
+ {%- set tool_call = tool_call.function %}
59
+ {%- endif %}
60
+ {{- '<tool_call>\n{"name": "' }}
61
+ {{- tool_call.name }}
62
+ {{- '", "arguments": ' }}
63
+ {%- if tool_call.arguments is string %}
64
+ {{- tool_call.arguments }}
65
+ {%- else %}
66
+ {{- tool_call.arguments | tojson }}
67
+ {%- endif %}
68
+ {{- '}\n</tool_call>' }}
69
+ {%- endfor %}
70
+ {%- endif %}
71
+ {{- '<|im_end|>\n' }}
72
+ {%- elif message.role == "tool" %}
73
+ {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
74
+ {{- '<|im_start|>user' }}
75
+ {%- endif %}
76
+ {{- '\n<tool_response>\n' }}
77
+ {{- content }}
78
+ {{- '\n</tool_response>' }}
79
+ {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
80
+ {{- '<|im_end|>\n' }}
81
+ {%- endif %}
82
+ {%- endif %}
83
+ {%- endfor %}
84
+ {%- if add_generation_prompt %}
85
+ {{- '<|im_start|>assistant\n' }}
86
+ {%- if enable_thinking is defined and enable_thinking is false %}
87
+ {{- '<think>\n\n</think>\n\n' }}
88
+ {%- endif %}
89
+ {%- endif %}
Conifer/PD/Qwen3-8B/final/final_eval_metrics.json ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ {
2
+ "eval_loss": 0.035394057631492615,
3
+ "eval_runtime": 110.9615,
4
+ "eval_samples_per_second": 13.194,
5
+ "eval_steps_per_second": 0.829,
6
+ "epoch": 3.0
7
+ }
Conifer/PD/Qwen3-8B/final/tokenizer_config.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
+ "<|im_start|>",
10
+ "<|im_end|>",
11
+ "<|object_ref_start|>",
12
+ "<|object_ref_end|>",
13
+ "<|box_start|>",
14
+ "<|box_end|>",
15
+ "<|quad_start|>",
16
+ "<|quad_end|>",
17
+ "<|vision_start|>",
18
+ "<|vision_end|>",
19
+ "<|vision_pad|>",
20
+ "<|image_pad|>",
21
+ "<|video_pad|>"
22
+ ],
23
+ "is_local": true,
24
+ "local_files_only": false,
25
+ "model_max_length": 131072,
26
+ "pad_token": "<|endoftext|>",
27
+ "split_special_tokens": false,
28
+ "tokenizer_class": "Qwen2Tokenizer",
29
+ "unk_token": null
30
+ }
Conifer/PD/Qwen3-8B/final/train_summary.json ADDED
@@ -0,0 +1,12 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "global_step": 618,
3
+ "training_loss": 0.042559183145418136,
4
+ "metrics": {
5
+ "train_runtime": 5439.3922,
6
+ "train_samples_per_second": 7.249,
7
+ "train_steps_per_second": 0.114,
8
+ "total_flos": 3.3504636406868214e+18,
9
+ "train_loss": 0.042559183145418136,
10
+ "epoch": 3.0
11
+ }
12
+ }