K2triinK commited on
Commit
5505b1c
·
verified ·
1 Parent(s): 99bed58

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. substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-1632/chat_template.jinja +154 -0
  2. substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-1632/tokenizer_config.json +31 -0
  3. substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-1632/trainer_state.json +442 -0
  4. substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2040/README.md +209 -0
  5. substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2040/adapter_config.json +48 -0
  6. substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2040/chat_template.jinja +154 -0
  7. substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2040/tokenizer_config.json +31 -0
  8. substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2040/trainer_state.json +544 -0
  9. substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2448/README.md +209 -0
  10. substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2448/adapter_config.json +48 -0
  11. substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2448/chat_template.jinja +154 -0
  12. substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2448/tokenizer_config.json +31 -0
  13. substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2448/trainer_state.json +646 -0
  14. substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2856/README.md +209 -0
  15. substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2856/adapter_config.json +48 -0
  16. substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2856/chat_template.jinja +154 -0
  17. substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2856/tokenizer_config.json +31 -0
  18. substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2856/trainer_state.json +758 -0
  19. substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3264/README.md +209 -0
  20. substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3264/adapter_config.json +48 -0
  21. substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3264/chat_template.jinja +154 -0
  22. substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3264/tokenizer_config.json +31 -0
  23. substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3264/trainer_state.json +860 -0
  24. substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3672/README.md +209 -0
  25. substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3672/adapter_config.json +48 -0
  26. substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3672/chat_template.jinja +154 -0
  27. substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3672/tokenizer_config.json +31 -0
  28. substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3672/trainer_state.json +962 -0
  29. substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-408/README.md +209 -0
  30. substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-408/adapter_config.json +48 -0
  31. substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-408/trainer_state.json +136 -0
  32. substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test2/checkpoint-1632/README.md +209 -0
  33. systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1371/README.md +209 -0
  34. systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1371/adapter_config.json +48 -0
  35. systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1371/chat_template.jinja +154 -0
  36. systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1371/tokenizer_config.json +31 -0
  37. systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1371/trainer_state.json +337 -0
  38. systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1828/README.md +209 -0
  39. systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1828/adapter_config.json +48 -0
  40. systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1828/chat_template.jinja +154 -0
  41. systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1828/tokenizer_config.json +31 -0
  42. systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1828/trainer_state.json +438 -0
  43. systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-2285/README.md +209 -0
  44. systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-2285/adapter_config.json +48 -0
  45. systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-2285/chat_template.jinja +154 -0
  46. systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-2285/tokenizer_config.json +31 -0
  47. systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-2285/trainer_state.json +539 -0
  48. systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-2742/README.md +209 -0
  49. systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-2742/adapter_config.json +48 -0
  50. systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-2742/chat_template.jinja +154 -0
substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-1632/chat_template.jinja ADDED
@@ -0,0 +1,154 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {%- set image_count = namespace(value=0) %}
2
+ {%- set video_count = namespace(value=0) %}
3
+ {%- macro render_content(content, do_vision_count, is_system_content=false) %}
4
+ {%- if content is string %}
5
+ {{- content }}
6
+ {%- elif content is iterable and content is not mapping %}
7
+ {%- for item in content %}
8
+ {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
9
+ {%- if is_system_content %}
10
+ {{- raise_exception('System message cannot contain images.') }}
11
+ {%- endif %}
12
+ {%- if do_vision_count %}
13
+ {%- set image_count.value = image_count.value + 1 %}
14
+ {%- endif %}
15
+ {%- if add_vision_id %}
16
+ {{- 'Picture ' ~ image_count.value ~ ': ' }}
17
+ {%- endif %}
18
+ {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
19
+ {%- elif 'video' in item or item.type == 'video' %}
20
+ {%- if is_system_content %}
21
+ {{- raise_exception('System message cannot contain videos.') }}
22
+ {%- endif %}
23
+ {%- if do_vision_count %}
24
+ {%- set video_count.value = video_count.value + 1 %}
25
+ {%- endif %}
26
+ {%- if add_vision_id %}
27
+ {{- 'Video ' ~ video_count.value ~ ': ' }}
28
+ {%- endif %}
29
+ {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
30
+ {%- elif 'text' in item %}
31
+ {{- item.text }}
32
+ {%- else %}
33
+ {{- raise_exception('Unexpected item type in content.') }}
34
+ {%- endif %}
35
+ {%- endfor %}
36
+ {%- elif content is none or content is undefined %}
37
+ {{- '' }}
38
+ {%- else %}
39
+ {{- raise_exception('Unexpected content type.') }}
40
+ {%- endif %}
41
+ {%- endmacro %}
42
+ {%- if not messages %}
43
+ {{- raise_exception('No messages provided.') }}
44
+ {%- endif %}
45
+ {%- if tools and tools is iterable and tools is not mapping %}
46
+ {{- '<|im_start|>system\n' }}
47
+ {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
48
+ {%- for tool in tools %}
49
+ {{- "\n" }}
50
+ {{- tool | tojson }}
51
+ {%- endfor %}
52
+ {{- "\n</tools>" }}
53
+ {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
54
+ {%- if messages[0].role == 'system' %}
55
+ {%- set content = render_content(messages[0].content, false, true)|trim %}
56
+ {%- if content %}
57
+ {{- '\n\n' + content }}
58
+ {%- endif %}
59
+ {%- endif %}
60
+ {{- '<|im_end|>\n' }}
61
+ {%- else %}
62
+ {%- if messages[0].role == 'system' %}
63
+ {%- set content = render_content(messages[0].content, false, true)|trim %}
64
+ {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
65
+ {%- endif %}
66
+ {%- endif %}
67
+ {%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
68
+ {%- for message in messages[::-1] %}
69
+ {%- set index = (messages|length - 1) - loop.index0 %}
70
+ {%- if ns.multi_step_tool and message.role == "user" %}
71
+ {%- set content = render_content(message.content, false)|trim %}
72
+ {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
73
+ {%- set ns.multi_step_tool = false %}
74
+ {%- set ns.last_query_index = index %}
75
+ {%- endif %}
76
+ {%- endif %}
77
+ {%- endfor %}
78
+ {%- if ns.multi_step_tool %}
79
+ {{- raise_exception('No user query found in messages.') }}
80
+ {%- endif %}
81
+ {%- for message in messages %}
82
+ {%- set content = render_content(message.content, true)|trim %}
83
+ {%- if message.role == "system" %}
84
+ {%- if not loop.first %}
85
+ {{- raise_exception('System message must be at the beginning.') }}
86
+ {%- endif %}
87
+ {%- elif message.role == "user" %}
88
+ {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
89
+ {%- elif message.role == "assistant" %}
90
+ {%- set reasoning_content = '' %}
91
+ {%- if message.reasoning_content is string %}
92
+ {%- set reasoning_content = message.reasoning_content %}
93
+ {%- else %}
94
+ {%- if '</think>' in content %}
95
+ {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
96
+ {%- set content = content.split('</think>')[-1].lstrip('\n') %}
97
+ {%- endif %}
98
+ {%- endif %}
99
+ {%- set reasoning_content = reasoning_content|trim %}
100
+ {%- if loop.index0 > ns.last_query_index %}
101
+ {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
102
+ {%- else %}
103
+ {{- '<|im_start|>' + message.role + '\n' + content }}
104
+ {%- endif %}
105
+ {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
106
+ {%- for tool_call in message.tool_calls %}
107
+ {%- if tool_call.function is defined %}
108
+ {%- set tool_call = tool_call.function %}
109
+ {%- endif %}
110
+ {%- if loop.first %}
111
+ {%- if content|trim %}
112
+ {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
113
+ {%- else %}
114
+ {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
115
+ {%- endif %}
116
+ {%- else %}
117
+ {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
118
+ {%- endif %}
119
+ {%- if tool_call.arguments is defined %}
120
+ {%- for args_name, args_value in tool_call.arguments|items %}
121
+ {{- '<parameter=' + args_name + '>\n' }}
122
+ {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
123
+ {{- args_value }}
124
+ {{- '\n</parameter>\n' }}
125
+ {%- endfor %}
126
+ {%- endif %}
127
+ {{- '</function>\n</tool_call>' }}
128
+ {%- endfor %}
129
+ {%- endif %}
130
+ {{- '<|im_end|>\n' }}
131
+ {%- elif message.role == "tool" %}
132
+ {%- if loop.previtem and loop.previtem.role != "tool" %}
133
+ {{- '<|im_start|>user' }}
134
+ {%- endif %}
135
+ {{- '\n<tool_response>\n' }}
136
+ {{- content }}
137
+ {{- '\n</tool_response>' }}
138
+ {%- if not loop.last and loop.nextitem.role != "tool" %}
139
+ {{- '<|im_end|>\n' }}
140
+ {%- elif loop.last %}
141
+ {{- '<|im_end|>\n' }}
142
+ {%- endif %}
143
+ {%- else %}
144
+ {{- raise_exception('Unexpected message role.') }}
145
+ {%- endif %}
146
+ {%- endfor %}
147
+ {%- if add_generation_prompt %}
148
+ {{- '<|im_start|>assistant\n' }}
149
+ {%- if enable_thinking is defined and enable_thinking is true %}
150
+ {{- '<think>\n' }}
151
+ {%- else %}
152
+ {{- '<think>\n\n</think>\n\n' }}
153
+ {%- endif %}
154
+ {%- endif %}
substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-1632/tokenizer_config.json ADDED
@@ -0,0 +1,31 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "audio_bos_token": "<|audio_start|>",
4
+ "audio_eos_token": "<|audio_end|>",
5
+ "audio_token": "<|audio_pad|>",
6
+ "backend": "tokenizers",
7
+ "bos_token": null,
8
+ "clean_up_tokenization_spaces": false,
9
+ "eos_token": "<|endoftext|>",
10
+ "errors": "replace",
11
+ "image_token": "<|image_pad|>",
12
+ "is_local": false,
13
+ "model_max_length": 262144,
14
+ "model_specific_special_tokens": {
15
+ "audio_bos_token": "<|audio_start|>",
16
+ "audio_eos_token": "<|audio_end|>",
17
+ "audio_token": "<|audio_pad|>",
18
+ "image_token": "<|image_pad|>",
19
+ "video_token": "<|video_pad|>",
20
+ "vision_bos_token": "<|vision_start|>",
21
+ "vision_eos_token": "<|vision_end|>"
22
+ },
23
+ "pad_token": "<|endoftext|>",
24
+ "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
25
+ "split_special_tokens": false,
26
+ "tokenizer_class": "TokenizersBackend",
27
+ "unk_token": null,
28
+ "video_token": "<|video_pad|>",
29
+ "vision_bos_token": "<|vision_start|>",
30
+ "vision_eos_token": "<|vision_end|>"
31
+ }
substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-1632/trainer_state.json ADDED
@@ -0,0 +1,442 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 4.0,
6
+ "eval_steps": 500,
7
+ "global_step": 1632,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "entropy": 2.2408065378665922,
14
+ "epoch": 0.12269938650306748,
15
+ "grad_norm": 1.6718658208847046,
16
+ "learning_rate": 4.487804970912036e-05,
17
+ "loss": 2.129624786376953,
18
+ "mean_token_accuracy": 0.5743644836544991,
19
+ "num_tokens": 134329.0,
20
+ "step": 50
21
+ },
22
+ {
23
+ "entropy": 1.1078504550457,
24
+ "epoch": 0.24539877300613497,
25
+ "grad_norm": 2.4091813564300537,
26
+ "learning_rate": 9.067197798373296e-05,
27
+ "loss": 1.0346390533447265,
28
+ "mean_token_accuracy": 0.7085381114482879,
29
+ "num_tokens": 262407.0,
30
+ "step": 100
31
+ },
32
+ {
33
+ "entropy": 0.9421598988771439,
34
+ "epoch": 0.36809815950920244,
35
+ "grad_norm": 1.3913646936416626,
36
+ "learning_rate": 0.00013646590625834558,
37
+ "loss": 0.8716551208496094,
38
+ "mean_token_accuracy": 0.7449040985107422,
39
+ "num_tokens": 401136.0,
40
+ "step": 150
41
+ },
42
+ {
43
+ "entropy": 0.8723550146818161,
44
+ "epoch": 0.49079754601226994,
45
+ "grad_norm": 1.1664468050003052,
46
+ "learning_rate": 0.0001822598345329582,
47
+ "loss": 0.8095977783203125,
48
+ "mean_token_accuracy": 0.7585240352153778,
49
+ "num_tokens": 531647.0,
50
+ "step": 200
51
+ },
52
+ {
53
+ "entropy": 0.8466519421339035,
54
+ "epoch": 0.6134969325153374,
55
+ "grad_norm": 1.0145421028137207,
56
+ "learning_rate": 0.00022805376280757083,
57
+ "loss": 0.7830857849121093,
58
+ "mean_token_accuracy": 0.7661705195903779,
59
+ "num_tokens": 666127.0,
60
+ "step": 250
61
+ },
62
+ {
63
+ "entropy": 0.8180295366048813,
64
+ "epoch": 0.7361963190184049,
65
+ "grad_norm": 0.9731705784797668,
66
+ "learning_rate": 0.00027384769108218336,
67
+ "loss": 0.7573603820800782,
68
+ "mean_token_accuracy": 0.7704453605413437,
69
+ "num_tokens": 799292.0,
70
+ "step": 300
71
+ },
72
+ {
73
+ "entropy": 0.7990227991342544,
74
+ "epoch": 0.8588957055214724,
75
+ "grad_norm": 1.0642374753952026,
76
+ "learning_rate": 0.000319641619356796,
77
+ "loss": 0.73709228515625,
78
+ "mean_token_accuracy": 0.7752733880281448,
79
+ "num_tokens": 930614.0,
80
+ "step": 350
81
+ },
82
+ {
83
+ "entropy": 0.7899087685346603,
84
+ "epoch": 0.9815950920245399,
85
+ "grad_norm": 0.880613386631012,
86
+ "learning_rate": 0.0003654355476314086,
87
+ "loss": 0.7347319793701171,
88
+ "mean_token_accuracy": 0.7766731631755829,
89
+ "num_tokens": 1061654.0,
90
+ "step": 400
91
+ },
92
+ {
93
+ "epoch": 1.0,
94
+ "eval_entropy": 0.83522472347532,
95
+ "eval_mean_token_accuracy": 0.7484098076820374,
96
+ "eval_not_syn_loss": 0.7872886061668396,
97
+ "eval_not_syn_runtime": 54.8697,
98
+ "eval_not_syn_samples_per_second": 25.424,
99
+ "eval_not_syn_steps_per_second": 3.189,
100
+ "eval_num_tokens": 1080196.0,
101
+ "step": 408
102
+ },
103
+ {
104
+ "epoch": 1.0,
105
+ "eval_entropy": 0.7784442271505083,
106
+ "eval_mean_token_accuracy": 0.8021791185651507,
107
+ "eval_num_tokens": 1080196.0,
108
+ "eval_syn_loss": 0.6980098485946655,
109
+ "eval_syn_runtime": 56.6342,
110
+ "eval_syn_samples_per_second": 24.632,
111
+ "eval_syn_steps_per_second": 3.09,
112
+ "step": 408
113
+ },
114
+ {
115
+ "entropy": 0.75672573694075,
116
+ "epoch": 1.1030674846625768,
117
+ "grad_norm": 0.9244198203086853,
118
+ "learning_rate": 0.00037356351883456205,
119
+ "loss": 0.7048745727539063,
120
+ "mean_token_accuracy": 0.783016896609104,
121
+ "num_tokens": 1199170.0,
122
+ "step": 450
123
+ },
124
+ {
125
+ "entropy": 0.7406797379255294,
126
+ "epoch": 1.2257668711656442,
127
+ "grad_norm": 0.6744896769523621,
128
+ "learning_rate": 0.00037311248151765587,
129
+ "loss": 0.6860730743408203,
130
+ "mean_token_accuracy": 0.787369327545166,
131
+ "num_tokens": 1334298.0,
132
+ "step": 500
133
+ },
134
+ {
135
+ "entropy": 0.7424973237514496,
136
+ "epoch": 1.3484662576687116,
137
+ "grad_norm": 0.92305588722229,
138
+ "learning_rate": 0.000372320629215228,
139
+ "loss": 0.6855724334716797,
140
+ "mean_token_accuracy": 0.7871134179830551,
141
+ "num_tokens": 1463189.0,
142
+ "step": 550
143
+ },
144
+ {
145
+ "entropy": 0.7502243828773498,
146
+ "epoch": 1.471165644171779,
147
+ "grad_norm": 0.8259047269821167,
148
+ "learning_rate": 0.00037118941074037944,
149
+ "loss": 0.6949834442138672,
150
+ "mean_token_accuracy": 0.7851914083957672,
151
+ "num_tokens": 1590310.0,
152
+ "step": 600
153
+ },
154
+ {
155
+ "entropy": 0.7233260518312454,
156
+ "epoch": 1.5938650306748468,
157
+ "grad_norm": 0.882722795009613,
158
+ "learning_rate": 0.00036972089582775814,
159
+ "loss": 0.6649005889892579,
160
+ "mean_token_accuracy": 0.7898879665136337,
161
+ "num_tokens": 1721836.0,
162
+ "step": 650
163
+ },
164
+ {
165
+ "entropy": 0.7162708967924118,
166
+ "epoch": 1.716564417177914,
167
+ "grad_norm": 0.6593027114868164,
168
+ "learning_rate": 0.0003679177713466678,
169
+ "loss": 0.6653710174560546,
170
+ "mean_token_accuracy": 0.7905678844451904,
171
+ "num_tokens": 1853507.0,
172
+ "step": 700
173
+ },
174
+ {
175
+ "entropy": 0.7150520020723343,
176
+ "epoch": 1.8392638036809816,
177
+ "grad_norm": 0.665340781211853,
178
+ "learning_rate": 0.0003657833363850354,
179
+ "loss": 0.6600938415527344,
180
+ "mean_token_accuracy": 0.7918649202585221,
181
+ "num_tokens": 1988113.0,
182
+ "step": 750
183
+ },
184
+ {
185
+ "entropy": 0.6993386310338974,
186
+ "epoch": 1.961963190184049,
187
+ "grad_norm": 0.6502187252044678,
188
+ "learning_rate": 0.00036332149621323294,
189
+ "loss": 0.6469607543945313,
190
+ "mean_token_accuracy": 0.7950778317451477,
191
+ "num_tokens": 2121314.0,
192
+ "step": 800
193
+ },
194
+ {
195
+ "epoch": 2.0,
196
+ "eval_entropy": 0.7315139893123082,
197
+ "eval_mean_token_accuracy": 0.7713504004478454,
198
+ "eval_not_syn_loss": 0.699893593788147,
199
+ "eval_not_syn_runtime": 55.4135,
200
+ "eval_not_syn_samples_per_second": 25.174,
201
+ "eval_not_syn_steps_per_second": 3.158,
202
+ "eval_num_tokens": 2160392.0,
203
+ "step": 816
204
+ },
205
+ {
206
+ "epoch": 2.0,
207
+ "eval_entropy": 0.6848451798302787,
208
+ "eval_mean_token_accuracy": 0.814895657471248,
209
+ "eval_num_tokens": 2160392.0,
210
+ "eval_syn_loss": 0.6415576934814453,
211
+ "eval_syn_runtime": 56.4735,
212
+ "eval_syn_samples_per_second": 24.702,
213
+ "eval_syn_steps_per_second": 3.099,
214
+ "step": 816
215
+ },
216
+ {
217
+ "entropy": 0.6596363644407253,
218
+ "epoch": 2.083435582822086,
219
+ "grad_norm": 0.6647700071334839,
220
+ "learning_rate": 0.00036053675513879666,
221
+ "loss": 0.599151611328125,
222
+ "mean_token_accuracy": 0.8071986864311527,
223
+ "num_tokens": 2253179.0,
224
+ "step": 850
225
+ },
226
+ {
227
+ "entropy": 0.6626429110765457,
228
+ "epoch": 2.2061349693251535,
229
+ "grad_norm": 0.7356868386268616,
230
+ "learning_rate": 0.00035743420826511783,
231
+ "loss": 0.6004195404052735,
232
+ "mean_token_accuracy": 0.805362731218338,
233
+ "num_tokens": 2379965.0,
234
+ "step": 900
235
+ },
236
+ {
237
+ "entropy": 0.6503470712900161,
238
+ "epoch": 2.3288343558282207,
239
+ "grad_norm": 1.095608115196228,
240
+ "learning_rate": 0.0003540195321691835,
241
+ "loss": 0.5951333999633789,
242
+ "mean_token_accuracy": 0.807324025630951,
243
+ "num_tokens": 2515593.0,
244
+ "step": 950
245
+ },
246
+ {
247
+ "entropy": 0.660577870607376,
248
+ "epoch": 2.4515337423312884,
249
+ "grad_norm": 0.6602179408073425,
250
+ "learning_rate": 0.00035029897451542325,
251
+ "loss": 0.6056365585327148,
252
+ "mean_token_accuracy": 0.8057438349723816,
253
+ "num_tokens": 2642790.0,
254
+ "step": 1000
255
+ },
256
+ {
257
+ "entropy": 0.6555853113532066,
258
+ "epoch": 2.574233128834356,
259
+ "grad_norm": 0.7459151148796082,
260
+ "learning_rate": 0.00034627934262466606,
261
+ "loss": 0.5963204193115235,
262
+ "mean_token_accuracy": 0.8060598099231719,
263
+ "num_tokens": 2774396.0,
264
+ "step": 1050
265
+ },
266
+ {
267
+ "entropy": 0.6505853188037872,
268
+ "epoch": 2.6969325153374233,
269
+ "grad_norm": 0.6359792947769165,
270
+ "learning_rate": 0.00034196799101912103,
271
+ "loss": 0.5993065643310547,
272
+ "mean_token_accuracy": 0.806280387043953,
273
+ "num_tokens": 2912102.0,
274
+ "step": 1100
275
+ },
276
+ {
277
+ "entropy": 0.6509970253705979,
278
+ "epoch": 2.819631901840491,
279
+ "grad_norm": 0.8086174726486206,
280
+ "learning_rate": 0.0003373728079661707,
281
+ "loss": 0.6001268005371094,
282
+ "mean_token_accuracy": 0.806866540312767,
283
+ "num_tokens": 3042987.0,
284
+ "step": 1150
285
+ },
286
+ {
287
+ "entropy": 0.6404844307899475,
288
+ "epoch": 2.942331288343558,
289
+ "grad_norm": 0.7038407325744629,
290
+ "learning_rate": 0.0003325022010455975,
291
+ "loss": 0.5870526123046875,
292
+ "mean_token_accuracy": 0.8104586935043335,
293
+ "num_tokens": 3178044.0,
294
+ "step": 1200
295
+ },
296
+ {
297
+ "epoch": 3.0,
298
+ "eval_entropy": 0.6645996727262224,
299
+ "eval_mean_token_accuracy": 0.798654237134116,
300
+ "eval_not_syn_loss": 0.6586771011352539,
301
+ "eval_not_syn_runtime": 55.1088,
302
+ "eval_not_syn_samples_per_second": 25.314,
303
+ "eval_not_syn_steps_per_second": 3.176,
304
+ "eval_num_tokens": 3240588.0,
305
+ "step": 1224
306
+ },
307
+ {
308
+ "epoch": 3.0,
309
+ "eval_entropy": 0.6233607809884207,
310
+ "eval_mean_token_accuracy": 0.7977249867575509,
311
+ "eval_num_tokens": 3240588.0,
312
+ "eval_syn_loss": 0.6288298964500427,
313
+ "eval_syn_runtime": 56.6081,
314
+ "eval_syn_samples_per_second": 24.643,
315
+ "eval_syn_steps_per_second": 3.091,
316
+ "step": 1224
317
+ },
318
+ {
319
+ "entropy": 0.6081961104364106,
320
+ "epoch": 3.063803680981595,
321
+ "grad_norm": 0.6161203980445862,
322
+ "learning_rate": 0.0003273650817666497,
323
+ "loss": 0.5481716537475586,
324
+ "mean_token_accuracy": 0.8175542035488167,
325
+ "num_tokens": 3312797.0,
326
+ "step": 1250
327
+ },
328
+ {
329
+ "entropy": 0.5877064895629883,
330
+ "epoch": 3.1865030674846624,
331
+ "grad_norm": 0.6909636855125427,
332
+ "learning_rate": 0.00032197084926309234,
333
+ "loss": 0.5224150466918945,
334
+ "mean_token_accuracy": 0.8250984263420105,
335
+ "num_tokens": 3441629.0,
336
+ "step": 1300
337
+ },
338
+ {
339
+ "entropy": 0.5952656552195549,
340
+ "epoch": 3.30920245398773,
341
+ "grad_norm": 0.8147026300430298,
342
+ "learning_rate": 0.0003163293730960758,
343
+ "loss": 0.5349474334716797,
344
+ "mean_token_accuracy": 0.8213064837455749,
345
+ "num_tokens": 3576093.0,
346
+ "step": 1350
347
+ },
348
+ {
349
+ "entropy": 0.5937123569846153,
350
+ "epoch": 3.4319018404907977,
351
+ "grad_norm": 0.8690970540046692,
352
+ "learning_rate": 0.0003104509751962869,
353
+ "loss": 0.5310481262207031,
354
+ "mean_token_accuracy": 0.8229135835170746,
355
+ "num_tokens": 3705912.0,
356
+ "step": 1400
357
+ },
358
+ {
359
+ "entropy": 0.6067730465531349,
360
+ "epoch": 3.554601226993865,
361
+ "grad_norm": 0.7433330416679382,
362
+ "learning_rate": 0.0003043464109784207,
363
+ "loss": 0.5450045394897461,
364
+ "mean_token_accuracy": 0.8205251175165177,
365
+ "num_tokens": 3832376.0,
366
+ "step": 1450
367
+ },
368
+ {
369
+ "entropy": 0.5926914718747139,
370
+ "epoch": 3.6773006134969326,
371
+ "grad_norm": 0.7265587449073792,
372
+ "learning_rate": 0.00029802684966252974,
373
+ "loss": 0.5330168914794922,
374
+ "mean_token_accuracy": 0.821294127702713,
375
+ "num_tokens": 3967273.0,
376
+ "step": 1500
377
+ },
378
+ {
379
+ "entropy": 0.5881199997663498,
380
+ "epoch": 3.8,
381
+ "grad_norm": 0.7681817412376404,
382
+ "learning_rate": 0.00029150385383825187,
383
+ "loss": 0.5288929367065429,
384
+ "mean_token_accuracy": 0.8234161508083343,
385
+ "num_tokens": 4102434.0,
386
+ "step": 1550
387
+ },
388
+ {
389
+ "entropy": 0.6008686199784279,
390
+ "epoch": 3.9226993865030675,
391
+ "grad_norm": 0.9384056329727173,
392
+ "learning_rate": 0.0002847893583093113,
393
+ "loss": 0.5437511444091797,
394
+ "mean_token_accuracy": 0.8201366275548935,
395
+ "num_tokens": 4234037.0,
396
+ "step": 1600
397
+ },
398
+ {
399
+ "epoch": 4.0,
400
+ "eval_entropy": 0.6218148783275059,
401
+ "eval_mean_token_accuracy": 0.783796865599496,
402
+ "eval_not_syn_loss": 0.6558170318603516,
403
+ "eval_not_syn_runtime": 55.429,
404
+ "eval_not_syn_samples_per_second": 25.167,
405
+ "eval_not_syn_steps_per_second": 3.157,
406
+ "eval_num_tokens": 4320784.0,
407
+ "step": 1632
408
+ },
409
+ {
410
+ "epoch": 4.0,
411
+ "eval_entropy": 0.5833754854542869,
412
+ "eval_mean_token_accuracy": 0.8278223422595433,
413
+ "eval_num_tokens": 4320784.0,
414
+ "eval_syn_loss": 0.604770302772522,
415
+ "eval_syn_runtime": 56.9593,
416
+ "eval_syn_samples_per_second": 24.491,
417
+ "eval_syn_steps_per_second": 3.072,
418
+ "step": 1632
419
+ }
420
+ ],
421
+ "logging_steps": 50,
422
+ "max_steps": 4080,
423
+ "num_input_tokens_seen": 0,
424
+ "num_train_epochs": 10,
425
+ "save_steps": 500,
426
+ "stateful_callbacks": {
427
+ "TrainerControl": {
428
+ "args": {
429
+ "should_epoch_stop": false,
430
+ "should_evaluate": false,
431
+ "should_log": false,
432
+ "should_save": true,
433
+ "should_training_stop": false
434
+ },
435
+ "attributes": {}
436
+ }
437
+ },
438
+ "total_flos": 8.61560698569166e+16,
439
+ "train_batch_size": 8,
440
+ "trial_name": null,
441
+ "trial_params": null
442
+ }
substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2040/README.md ADDED
@@ -0,0 +1,209 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: Qwen/Qwen3.5-2B-Base
3
+ library_name: peft
4
+ pipeline_tag: text-generation
5
+ tags:
6
+ - base_model:adapter:Qwen/Qwen3.5-2B-Base
7
+ - lora
8
+ - sft
9
+ - transformers
10
+ - trl
11
+ ---
12
+
13
+ # Model Card for Model ID
14
+
15
+ <!-- Provide a quick summary of what the model is/does. -->
16
+
17
+
18
+
19
+ ## Model Details
20
+
21
+ ### Model Description
22
+
23
+ <!-- Provide a longer summary of what this model is. -->
24
+
25
+
26
+
27
+ - **Developed by:** [More Information Needed]
28
+ - **Funded by [optional]:** [More Information Needed]
29
+ - **Shared by [optional]:** [More Information Needed]
30
+ - **Model type:** [More Information Needed]
31
+ - **Language(s) (NLP):** [More Information Needed]
32
+ - **License:** [More Information Needed]
33
+ - **Finetuned from model [optional]:** [More Information Needed]
34
+
35
+ ### Model Sources [optional]
36
+
37
+ <!-- Provide the basic links for the model. -->
38
+
39
+ - **Repository:** [More Information Needed]
40
+ - **Paper [optional]:** [More Information Needed]
41
+ - **Demo [optional]:** [More Information Needed]
42
+
43
+ ## Uses
44
+
45
+ <!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
46
+
47
+ ### Direct Use
48
+
49
+ <!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
50
+
51
+ [More Information Needed]
52
+
53
+ ### Downstream Use [optional]
54
+
55
+ <!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
56
+
57
+ [More Information Needed]
58
+
59
+ ### Out-of-Scope Use
60
+
61
+ <!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
62
+
63
+ [More Information Needed]
64
+
65
+ ## Bias, Risks, and Limitations
66
+
67
+ <!-- This section is meant to convey both technical and sociotechnical limitations. -->
68
+
69
+ [More Information Needed]
70
+
71
+ ### Recommendations
72
+
73
+ <!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
74
+
75
+ Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
76
+
77
+ ## How to Get Started with the Model
78
+
79
+ Use the code below to get started with the model.
80
+
81
+ [More Information Needed]
82
+
83
+ ## Training Details
84
+
85
+ ### Training Data
86
+
87
+ <!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
88
+
89
+ [More Information Needed]
90
+
91
+ ### Training Procedure
92
+
93
+ <!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
94
+
95
+ #### Preprocessing [optional]
96
+
97
+ [More Information Needed]
98
+
99
+
100
+ #### Training Hyperparameters
101
+
102
+ - **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
103
+
104
+ #### Speeds, Sizes, Times [optional]
105
+
106
+ <!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
107
+
108
+ [More Information Needed]
109
+
110
+ ## Evaluation
111
+
112
+ <!-- This section describes the evaluation protocols and provides the results. -->
113
+
114
+ ### Testing Data, Factors & Metrics
115
+
116
+ #### Testing Data
117
+
118
+ <!-- This should link to a Dataset Card if possible. -->
119
+
120
+ [More Information Needed]
121
+
122
+ #### Factors
123
+
124
+ <!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
125
+
126
+ [More Information Needed]
127
+
128
+ #### Metrics
129
+
130
+ <!-- These are the evaluation metrics being used, ideally with a description of why. -->
131
+
132
+ [More Information Needed]
133
+
134
+ ### Results
135
+
136
+ [More Information Needed]
137
+
138
+ #### Summary
139
+
140
+
141
+
142
+ ## Model Examination [optional]
143
+
144
+ <!-- Relevant interpretability work for the model goes here -->
145
+
146
+ [More Information Needed]
147
+
148
+ ## Environmental Impact
149
+
150
+ <!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
151
+
152
+ Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
153
+
154
+ - **Hardware Type:** [More Information Needed]
155
+ - **Hours used:** [More Information Needed]
156
+ - **Cloud Provider:** [More Information Needed]
157
+ - **Compute Region:** [More Information Needed]
158
+ - **Carbon Emitted:** [More Information Needed]
159
+
160
+ ## Technical Specifications [optional]
161
+
162
+ ### Model Architecture and Objective
163
+
164
+ [More Information Needed]
165
+
166
+ ### Compute Infrastructure
167
+
168
+ [More Information Needed]
169
+
170
+ #### Hardware
171
+
172
+ [More Information Needed]
173
+
174
+ #### Software
175
+
176
+ [More Information Needed]
177
+
178
+ ## Citation [optional]
179
+
180
+ <!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
181
+
182
+ **BibTeX:**
183
+
184
+ [More Information Needed]
185
+
186
+ **APA:**
187
+
188
+ [More Information Needed]
189
+
190
+ ## Glossary [optional]
191
+
192
+ <!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
193
+
194
+ [More Information Needed]
195
+
196
+ ## More Information [optional]
197
+
198
+ [More Information Needed]
199
+
200
+ ## Model Card Authors [optional]
201
+
202
+ [More Information Needed]
203
+
204
+ ## Model Card Contact
205
+
206
+ [More Information Needed]
207
+ ### Framework versions
208
+
209
+ - PEFT 0.19.1
substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2040/adapter_config.json ADDED
@@ -0,0 +1,48 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alora_invocation_tokens": null,
3
+ "alpha_pattern": {},
4
+ "arrow_config": null,
5
+ "auto_mapping": null,
6
+ "base_model_name_or_path": "Qwen/Qwen3.5-2B-Base",
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": 32,
20
+ "lora_bias": false,
21
+ "lora_dropout": 0.08602048083239339,
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": 16,
30
+ "rank_pattern": {},
31
+ "revision": null,
32
+ "target_modules": [
33
+ "q_proj",
34
+ "k_proj",
35
+ "gate_proj",
36
+ "up_proj",
37
+ "o_proj",
38
+ "down_proj",
39
+ "v_proj"
40
+ ],
41
+ "target_parameters": null,
42
+ "task_type": "CAUSAL_LM",
43
+ "trainable_token_indices": null,
44
+ "use_bdlora": null,
45
+ "use_dora": false,
46
+ "use_qalora": false,
47
+ "use_rslora": false
48
+ }
substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2040/chat_template.jinja ADDED
@@ -0,0 +1,154 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {%- set image_count = namespace(value=0) %}
2
+ {%- set video_count = namespace(value=0) %}
3
+ {%- macro render_content(content, do_vision_count, is_system_content=false) %}
4
+ {%- if content is string %}
5
+ {{- content }}
6
+ {%- elif content is iterable and content is not mapping %}
7
+ {%- for item in content %}
8
+ {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
9
+ {%- if is_system_content %}
10
+ {{- raise_exception('System message cannot contain images.') }}
11
+ {%- endif %}
12
+ {%- if do_vision_count %}
13
+ {%- set image_count.value = image_count.value + 1 %}
14
+ {%- endif %}
15
+ {%- if add_vision_id %}
16
+ {{- 'Picture ' ~ image_count.value ~ ': ' }}
17
+ {%- endif %}
18
+ {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
19
+ {%- elif 'video' in item or item.type == 'video' %}
20
+ {%- if is_system_content %}
21
+ {{- raise_exception('System message cannot contain videos.') }}
22
+ {%- endif %}
23
+ {%- if do_vision_count %}
24
+ {%- set video_count.value = video_count.value + 1 %}
25
+ {%- endif %}
26
+ {%- if add_vision_id %}
27
+ {{- 'Video ' ~ video_count.value ~ ': ' }}
28
+ {%- endif %}
29
+ {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
30
+ {%- elif 'text' in item %}
31
+ {{- item.text }}
32
+ {%- else %}
33
+ {{- raise_exception('Unexpected item type in content.') }}
34
+ {%- endif %}
35
+ {%- endfor %}
36
+ {%- elif content is none or content is undefined %}
37
+ {{- '' }}
38
+ {%- else %}
39
+ {{- raise_exception('Unexpected content type.') }}
40
+ {%- endif %}
41
+ {%- endmacro %}
42
+ {%- if not messages %}
43
+ {{- raise_exception('No messages provided.') }}
44
+ {%- endif %}
45
+ {%- if tools and tools is iterable and tools is not mapping %}
46
+ {{- '<|im_start|>system\n' }}
47
+ {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
48
+ {%- for tool in tools %}
49
+ {{- "\n" }}
50
+ {{- tool | tojson }}
51
+ {%- endfor %}
52
+ {{- "\n</tools>" }}
53
+ {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
54
+ {%- if messages[0].role == 'system' %}
55
+ {%- set content = render_content(messages[0].content, false, true)|trim %}
56
+ {%- if content %}
57
+ {{- '\n\n' + content }}
58
+ {%- endif %}
59
+ {%- endif %}
60
+ {{- '<|im_end|>\n' }}
61
+ {%- else %}
62
+ {%- if messages[0].role == 'system' %}
63
+ {%- set content = render_content(messages[0].content, false, true)|trim %}
64
+ {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
65
+ {%- endif %}
66
+ {%- endif %}
67
+ {%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
68
+ {%- for message in messages[::-1] %}
69
+ {%- set index = (messages|length - 1) - loop.index0 %}
70
+ {%- if ns.multi_step_tool and message.role == "user" %}
71
+ {%- set content = render_content(message.content, false)|trim %}
72
+ {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
73
+ {%- set ns.multi_step_tool = false %}
74
+ {%- set ns.last_query_index = index %}
75
+ {%- endif %}
76
+ {%- endif %}
77
+ {%- endfor %}
78
+ {%- if ns.multi_step_tool %}
79
+ {{- raise_exception('No user query found in messages.') }}
80
+ {%- endif %}
81
+ {%- for message in messages %}
82
+ {%- set content = render_content(message.content, true)|trim %}
83
+ {%- if message.role == "system" %}
84
+ {%- if not loop.first %}
85
+ {{- raise_exception('System message must be at the beginning.') }}
86
+ {%- endif %}
87
+ {%- elif message.role == "user" %}
88
+ {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
89
+ {%- elif message.role == "assistant" %}
90
+ {%- set reasoning_content = '' %}
91
+ {%- if message.reasoning_content is string %}
92
+ {%- set reasoning_content = message.reasoning_content %}
93
+ {%- else %}
94
+ {%- if '</think>' in content %}
95
+ {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
96
+ {%- set content = content.split('</think>')[-1].lstrip('\n') %}
97
+ {%- endif %}
98
+ {%- endif %}
99
+ {%- set reasoning_content = reasoning_content|trim %}
100
+ {%- if loop.index0 > ns.last_query_index %}
101
+ {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
102
+ {%- else %}
103
+ {{- '<|im_start|>' + message.role + '\n' + content }}
104
+ {%- endif %}
105
+ {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
106
+ {%- for tool_call in message.tool_calls %}
107
+ {%- if tool_call.function is defined %}
108
+ {%- set tool_call = tool_call.function %}
109
+ {%- endif %}
110
+ {%- if loop.first %}
111
+ {%- if content|trim %}
112
+ {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
113
+ {%- else %}
114
+ {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
115
+ {%- endif %}
116
+ {%- else %}
117
+ {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
118
+ {%- endif %}
119
+ {%- if tool_call.arguments is defined %}
120
+ {%- for args_name, args_value in tool_call.arguments|items %}
121
+ {{- '<parameter=' + args_name + '>\n' }}
122
+ {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
123
+ {{- args_value }}
124
+ {{- '\n</parameter>\n' }}
125
+ {%- endfor %}
126
+ {%- endif %}
127
+ {{- '</function>\n</tool_call>' }}
128
+ {%- endfor %}
129
+ {%- endif %}
130
+ {{- '<|im_end|>\n' }}
131
+ {%- elif message.role == "tool" %}
132
+ {%- if loop.previtem and loop.previtem.role != "tool" %}
133
+ {{- '<|im_start|>user' }}
134
+ {%- endif %}
135
+ {{- '\n<tool_response>\n' }}
136
+ {{- content }}
137
+ {{- '\n</tool_response>' }}
138
+ {%- if not loop.last and loop.nextitem.role != "tool" %}
139
+ {{- '<|im_end|>\n' }}
140
+ {%- elif loop.last %}
141
+ {{- '<|im_end|>\n' }}
142
+ {%- endif %}
143
+ {%- else %}
144
+ {{- raise_exception('Unexpected message role.') }}
145
+ {%- endif %}
146
+ {%- endfor %}
147
+ {%- if add_generation_prompt %}
148
+ {{- '<|im_start|>assistant\n' }}
149
+ {%- if enable_thinking is defined and enable_thinking is true %}
150
+ {{- '<think>\n' }}
151
+ {%- else %}
152
+ {{- '<think>\n\n</think>\n\n' }}
153
+ {%- endif %}
154
+ {%- endif %}
substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2040/tokenizer_config.json ADDED
@@ -0,0 +1,31 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "audio_bos_token": "<|audio_start|>",
4
+ "audio_eos_token": "<|audio_end|>",
5
+ "audio_token": "<|audio_pad|>",
6
+ "backend": "tokenizers",
7
+ "bos_token": null,
8
+ "clean_up_tokenization_spaces": false,
9
+ "eos_token": "<|endoftext|>",
10
+ "errors": "replace",
11
+ "image_token": "<|image_pad|>",
12
+ "is_local": false,
13
+ "model_max_length": 262144,
14
+ "model_specific_special_tokens": {
15
+ "audio_bos_token": "<|audio_start|>",
16
+ "audio_eos_token": "<|audio_end|>",
17
+ "audio_token": "<|audio_pad|>",
18
+ "image_token": "<|image_pad|>",
19
+ "video_token": "<|video_pad|>",
20
+ "vision_bos_token": "<|vision_start|>",
21
+ "vision_eos_token": "<|vision_end|>"
22
+ },
23
+ "pad_token": "<|endoftext|>",
24
+ "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
25
+ "split_special_tokens": false,
26
+ "tokenizer_class": "TokenizersBackend",
27
+ "unk_token": null,
28
+ "video_token": "<|video_pad|>",
29
+ "vision_bos_token": "<|vision_start|>",
30
+ "vision_eos_token": "<|vision_end|>"
31
+ }
substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2040/trainer_state.json ADDED
@@ -0,0 +1,544 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 5.0,
6
+ "eval_steps": 500,
7
+ "global_step": 2040,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "entropy": 2.2408065378665922,
14
+ "epoch": 0.12269938650306748,
15
+ "grad_norm": 1.6718658208847046,
16
+ "learning_rate": 4.487804970912036e-05,
17
+ "loss": 2.129624786376953,
18
+ "mean_token_accuracy": 0.5743644836544991,
19
+ "num_tokens": 134329.0,
20
+ "step": 50
21
+ },
22
+ {
23
+ "entropy": 1.1078504550457,
24
+ "epoch": 0.24539877300613497,
25
+ "grad_norm": 2.4091813564300537,
26
+ "learning_rate": 9.067197798373296e-05,
27
+ "loss": 1.0346390533447265,
28
+ "mean_token_accuracy": 0.7085381114482879,
29
+ "num_tokens": 262407.0,
30
+ "step": 100
31
+ },
32
+ {
33
+ "entropy": 0.9421598988771439,
34
+ "epoch": 0.36809815950920244,
35
+ "grad_norm": 1.3913646936416626,
36
+ "learning_rate": 0.00013646590625834558,
37
+ "loss": 0.8716551208496094,
38
+ "mean_token_accuracy": 0.7449040985107422,
39
+ "num_tokens": 401136.0,
40
+ "step": 150
41
+ },
42
+ {
43
+ "entropy": 0.8723550146818161,
44
+ "epoch": 0.49079754601226994,
45
+ "grad_norm": 1.1664468050003052,
46
+ "learning_rate": 0.0001822598345329582,
47
+ "loss": 0.8095977783203125,
48
+ "mean_token_accuracy": 0.7585240352153778,
49
+ "num_tokens": 531647.0,
50
+ "step": 200
51
+ },
52
+ {
53
+ "entropy": 0.8466519421339035,
54
+ "epoch": 0.6134969325153374,
55
+ "grad_norm": 1.0145421028137207,
56
+ "learning_rate": 0.00022805376280757083,
57
+ "loss": 0.7830857849121093,
58
+ "mean_token_accuracy": 0.7661705195903779,
59
+ "num_tokens": 666127.0,
60
+ "step": 250
61
+ },
62
+ {
63
+ "entropy": 0.8180295366048813,
64
+ "epoch": 0.7361963190184049,
65
+ "grad_norm": 0.9731705784797668,
66
+ "learning_rate": 0.00027384769108218336,
67
+ "loss": 0.7573603820800782,
68
+ "mean_token_accuracy": 0.7704453605413437,
69
+ "num_tokens": 799292.0,
70
+ "step": 300
71
+ },
72
+ {
73
+ "entropy": 0.7990227991342544,
74
+ "epoch": 0.8588957055214724,
75
+ "grad_norm": 1.0642374753952026,
76
+ "learning_rate": 0.000319641619356796,
77
+ "loss": 0.73709228515625,
78
+ "mean_token_accuracy": 0.7752733880281448,
79
+ "num_tokens": 930614.0,
80
+ "step": 350
81
+ },
82
+ {
83
+ "entropy": 0.7899087685346603,
84
+ "epoch": 0.9815950920245399,
85
+ "grad_norm": 0.880613386631012,
86
+ "learning_rate": 0.0003654355476314086,
87
+ "loss": 0.7347319793701171,
88
+ "mean_token_accuracy": 0.7766731631755829,
89
+ "num_tokens": 1061654.0,
90
+ "step": 400
91
+ },
92
+ {
93
+ "epoch": 1.0,
94
+ "eval_entropy": 0.83522472347532,
95
+ "eval_mean_token_accuracy": 0.7484098076820374,
96
+ "eval_not_syn_loss": 0.7872886061668396,
97
+ "eval_not_syn_runtime": 54.8697,
98
+ "eval_not_syn_samples_per_second": 25.424,
99
+ "eval_not_syn_steps_per_second": 3.189,
100
+ "eval_num_tokens": 1080196.0,
101
+ "step": 408
102
+ },
103
+ {
104
+ "epoch": 1.0,
105
+ "eval_entropy": 0.7784442271505083,
106
+ "eval_mean_token_accuracy": 0.8021791185651507,
107
+ "eval_num_tokens": 1080196.0,
108
+ "eval_syn_loss": 0.6980098485946655,
109
+ "eval_syn_runtime": 56.6342,
110
+ "eval_syn_samples_per_second": 24.632,
111
+ "eval_syn_steps_per_second": 3.09,
112
+ "step": 408
113
+ },
114
+ {
115
+ "entropy": 0.75672573694075,
116
+ "epoch": 1.1030674846625768,
117
+ "grad_norm": 0.9244198203086853,
118
+ "learning_rate": 0.00037356351883456205,
119
+ "loss": 0.7048745727539063,
120
+ "mean_token_accuracy": 0.783016896609104,
121
+ "num_tokens": 1199170.0,
122
+ "step": 450
123
+ },
124
+ {
125
+ "entropy": 0.7406797379255294,
126
+ "epoch": 1.2257668711656442,
127
+ "grad_norm": 0.6744896769523621,
128
+ "learning_rate": 0.00037311248151765587,
129
+ "loss": 0.6860730743408203,
130
+ "mean_token_accuracy": 0.787369327545166,
131
+ "num_tokens": 1334298.0,
132
+ "step": 500
133
+ },
134
+ {
135
+ "entropy": 0.7424973237514496,
136
+ "epoch": 1.3484662576687116,
137
+ "grad_norm": 0.92305588722229,
138
+ "learning_rate": 0.000372320629215228,
139
+ "loss": 0.6855724334716797,
140
+ "mean_token_accuracy": 0.7871134179830551,
141
+ "num_tokens": 1463189.0,
142
+ "step": 550
143
+ },
144
+ {
145
+ "entropy": 0.7502243828773498,
146
+ "epoch": 1.471165644171779,
147
+ "grad_norm": 0.8259047269821167,
148
+ "learning_rate": 0.00037118941074037944,
149
+ "loss": 0.6949834442138672,
150
+ "mean_token_accuracy": 0.7851914083957672,
151
+ "num_tokens": 1590310.0,
152
+ "step": 600
153
+ },
154
+ {
155
+ "entropy": 0.7233260518312454,
156
+ "epoch": 1.5938650306748468,
157
+ "grad_norm": 0.882722795009613,
158
+ "learning_rate": 0.00036972089582775814,
159
+ "loss": 0.6649005889892579,
160
+ "mean_token_accuracy": 0.7898879665136337,
161
+ "num_tokens": 1721836.0,
162
+ "step": 650
163
+ },
164
+ {
165
+ "entropy": 0.7162708967924118,
166
+ "epoch": 1.716564417177914,
167
+ "grad_norm": 0.6593027114868164,
168
+ "learning_rate": 0.0003679177713466678,
169
+ "loss": 0.6653710174560546,
170
+ "mean_token_accuracy": 0.7905678844451904,
171
+ "num_tokens": 1853507.0,
172
+ "step": 700
173
+ },
174
+ {
175
+ "entropy": 0.7150520020723343,
176
+ "epoch": 1.8392638036809816,
177
+ "grad_norm": 0.665340781211853,
178
+ "learning_rate": 0.0003657833363850354,
179
+ "loss": 0.6600938415527344,
180
+ "mean_token_accuracy": 0.7918649202585221,
181
+ "num_tokens": 1988113.0,
182
+ "step": 750
183
+ },
184
+ {
185
+ "entropy": 0.6993386310338974,
186
+ "epoch": 1.961963190184049,
187
+ "grad_norm": 0.6502187252044678,
188
+ "learning_rate": 0.00036332149621323294,
189
+ "loss": 0.6469607543945313,
190
+ "mean_token_accuracy": 0.7950778317451477,
191
+ "num_tokens": 2121314.0,
192
+ "step": 800
193
+ },
194
+ {
195
+ "epoch": 2.0,
196
+ "eval_entropy": 0.7315139893123082,
197
+ "eval_mean_token_accuracy": 0.7713504004478454,
198
+ "eval_not_syn_loss": 0.699893593788147,
199
+ "eval_not_syn_runtime": 55.4135,
200
+ "eval_not_syn_samples_per_second": 25.174,
201
+ "eval_not_syn_steps_per_second": 3.158,
202
+ "eval_num_tokens": 2160392.0,
203
+ "step": 816
204
+ },
205
+ {
206
+ "epoch": 2.0,
207
+ "eval_entropy": 0.6848451798302787,
208
+ "eval_mean_token_accuracy": 0.814895657471248,
209
+ "eval_num_tokens": 2160392.0,
210
+ "eval_syn_loss": 0.6415576934814453,
211
+ "eval_syn_runtime": 56.4735,
212
+ "eval_syn_samples_per_second": 24.702,
213
+ "eval_syn_steps_per_second": 3.099,
214
+ "step": 816
215
+ },
216
+ {
217
+ "entropy": 0.6596363644407253,
218
+ "epoch": 2.083435582822086,
219
+ "grad_norm": 0.6647700071334839,
220
+ "learning_rate": 0.00036053675513879666,
221
+ "loss": 0.599151611328125,
222
+ "mean_token_accuracy": 0.8071986864311527,
223
+ "num_tokens": 2253179.0,
224
+ "step": 850
225
+ },
226
+ {
227
+ "entropy": 0.6626429110765457,
228
+ "epoch": 2.2061349693251535,
229
+ "grad_norm": 0.7356868386268616,
230
+ "learning_rate": 0.00035743420826511783,
231
+ "loss": 0.6004195404052735,
232
+ "mean_token_accuracy": 0.805362731218338,
233
+ "num_tokens": 2379965.0,
234
+ "step": 900
235
+ },
236
+ {
237
+ "entropy": 0.6503470712900161,
238
+ "epoch": 2.3288343558282207,
239
+ "grad_norm": 1.095608115196228,
240
+ "learning_rate": 0.0003540195321691835,
241
+ "loss": 0.5951333999633789,
242
+ "mean_token_accuracy": 0.807324025630951,
243
+ "num_tokens": 2515593.0,
244
+ "step": 950
245
+ },
246
+ {
247
+ "entropy": 0.660577870607376,
248
+ "epoch": 2.4515337423312884,
249
+ "grad_norm": 0.6602179408073425,
250
+ "learning_rate": 0.00035029897451542325,
251
+ "loss": 0.6056365585327148,
252
+ "mean_token_accuracy": 0.8057438349723816,
253
+ "num_tokens": 2642790.0,
254
+ "step": 1000
255
+ },
256
+ {
257
+ "entropy": 0.6555853113532066,
258
+ "epoch": 2.574233128834356,
259
+ "grad_norm": 0.7459151148796082,
260
+ "learning_rate": 0.00034627934262466606,
261
+ "loss": 0.5963204193115235,
262
+ "mean_token_accuracy": 0.8060598099231719,
263
+ "num_tokens": 2774396.0,
264
+ "step": 1050
265
+ },
266
+ {
267
+ "entropy": 0.6505853188037872,
268
+ "epoch": 2.6969325153374233,
269
+ "grad_norm": 0.6359792947769165,
270
+ "learning_rate": 0.00034196799101912103,
271
+ "loss": 0.5993065643310547,
272
+ "mean_token_accuracy": 0.806280387043953,
273
+ "num_tokens": 2912102.0,
274
+ "step": 1100
275
+ },
276
+ {
277
+ "entropy": 0.6509970253705979,
278
+ "epoch": 2.819631901840491,
279
+ "grad_norm": 0.8086174726486206,
280
+ "learning_rate": 0.0003373728079661707,
281
+ "loss": 0.6001268005371094,
282
+ "mean_token_accuracy": 0.806866540312767,
283
+ "num_tokens": 3042987.0,
284
+ "step": 1150
285
+ },
286
+ {
287
+ "entropy": 0.6404844307899475,
288
+ "epoch": 2.942331288343558,
289
+ "grad_norm": 0.7038407325744629,
290
+ "learning_rate": 0.0003325022010455975,
291
+ "loss": 0.5870526123046875,
292
+ "mean_token_accuracy": 0.8104586935043335,
293
+ "num_tokens": 3178044.0,
294
+ "step": 1200
295
+ },
296
+ {
297
+ "epoch": 3.0,
298
+ "eval_entropy": 0.6645996727262224,
299
+ "eval_mean_token_accuracy": 0.798654237134116,
300
+ "eval_not_syn_loss": 0.6586771011352539,
301
+ "eval_not_syn_runtime": 55.1088,
302
+ "eval_not_syn_samples_per_second": 25.314,
303
+ "eval_not_syn_steps_per_second": 3.176,
304
+ "eval_num_tokens": 3240588.0,
305
+ "step": 1224
306
+ },
307
+ {
308
+ "epoch": 3.0,
309
+ "eval_entropy": 0.6233607809884207,
310
+ "eval_mean_token_accuracy": 0.7977249867575509,
311
+ "eval_num_tokens": 3240588.0,
312
+ "eval_syn_loss": 0.6288298964500427,
313
+ "eval_syn_runtime": 56.6081,
314
+ "eval_syn_samples_per_second": 24.643,
315
+ "eval_syn_steps_per_second": 3.091,
316
+ "step": 1224
317
+ },
318
+ {
319
+ "entropy": 0.6081961104364106,
320
+ "epoch": 3.063803680981595,
321
+ "grad_norm": 0.6161203980445862,
322
+ "learning_rate": 0.0003273650817666497,
323
+ "loss": 0.5481716537475586,
324
+ "mean_token_accuracy": 0.8175542035488167,
325
+ "num_tokens": 3312797.0,
326
+ "step": 1250
327
+ },
328
+ {
329
+ "entropy": 0.5877064895629883,
330
+ "epoch": 3.1865030674846624,
331
+ "grad_norm": 0.6909636855125427,
332
+ "learning_rate": 0.00032197084926309234,
333
+ "loss": 0.5224150466918945,
334
+ "mean_token_accuracy": 0.8250984263420105,
335
+ "num_tokens": 3441629.0,
336
+ "step": 1300
337
+ },
338
+ {
339
+ "entropy": 0.5952656552195549,
340
+ "epoch": 3.30920245398773,
341
+ "grad_norm": 0.8147026300430298,
342
+ "learning_rate": 0.0003163293730960758,
343
+ "loss": 0.5349474334716797,
344
+ "mean_token_accuracy": 0.8213064837455749,
345
+ "num_tokens": 3576093.0,
346
+ "step": 1350
347
+ },
348
+ {
349
+ "entropy": 0.5937123569846153,
350
+ "epoch": 3.4319018404907977,
351
+ "grad_norm": 0.8690970540046692,
352
+ "learning_rate": 0.0003104509751962869,
353
+ "loss": 0.5310481262207031,
354
+ "mean_token_accuracy": 0.8229135835170746,
355
+ "num_tokens": 3705912.0,
356
+ "step": 1400
357
+ },
358
+ {
359
+ "entropy": 0.6067730465531349,
360
+ "epoch": 3.554601226993865,
361
+ "grad_norm": 0.7433330416679382,
362
+ "learning_rate": 0.0003043464109784207,
363
+ "loss": 0.5450045394897461,
364
+ "mean_token_accuracy": 0.8205251175165177,
365
+ "num_tokens": 3832376.0,
366
+ "step": 1450
367
+ },
368
+ {
369
+ "entropy": 0.5926914718747139,
370
+ "epoch": 3.6773006134969326,
371
+ "grad_norm": 0.7265587449073792,
372
+ "learning_rate": 0.00029802684966252974,
373
+ "loss": 0.5330168914794922,
374
+ "mean_token_accuracy": 0.821294127702713,
375
+ "num_tokens": 3967273.0,
376
+ "step": 1500
377
+ },
378
+ {
379
+ "entropy": 0.5881199997663498,
380
+ "epoch": 3.8,
381
+ "grad_norm": 0.7681817412376404,
382
+ "learning_rate": 0.00029150385383825187,
383
+ "loss": 0.5288929367065429,
384
+ "mean_token_accuracy": 0.8234161508083343,
385
+ "num_tokens": 4102434.0,
386
+ "step": 1550
387
+ },
388
+ {
389
+ "entropy": 0.6008686199784279,
390
+ "epoch": 3.9226993865030675,
391
+ "grad_norm": 0.9384056329727173,
392
+ "learning_rate": 0.0002847893583093113,
393
+ "loss": 0.5437511444091797,
394
+ "mean_token_accuracy": 0.8201366275548935,
395
+ "num_tokens": 4234037.0,
396
+ "step": 1600
397
+ },
398
+ {
399
+ "epoch": 4.0,
400
+ "eval_entropy": 0.6218148783275059,
401
+ "eval_mean_token_accuracy": 0.783796865599496,
402
+ "eval_not_syn_loss": 0.6558170318603516,
403
+ "eval_not_syn_runtime": 55.429,
404
+ "eval_not_syn_samples_per_second": 25.167,
405
+ "eval_not_syn_steps_per_second": 3.157,
406
+ "eval_num_tokens": 4320784.0,
407
+ "step": 1632
408
+ },
409
+ {
410
+ "epoch": 4.0,
411
+ "eval_entropy": 0.5833754854542869,
412
+ "eval_mean_token_accuracy": 0.8278223422595433,
413
+ "eval_num_tokens": 4320784.0,
414
+ "eval_syn_loss": 0.604770302772522,
415
+ "eval_syn_runtime": 56.9593,
416
+ "eval_syn_samples_per_second": 24.491,
417
+ "eval_syn_steps_per_second": 3.072,
418
+ "step": 1632
419
+ },
420
+ {
421
+ "entropy": 0.559690699733869,
422
+ "epoch": 4.044171779141104,
423
+ "grad_norm": 0.65739905834198,
424
+ "learning_rate": 0.0002778956482569963,
425
+ "loss": 0.49702674865722657,
426
+ "mean_token_accuracy": 0.8317034166268628,
427
+ "num_tokens": 4368268.0,
428
+ "step": 1650
429
+ },
430
+ {
431
+ "entropy": 0.5401161375641823,
432
+ "epoch": 4.166871165644172,
433
+ "grad_norm": 1.5176353454589844,
434
+ "learning_rate": 0.0002708353367625691,
435
+ "loss": 0.4598746109008789,
436
+ "mean_token_accuracy": 0.8412033343315124,
437
+ "num_tokens": 4494525.0,
438
+ "step": 1700
439
+ },
440
+ {
441
+ "entropy": 0.5299916860461235,
442
+ "epoch": 4.289570552147239,
443
+ "grad_norm": 0.685965359210968,
444
+ "learning_rate": 0.0002636213417297333,
445
+ "loss": 0.45663665771484374,
446
+ "mean_token_accuracy": 0.8412413811683654,
447
+ "num_tokens": 4631985.0,
448
+ "step": 1750
449
+ },
450
+ {
451
+ "entropy": 0.5352088099718094,
452
+ "epoch": 4.412269938650307,
453
+ "grad_norm": 0.8471158742904663,
454
+ "learning_rate": 0.0002562668622493824,
455
+ "loss": 0.4625707626342773,
456
+ "mean_token_accuracy": 0.8400991547107697,
457
+ "num_tokens": 4766323.0,
458
+ "step": 1800
459
+ },
460
+ {
461
+ "entropy": 0.5425067806243896,
462
+ "epoch": 4.534969325153375,
463
+ "grad_norm": 0.8595437407493591,
464
+ "learning_rate": 0.00024878535444987474,
465
+ "loss": 0.47124954223632814,
466
+ "mean_token_accuracy": 0.8376108628511428,
467
+ "num_tokens": 4895606.0,
468
+ "step": 1850
469
+ },
470
+ {
471
+ "entropy": 0.5419312778115273,
472
+ "epoch": 4.6576687116564415,
473
+ "grad_norm": 0.8436959981918335,
474
+ "learning_rate": 0.00024119050687701818,
475
+ "loss": 0.4708108901977539,
476
+ "mean_token_accuracy": 0.8390797019004822,
477
+ "num_tokens": 5025912.0,
478
+ "step": 1900
479
+ },
480
+ {
481
+ "entropy": 0.5392271488904953,
482
+ "epoch": 4.780368098159509,
483
+ "grad_norm": 0.8691944479942322,
484
+ "learning_rate": 0.00023349621544881404,
485
+ "loss": 0.46682029724121094,
486
+ "mean_token_accuracy": 0.8385172665119172,
487
+ "num_tokens": 5162972.0,
488
+ "step": 1950
489
+ },
490
+ {
491
+ "entropy": 0.5332251599431038,
492
+ "epoch": 4.903067484662577,
493
+ "grad_norm": 0.6814598441123962,
494
+ "learning_rate": 0.00022571655803078078,
495
+ "loss": 0.4687493896484375,
496
+ "mean_token_accuracy": 0.8388861840963364,
497
+ "num_tokens": 5297935.0,
498
+ "step": 2000
499
+ },
500
+ {
501
+ "epoch": 5.0,
502
+ "eval_entropy": 0.5847777877535139,
503
+ "eval_mean_token_accuracy": 0.7782382583618164,
504
+ "eval_not_syn_loss": 0.6678160429000854,
505
+ "eval_not_syn_runtime": 55.1467,
506
+ "eval_not_syn_samples_per_second": 25.296,
507
+ "eval_not_syn_steps_per_second": 3.173,
508
+ "eval_num_tokens": 5400980.0,
509
+ "step": 2040
510
+ },
511
+ {
512
+ "epoch": 5.0,
513
+ "eval_entropy": 0.5505962218557086,
514
+ "eval_mean_token_accuracy": 0.8308270008223397,
515
+ "eval_num_tokens": 5400980.0,
516
+ "eval_syn_loss": 0.606867253780365,
517
+ "eval_syn_runtime": 56.9768,
518
+ "eval_syn_samples_per_second": 24.484,
519
+ "eval_syn_steps_per_second": 3.071,
520
+ "step": 2040
521
+ }
522
+ ],
523
+ "logging_steps": 50,
524
+ "max_steps": 4080,
525
+ "num_input_tokens_seen": 0,
526
+ "num_train_epochs": 10,
527
+ "save_steps": 500,
528
+ "stateful_callbacks": {
529
+ "TrainerControl": {
530
+ "args": {
531
+ "should_epoch_stop": false,
532
+ "should_evaluate": false,
533
+ "should_log": false,
534
+ "should_save": true,
535
+ "should_training_stop": false
536
+ },
537
+ "attributes": {}
538
+ }
539
+ },
540
+ "total_flos": 1.0759747166537664e+17,
541
+ "train_batch_size": 8,
542
+ "trial_name": null,
543
+ "trial_params": null
544
+ }
substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2448/README.md ADDED
@@ -0,0 +1,209 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: Qwen/Qwen3.5-2B-Base
3
+ library_name: peft
4
+ pipeline_tag: text-generation
5
+ tags:
6
+ - base_model:adapter:Qwen/Qwen3.5-2B-Base
7
+ - lora
8
+ - sft
9
+ - transformers
10
+ - trl
11
+ ---
12
+
13
+ # Model Card for Model ID
14
+
15
+ <!-- Provide a quick summary of what the model is/does. -->
16
+
17
+
18
+
19
+ ## Model Details
20
+
21
+ ### Model Description
22
+
23
+ <!-- Provide a longer summary of what this model is. -->
24
+
25
+
26
+
27
+ - **Developed by:** [More Information Needed]
28
+ - **Funded by [optional]:** [More Information Needed]
29
+ - **Shared by [optional]:** [More Information Needed]
30
+ - **Model type:** [More Information Needed]
31
+ - **Language(s) (NLP):** [More Information Needed]
32
+ - **License:** [More Information Needed]
33
+ - **Finetuned from model [optional]:** [More Information Needed]
34
+
35
+ ### Model Sources [optional]
36
+
37
+ <!-- Provide the basic links for the model. -->
38
+
39
+ - **Repository:** [More Information Needed]
40
+ - **Paper [optional]:** [More Information Needed]
41
+ - **Demo [optional]:** [More Information Needed]
42
+
43
+ ## Uses
44
+
45
+ <!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
46
+
47
+ ### Direct Use
48
+
49
+ <!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
50
+
51
+ [More Information Needed]
52
+
53
+ ### Downstream Use [optional]
54
+
55
+ <!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
56
+
57
+ [More Information Needed]
58
+
59
+ ### Out-of-Scope Use
60
+
61
+ <!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
62
+
63
+ [More Information Needed]
64
+
65
+ ## Bias, Risks, and Limitations
66
+
67
+ <!-- This section is meant to convey both technical and sociotechnical limitations. -->
68
+
69
+ [More Information Needed]
70
+
71
+ ### Recommendations
72
+
73
+ <!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
74
+
75
+ Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
76
+
77
+ ## How to Get Started with the Model
78
+
79
+ Use the code below to get started with the model.
80
+
81
+ [More Information Needed]
82
+
83
+ ## Training Details
84
+
85
+ ### Training Data
86
+
87
+ <!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
88
+
89
+ [More Information Needed]
90
+
91
+ ### Training Procedure
92
+
93
+ <!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
94
+
95
+ #### Preprocessing [optional]
96
+
97
+ [More Information Needed]
98
+
99
+
100
+ #### Training Hyperparameters
101
+
102
+ - **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
103
+
104
+ #### Speeds, Sizes, Times [optional]
105
+
106
+ <!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
107
+
108
+ [More Information Needed]
109
+
110
+ ## Evaluation
111
+
112
+ <!-- This section describes the evaluation protocols and provides the results. -->
113
+
114
+ ### Testing Data, Factors & Metrics
115
+
116
+ #### Testing Data
117
+
118
+ <!-- This should link to a Dataset Card if possible. -->
119
+
120
+ [More Information Needed]
121
+
122
+ #### Factors
123
+
124
+ <!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
125
+
126
+ [More Information Needed]
127
+
128
+ #### Metrics
129
+
130
+ <!-- These are the evaluation metrics being used, ideally with a description of why. -->
131
+
132
+ [More Information Needed]
133
+
134
+ ### Results
135
+
136
+ [More Information Needed]
137
+
138
+ #### Summary
139
+
140
+
141
+
142
+ ## Model Examination [optional]
143
+
144
+ <!-- Relevant interpretability work for the model goes here -->
145
+
146
+ [More Information Needed]
147
+
148
+ ## Environmental Impact
149
+
150
+ <!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
151
+
152
+ Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
153
+
154
+ - **Hardware Type:** [More Information Needed]
155
+ - **Hours used:** [More Information Needed]
156
+ - **Cloud Provider:** [More Information Needed]
157
+ - **Compute Region:** [More Information Needed]
158
+ - **Carbon Emitted:** [More Information Needed]
159
+
160
+ ## Technical Specifications [optional]
161
+
162
+ ### Model Architecture and Objective
163
+
164
+ [More Information Needed]
165
+
166
+ ### Compute Infrastructure
167
+
168
+ [More Information Needed]
169
+
170
+ #### Hardware
171
+
172
+ [More Information Needed]
173
+
174
+ #### Software
175
+
176
+ [More Information Needed]
177
+
178
+ ## Citation [optional]
179
+
180
+ <!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
181
+
182
+ **BibTeX:**
183
+
184
+ [More Information Needed]
185
+
186
+ **APA:**
187
+
188
+ [More Information Needed]
189
+
190
+ ## Glossary [optional]
191
+
192
+ <!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
193
+
194
+ [More Information Needed]
195
+
196
+ ## More Information [optional]
197
+
198
+ [More Information Needed]
199
+
200
+ ## Model Card Authors [optional]
201
+
202
+ [More Information Needed]
203
+
204
+ ## Model Card Contact
205
+
206
+ [More Information Needed]
207
+ ### Framework versions
208
+
209
+ - PEFT 0.19.1
substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2448/adapter_config.json ADDED
@@ -0,0 +1,48 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alora_invocation_tokens": null,
3
+ "alpha_pattern": {},
4
+ "arrow_config": null,
5
+ "auto_mapping": null,
6
+ "base_model_name_or_path": "Qwen/Qwen3.5-2B-Base",
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": 32,
20
+ "lora_bias": false,
21
+ "lora_dropout": 0.08602048083239339,
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": 16,
30
+ "rank_pattern": {},
31
+ "revision": null,
32
+ "target_modules": [
33
+ "q_proj",
34
+ "k_proj",
35
+ "gate_proj",
36
+ "up_proj",
37
+ "o_proj",
38
+ "down_proj",
39
+ "v_proj"
40
+ ],
41
+ "target_parameters": null,
42
+ "task_type": "CAUSAL_LM",
43
+ "trainable_token_indices": null,
44
+ "use_bdlora": null,
45
+ "use_dora": false,
46
+ "use_qalora": false,
47
+ "use_rslora": false
48
+ }
substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2448/chat_template.jinja ADDED
@@ -0,0 +1,154 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {%- set image_count = namespace(value=0) %}
2
+ {%- set video_count = namespace(value=0) %}
3
+ {%- macro render_content(content, do_vision_count, is_system_content=false) %}
4
+ {%- if content is string %}
5
+ {{- content }}
6
+ {%- elif content is iterable and content is not mapping %}
7
+ {%- for item in content %}
8
+ {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
9
+ {%- if is_system_content %}
10
+ {{- raise_exception('System message cannot contain images.') }}
11
+ {%- endif %}
12
+ {%- if do_vision_count %}
13
+ {%- set image_count.value = image_count.value + 1 %}
14
+ {%- endif %}
15
+ {%- if add_vision_id %}
16
+ {{- 'Picture ' ~ image_count.value ~ ': ' }}
17
+ {%- endif %}
18
+ {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
19
+ {%- elif 'video' in item or item.type == 'video' %}
20
+ {%- if is_system_content %}
21
+ {{- raise_exception('System message cannot contain videos.') }}
22
+ {%- endif %}
23
+ {%- if do_vision_count %}
24
+ {%- set video_count.value = video_count.value + 1 %}
25
+ {%- endif %}
26
+ {%- if add_vision_id %}
27
+ {{- 'Video ' ~ video_count.value ~ ': ' }}
28
+ {%- endif %}
29
+ {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
30
+ {%- elif 'text' in item %}
31
+ {{- item.text }}
32
+ {%- else %}
33
+ {{- raise_exception('Unexpected item type in content.') }}
34
+ {%- endif %}
35
+ {%- endfor %}
36
+ {%- elif content is none or content is undefined %}
37
+ {{- '' }}
38
+ {%- else %}
39
+ {{- raise_exception('Unexpected content type.') }}
40
+ {%- endif %}
41
+ {%- endmacro %}
42
+ {%- if not messages %}
43
+ {{- raise_exception('No messages provided.') }}
44
+ {%- endif %}
45
+ {%- if tools and tools is iterable and tools is not mapping %}
46
+ {{- '<|im_start|>system\n' }}
47
+ {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
48
+ {%- for tool in tools %}
49
+ {{- "\n" }}
50
+ {{- tool | tojson }}
51
+ {%- endfor %}
52
+ {{- "\n</tools>" }}
53
+ {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
54
+ {%- if messages[0].role == 'system' %}
55
+ {%- set content = render_content(messages[0].content, false, true)|trim %}
56
+ {%- if content %}
57
+ {{- '\n\n' + content }}
58
+ {%- endif %}
59
+ {%- endif %}
60
+ {{- '<|im_end|>\n' }}
61
+ {%- else %}
62
+ {%- if messages[0].role == 'system' %}
63
+ {%- set content = render_content(messages[0].content, false, true)|trim %}
64
+ {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
65
+ {%- endif %}
66
+ {%- endif %}
67
+ {%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
68
+ {%- for message in messages[::-1] %}
69
+ {%- set index = (messages|length - 1) - loop.index0 %}
70
+ {%- if ns.multi_step_tool and message.role == "user" %}
71
+ {%- set content = render_content(message.content, false)|trim %}
72
+ {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
73
+ {%- set ns.multi_step_tool = false %}
74
+ {%- set ns.last_query_index = index %}
75
+ {%- endif %}
76
+ {%- endif %}
77
+ {%- endfor %}
78
+ {%- if ns.multi_step_tool %}
79
+ {{- raise_exception('No user query found in messages.') }}
80
+ {%- endif %}
81
+ {%- for message in messages %}
82
+ {%- set content = render_content(message.content, true)|trim %}
83
+ {%- if message.role == "system" %}
84
+ {%- if not loop.first %}
85
+ {{- raise_exception('System message must be at the beginning.') }}
86
+ {%- endif %}
87
+ {%- elif message.role == "user" %}
88
+ {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
89
+ {%- elif message.role == "assistant" %}
90
+ {%- set reasoning_content = '' %}
91
+ {%- if message.reasoning_content is string %}
92
+ {%- set reasoning_content = message.reasoning_content %}
93
+ {%- else %}
94
+ {%- if '</think>' in content %}
95
+ {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
96
+ {%- set content = content.split('</think>')[-1].lstrip('\n') %}
97
+ {%- endif %}
98
+ {%- endif %}
99
+ {%- set reasoning_content = reasoning_content|trim %}
100
+ {%- if loop.index0 > ns.last_query_index %}
101
+ {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
102
+ {%- else %}
103
+ {{- '<|im_start|>' + message.role + '\n' + content }}
104
+ {%- endif %}
105
+ {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
106
+ {%- for tool_call in message.tool_calls %}
107
+ {%- if tool_call.function is defined %}
108
+ {%- set tool_call = tool_call.function %}
109
+ {%- endif %}
110
+ {%- if loop.first %}
111
+ {%- if content|trim %}
112
+ {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
113
+ {%- else %}
114
+ {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
115
+ {%- endif %}
116
+ {%- else %}
117
+ {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
118
+ {%- endif %}
119
+ {%- if tool_call.arguments is defined %}
120
+ {%- for args_name, args_value in tool_call.arguments|items %}
121
+ {{- '<parameter=' + args_name + '>\n' }}
122
+ {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
123
+ {{- args_value }}
124
+ {{- '\n</parameter>\n' }}
125
+ {%- endfor %}
126
+ {%- endif %}
127
+ {{- '</function>\n</tool_call>' }}
128
+ {%- endfor %}
129
+ {%- endif %}
130
+ {{- '<|im_end|>\n' }}
131
+ {%- elif message.role == "tool" %}
132
+ {%- if loop.previtem and loop.previtem.role != "tool" %}
133
+ {{- '<|im_start|>user' }}
134
+ {%- endif %}
135
+ {{- '\n<tool_response>\n' }}
136
+ {{- content }}
137
+ {{- '\n</tool_response>' }}
138
+ {%- if not loop.last and loop.nextitem.role != "tool" %}
139
+ {{- '<|im_end|>\n' }}
140
+ {%- elif loop.last %}
141
+ {{- '<|im_end|>\n' }}
142
+ {%- endif %}
143
+ {%- else %}
144
+ {{- raise_exception('Unexpected message role.') }}
145
+ {%- endif %}
146
+ {%- endfor %}
147
+ {%- if add_generation_prompt %}
148
+ {{- '<|im_start|>assistant\n' }}
149
+ {%- if enable_thinking is defined and enable_thinking is true %}
150
+ {{- '<think>\n' }}
151
+ {%- else %}
152
+ {{- '<think>\n\n</think>\n\n' }}
153
+ {%- endif %}
154
+ {%- endif %}
substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2448/tokenizer_config.json ADDED
@@ -0,0 +1,31 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "audio_bos_token": "<|audio_start|>",
4
+ "audio_eos_token": "<|audio_end|>",
5
+ "audio_token": "<|audio_pad|>",
6
+ "backend": "tokenizers",
7
+ "bos_token": null,
8
+ "clean_up_tokenization_spaces": false,
9
+ "eos_token": "<|endoftext|>",
10
+ "errors": "replace",
11
+ "image_token": "<|image_pad|>",
12
+ "is_local": false,
13
+ "model_max_length": 262144,
14
+ "model_specific_special_tokens": {
15
+ "audio_bos_token": "<|audio_start|>",
16
+ "audio_eos_token": "<|audio_end|>",
17
+ "audio_token": "<|audio_pad|>",
18
+ "image_token": "<|image_pad|>",
19
+ "video_token": "<|video_pad|>",
20
+ "vision_bos_token": "<|vision_start|>",
21
+ "vision_eos_token": "<|vision_end|>"
22
+ },
23
+ "pad_token": "<|endoftext|>",
24
+ "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
25
+ "split_special_tokens": false,
26
+ "tokenizer_class": "TokenizersBackend",
27
+ "unk_token": null,
28
+ "video_token": "<|video_pad|>",
29
+ "vision_bos_token": "<|vision_start|>",
30
+ "vision_eos_token": "<|vision_end|>"
31
+ }
substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2448/trainer_state.json ADDED
@@ -0,0 +1,646 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 6.0,
6
+ "eval_steps": 500,
7
+ "global_step": 2448,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "entropy": 2.2408065378665922,
14
+ "epoch": 0.12269938650306748,
15
+ "grad_norm": 1.6718658208847046,
16
+ "learning_rate": 4.487804970912036e-05,
17
+ "loss": 2.129624786376953,
18
+ "mean_token_accuracy": 0.5743644836544991,
19
+ "num_tokens": 134329.0,
20
+ "step": 50
21
+ },
22
+ {
23
+ "entropy": 1.1078504550457,
24
+ "epoch": 0.24539877300613497,
25
+ "grad_norm": 2.4091813564300537,
26
+ "learning_rate": 9.067197798373296e-05,
27
+ "loss": 1.0346390533447265,
28
+ "mean_token_accuracy": 0.7085381114482879,
29
+ "num_tokens": 262407.0,
30
+ "step": 100
31
+ },
32
+ {
33
+ "entropy": 0.9421598988771439,
34
+ "epoch": 0.36809815950920244,
35
+ "grad_norm": 1.3913646936416626,
36
+ "learning_rate": 0.00013646590625834558,
37
+ "loss": 0.8716551208496094,
38
+ "mean_token_accuracy": 0.7449040985107422,
39
+ "num_tokens": 401136.0,
40
+ "step": 150
41
+ },
42
+ {
43
+ "entropy": 0.8723550146818161,
44
+ "epoch": 0.49079754601226994,
45
+ "grad_norm": 1.1664468050003052,
46
+ "learning_rate": 0.0001822598345329582,
47
+ "loss": 0.8095977783203125,
48
+ "mean_token_accuracy": 0.7585240352153778,
49
+ "num_tokens": 531647.0,
50
+ "step": 200
51
+ },
52
+ {
53
+ "entropy": 0.8466519421339035,
54
+ "epoch": 0.6134969325153374,
55
+ "grad_norm": 1.0145421028137207,
56
+ "learning_rate": 0.00022805376280757083,
57
+ "loss": 0.7830857849121093,
58
+ "mean_token_accuracy": 0.7661705195903779,
59
+ "num_tokens": 666127.0,
60
+ "step": 250
61
+ },
62
+ {
63
+ "entropy": 0.8180295366048813,
64
+ "epoch": 0.7361963190184049,
65
+ "grad_norm": 0.9731705784797668,
66
+ "learning_rate": 0.00027384769108218336,
67
+ "loss": 0.7573603820800782,
68
+ "mean_token_accuracy": 0.7704453605413437,
69
+ "num_tokens": 799292.0,
70
+ "step": 300
71
+ },
72
+ {
73
+ "entropy": 0.7990227991342544,
74
+ "epoch": 0.8588957055214724,
75
+ "grad_norm": 1.0642374753952026,
76
+ "learning_rate": 0.000319641619356796,
77
+ "loss": 0.73709228515625,
78
+ "mean_token_accuracy": 0.7752733880281448,
79
+ "num_tokens": 930614.0,
80
+ "step": 350
81
+ },
82
+ {
83
+ "entropy": 0.7899087685346603,
84
+ "epoch": 0.9815950920245399,
85
+ "grad_norm": 0.880613386631012,
86
+ "learning_rate": 0.0003654355476314086,
87
+ "loss": 0.7347319793701171,
88
+ "mean_token_accuracy": 0.7766731631755829,
89
+ "num_tokens": 1061654.0,
90
+ "step": 400
91
+ },
92
+ {
93
+ "epoch": 1.0,
94
+ "eval_entropy": 0.83522472347532,
95
+ "eval_mean_token_accuracy": 0.7484098076820374,
96
+ "eval_not_syn_loss": 0.7872886061668396,
97
+ "eval_not_syn_runtime": 54.8697,
98
+ "eval_not_syn_samples_per_second": 25.424,
99
+ "eval_not_syn_steps_per_second": 3.189,
100
+ "eval_num_tokens": 1080196.0,
101
+ "step": 408
102
+ },
103
+ {
104
+ "epoch": 1.0,
105
+ "eval_entropy": 0.7784442271505083,
106
+ "eval_mean_token_accuracy": 0.8021791185651507,
107
+ "eval_num_tokens": 1080196.0,
108
+ "eval_syn_loss": 0.6980098485946655,
109
+ "eval_syn_runtime": 56.6342,
110
+ "eval_syn_samples_per_second": 24.632,
111
+ "eval_syn_steps_per_second": 3.09,
112
+ "step": 408
113
+ },
114
+ {
115
+ "entropy": 0.75672573694075,
116
+ "epoch": 1.1030674846625768,
117
+ "grad_norm": 0.9244198203086853,
118
+ "learning_rate": 0.00037356351883456205,
119
+ "loss": 0.7048745727539063,
120
+ "mean_token_accuracy": 0.783016896609104,
121
+ "num_tokens": 1199170.0,
122
+ "step": 450
123
+ },
124
+ {
125
+ "entropy": 0.7406797379255294,
126
+ "epoch": 1.2257668711656442,
127
+ "grad_norm": 0.6744896769523621,
128
+ "learning_rate": 0.00037311248151765587,
129
+ "loss": 0.6860730743408203,
130
+ "mean_token_accuracy": 0.787369327545166,
131
+ "num_tokens": 1334298.0,
132
+ "step": 500
133
+ },
134
+ {
135
+ "entropy": 0.7424973237514496,
136
+ "epoch": 1.3484662576687116,
137
+ "grad_norm": 0.92305588722229,
138
+ "learning_rate": 0.000372320629215228,
139
+ "loss": 0.6855724334716797,
140
+ "mean_token_accuracy": 0.7871134179830551,
141
+ "num_tokens": 1463189.0,
142
+ "step": 550
143
+ },
144
+ {
145
+ "entropy": 0.7502243828773498,
146
+ "epoch": 1.471165644171779,
147
+ "grad_norm": 0.8259047269821167,
148
+ "learning_rate": 0.00037118941074037944,
149
+ "loss": 0.6949834442138672,
150
+ "mean_token_accuracy": 0.7851914083957672,
151
+ "num_tokens": 1590310.0,
152
+ "step": 600
153
+ },
154
+ {
155
+ "entropy": 0.7233260518312454,
156
+ "epoch": 1.5938650306748468,
157
+ "grad_norm": 0.882722795009613,
158
+ "learning_rate": 0.00036972089582775814,
159
+ "loss": 0.6649005889892579,
160
+ "mean_token_accuracy": 0.7898879665136337,
161
+ "num_tokens": 1721836.0,
162
+ "step": 650
163
+ },
164
+ {
165
+ "entropy": 0.7162708967924118,
166
+ "epoch": 1.716564417177914,
167
+ "grad_norm": 0.6593027114868164,
168
+ "learning_rate": 0.0003679177713466678,
169
+ "loss": 0.6653710174560546,
170
+ "mean_token_accuracy": 0.7905678844451904,
171
+ "num_tokens": 1853507.0,
172
+ "step": 700
173
+ },
174
+ {
175
+ "entropy": 0.7150520020723343,
176
+ "epoch": 1.8392638036809816,
177
+ "grad_norm": 0.665340781211853,
178
+ "learning_rate": 0.0003657833363850354,
179
+ "loss": 0.6600938415527344,
180
+ "mean_token_accuracy": 0.7918649202585221,
181
+ "num_tokens": 1988113.0,
182
+ "step": 750
183
+ },
184
+ {
185
+ "entropy": 0.6993386310338974,
186
+ "epoch": 1.961963190184049,
187
+ "grad_norm": 0.6502187252044678,
188
+ "learning_rate": 0.00036332149621323294,
189
+ "loss": 0.6469607543945313,
190
+ "mean_token_accuracy": 0.7950778317451477,
191
+ "num_tokens": 2121314.0,
192
+ "step": 800
193
+ },
194
+ {
195
+ "epoch": 2.0,
196
+ "eval_entropy": 0.7315139893123082,
197
+ "eval_mean_token_accuracy": 0.7713504004478454,
198
+ "eval_not_syn_loss": 0.699893593788147,
199
+ "eval_not_syn_runtime": 55.4135,
200
+ "eval_not_syn_samples_per_second": 25.174,
201
+ "eval_not_syn_steps_per_second": 3.158,
202
+ "eval_num_tokens": 2160392.0,
203
+ "step": 816
204
+ },
205
+ {
206
+ "epoch": 2.0,
207
+ "eval_entropy": 0.6848451798302787,
208
+ "eval_mean_token_accuracy": 0.814895657471248,
209
+ "eval_num_tokens": 2160392.0,
210
+ "eval_syn_loss": 0.6415576934814453,
211
+ "eval_syn_runtime": 56.4735,
212
+ "eval_syn_samples_per_second": 24.702,
213
+ "eval_syn_steps_per_second": 3.099,
214
+ "step": 816
215
+ },
216
+ {
217
+ "entropy": 0.6596363644407253,
218
+ "epoch": 2.083435582822086,
219
+ "grad_norm": 0.6647700071334839,
220
+ "learning_rate": 0.00036053675513879666,
221
+ "loss": 0.599151611328125,
222
+ "mean_token_accuracy": 0.8071986864311527,
223
+ "num_tokens": 2253179.0,
224
+ "step": 850
225
+ },
226
+ {
227
+ "entropy": 0.6626429110765457,
228
+ "epoch": 2.2061349693251535,
229
+ "grad_norm": 0.7356868386268616,
230
+ "learning_rate": 0.00035743420826511783,
231
+ "loss": 0.6004195404052735,
232
+ "mean_token_accuracy": 0.805362731218338,
233
+ "num_tokens": 2379965.0,
234
+ "step": 900
235
+ },
236
+ {
237
+ "entropy": 0.6503470712900161,
238
+ "epoch": 2.3288343558282207,
239
+ "grad_norm": 1.095608115196228,
240
+ "learning_rate": 0.0003540195321691835,
241
+ "loss": 0.5951333999633789,
242
+ "mean_token_accuracy": 0.807324025630951,
243
+ "num_tokens": 2515593.0,
244
+ "step": 950
245
+ },
246
+ {
247
+ "entropy": 0.660577870607376,
248
+ "epoch": 2.4515337423312884,
249
+ "grad_norm": 0.6602179408073425,
250
+ "learning_rate": 0.00035029897451542325,
251
+ "loss": 0.6056365585327148,
252
+ "mean_token_accuracy": 0.8057438349723816,
253
+ "num_tokens": 2642790.0,
254
+ "step": 1000
255
+ },
256
+ {
257
+ "entropy": 0.6555853113532066,
258
+ "epoch": 2.574233128834356,
259
+ "grad_norm": 0.7459151148796082,
260
+ "learning_rate": 0.00034627934262466606,
261
+ "loss": 0.5963204193115235,
262
+ "mean_token_accuracy": 0.8060598099231719,
263
+ "num_tokens": 2774396.0,
264
+ "step": 1050
265
+ },
266
+ {
267
+ "entropy": 0.6505853188037872,
268
+ "epoch": 2.6969325153374233,
269
+ "grad_norm": 0.6359792947769165,
270
+ "learning_rate": 0.00034196799101912103,
271
+ "loss": 0.5993065643310547,
272
+ "mean_token_accuracy": 0.806280387043953,
273
+ "num_tokens": 2912102.0,
274
+ "step": 1100
275
+ },
276
+ {
277
+ "entropy": 0.6509970253705979,
278
+ "epoch": 2.819631901840491,
279
+ "grad_norm": 0.8086174726486206,
280
+ "learning_rate": 0.0003373728079661707,
281
+ "loss": 0.6001268005371094,
282
+ "mean_token_accuracy": 0.806866540312767,
283
+ "num_tokens": 3042987.0,
284
+ "step": 1150
285
+ },
286
+ {
287
+ "entropy": 0.6404844307899475,
288
+ "epoch": 2.942331288343558,
289
+ "grad_norm": 0.7038407325744629,
290
+ "learning_rate": 0.0003325022010455975,
291
+ "loss": 0.5870526123046875,
292
+ "mean_token_accuracy": 0.8104586935043335,
293
+ "num_tokens": 3178044.0,
294
+ "step": 1200
295
+ },
296
+ {
297
+ "epoch": 3.0,
298
+ "eval_entropy": 0.6645996727262224,
299
+ "eval_mean_token_accuracy": 0.798654237134116,
300
+ "eval_not_syn_loss": 0.6586771011352539,
301
+ "eval_not_syn_runtime": 55.1088,
302
+ "eval_not_syn_samples_per_second": 25.314,
303
+ "eval_not_syn_steps_per_second": 3.176,
304
+ "eval_num_tokens": 3240588.0,
305
+ "step": 1224
306
+ },
307
+ {
308
+ "epoch": 3.0,
309
+ "eval_entropy": 0.6233607809884207,
310
+ "eval_mean_token_accuracy": 0.7977249867575509,
311
+ "eval_num_tokens": 3240588.0,
312
+ "eval_syn_loss": 0.6288298964500427,
313
+ "eval_syn_runtime": 56.6081,
314
+ "eval_syn_samples_per_second": 24.643,
315
+ "eval_syn_steps_per_second": 3.091,
316
+ "step": 1224
317
+ },
318
+ {
319
+ "entropy": 0.6081961104364106,
320
+ "epoch": 3.063803680981595,
321
+ "grad_norm": 0.6161203980445862,
322
+ "learning_rate": 0.0003273650817666497,
323
+ "loss": 0.5481716537475586,
324
+ "mean_token_accuracy": 0.8175542035488167,
325
+ "num_tokens": 3312797.0,
326
+ "step": 1250
327
+ },
328
+ {
329
+ "entropy": 0.5877064895629883,
330
+ "epoch": 3.1865030674846624,
331
+ "grad_norm": 0.6909636855125427,
332
+ "learning_rate": 0.00032197084926309234,
333
+ "loss": 0.5224150466918945,
334
+ "mean_token_accuracy": 0.8250984263420105,
335
+ "num_tokens": 3441629.0,
336
+ "step": 1300
337
+ },
338
+ {
339
+ "entropy": 0.5952656552195549,
340
+ "epoch": 3.30920245398773,
341
+ "grad_norm": 0.8147026300430298,
342
+ "learning_rate": 0.0003163293730960758,
343
+ "loss": 0.5349474334716797,
344
+ "mean_token_accuracy": 0.8213064837455749,
345
+ "num_tokens": 3576093.0,
346
+ "step": 1350
347
+ },
348
+ {
349
+ "entropy": 0.5937123569846153,
350
+ "epoch": 3.4319018404907977,
351
+ "grad_norm": 0.8690970540046692,
352
+ "learning_rate": 0.0003104509751962869,
353
+ "loss": 0.5310481262207031,
354
+ "mean_token_accuracy": 0.8229135835170746,
355
+ "num_tokens": 3705912.0,
356
+ "step": 1400
357
+ },
358
+ {
359
+ "entropy": 0.6067730465531349,
360
+ "epoch": 3.554601226993865,
361
+ "grad_norm": 0.7433330416679382,
362
+ "learning_rate": 0.0003043464109784207,
363
+ "loss": 0.5450045394897461,
364
+ "mean_token_accuracy": 0.8205251175165177,
365
+ "num_tokens": 3832376.0,
366
+ "step": 1450
367
+ },
368
+ {
369
+ "entropy": 0.5926914718747139,
370
+ "epoch": 3.6773006134969326,
371
+ "grad_norm": 0.7265587449073792,
372
+ "learning_rate": 0.00029802684966252974,
373
+ "loss": 0.5330168914794922,
374
+ "mean_token_accuracy": 0.821294127702713,
375
+ "num_tokens": 3967273.0,
376
+ "step": 1500
377
+ },
378
+ {
379
+ "entropy": 0.5881199997663498,
380
+ "epoch": 3.8,
381
+ "grad_norm": 0.7681817412376404,
382
+ "learning_rate": 0.00029150385383825187,
383
+ "loss": 0.5288929367065429,
384
+ "mean_token_accuracy": 0.8234161508083343,
385
+ "num_tokens": 4102434.0,
386
+ "step": 1550
387
+ },
388
+ {
389
+ "entropy": 0.6008686199784279,
390
+ "epoch": 3.9226993865030675,
391
+ "grad_norm": 0.9384056329727173,
392
+ "learning_rate": 0.0002847893583093113,
393
+ "loss": 0.5437511444091797,
394
+ "mean_token_accuracy": 0.8201366275548935,
395
+ "num_tokens": 4234037.0,
396
+ "step": 1600
397
+ },
398
+ {
399
+ "epoch": 4.0,
400
+ "eval_entropy": 0.6218148783275059,
401
+ "eval_mean_token_accuracy": 0.783796865599496,
402
+ "eval_not_syn_loss": 0.6558170318603516,
403
+ "eval_not_syn_runtime": 55.429,
404
+ "eval_not_syn_samples_per_second": 25.167,
405
+ "eval_not_syn_steps_per_second": 3.157,
406
+ "eval_num_tokens": 4320784.0,
407
+ "step": 1632
408
+ },
409
+ {
410
+ "epoch": 4.0,
411
+ "eval_entropy": 0.5833754854542869,
412
+ "eval_mean_token_accuracy": 0.8278223422595433,
413
+ "eval_num_tokens": 4320784.0,
414
+ "eval_syn_loss": 0.604770302772522,
415
+ "eval_syn_runtime": 56.9593,
416
+ "eval_syn_samples_per_second": 24.491,
417
+ "eval_syn_steps_per_second": 3.072,
418
+ "step": 1632
419
+ },
420
+ {
421
+ "entropy": 0.559690699733869,
422
+ "epoch": 4.044171779141104,
423
+ "grad_norm": 0.65739905834198,
424
+ "learning_rate": 0.0002778956482569963,
425
+ "loss": 0.49702674865722657,
426
+ "mean_token_accuracy": 0.8317034166268628,
427
+ "num_tokens": 4368268.0,
428
+ "step": 1650
429
+ },
430
+ {
431
+ "entropy": 0.5401161375641823,
432
+ "epoch": 4.166871165644172,
433
+ "grad_norm": 1.5176353454589844,
434
+ "learning_rate": 0.0002708353367625691,
435
+ "loss": 0.4598746109008789,
436
+ "mean_token_accuracy": 0.8412033343315124,
437
+ "num_tokens": 4494525.0,
438
+ "step": 1700
439
+ },
440
+ {
441
+ "entropy": 0.5299916860461235,
442
+ "epoch": 4.289570552147239,
443
+ "grad_norm": 0.685965359210968,
444
+ "learning_rate": 0.0002636213417297333,
445
+ "loss": 0.45663665771484374,
446
+ "mean_token_accuracy": 0.8412413811683654,
447
+ "num_tokens": 4631985.0,
448
+ "step": 1750
449
+ },
450
+ {
451
+ "entropy": 0.5352088099718094,
452
+ "epoch": 4.412269938650307,
453
+ "grad_norm": 0.8471158742904663,
454
+ "learning_rate": 0.0002562668622493824,
455
+ "loss": 0.4625707626342773,
456
+ "mean_token_accuracy": 0.8400991547107697,
457
+ "num_tokens": 4766323.0,
458
+ "step": 1800
459
+ },
460
+ {
461
+ "entropy": 0.5425067806243896,
462
+ "epoch": 4.534969325153375,
463
+ "grad_norm": 0.8595437407493591,
464
+ "learning_rate": 0.00024878535444987474,
465
+ "loss": 0.47124954223632814,
466
+ "mean_token_accuracy": 0.8376108628511428,
467
+ "num_tokens": 4895606.0,
468
+ "step": 1850
469
+ },
470
+ {
471
+ "entropy": 0.5419312778115273,
472
+ "epoch": 4.6576687116564415,
473
+ "grad_norm": 0.8436959981918335,
474
+ "learning_rate": 0.00024119050687701818,
475
+ "loss": 0.4708108901977539,
476
+ "mean_token_accuracy": 0.8390797019004822,
477
+ "num_tokens": 5025912.0,
478
+ "step": 1900
479
+ },
480
+ {
481
+ "entropy": 0.5392271488904953,
482
+ "epoch": 4.780368098159509,
483
+ "grad_norm": 0.8691944479942322,
484
+ "learning_rate": 0.00023349621544881404,
485
+ "loss": 0.46682029724121094,
486
+ "mean_token_accuracy": 0.8385172665119172,
487
+ "num_tokens": 5162972.0,
488
+ "step": 1950
489
+ },
490
+ {
491
+ "entropy": 0.5332251599431038,
492
+ "epoch": 4.903067484662577,
493
+ "grad_norm": 0.6814598441123962,
494
+ "learning_rate": 0.00022571655803078078,
495
+ "loss": 0.4687493896484375,
496
+ "mean_token_accuracy": 0.8388861840963364,
497
+ "num_tokens": 5297935.0,
498
+ "step": 2000
499
+ },
500
+ {
501
+ "epoch": 5.0,
502
+ "eval_entropy": 0.5847777877535139,
503
+ "eval_mean_token_accuracy": 0.7782382583618164,
504
+ "eval_not_syn_loss": 0.6678160429000854,
505
+ "eval_not_syn_runtime": 55.1467,
506
+ "eval_not_syn_samples_per_second": 25.296,
507
+ "eval_not_syn_steps_per_second": 3.173,
508
+ "eval_num_tokens": 5400980.0,
509
+ "step": 2040
510
+ },
511
+ {
512
+ "epoch": 5.0,
513
+ "eval_entropy": 0.5505962218557086,
514
+ "eval_mean_token_accuracy": 0.8308270008223397,
515
+ "eval_num_tokens": 5400980.0,
516
+ "eval_syn_loss": 0.606867253780365,
517
+ "eval_syn_runtime": 56.9768,
518
+ "eval_syn_samples_per_second": 24.484,
519
+ "eval_syn_steps_per_second": 3.071,
520
+ "step": 2040
521
+ },
522
+ {
523
+ "entropy": 0.533086326688227,
524
+ "epoch": 5.024539877300613,
525
+ "grad_norm": 0.9453939199447632,
526
+ "learning_rate": 0.00021786576867837776,
527
+ "loss": 0.45802555084228513,
528
+ "mean_token_accuracy": 0.8429379854539428,
529
+ "num_tokens": 5427362.0,
530
+ "step": 2050
531
+ },
532
+ {
533
+ "entropy": 0.46721911430358887,
534
+ "epoch": 5.147239263803681,
535
+ "grad_norm": 0.9237910509109497,
536
+ "learning_rate": 0.00020995821159365647,
537
+ "loss": 0.37914146423339845,
538
+ "mean_token_accuracy": 0.8648965793848038,
539
+ "num_tokens": 5564186.0,
540
+ "step": 2100
541
+ },
542
+ {
543
+ "entropy": 0.4682381117343903,
544
+ "epoch": 5.269938650306749,
545
+ "grad_norm": 0.7424130439758301,
546
+ "learning_rate": 0.0002020083548437879,
547
+ "loss": 0.3859384536743164,
548
+ "mean_token_accuracy": 0.8622810471057892,
549
+ "num_tokens": 5694728.0,
550
+ "step": 2150
551
+ },
552
+ {
553
+ "entropy": 0.464998793900013,
554
+ "epoch": 5.392638036809816,
555
+ "grad_norm": 0.9224902987480164,
556
+ "learning_rate": 0.0001940307438895544,
557
+ "loss": 0.3839593505859375,
558
+ "mean_token_accuracy": 0.8627353543043137,
559
+ "num_tokens": 5832581.0,
560
+ "step": 2200
561
+ },
562
+ {
563
+ "entropy": 0.4753906837105751,
564
+ "epoch": 5.515337423312883,
565
+ "grad_norm": 0.8931427001953125,
566
+ "learning_rate": 0.0001860399749722367,
567
+ "loss": 0.393004264831543,
568
+ "mean_token_accuracy": 0.8586806547641754,
569
+ "num_tokens": 5962417.0,
570
+ "step": 2250
571
+ },
572
+ {
573
+ "entropy": 0.4671635267138481,
574
+ "epoch": 5.638036809815951,
575
+ "grad_norm": 0.9559172987937927,
576
+ "learning_rate": 0.00017805066840759193,
577
+ "loss": 0.38852767944335936,
578
+ "mean_token_accuracy": 0.8615026396512985,
579
+ "num_tokens": 6092338.0,
580
+ "step": 2300
581
+ },
582
+ {
583
+ "entropy": 0.4746100226044655,
584
+ "epoch": 5.7607361963190185,
585
+ "grad_norm": 0.9070253968238831,
586
+ "learning_rate": 0.00017007744183578254,
587
+ "loss": 0.3960166168212891,
588
+ "mean_token_accuracy": 0.8591635781526565,
589
+ "num_tokens": 6226327.0,
590
+ "step": 2350
591
+ },
592
+ {
593
+ "entropy": 0.4698092398047447,
594
+ "epoch": 5.883435582822086,
595
+ "grad_norm": 0.8354232907295227,
596
+ "learning_rate": 0.0001621348834762011,
597
+ "loss": 0.39321571350097656,
598
+ "mean_token_accuracy": 0.8603039687871933,
599
+ "num_tokens": 6359083.0,
600
+ "step": 2400
601
+ },
602
+ {
603
+ "epoch": 6.0,
604
+ "eval_entropy": 0.536405998979296,
605
+ "eval_mean_token_accuracy": 0.790239520413535,
606
+ "eval_not_syn_loss": 0.6884409785270691,
607
+ "eval_not_syn_runtime": 54.9927,
608
+ "eval_not_syn_samples_per_second": 25.367,
609
+ "eval_not_syn_steps_per_second": 3.182,
610
+ "eval_num_tokens": 6481176.0,
611
+ "step": 2448
612
+ },
613
+ {
614
+ "epoch": 6.0,
615
+ "eval_entropy": 0.5039791010107313,
616
+ "eval_mean_token_accuracy": 0.8163431504794529,
617
+ "eval_num_tokens": 6481176.0,
618
+ "eval_syn_loss": 0.6377155184745789,
619
+ "eval_syn_runtime": 57.196,
620
+ "eval_syn_samples_per_second": 24.39,
621
+ "eval_syn_steps_per_second": 3.06,
622
+ "step": 2448
623
+ }
624
+ ],
625
+ "logging_steps": 50,
626
+ "max_steps": 4080,
627
+ "num_input_tokens_seen": 0,
628
+ "num_train_epochs": 10,
629
+ "save_steps": 500,
630
+ "stateful_callbacks": {
631
+ "TrainerControl": {
632
+ "args": {
633
+ "should_epoch_stop": false,
634
+ "should_evaluate": false,
635
+ "should_log": false,
636
+ "should_save": true,
637
+ "should_training_stop": false
638
+ },
639
+ "attributes": {}
640
+ }
641
+ },
642
+ "total_flos": 1.2906721035504346e+17,
643
+ "train_batch_size": 8,
644
+ "trial_name": null,
645
+ "trial_params": null
646
+ }
substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2856/README.md ADDED
@@ -0,0 +1,209 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: Qwen/Qwen3.5-2B-Base
3
+ library_name: peft
4
+ pipeline_tag: text-generation
5
+ tags:
6
+ - base_model:adapter:Qwen/Qwen3.5-2B-Base
7
+ - lora
8
+ - sft
9
+ - transformers
10
+ - trl
11
+ ---
12
+
13
+ # Model Card for Model ID
14
+
15
+ <!-- Provide a quick summary of what the model is/does. -->
16
+
17
+
18
+
19
+ ## Model Details
20
+
21
+ ### Model Description
22
+
23
+ <!-- Provide a longer summary of what this model is. -->
24
+
25
+
26
+
27
+ - **Developed by:** [More Information Needed]
28
+ - **Funded by [optional]:** [More Information Needed]
29
+ - **Shared by [optional]:** [More Information Needed]
30
+ - **Model type:** [More Information Needed]
31
+ - **Language(s) (NLP):** [More Information Needed]
32
+ - **License:** [More Information Needed]
33
+ - **Finetuned from model [optional]:** [More Information Needed]
34
+
35
+ ### Model Sources [optional]
36
+
37
+ <!-- Provide the basic links for the model. -->
38
+
39
+ - **Repository:** [More Information Needed]
40
+ - **Paper [optional]:** [More Information Needed]
41
+ - **Demo [optional]:** [More Information Needed]
42
+
43
+ ## Uses
44
+
45
+ <!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
46
+
47
+ ### Direct Use
48
+
49
+ <!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
50
+
51
+ [More Information Needed]
52
+
53
+ ### Downstream Use [optional]
54
+
55
+ <!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
56
+
57
+ [More Information Needed]
58
+
59
+ ### Out-of-Scope Use
60
+
61
+ <!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
62
+
63
+ [More Information Needed]
64
+
65
+ ## Bias, Risks, and Limitations
66
+
67
+ <!-- This section is meant to convey both technical and sociotechnical limitations. -->
68
+
69
+ [More Information Needed]
70
+
71
+ ### Recommendations
72
+
73
+ <!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
74
+
75
+ Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
76
+
77
+ ## How to Get Started with the Model
78
+
79
+ Use the code below to get started with the model.
80
+
81
+ [More Information Needed]
82
+
83
+ ## Training Details
84
+
85
+ ### Training Data
86
+
87
+ <!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
88
+
89
+ [More Information Needed]
90
+
91
+ ### Training Procedure
92
+
93
+ <!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
94
+
95
+ #### Preprocessing [optional]
96
+
97
+ [More Information Needed]
98
+
99
+
100
+ #### Training Hyperparameters
101
+
102
+ - **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
103
+
104
+ #### Speeds, Sizes, Times [optional]
105
+
106
+ <!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
107
+
108
+ [More Information Needed]
109
+
110
+ ## Evaluation
111
+
112
+ <!-- This section describes the evaluation protocols and provides the results. -->
113
+
114
+ ### Testing Data, Factors & Metrics
115
+
116
+ #### Testing Data
117
+
118
+ <!-- This should link to a Dataset Card if possible. -->
119
+
120
+ [More Information Needed]
121
+
122
+ #### Factors
123
+
124
+ <!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
125
+
126
+ [More Information Needed]
127
+
128
+ #### Metrics
129
+
130
+ <!-- These are the evaluation metrics being used, ideally with a description of why. -->
131
+
132
+ [More Information Needed]
133
+
134
+ ### Results
135
+
136
+ [More Information Needed]
137
+
138
+ #### Summary
139
+
140
+
141
+
142
+ ## Model Examination [optional]
143
+
144
+ <!-- Relevant interpretability work for the model goes here -->
145
+
146
+ [More Information Needed]
147
+
148
+ ## Environmental Impact
149
+
150
+ <!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
151
+
152
+ Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
153
+
154
+ - **Hardware Type:** [More Information Needed]
155
+ - **Hours used:** [More Information Needed]
156
+ - **Cloud Provider:** [More Information Needed]
157
+ - **Compute Region:** [More Information Needed]
158
+ - **Carbon Emitted:** [More Information Needed]
159
+
160
+ ## Technical Specifications [optional]
161
+
162
+ ### Model Architecture and Objective
163
+
164
+ [More Information Needed]
165
+
166
+ ### Compute Infrastructure
167
+
168
+ [More Information Needed]
169
+
170
+ #### Hardware
171
+
172
+ [More Information Needed]
173
+
174
+ #### Software
175
+
176
+ [More Information Needed]
177
+
178
+ ## Citation [optional]
179
+
180
+ <!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
181
+
182
+ **BibTeX:**
183
+
184
+ [More Information Needed]
185
+
186
+ **APA:**
187
+
188
+ [More Information Needed]
189
+
190
+ ## Glossary [optional]
191
+
192
+ <!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
193
+
194
+ [More Information Needed]
195
+
196
+ ## More Information [optional]
197
+
198
+ [More Information Needed]
199
+
200
+ ## Model Card Authors [optional]
201
+
202
+ [More Information Needed]
203
+
204
+ ## Model Card Contact
205
+
206
+ [More Information Needed]
207
+ ### Framework versions
208
+
209
+ - PEFT 0.19.1
substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2856/adapter_config.json ADDED
@@ -0,0 +1,48 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alora_invocation_tokens": null,
3
+ "alpha_pattern": {},
4
+ "arrow_config": null,
5
+ "auto_mapping": null,
6
+ "base_model_name_or_path": "Qwen/Qwen3.5-2B-Base",
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": 32,
20
+ "lora_bias": false,
21
+ "lora_dropout": 0.08602048083239339,
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": 16,
30
+ "rank_pattern": {},
31
+ "revision": null,
32
+ "target_modules": [
33
+ "q_proj",
34
+ "k_proj",
35
+ "gate_proj",
36
+ "up_proj",
37
+ "o_proj",
38
+ "down_proj",
39
+ "v_proj"
40
+ ],
41
+ "target_parameters": null,
42
+ "task_type": "CAUSAL_LM",
43
+ "trainable_token_indices": null,
44
+ "use_bdlora": null,
45
+ "use_dora": false,
46
+ "use_qalora": false,
47
+ "use_rslora": false
48
+ }
substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2856/chat_template.jinja ADDED
@@ -0,0 +1,154 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {%- set image_count = namespace(value=0) %}
2
+ {%- set video_count = namespace(value=0) %}
3
+ {%- macro render_content(content, do_vision_count, is_system_content=false) %}
4
+ {%- if content is string %}
5
+ {{- content }}
6
+ {%- elif content is iterable and content is not mapping %}
7
+ {%- for item in content %}
8
+ {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
9
+ {%- if is_system_content %}
10
+ {{- raise_exception('System message cannot contain images.') }}
11
+ {%- endif %}
12
+ {%- if do_vision_count %}
13
+ {%- set image_count.value = image_count.value + 1 %}
14
+ {%- endif %}
15
+ {%- if add_vision_id %}
16
+ {{- 'Picture ' ~ image_count.value ~ ': ' }}
17
+ {%- endif %}
18
+ {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
19
+ {%- elif 'video' in item or item.type == 'video' %}
20
+ {%- if is_system_content %}
21
+ {{- raise_exception('System message cannot contain videos.') }}
22
+ {%- endif %}
23
+ {%- if do_vision_count %}
24
+ {%- set video_count.value = video_count.value + 1 %}
25
+ {%- endif %}
26
+ {%- if add_vision_id %}
27
+ {{- 'Video ' ~ video_count.value ~ ': ' }}
28
+ {%- endif %}
29
+ {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
30
+ {%- elif 'text' in item %}
31
+ {{- item.text }}
32
+ {%- else %}
33
+ {{- raise_exception('Unexpected item type in content.') }}
34
+ {%- endif %}
35
+ {%- endfor %}
36
+ {%- elif content is none or content is undefined %}
37
+ {{- '' }}
38
+ {%- else %}
39
+ {{- raise_exception('Unexpected content type.') }}
40
+ {%- endif %}
41
+ {%- endmacro %}
42
+ {%- if not messages %}
43
+ {{- raise_exception('No messages provided.') }}
44
+ {%- endif %}
45
+ {%- if tools and tools is iterable and tools is not mapping %}
46
+ {{- '<|im_start|>system\n' }}
47
+ {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
48
+ {%- for tool in tools %}
49
+ {{- "\n" }}
50
+ {{- tool | tojson }}
51
+ {%- endfor %}
52
+ {{- "\n</tools>" }}
53
+ {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
54
+ {%- if messages[0].role == 'system' %}
55
+ {%- set content = render_content(messages[0].content, false, true)|trim %}
56
+ {%- if content %}
57
+ {{- '\n\n' + content }}
58
+ {%- endif %}
59
+ {%- endif %}
60
+ {{- '<|im_end|>\n' }}
61
+ {%- else %}
62
+ {%- if messages[0].role == 'system' %}
63
+ {%- set content = render_content(messages[0].content, false, true)|trim %}
64
+ {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
65
+ {%- endif %}
66
+ {%- endif %}
67
+ {%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
68
+ {%- for message in messages[::-1] %}
69
+ {%- set index = (messages|length - 1) - loop.index0 %}
70
+ {%- if ns.multi_step_tool and message.role == "user" %}
71
+ {%- set content = render_content(message.content, false)|trim %}
72
+ {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
73
+ {%- set ns.multi_step_tool = false %}
74
+ {%- set ns.last_query_index = index %}
75
+ {%- endif %}
76
+ {%- endif %}
77
+ {%- endfor %}
78
+ {%- if ns.multi_step_tool %}
79
+ {{- raise_exception('No user query found in messages.') }}
80
+ {%- endif %}
81
+ {%- for message in messages %}
82
+ {%- set content = render_content(message.content, true)|trim %}
83
+ {%- if message.role == "system" %}
84
+ {%- if not loop.first %}
85
+ {{- raise_exception('System message must be at the beginning.') }}
86
+ {%- endif %}
87
+ {%- elif message.role == "user" %}
88
+ {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
89
+ {%- elif message.role == "assistant" %}
90
+ {%- set reasoning_content = '' %}
91
+ {%- if message.reasoning_content is string %}
92
+ {%- set reasoning_content = message.reasoning_content %}
93
+ {%- else %}
94
+ {%- if '</think>' in content %}
95
+ {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
96
+ {%- set content = content.split('</think>')[-1].lstrip('\n') %}
97
+ {%- endif %}
98
+ {%- endif %}
99
+ {%- set reasoning_content = reasoning_content|trim %}
100
+ {%- if loop.index0 > ns.last_query_index %}
101
+ {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
102
+ {%- else %}
103
+ {{- '<|im_start|>' + message.role + '\n' + content }}
104
+ {%- endif %}
105
+ {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
106
+ {%- for tool_call in message.tool_calls %}
107
+ {%- if tool_call.function is defined %}
108
+ {%- set tool_call = tool_call.function %}
109
+ {%- endif %}
110
+ {%- if loop.first %}
111
+ {%- if content|trim %}
112
+ {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
113
+ {%- else %}
114
+ {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
115
+ {%- endif %}
116
+ {%- else %}
117
+ {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
118
+ {%- endif %}
119
+ {%- if tool_call.arguments is defined %}
120
+ {%- for args_name, args_value in tool_call.arguments|items %}
121
+ {{- '<parameter=' + args_name + '>\n' }}
122
+ {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
123
+ {{- args_value }}
124
+ {{- '\n</parameter>\n' }}
125
+ {%- endfor %}
126
+ {%- endif %}
127
+ {{- '</function>\n</tool_call>' }}
128
+ {%- endfor %}
129
+ {%- endif %}
130
+ {{- '<|im_end|>\n' }}
131
+ {%- elif message.role == "tool" %}
132
+ {%- if loop.previtem and loop.previtem.role != "tool" %}
133
+ {{- '<|im_start|>user' }}
134
+ {%- endif %}
135
+ {{- '\n<tool_response>\n' }}
136
+ {{- content }}
137
+ {{- '\n</tool_response>' }}
138
+ {%- if not loop.last and loop.nextitem.role != "tool" %}
139
+ {{- '<|im_end|>\n' }}
140
+ {%- elif loop.last %}
141
+ {{- '<|im_end|>\n' }}
142
+ {%- endif %}
143
+ {%- else %}
144
+ {{- raise_exception('Unexpected message role.') }}
145
+ {%- endif %}
146
+ {%- endfor %}
147
+ {%- if add_generation_prompt %}
148
+ {{- '<|im_start|>assistant\n' }}
149
+ {%- if enable_thinking is defined and enable_thinking is true %}
150
+ {{- '<think>\n' }}
151
+ {%- else %}
152
+ {{- '<think>\n\n</think>\n\n' }}
153
+ {%- endif %}
154
+ {%- endif %}
substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2856/tokenizer_config.json ADDED
@@ -0,0 +1,31 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "audio_bos_token": "<|audio_start|>",
4
+ "audio_eos_token": "<|audio_end|>",
5
+ "audio_token": "<|audio_pad|>",
6
+ "backend": "tokenizers",
7
+ "bos_token": null,
8
+ "clean_up_tokenization_spaces": false,
9
+ "eos_token": "<|endoftext|>",
10
+ "errors": "replace",
11
+ "image_token": "<|image_pad|>",
12
+ "is_local": false,
13
+ "model_max_length": 262144,
14
+ "model_specific_special_tokens": {
15
+ "audio_bos_token": "<|audio_start|>",
16
+ "audio_eos_token": "<|audio_end|>",
17
+ "audio_token": "<|audio_pad|>",
18
+ "image_token": "<|image_pad|>",
19
+ "video_token": "<|video_pad|>",
20
+ "vision_bos_token": "<|vision_start|>",
21
+ "vision_eos_token": "<|vision_end|>"
22
+ },
23
+ "pad_token": "<|endoftext|>",
24
+ "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
25
+ "split_special_tokens": false,
26
+ "tokenizer_class": "TokenizersBackend",
27
+ "unk_token": null,
28
+ "video_token": "<|video_pad|>",
29
+ "vision_bos_token": "<|vision_start|>",
30
+ "vision_eos_token": "<|vision_end|>"
31
+ }
substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2856/trainer_state.json ADDED
@@ -0,0 +1,758 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 7.0,
6
+ "eval_steps": 500,
7
+ "global_step": 2856,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "entropy": 2.2408065378665922,
14
+ "epoch": 0.12269938650306748,
15
+ "grad_norm": 1.6718658208847046,
16
+ "learning_rate": 4.487804970912036e-05,
17
+ "loss": 2.129624786376953,
18
+ "mean_token_accuracy": 0.5743644836544991,
19
+ "num_tokens": 134329.0,
20
+ "step": 50
21
+ },
22
+ {
23
+ "entropy": 1.1078504550457,
24
+ "epoch": 0.24539877300613497,
25
+ "grad_norm": 2.4091813564300537,
26
+ "learning_rate": 9.067197798373296e-05,
27
+ "loss": 1.0346390533447265,
28
+ "mean_token_accuracy": 0.7085381114482879,
29
+ "num_tokens": 262407.0,
30
+ "step": 100
31
+ },
32
+ {
33
+ "entropy": 0.9421598988771439,
34
+ "epoch": 0.36809815950920244,
35
+ "grad_norm": 1.3913646936416626,
36
+ "learning_rate": 0.00013646590625834558,
37
+ "loss": 0.8716551208496094,
38
+ "mean_token_accuracy": 0.7449040985107422,
39
+ "num_tokens": 401136.0,
40
+ "step": 150
41
+ },
42
+ {
43
+ "entropy": 0.8723550146818161,
44
+ "epoch": 0.49079754601226994,
45
+ "grad_norm": 1.1664468050003052,
46
+ "learning_rate": 0.0001822598345329582,
47
+ "loss": 0.8095977783203125,
48
+ "mean_token_accuracy": 0.7585240352153778,
49
+ "num_tokens": 531647.0,
50
+ "step": 200
51
+ },
52
+ {
53
+ "entropy": 0.8466519421339035,
54
+ "epoch": 0.6134969325153374,
55
+ "grad_norm": 1.0145421028137207,
56
+ "learning_rate": 0.00022805376280757083,
57
+ "loss": 0.7830857849121093,
58
+ "mean_token_accuracy": 0.7661705195903779,
59
+ "num_tokens": 666127.0,
60
+ "step": 250
61
+ },
62
+ {
63
+ "entropy": 0.8180295366048813,
64
+ "epoch": 0.7361963190184049,
65
+ "grad_norm": 0.9731705784797668,
66
+ "learning_rate": 0.00027384769108218336,
67
+ "loss": 0.7573603820800782,
68
+ "mean_token_accuracy": 0.7704453605413437,
69
+ "num_tokens": 799292.0,
70
+ "step": 300
71
+ },
72
+ {
73
+ "entropy": 0.7990227991342544,
74
+ "epoch": 0.8588957055214724,
75
+ "grad_norm": 1.0642374753952026,
76
+ "learning_rate": 0.000319641619356796,
77
+ "loss": 0.73709228515625,
78
+ "mean_token_accuracy": 0.7752733880281448,
79
+ "num_tokens": 930614.0,
80
+ "step": 350
81
+ },
82
+ {
83
+ "entropy": 0.7899087685346603,
84
+ "epoch": 0.9815950920245399,
85
+ "grad_norm": 0.880613386631012,
86
+ "learning_rate": 0.0003654355476314086,
87
+ "loss": 0.7347319793701171,
88
+ "mean_token_accuracy": 0.7766731631755829,
89
+ "num_tokens": 1061654.0,
90
+ "step": 400
91
+ },
92
+ {
93
+ "epoch": 1.0,
94
+ "eval_entropy": 0.83522472347532,
95
+ "eval_mean_token_accuracy": 0.7484098076820374,
96
+ "eval_not_syn_loss": 0.7872886061668396,
97
+ "eval_not_syn_runtime": 54.8697,
98
+ "eval_not_syn_samples_per_second": 25.424,
99
+ "eval_not_syn_steps_per_second": 3.189,
100
+ "eval_num_tokens": 1080196.0,
101
+ "step": 408
102
+ },
103
+ {
104
+ "epoch": 1.0,
105
+ "eval_entropy": 0.7784442271505083,
106
+ "eval_mean_token_accuracy": 0.8021791185651507,
107
+ "eval_num_tokens": 1080196.0,
108
+ "eval_syn_loss": 0.6980098485946655,
109
+ "eval_syn_runtime": 56.6342,
110
+ "eval_syn_samples_per_second": 24.632,
111
+ "eval_syn_steps_per_second": 3.09,
112
+ "step": 408
113
+ },
114
+ {
115
+ "entropy": 0.75672573694075,
116
+ "epoch": 1.1030674846625768,
117
+ "grad_norm": 0.9244198203086853,
118
+ "learning_rate": 0.00037356351883456205,
119
+ "loss": 0.7048745727539063,
120
+ "mean_token_accuracy": 0.783016896609104,
121
+ "num_tokens": 1199170.0,
122
+ "step": 450
123
+ },
124
+ {
125
+ "entropy": 0.7406797379255294,
126
+ "epoch": 1.2257668711656442,
127
+ "grad_norm": 0.6744896769523621,
128
+ "learning_rate": 0.00037311248151765587,
129
+ "loss": 0.6860730743408203,
130
+ "mean_token_accuracy": 0.787369327545166,
131
+ "num_tokens": 1334298.0,
132
+ "step": 500
133
+ },
134
+ {
135
+ "entropy": 0.7424973237514496,
136
+ "epoch": 1.3484662576687116,
137
+ "grad_norm": 0.92305588722229,
138
+ "learning_rate": 0.000372320629215228,
139
+ "loss": 0.6855724334716797,
140
+ "mean_token_accuracy": 0.7871134179830551,
141
+ "num_tokens": 1463189.0,
142
+ "step": 550
143
+ },
144
+ {
145
+ "entropy": 0.7502243828773498,
146
+ "epoch": 1.471165644171779,
147
+ "grad_norm": 0.8259047269821167,
148
+ "learning_rate": 0.00037118941074037944,
149
+ "loss": 0.6949834442138672,
150
+ "mean_token_accuracy": 0.7851914083957672,
151
+ "num_tokens": 1590310.0,
152
+ "step": 600
153
+ },
154
+ {
155
+ "entropy": 0.7233260518312454,
156
+ "epoch": 1.5938650306748468,
157
+ "grad_norm": 0.882722795009613,
158
+ "learning_rate": 0.00036972089582775814,
159
+ "loss": 0.6649005889892579,
160
+ "mean_token_accuracy": 0.7898879665136337,
161
+ "num_tokens": 1721836.0,
162
+ "step": 650
163
+ },
164
+ {
165
+ "entropy": 0.7162708967924118,
166
+ "epoch": 1.716564417177914,
167
+ "grad_norm": 0.6593027114868164,
168
+ "learning_rate": 0.0003679177713466678,
169
+ "loss": 0.6653710174560546,
170
+ "mean_token_accuracy": 0.7905678844451904,
171
+ "num_tokens": 1853507.0,
172
+ "step": 700
173
+ },
174
+ {
175
+ "entropy": 0.7150520020723343,
176
+ "epoch": 1.8392638036809816,
177
+ "grad_norm": 0.665340781211853,
178
+ "learning_rate": 0.0003657833363850354,
179
+ "loss": 0.6600938415527344,
180
+ "mean_token_accuracy": 0.7918649202585221,
181
+ "num_tokens": 1988113.0,
182
+ "step": 750
183
+ },
184
+ {
185
+ "entropy": 0.6993386310338974,
186
+ "epoch": 1.961963190184049,
187
+ "grad_norm": 0.6502187252044678,
188
+ "learning_rate": 0.00036332149621323294,
189
+ "loss": 0.6469607543945313,
190
+ "mean_token_accuracy": 0.7950778317451477,
191
+ "num_tokens": 2121314.0,
192
+ "step": 800
193
+ },
194
+ {
195
+ "epoch": 2.0,
196
+ "eval_entropy": 0.7315139893123082,
197
+ "eval_mean_token_accuracy": 0.7713504004478454,
198
+ "eval_not_syn_loss": 0.699893593788147,
199
+ "eval_not_syn_runtime": 55.4135,
200
+ "eval_not_syn_samples_per_second": 25.174,
201
+ "eval_not_syn_steps_per_second": 3.158,
202
+ "eval_num_tokens": 2160392.0,
203
+ "step": 816
204
+ },
205
+ {
206
+ "epoch": 2.0,
207
+ "eval_entropy": 0.6848451798302787,
208
+ "eval_mean_token_accuracy": 0.814895657471248,
209
+ "eval_num_tokens": 2160392.0,
210
+ "eval_syn_loss": 0.6415576934814453,
211
+ "eval_syn_runtime": 56.4735,
212
+ "eval_syn_samples_per_second": 24.702,
213
+ "eval_syn_steps_per_second": 3.099,
214
+ "step": 816
215
+ },
216
+ {
217
+ "entropy": 0.6596363644407253,
218
+ "epoch": 2.083435582822086,
219
+ "grad_norm": 0.6647700071334839,
220
+ "learning_rate": 0.00036053675513879666,
221
+ "loss": 0.599151611328125,
222
+ "mean_token_accuracy": 0.8071986864311527,
223
+ "num_tokens": 2253179.0,
224
+ "step": 850
225
+ },
226
+ {
227
+ "entropy": 0.6626429110765457,
228
+ "epoch": 2.2061349693251535,
229
+ "grad_norm": 0.7356868386268616,
230
+ "learning_rate": 0.00035743420826511783,
231
+ "loss": 0.6004195404052735,
232
+ "mean_token_accuracy": 0.805362731218338,
233
+ "num_tokens": 2379965.0,
234
+ "step": 900
235
+ },
236
+ {
237
+ "entropy": 0.6503470712900161,
238
+ "epoch": 2.3288343558282207,
239
+ "grad_norm": 1.095608115196228,
240
+ "learning_rate": 0.0003540195321691835,
241
+ "loss": 0.5951333999633789,
242
+ "mean_token_accuracy": 0.807324025630951,
243
+ "num_tokens": 2515593.0,
244
+ "step": 950
245
+ },
246
+ {
247
+ "entropy": 0.660577870607376,
248
+ "epoch": 2.4515337423312884,
249
+ "grad_norm": 0.6602179408073425,
250
+ "learning_rate": 0.00035029897451542325,
251
+ "loss": 0.6056365585327148,
252
+ "mean_token_accuracy": 0.8057438349723816,
253
+ "num_tokens": 2642790.0,
254
+ "step": 1000
255
+ },
256
+ {
257
+ "entropy": 0.6555853113532066,
258
+ "epoch": 2.574233128834356,
259
+ "grad_norm": 0.7459151148796082,
260
+ "learning_rate": 0.00034627934262466606,
261
+ "loss": 0.5963204193115235,
262
+ "mean_token_accuracy": 0.8060598099231719,
263
+ "num_tokens": 2774396.0,
264
+ "step": 1050
265
+ },
266
+ {
267
+ "entropy": 0.6505853188037872,
268
+ "epoch": 2.6969325153374233,
269
+ "grad_norm": 0.6359792947769165,
270
+ "learning_rate": 0.00034196799101912103,
271
+ "loss": 0.5993065643310547,
272
+ "mean_token_accuracy": 0.806280387043953,
273
+ "num_tokens": 2912102.0,
274
+ "step": 1100
275
+ },
276
+ {
277
+ "entropy": 0.6509970253705979,
278
+ "epoch": 2.819631901840491,
279
+ "grad_norm": 0.8086174726486206,
280
+ "learning_rate": 0.0003373728079661707,
281
+ "loss": 0.6001268005371094,
282
+ "mean_token_accuracy": 0.806866540312767,
283
+ "num_tokens": 3042987.0,
284
+ "step": 1150
285
+ },
286
+ {
287
+ "entropy": 0.6404844307899475,
288
+ "epoch": 2.942331288343558,
289
+ "grad_norm": 0.7038407325744629,
290
+ "learning_rate": 0.0003325022010455975,
291
+ "loss": 0.5870526123046875,
292
+ "mean_token_accuracy": 0.8104586935043335,
293
+ "num_tokens": 3178044.0,
294
+ "step": 1200
295
+ },
296
+ {
297
+ "epoch": 3.0,
298
+ "eval_entropy": 0.6645996727262224,
299
+ "eval_mean_token_accuracy": 0.798654237134116,
300
+ "eval_not_syn_loss": 0.6586771011352539,
301
+ "eval_not_syn_runtime": 55.1088,
302
+ "eval_not_syn_samples_per_second": 25.314,
303
+ "eval_not_syn_steps_per_second": 3.176,
304
+ "eval_num_tokens": 3240588.0,
305
+ "step": 1224
306
+ },
307
+ {
308
+ "epoch": 3.0,
309
+ "eval_entropy": 0.6233607809884207,
310
+ "eval_mean_token_accuracy": 0.7977249867575509,
311
+ "eval_num_tokens": 3240588.0,
312
+ "eval_syn_loss": 0.6288298964500427,
313
+ "eval_syn_runtime": 56.6081,
314
+ "eval_syn_samples_per_second": 24.643,
315
+ "eval_syn_steps_per_second": 3.091,
316
+ "step": 1224
317
+ },
318
+ {
319
+ "entropy": 0.6081961104364106,
320
+ "epoch": 3.063803680981595,
321
+ "grad_norm": 0.6161203980445862,
322
+ "learning_rate": 0.0003273650817666497,
323
+ "loss": 0.5481716537475586,
324
+ "mean_token_accuracy": 0.8175542035488167,
325
+ "num_tokens": 3312797.0,
326
+ "step": 1250
327
+ },
328
+ {
329
+ "entropy": 0.5877064895629883,
330
+ "epoch": 3.1865030674846624,
331
+ "grad_norm": 0.6909636855125427,
332
+ "learning_rate": 0.00032197084926309234,
333
+ "loss": 0.5224150466918945,
334
+ "mean_token_accuracy": 0.8250984263420105,
335
+ "num_tokens": 3441629.0,
336
+ "step": 1300
337
+ },
338
+ {
339
+ "entropy": 0.5952656552195549,
340
+ "epoch": 3.30920245398773,
341
+ "grad_norm": 0.8147026300430298,
342
+ "learning_rate": 0.0003163293730960758,
343
+ "loss": 0.5349474334716797,
344
+ "mean_token_accuracy": 0.8213064837455749,
345
+ "num_tokens": 3576093.0,
346
+ "step": 1350
347
+ },
348
+ {
349
+ "entropy": 0.5937123569846153,
350
+ "epoch": 3.4319018404907977,
351
+ "grad_norm": 0.8690970540046692,
352
+ "learning_rate": 0.0003104509751962869,
353
+ "loss": 0.5310481262207031,
354
+ "mean_token_accuracy": 0.8229135835170746,
355
+ "num_tokens": 3705912.0,
356
+ "step": 1400
357
+ },
358
+ {
359
+ "entropy": 0.6067730465531349,
360
+ "epoch": 3.554601226993865,
361
+ "grad_norm": 0.7433330416679382,
362
+ "learning_rate": 0.0003043464109784207,
363
+ "loss": 0.5450045394897461,
364
+ "mean_token_accuracy": 0.8205251175165177,
365
+ "num_tokens": 3832376.0,
366
+ "step": 1450
367
+ },
368
+ {
369
+ "entropy": 0.5926914718747139,
370
+ "epoch": 3.6773006134969326,
371
+ "grad_norm": 0.7265587449073792,
372
+ "learning_rate": 0.00029802684966252974,
373
+ "loss": 0.5330168914794922,
374
+ "mean_token_accuracy": 0.821294127702713,
375
+ "num_tokens": 3967273.0,
376
+ "step": 1500
377
+ },
378
+ {
379
+ "entropy": 0.5881199997663498,
380
+ "epoch": 3.8,
381
+ "grad_norm": 0.7681817412376404,
382
+ "learning_rate": 0.00029150385383825187,
383
+ "loss": 0.5288929367065429,
384
+ "mean_token_accuracy": 0.8234161508083343,
385
+ "num_tokens": 4102434.0,
386
+ "step": 1550
387
+ },
388
+ {
389
+ "entropy": 0.6008686199784279,
390
+ "epoch": 3.9226993865030675,
391
+ "grad_norm": 0.9384056329727173,
392
+ "learning_rate": 0.0002847893583093113,
393
+ "loss": 0.5437511444091797,
394
+ "mean_token_accuracy": 0.8201366275548935,
395
+ "num_tokens": 4234037.0,
396
+ "step": 1600
397
+ },
398
+ {
399
+ "epoch": 4.0,
400
+ "eval_entropy": 0.6218148783275059,
401
+ "eval_mean_token_accuracy": 0.783796865599496,
402
+ "eval_not_syn_loss": 0.6558170318603516,
403
+ "eval_not_syn_runtime": 55.429,
404
+ "eval_not_syn_samples_per_second": 25.167,
405
+ "eval_not_syn_steps_per_second": 3.157,
406
+ "eval_num_tokens": 4320784.0,
407
+ "step": 1632
408
+ },
409
+ {
410
+ "epoch": 4.0,
411
+ "eval_entropy": 0.5833754854542869,
412
+ "eval_mean_token_accuracy": 0.8278223422595433,
413
+ "eval_num_tokens": 4320784.0,
414
+ "eval_syn_loss": 0.604770302772522,
415
+ "eval_syn_runtime": 56.9593,
416
+ "eval_syn_samples_per_second": 24.491,
417
+ "eval_syn_steps_per_second": 3.072,
418
+ "step": 1632
419
+ },
420
+ {
421
+ "entropy": 0.559690699733869,
422
+ "epoch": 4.044171779141104,
423
+ "grad_norm": 0.65739905834198,
424
+ "learning_rate": 0.0002778956482569963,
425
+ "loss": 0.49702674865722657,
426
+ "mean_token_accuracy": 0.8317034166268628,
427
+ "num_tokens": 4368268.0,
428
+ "step": 1650
429
+ },
430
+ {
431
+ "entropy": 0.5401161375641823,
432
+ "epoch": 4.166871165644172,
433
+ "grad_norm": 1.5176353454589844,
434
+ "learning_rate": 0.0002708353367625691,
435
+ "loss": 0.4598746109008789,
436
+ "mean_token_accuracy": 0.8412033343315124,
437
+ "num_tokens": 4494525.0,
438
+ "step": 1700
439
+ },
440
+ {
441
+ "entropy": 0.5299916860461235,
442
+ "epoch": 4.289570552147239,
443
+ "grad_norm": 0.685965359210968,
444
+ "learning_rate": 0.0002636213417297333,
445
+ "loss": 0.45663665771484374,
446
+ "mean_token_accuracy": 0.8412413811683654,
447
+ "num_tokens": 4631985.0,
448
+ "step": 1750
449
+ },
450
+ {
451
+ "entropy": 0.5352088099718094,
452
+ "epoch": 4.412269938650307,
453
+ "grad_norm": 0.8471158742904663,
454
+ "learning_rate": 0.0002562668622493824,
455
+ "loss": 0.4625707626342773,
456
+ "mean_token_accuracy": 0.8400991547107697,
457
+ "num_tokens": 4766323.0,
458
+ "step": 1800
459
+ },
460
+ {
461
+ "entropy": 0.5425067806243896,
462
+ "epoch": 4.534969325153375,
463
+ "grad_norm": 0.8595437407493591,
464
+ "learning_rate": 0.00024878535444987474,
465
+ "loss": 0.47124954223632814,
466
+ "mean_token_accuracy": 0.8376108628511428,
467
+ "num_tokens": 4895606.0,
468
+ "step": 1850
469
+ },
470
+ {
471
+ "entropy": 0.5419312778115273,
472
+ "epoch": 4.6576687116564415,
473
+ "grad_norm": 0.8436959981918335,
474
+ "learning_rate": 0.00024119050687701818,
475
+ "loss": 0.4708108901977539,
476
+ "mean_token_accuracy": 0.8390797019004822,
477
+ "num_tokens": 5025912.0,
478
+ "step": 1900
479
+ },
480
+ {
481
+ "entropy": 0.5392271488904953,
482
+ "epoch": 4.780368098159509,
483
+ "grad_norm": 0.8691944479942322,
484
+ "learning_rate": 0.00023349621544881404,
485
+ "loss": 0.46682029724121094,
486
+ "mean_token_accuracy": 0.8385172665119172,
487
+ "num_tokens": 5162972.0,
488
+ "step": 1950
489
+ },
490
+ {
491
+ "entropy": 0.5332251599431038,
492
+ "epoch": 4.903067484662577,
493
+ "grad_norm": 0.6814598441123962,
494
+ "learning_rate": 0.00022571655803078078,
495
+ "loss": 0.4687493896484375,
496
+ "mean_token_accuracy": 0.8388861840963364,
497
+ "num_tokens": 5297935.0,
498
+ "step": 2000
499
+ },
500
+ {
501
+ "epoch": 5.0,
502
+ "eval_entropy": 0.5847777877535139,
503
+ "eval_mean_token_accuracy": 0.7782382583618164,
504
+ "eval_not_syn_loss": 0.6678160429000854,
505
+ "eval_not_syn_runtime": 55.1467,
506
+ "eval_not_syn_samples_per_second": 25.296,
507
+ "eval_not_syn_steps_per_second": 3.173,
508
+ "eval_num_tokens": 5400980.0,
509
+ "step": 2040
510
+ },
511
+ {
512
+ "epoch": 5.0,
513
+ "eval_entropy": 0.5505962218557086,
514
+ "eval_mean_token_accuracy": 0.8308270008223397,
515
+ "eval_num_tokens": 5400980.0,
516
+ "eval_syn_loss": 0.606867253780365,
517
+ "eval_syn_runtime": 56.9768,
518
+ "eval_syn_samples_per_second": 24.484,
519
+ "eval_syn_steps_per_second": 3.071,
520
+ "step": 2040
521
+ },
522
+ {
523
+ "entropy": 0.533086326688227,
524
+ "epoch": 5.024539877300613,
525
+ "grad_norm": 0.9453939199447632,
526
+ "learning_rate": 0.00021786576867837776,
527
+ "loss": 0.45802555084228513,
528
+ "mean_token_accuracy": 0.8429379854539428,
529
+ "num_tokens": 5427362.0,
530
+ "step": 2050
531
+ },
532
+ {
533
+ "entropy": 0.46721911430358887,
534
+ "epoch": 5.147239263803681,
535
+ "grad_norm": 0.9237910509109497,
536
+ "learning_rate": 0.00020995821159365647,
537
+ "loss": 0.37914146423339845,
538
+ "mean_token_accuracy": 0.8648965793848038,
539
+ "num_tokens": 5564186.0,
540
+ "step": 2100
541
+ },
542
+ {
543
+ "entropy": 0.4682381117343903,
544
+ "epoch": 5.269938650306749,
545
+ "grad_norm": 0.7424130439758301,
546
+ "learning_rate": 0.0002020083548437879,
547
+ "loss": 0.3859384536743164,
548
+ "mean_token_accuracy": 0.8622810471057892,
549
+ "num_tokens": 5694728.0,
550
+ "step": 2150
551
+ },
552
+ {
553
+ "entropy": 0.464998793900013,
554
+ "epoch": 5.392638036809816,
555
+ "grad_norm": 0.9224902987480164,
556
+ "learning_rate": 0.0001940307438895544,
557
+ "loss": 0.3839593505859375,
558
+ "mean_token_accuracy": 0.8627353543043137,
559
+ "num_tokens": 5832581.0,
560
+ "step": 2200
561
+ },
562
+ {
563
+ "entropy": 0.4753906837105751,
564
+ "epoch": 5.515337423312883,
565
+ "grad_norm": 0.8931427001953125,
566
+ "learning_rate": 0.0001860399749722367,
567
+ "loss": 0.393004264831543,
568
+ "mean_token_accuracy": 0.8586806547641754,
569
+ "num_tokens": 5962417.0,
570
+ "step": 2250
571
+ },
572
+ {
573
+ "entropy": 0.4671635267138481,
574
+ "epoch": 5.638036809815951,
575
+ "grad_norm": 0.9559172987937927,
576
+ "learning_rate": 0.00017805066840759193,
577
+ "loss": 0.38852767944335936,
578
+ "mean_token_accuracy": 0.8615026396512985,
579
+ "num_tokens": 6092338.0,
580
+ "step": 2300
581
+ },
582
+ {
583
+ "entropy": 0.4746100226044655,
584
+ "epoch": 5.7607361963190185,
585
+ "grad_norm": 0.9070253968238831,
586
+ "learning_rate": 0.00017007744183578254,
587
+ "loss": 0.3960166168212891,
588
+ "mean_token_accuracy": 0.8591635781526565,
589
+ "num_tokens": 6226327.0,
590
+ "step": 2350
591
+ },
592
+ {
593
+ "entropy": 0.4698092398047447,
594
+ "epoch": 5.883435582822086,
595
+ "grad_norm": 0.8354232907295227,
596
+ "learning_rate": 0.0001621348834762011,
597
+ "loss": 0.39321571350097656,
598
+ "mean_token_accuracy": 0.8603039687871933,
599
+ "num_tokens": 6359083.0,
600
+ "step": 2400
601
+ },
602
+ {
603
+ "epoch": 6.0,
604
+ "eval_entropy": 0.536405998979296,
605
+ "eval_mean_token_accuracy": 0.790239520413535,
606
+ "eval_not_syn_loss": 0.6884409785270691,
607
+ "eval_not_syn_runtime": 54.9927,
608
+ "eval_not_syn_samples_per_second": 25.367,
609
+ "eval_not_syn_steps_per_second": 3.182,
610
+ "eval_num_tokens": 6481176.0,
611
+ "step": 2448
612
+ },
613
+ {
614
+ "epoch": 6.0,
615
+ "eval_entropy": 0.5039791010107313,
616
+ "eval_mean_token_accuracy": 0.8163431504794529,
617
+ "eval_num_tokens": 6481176.0,
618
+ "eval_syn_loss": 0.6377155184745789,
619
+ "eval_syn_runtime": 57.196,
620
+ "eval_syn_samples_per_second": 24.39,
621
+ "eval_syn_steps_per_second": 3.06,
622
+ "step": 2448
623
+ },
624
+ {
625
+ "entropy": 0.4671963176341972,
626
+ "epoch": 6.004907975460123,
627
+ "grad_norm": 0.8445424437522888,
628
+ "learning_rate": 0.00015423752543612516,
629
+ "loss": 0.3872825241088867,
630
+ "mean_token_accuracy": 0.8618519083418027,
631
+ "num_tokens": 6486549.0,
632
+ "step": 2450
633
+ },
634
+ {
635
+ "entropy": 0.3890652832388878,
636
+ "epoch": 6.12760736196319,
637
+ "grad_norm": 0.725812554359436,
638
+ "learning_rate": 0.00014639981712203683,
639
+ "loss": 0.2923568153381348,
640
+ "mean_token_accuracy": 0.8928290349245072,
641
+ "num_tokens": 6622384.0,
642
+ "step": 2500
643
+ },
644
+ {
645
+ "entropy": 0.39649384289979933,
646
+ "epoch": 6.250306748466258,
647
+ "grad_norm": 0.9143152236938477,
648
+ "learning_rate": 0.000138636098802257,
649
+ "loss": 0.303057861328125,
650
+ "mean_token_accuracy": 0.8888710218667984,
651
+ "num_tokens": 6758637.0,
652
+ "step": 2550
653
+ },
654
+ {
655
+ "entropy": 0.4020465177297592,
656
+ "epoch": 6.373006134969325,
657
+ "grad_norm": 0.9407342076301575,
658
+ "learning_rate": 0.00013096057536926304,
659
+ "loss": 0.3073894691467285,
660
+ "mean_token_accuracy": 0.8894575881958008,
661
+ "num_tokens": 6888533.0,
662
+ "step": 2600
663
+ },
664
+ {
665
+ "entropy": 0.397317076921463,
666
+ "epoch": 6.495705521472392,
667
+ "grad_norm": 0.8517777323722839,
668
+ "learning_rate": 0.00012338729034969828,
669
+ "loss": 0.30508209228515626,
670
+ "mean_token_accuracy": 0.8886856293678284,
671
+ "num_tokens": 7023729.0,
672
+ "step": 2650
673
+ },
674
+ {
675
+ "entropy": 0.3915065750479698,
676
+ "epoch": 6.61840490797546,
677
+ "grad_norm": 0.862319827079773,
678
+ "learning_rate": 0.00011593010020962378,
679
+ "loss": 0.30237462997436526,
680
+ "mean_token_accuracy": 0.8897094148397445,
681
+ "num_tokens": 7153821.0,
682
+ "step": 2700
683
+ },
684
+ {
685
+ "entropy": 0.3977497002482414,
686
+ "epoch": 6.741104294478528,
687
+ "grad_norm": 0.8821370601654053,
688
+ "learning_rate": 0.00010860264900202563,
689
+ "loss": 0.3067088508605957,
690
+ "mean_token_accuracy": 0.8873285222053527,
691
+ "num_tokens": 7283513.0,
692
+ "step": 2750
693
+ },
694
+ {
695
+ "entropy": 0.4002616259455681,
696
+ "epoch": 6.863803680981595,
697
+ "grad_norm": 1.069462776184082,
698
+ "learning_rate": 0.0001014183434029644,
699
+ "loss": 0.30735193252563475,
700
+ "mean_token_accuracy": 0.8880347841978073,
701
+ "num_tokens": 7411578.0,
702
+ "step": 2800
703
+ },
704
+ {
705
+ "entropy": 0.3951660317182541,
706
+ "epoch": 6.986503067484662,
707
+ "grad_norm": 0.8845367431640625,
708
+ "learning_rate": 9.439032818204105e-05,
709
+ "loss": 0.30523622512817383,
710
+ "mean_token_accuracy": 0.8894886714220047,
711
+ "num_tokens": 7548356.0,
712
+ "step": 2850
713
+ },
714
+ {
715
+ "epoch": 7.0,
716
+ "eval_entropy": 0.4834100874832698,
717
+ "eval_mean_token_accuracy": 0.788761088166918,
718
+ "eval_not_syn_loss": 0.7379753589630127,
719
+ "eval_not_syn_runtime": 55.6309,
720
+ "eval_not_syn_samples_per_second": 25.076,
721
+ "eval_not_syn_steps_per_second": 3.146,
722
+ "eval_num_tokens": 7561372.0,
723
+ "step": 2856
724
+ },
725
+ {
726
+ "epoch": 7.0,
727
+ "eval_entropy": 0.453818735906056,
728
+ "eval_mean_token_accuracy": 0.8209771479879107,
729
+ "eval_num_tokens": 7561372.0,
730
+ "eval_syn_loss": 0.6787521243095398,
731
+ "eval_syn_runtime": 56.9966,
732
+ "eval_syn_samples_per_second": 24.475,
733
+ "eval_syn_steps_per_second": 3.07,
734
+ "step": 2856
735
+ }
736
+ ],
737
+ "logging_steps": 50,
738
+ "max_steps": 4080,
739
+ "num_input_tokens_seen": 0,
740
+ "num_train_epochs": 10,
741
+ "save_steps": 500,
742
+ "stateful_callbacks": {
743
+ "TrainerControl": {
744
+ "args": {
745
+ "should_epoch_stop": false,
746
+ "should_evaluate": false,
747
+ "should_log": false,
748
+ "should_save": true,
749
+ "should_training_stop": false
750
+ },
751
+ "attributes": {}
752
+ }
753
+ },
754
+ "total_flos": 1.5044315164249766e+17,
755
+ "train_batch_size": 8,
756
+ "trial_name": null,
757
+ "trial_params": null
758
+ }
substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3264/README.md ADDED
@@ -0,0 +1,209 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: Qwen/Qwen3.5-2B-Base
3
+ library_name: peft
4
+ pipeline_tag: text-generation
5
+ tags:
6
+ - base_model:adapter:Qwen/Qwen3.5-2B-Base
7
+ - lora
8
+ - sft
9
+ - transformers
10
+ - trl
11
+ ---
12
+
13
+ # Model Card for Model ID
14
+
15
+ <!-- Provide a quick summary of what the model is/does. -->
16
+
17
+
18
+
19
+ ## Model Details
20
+
21
+ ### Model Description
22
+
23
+ <!-- Provide a longer summary of what this model is. -->
24
+
25
+
26
+
27
+ - **Developed by:** [More Information Needed]
28
+ - **Funded by [optional]:** [More Information Needed]
29
+ - **Shared by [optional]:** [More Information Needed]
30
+ - **Model type:** [More Information Needed]
31
+ - **Language(s) (NLP):** [More Information Needed]
32
+ - **License:** [More Information Needed]
33
+ - **Finetuned from model [optional]:** [More Information Needed]
34
+
35
+ ### Model Sources [optional]
36
+
37
+ <!-- Provide the basic links for the model. -->
38
+
39
+ - **Repository:** [More Information Needed]
40
+ - **Paper [optional]:** [More Information Needed]
41
+ - **Demo [optional]:** [More Information Needed]
42
+
43
+ ## Uses
44
+
45
+ <!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
46
+
47
+ ### Direct Use
48
+
49
+ <!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
50
+
51
+ [More Information Needed]
52
+
53
+ ### Downstream Use [optional]
54
+
55
+ <!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
56
+
57
+ [More Information Needed]
58
+
59
+ ### Out-of-Scope Use
60
+
61
+ <!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
62
+
63
+ [More Information Needed]
64
+
65
+ ## Bias, Risks, and Limitations
66
+
67
+ <!-- This section is meant to convey both technical and sociotechnical limitations. -->
68
+
69
+ [More Information Needed]
70
+
71
+ ### Recommendations
72
+
73
+ <!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
74
+
75
+ Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
76
+
77
+ ## How to Get Started with the Model
78
+
79
+ Use the code below to get started with the model.
80
+
81
+ [More Information Needed]
82
+
83
+ ## Training Details
84
+
85
+ ### Training Data
86
+
87
+ <!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
88
+
89
+ [More Information Needed]
90
+
91
+ ### Training Procedure
92
+
93
+ <!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
94
+
95
+ #### Preprocessing [optional]
96
+
97
+ [More Information Needed]
98
+
99
+
100
+ #### Training Hyperparameters
101
+
102
+ - **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
103
+
104
+ #### Speeds, Sizes, Times [optional]
105
+
106
+ <!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
107
+
108
+ [More Information Needed]
109
+
110
+ ## Evaluation
111
+
112
+ <!-- This section describes the evaluation protocols and provides the results. -->
113
+
114
+ ### Testing Data, Factors & Metrics
115
+
116
+ #### Testing Data
117
+
118
+ <!-- This should link to a Dataset Card if possible. -->
119
+
120
+ [More Information Needed]
121
+
122
+ #### Factors
123
+
124
+ <!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
125
+
126
+ [More Information Needed]
127
+
128
+ #### Metrics
129
+
130
+ <!-- These are the evaluation metrics being used, ideally with a description of why. -->
131
+
132
+ [More Information Needed]
133
+
134
+ ### Results
135
+
136
+ [More Information Needed]
137
+
138
+ #### Summary
139
+
140
+
141
+
142
+ ## Model Examination [optional]
143
+
144
+ <!-- Relevant interpretability work for the model goes here -->
145
+
146
+ [More Information Needed]
147
+
148
+ ## Environmental Impact
149
+
150
+ <!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
151
+
152
+ Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
153
+
154
+ - **Hardware Type:** [More Information Needed]
155
+ - **Hours used:** [More Information Needed]
156
+ - **Cloud Provider:** [More Information Needed]
157
+ - **Compute Region:** [More Information Needed]
158
+ - **Carbon Emitted:** [More Information Needed]
159
+
160
+ ## Technical Specifications [optional]
161
+
162
+ ### Model Architecture and Objective
163
+
164
+ [More Information Needed]
165
+
166
+ ### Compute Infrastructure
167
+
168
+ [More Information Needed]
169
+
170
+ #### Hardware
171
+
172
+ [More Information Needed]
173
+
174
+ #### Software
175
+
176
+ [More Information Needed]
177
+
178
+ ## Citation [optional]
179
+
180
+ <!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
181
+
182
+ **BibTeX:**
183
+
184
+ [More Information Needed]
185
+
186
+ **APA:**
187
+
188
+ [More Information Needed]
189
+
190
+ ## Glossary [optional]
191
+
192
+ <!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
193
+
194
+ [More Information Needed]
195
+
196
+ ## More Information [optional]
197
+
198
+ [More Information Needed]
199
+
200
+ ## Model Card Authors [optional]
201
+
202
+ [More Information Needed]
203
+
204
+ ## Model Card Contact
205
+
206
+ [More Information Needed]
207
+ ### Framework versions
208
+
209
+ - PEFT 0.19.1
substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3264/adapter_config.json ADDED
@@ -0,0 +1,48 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alora_invocation_tokens": null,
3
+ "alpha_pattern": {},
4
+ "arrow_config": null,
5
+ "auto_mapping": null,
6
+ "base_model_name_or_path": "Qwen/Qwen3.5-2B-Base",
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": 32,
20
+ "lora_bias": false,
21
+ "lora_dropout": 0.08602048083239339,
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": 16,
30
+ "rank_pattern": {},
31
+ "revision": null,
32
+ "target_modules": [
33
+ "q_proj",
34
+ "k_proj",
35
+ "gate_proj",
36
+ "up_proj",
37
+ "o_proj",
38
+ "down_proj",
39
+ "v_proj"
40
+ ],
41
+ "target_parameters": null,
42
+ "task_type": "CAUSAL_LM",
43
+ "trainable_token_indices": null,
44
+ "use_bdlora": null,
45
+ "use_dora": false,
46
+ "use_qalora": false,
47
+ "use_rslora": false
48
+ }
substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3264/chat_template.jinja ADDED
@@ -0,0 +1,154 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {%- set image_count = namespace(value=0) %}
2
+ {%- set video_count = namespace(value=0) %}
3
+ {%- macro render_content(content, do_vision_count, is_system_content=false) %}
4
+ {%- if content is string %}
5
+ {{- content }}
6
+ {%- elif content is iterable and content is not mapping %}
7
+ {%- for item in content %}
8
+ {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
9
+ {%- if is_system_content %}
10
+ {{- raise_exception('System message cannot contain images.') }}
11
+ {%- endif %}
12
+ {%- if do_vision_count %}
13
+ {%- set image_count.value = image_count.value + 1 %}
14
+ {%- endif %}
15
+ {%- if add_vision_id %}
16
+ {{- 'Picture ' ~ image_count.value ~ ': ' }}
17
+ {%- endif %}
18
+ {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
19
+ {%- elif 'video' in item or item.type == 'video' %}
20
+ {%- if is_system_content %}
21
+ {{- raise_exception('System message cannot contain videos.') }}
22
+ {%- endif %}
23
+ {%- if do_vision_count %}
24
+ {%- set video_count.value = video_count.value + 1 %}
25
+ {%- endif %}
26
+ {%- if add_vision_id %}
27
+ {{- 'Video ' ~ video_count.value ~ ': ' }}
28
+ {%- endif %}
29
+ {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
30
+ {%- elif 'text' in item %}
31
+ {{- item.text }}
32
+ {%- else %}
33
+ {{- raise_exception('Unexpected item type in content.') }}
34
+ {%- endif %}
35
+ {%- endfor %}
36
+ {%- elif content is none or content is undefined %}
37
+ {{- '' }}
38
+ {%- else %}
39
+ {{- raise_exception('Unexpected content type.') }}
40
+ {%- endif %}
41
+ {%- endmacro %}
42
+ {%- if not messages %}
43
+ {{- raise_exception('No messages provided.') }}
44
+ {%- endif %}
45
+ {%- if tools and tools is iterable and tools is not mapping %}
46
+ {{- '<|im_start|>system\n' }}
47
+ {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
48
+ {%- for tool in tools %}
49
+ {{- "\n" }}
50
+ {{- tool | tojson }}
51
+ {%- endfor %}
52
+ {{- "\n</tools>" }}
53
+ {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
54
+ {%- if messages[0].role == 'system' %}
55
+ {%- set content = render_content(messages[0].content, false, true)|trim %}
56
+ {%- if content %}
57
+ {{- '\n\n' + content }}
58
+ {%- endif %}
59
+ {%- endif %}
60
+ {{- '<|im_end|>\n' }}
61
+ {%- else %}
62
+ {%- if messages[0].role == 'system' %}
63
+ {%- set content = render_content(messages[0].content, false, true)|trim %}
64
+ {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
65
+ {%- endif %}
66
+ {%- endif %}
67
+ {%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
68
+ {%- for message in messages[::-1] %}
69
+ {%- set index = (messages|length - 1) - loop.index0 %}
70
+ {%- if ns.multi_step_tool and message.role == "user" %}
71
+ {%- set content = render_content(message.content, false)|trim %}
72
+ {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
73
+ {%- set ns.multi_step_tool = false %}
74
+ {%- set ns.last_query_index = index %}
75
+ {%- endif %}
76
+ {%- endif %}
77
+ {%- endfor %}
78
+ {%- if ns.multi_step_tool %}
79
+ {{- raise_exception('No user query found in messages.') }}
80
+ {%- endif %}
81
+ {%- for message in messages %}
82
+ {%- set content = render_content(message.content, true)|trim %}
83
+ {%- if message.role == "system" %}
84
+ {%- if not loop.first %}
85
+ {{- raise_exception('System message must be at the beginning.') }}
86
+ {%- endif %}
87
+ {%- elif message.role == "user" %}
88
+ {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
89
+ {%- elif message.role == "assistant" %}
90
+ {%- set reasoning_content = '' %}
91
+ {%- if message.reasoning_content is string %}
92
+ {%- set reasoning_content = message.reasoning_content %}
93
+ {%- else %}
94
+ {%- if '</think>' in content %}
95
+ {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
96
+ {%- set content = content.split('</think>')[-1].lstrip('\n') %}
97
+ {%- endif %}
98
+ {%- endif %}
99
+ {%- set reasoning_content = reasoning_content|trim %}
100
+ {%- if loop.index0 > ns.last_query_index %}
101
+ {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
102
+ {%- else %}
103
+ {{- '<|im_start|>' + message.role + '\n' + content }}
104
+ {%- endif %}
105
+ {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
106
+ {%- for tool_call in message.tool_calls %}
107
+ {%- if tool_call.function is defined %}
108
+ {%- set tool_call = tool_call.function %}
109
+ {%- endif %}
110
+ {%- if loop.first %}
111
+ {%- if content|trim %}
112
+ {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
113
+ {%- else %}
114
+ {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
115
+ {%- endif %}
116
+ {%- else %}
117
+ {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
118
+ {%- endif %}
119
+ {%- if tool_call.arguments is defined %}
120
+ {%- for args_name, args_value in tool_call.arguments|items %}
121
+ {{- '<parameter=' + args_name + '>\n' }}
122
+ {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
123
+ {{- args_value }}
124
+ {{- '\n</parameter>\n' }}
125
+ {%- endfor %}
126
+ {%- endif %}
127
+ {{- '</function>\n</tool_call>' }}
128
+ {%- endfor %}
129
+ {%- endif %}
130
+ {{- '<|im_end|>\n' }}
131
+ {%- elif message.role == "tool" %}
132
+ {%- if loop.previtem and loop.previtem.role != "tool" %}
133
+ {{- '<|im_start|>user' }}
134
+ {%- endif %}
135
+ {{- '\n<tool_response>\n' }}
136
+ {{- content }}
137
+ {{- '\n</tool_response>' }}
138
+ {%- if not loop.last and loop.nextitem.role != "tool" %}
139
+ {{- '<|im_end|>\n' }}
140
+ {%- elif loop.last %}
141
+ {{- '<|im_end|>\n' }}
142
+ {%- endif %}
143
+ {%- else %}
144
+ {{- raise_exception('Unexpected message role.') }}
145
+ {%- endif %}
146
+ {%- endfor %}
147
+ {%- if add_generation_prompt %}
148
+ {{- '<|im_start|>assistant\n' }}
149
+ {%- if enable_thinking is defined and enable_thinking is true %}
150
+ {{- '<think>\n' }}
151
+ {%- else %}
152
+ {{- '<think>\n\n</think>\n\n' }}
153
+ {%- endif %}
154
+ {%- endif %}
substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3264/tokenizer_config.json ADDED
@@ -0,0 +1,31 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "audio_bos_token": "<|audio_start|>",
4
+ "audio_eos_token": "<|audio_end|>",
5
+ "audio_token": "<|audio_pad|>",
6
+ "backend": "tokenizers",
7
+ "bos_token": null,
8
+ "clean_up_tokenization_spaces": false,
9
+ "eos_token": "<|endoftext|>",
10
+ "errors": "replace",
11
+ "image_token": "<|image_pad|>",
12
+ "is_local": false,
13
+ "model_max_length": 262144,
14
+ "model_specific_special_tokens": {
15
+ "audio_bos_token": "<|audio_start|>",
16
+ "audio_eos_token": "<|audio_end|>",
17
+ "audio_token": "<|audio_pad|>",
18
+ "image_token": "<|image_pad|>",
19
+ "video_token": "<|video_pad|>",
20
+ "vision_bos_token": "<|vision_start|>",
21
+ "vision_eos_token": "<|vision_end|>"
22
+ },
23
+ "pad_token": "<|endoftext|>",
24
+ "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
25
+ "split_special_tokens": false,
26
+ "tokenizer_class": "TokenizersBackend",
27
+ "unk_token": null,
28
+ "video_token": "<|video_pad|>",
29
+ "vision_bos_token": "<|vision_start|>",
30
+ "vision_eos_token": "<|vision_end|>"
31
+ }
substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3264/trainer_state.json ADDED
@@ -0,0 +1,860 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 8.0,
6
+ "eval_steps": 500,
7
+ "global_step": 3264,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "entropy": 2.2408065378665922,
14
+ "epoch": 0.12269938650306748,
15
+ "grad_norm": 1.6718658208847046,
16
+ "learning_rate": 4.487804970912036e-05,
17
+ "loss": 2.129624786376953,
18
+ "mean_token_accuracy": 0.5743644836544991,
19
+ "num_tokens": 134329.0,
20
+ "step": 50
21
+ },
22
+ {
23
+ "entropy": 1.1078504550457,
24
+ "epoch": 0.24539877300613497,
25
+ "grad_norm": 2.4091813564300537,
26
+ "learning_rate": 9.067197798373296e-05,
27
+ "loss": 1.0346390533447265,
28
+ "mean_token_accuracy": 0.7085381114482879,
29
+ "num_tokens": 262407.0,
30
+ "step": 100
31
+ },
32
+ {
33
+ "entropy": 0.9421598988771439,
34
+ "epoch": 0.36809815950920244,
35
+ "grad_norm": 1.3913646936416626,
36
+ "learning_rate": 0.00013646590625834558,
37
+ "loss": 0.8716551208496094,
38
+ "mean_token_accuracy": 0.7449040985107422,
39
+ "num_tokens": 401136.0,
40
+ "step": 150
41
+ },
42
+ {
43
+ "entropy": 0.8723550146818161,
44
+ "epoch": 0.49079754601226994,
45
+ "grad_norm": 1.1664468050003052,
46
+ "learning_rate": 0.0001822598345329582,
47
+ "loss": 0.8095977783203125,
48
+ "mean_token_accuracy": 0.7585240352153778,
49
+ "num_tokens": 531647.0,
50
+ "step": 200
51
+ },
52
+ {
53
+ "entropy": 0.8466519421339035,
54
+ "epoch": 0.6134969325153374,
55
+ "grad_norm": 1.0145421028137207,
56
+ "learning_rate": 0.00022805376280757083,
57
+ "loss": 0.7830857849121093,
58
+ "mean_token_accuracy": 0.7661705195903779,
59
+ "num_tokens": 666127.0,
60
+ "step": 250
61
+ },
62
+ {
63
+ "entropy": 0.8180295366048813,
64
+ "epoch": 0.7361963190184049,
65
+ "grad_norm": 0.9731705784797668,
66
+ "learning_rate": 0.00027384769108218336,
67
+ "loss": 0.7573603820800782,
68
+ "mean_token_accuracy": 0.7704453605413437,
69
+ "num_tokens": 799292.0,
70
+ "step": 300
71
+ },
72
+ {
73
+ "entropy": 0.7990227991342544,
74
+ "epoch": 0.8588957055214724,
75
+ "grad_norm": 1.0642374753952026,
76
+ "learning_rate": 0.000319641619356796,
77
+ "loss": 0.73709228515625,
78
+ "mean_token_accuracy": 0.7752733880281448,
79
+ "num_tokens": 930614.0,
80
+ "step": 350
81
+ },
82
+ {
83
+ "entropy": 0.7899087685346603,
84
+ "epoch": 0.9815950920245399,
85
+ "grad_norm": 0.880613386631012,
86
+ "learning_rate": 0.0003654355476314086,
87
+ "loss": 0.7347319793701171,
88
+ "mean_token_accuracy": 0.7766731631755829,
89
+ "num_tokens": 1061654.0,
90
+ "step": 400
91
+ },
92
+ {
93
+ "epoch": 1.0,
94
+ "eval_entropy": 0.83522472347532,
95
+ "eval_mean_token_accuracy": 0.7484098076820374,
96
+ "eval_not_syn_loss": 0.7872886061668396,
97
+ "eval_not_syn_runtime": 54.8697,
98
+ "eval_not_syn_samples_per_second": 25.424,
99
+ "eval_not_syn_steps_per_second": 3.189,
100
+ "eval_num_tokens": 1080196.0,
101
+ "step": 408
102
+ },
103
+ {
104
+ "epoch": 1.0,
105
+ "eval_entropy": 0.7784442271505083,
106
+ "eval_mean_token_accuracy": 0.8021791185651507,
107
+ "eval_num_tokens": 1080196.0,
108
+ "eval_syn_loss": 0.6980098485946655,
109
+ "eval_syn_runtime": 56.6342,
110
+ "eval_syn_samples_per_second": 24.632,
111
+ "eval_syn_steps_per_second": 3.09,
112
+ "step": 408
113
+ },
114
+ {
115
+ "entropy": 0.75672573694075,
116
+ "epoch": 1.1030674846625768,
117
+ "grad_norm": 0.9244198203086853,
118
+ "learning_rate": 0.00037356351883456205,
119
+ "loss": 0.7048745727539063,
120
+ "mean_token_accuracy": 0.783016896609104,
121
+ "num_tokens": 1199170.0,
122
+ "step": 450
123
+ },
124
+ {
125
+ "entropy": 0.7406797379255294,
126
+ "epoch": 1.2257668711656442,
127
+ "grad_norm": 0.6744896769523621,
128
+ "learning_rate": 0.00037311248151765587,
129
+ "loss": 0.6860730743408203,
130
+ "mean_token_accuracy": 0.787369327545166,
131
+ "num_tokens": 1334298.0,
132
+ "step": 500
133
+ },
134
+ {
135
+ "entropy": 0.7424973237514496,
136
+ "epoch": 1.3484662576687116,
137
+ "grad_norm": 0.92305588722229,
138
+ "learning_rate": 0.000372320629215228,
139
+ "loss": 0.6855724334716797,
140
+ "mean_token_accuracy": 0.7871134179830551,
141
+ "num_tokens": 1463189.0,
142
+ "step": 550
143
+ },
144
+ {
145
+ "entropy": 0.7502243828773498,
146
+ "epoch": 1.471165644171779,
147
+ "grad_norm": 0.8259047269821167,
148
+ "learning_rate": 0.00037118941074037944,
149
+ "loss": 0.6949834442138672,
150
+ "mean_token_accuracy": 0.7851914083957672,
151
+ "num_tokens": 1590310.0,
152
+ "step": 600
153
+ },
154
+ {
155
+ "entropy": 0.7233260518312454,
156
+ "epoch": 1.5938650306748468,
157
+ "grad_norm": 0.882722795009613,
158
+ "learning_rate": 0.00036972089582775814,
159
+ "loss": 0.6649005889892579,
160
+ "mean_token_accuracy": 0.7898879665136337,
161
+ "num_tokens": 1721836.0,
162
+ "step": 650
163
+ },
164
+ {
165
+ "entropy": 0.7162708967924118,
166
+ "epoch": 1.716564417177914,
167
+ "grad_norm": 0.6593027114868164,
168
+ "learning_rate": 0.0003679177713466678,
169
+ "loss": 0.6653710174560546,
170
+ "mean_token_accuracy": 0.7905678844451904,
171
+ "num_tokens": 1853507.0,
172
+ "step": 700
173
+ },
174
+ {
175
+ "entropy": 0.7150520020723343,
176
+ "epoch": 1.8392638036809816,
177
+ "grad_norm": 0.665340781211853,
178
+ "learning_rate": 0.0003657833363850354,
179
+ "loss": 0.6600938415527344,
180
+ "mean_token_accuracy": 0.7918649202585221,
181
+ "num_tokens": 1988113.0,
182
+ "step": 750
183
+ },
184
+ {
185
+ "entropy": 0.6993386310338974,
186
+ "epoch": 1.961963190184049,
187
+ "grad_norm": 0.6502187252044678,
188
+ "learning_rate": 0.00036332149621323294,
189
+ "loss": 0.6469607543945313,
190
+ "mean_token_accuracy": 0.7950778317451477,
191
+ "num_tokens": 2121314.0,
192
+ "step": 800
193
+ },
194
+ {
195
+ "epoch": 2.0,
196
+ "eval_entropy": 0.7315139893123082,
197
+ "eval_mean_token_accuracy": 0.7713504004478454,
198
+ "eval_not_syn_loss": 0.699893593788147,
199
+ "eval_not_syn_runtime": 55.4135,
200
+ "eval_not_syn_samples_per_second": 25.174,
201
+ "eval_not_syn_steps_per_second": 3.158,
202
+ "eval_num_tokens": 2160392.0,
203
+ "step": 816
204
+ },
205
+ {
206
+ "epoch": 2.0,
207
+ "eval_entropy": 0.6848451798302787,
208
+ "eval_mean_token_accuracy": 0.814895657471248,
209
+ "eval_num_tokens": 2160392.0,
210
+ "eval_syn_loss": 0.6415576934814453,
211
+ "eval_syn_runtime": 56.4735,
212
+ "eval_syn_samples_per_second": 24.702,
213
+ "eval_syn_steps_per_second": 3.099,
214
+ "step": 816
215
+ },
216
+ {
217
+ "entropy": 0.6596363644407253,
218
+ "epoch": 2.083435582822086,
219
+ "grad_norm": 0.6647700071334839,
220
+ "learning_rate": 0.00036053675513879666,
221
+ "loss": 0.599151611328125,
222
+ "mean_token_accuracy": 0.8071986864311527,
223
+ "num_tokens": 2253179.0,
224
+ "step": 850
225
+ },
226
+ {
227
+ "entropy": 0.6626429110765457,
228
+ "epoch": 2.2061349693251535,
229
+ "grad_norm": 0.7356868386268616,
230
+ "learning_rate": 0.00035743420826511783,
231
+ "loss": 0.6004195404052735,
232
+ "mean_token_accuracy": 0.805362731218338,
233
+ "num_tokens": 2379965.0,
234
+ "step": 900
235
+ },
236
+ {
237
+ "entropy": 0.6503470712900161,
238
+ "epoch": 2.3288343558282207,
239
+ "grad_norm": 1.095608115196228,
240
+ "learning_rate": 0.0003540195321691835,
241
+ "loss": 0.5951333999633789,
242
+ "mean_token_accuracy": 0.807324025630951,
243
+ "num_tokens": 2515593.0,
244
+ "step": 950
245
+ },
246
+ {
247
+ "entropy": 0.660577870607376,
248
+ "epoch": 2.4515337423312884,
249
+ "grad_norm": 0.6602179408073425,
250
+ "learning_rate": 0.00035029897451542325,
251
+ "loss": 0.6056365585327148,
252
+ "mean_token_accuracy": 0.8057438349723816,
253
+ "num_tokens": 2642790.0,
254
+ "step": 1000
255
+ },
256
+ {
257
+ "entropy": 0.6555853113532066,
258
+ "epoch": 2.574233128834356,
259
+ "grad_norm": 0.7459151148796082,
260
+ "learning_rate": 0.00034627934262466606,
261
+ "loss": 0.5963204193115235,
262
+ "mean_token_accuracy": 0.8060598099231719,
263
+ "num_tokens": 2774396.0,
264
+ "step": 1050
265
+ },
266
+ {
267
+ "entropy": 0.6505853188037872,
268
+ "epoch": 2.6969325153374233,
269
+ "grad_norm": 0.6359792947769165,
270
+ "learning_rate": 0.00034196799101912103,
271
+ "loss": 0.5993065643310547,
272
+ "mean_token_accuracy": 0.806280387043953,
273
+ "num_tokens": 2912102.0,
274
+ "step": 1100
275
+ },
276
+ {
277
+ "entropy": 0.6509970253705979,
278
+ "epoch": 2.819631901840491,
279
+ "grad_norm": 0.8086174726486206,
280
+ "learning_rate": 0.0003373728079661707,
281
+ "loss": 0.6001268005371094,
282
+ "mean_token_accuracy": 0.806866540312767,
283
+ "num_tokens": 3042987.0,
284
+ "step": 1150
285
+ },
286
+ {
287
+ "entropy": 0.6404844307899475,
288
+ "epoch": 2.942331288343558,
289
+ "grad_norm": 0.7038407325744629,
290
+ "learning_rate": 0.0003325022010455975,
291
+ "loss": 0.5870526123046875,
292
+ "mean_token_accuracy": 0.8104586935043335,
293
+ "num_tokens": 3178044.0,
294
+ "step": 1200
295
+ },
296
+ {
297
+ "epoch": 3.0,
298
+ "eval_entropy": 0.6645996727262224,
299
+ "eval_mean_token_accuracy": 0.798654237134116,
300
+ "eval_not_syn_loss": 0.6586771011352539,
301
+ "eval_not_syn_runtime": 55.1088,
302
+ "eval_not_syn_samples_per_second": 25.314,
303
+ "eval_not_syn_steps_per_second": 3.176,
304
+ "eval_num_tokens": 3240588.0,
305
+ "step": 1224
306
+ },
307
+ {
308
+ "epoch": 3.0,
309
+ "eval_entropy": 0.6233607809884207,
310
+ "eval_mean_token_accuracy": 0.7977249867575509,
311
+ "eval_num_tokens": 3240588.0,
312
+ "eval_syn_loss": 0.6288298964500427,
313
+ "eval_syn_runtime": 56.6081,
314
+ "eval_syn_samples_per_second": 24.643,
315
+ "eval_syn_steps_per_second": 3.091,
316
+ "step": 1224
317
+ },
318
+ {
319
+ "entropy": 0.6081961104364106,
320
+ "epoch": 3.063803680981595,
321
+ "grad_norm": 0.6161203980445862,
322
+ "learning_rate": 0.0003273650817666497,
323
+ "loss": 0.5481716537475586,
324
+ "mean_token_accuracy": 0.8175542035488167,
325
+ "num_tokens": 3312797.0,
326
+ "step": 1250
327
+ },
328
+ {
329
+ "entropy": 0.5877064895629883,
330
+ "epoch": 3.1865030674846624,
331
+ "grad_norm": 0.6909636855125427,
332
+ "learning_rate": 0.00032197084926309234,
333
+ "loss": 0.5224150466918945,
334
+ "mean_token_accuracy": 0.8250984263420105,
335
+ "num_tokens": 3441629.0,
336
+ "step": 1300
337
+ },
338
+ {
339
+ "entropy": 0.5952656552195549,
340
+ "epoch": 3.30920245398773,
341
+ "grad_norm": 0.8147026300430298,
342
+ "learning_rate": 0.0003163293730960758,
343
+ "loss": 0.5349474334716797,
344
+ "mean_token_accuracy": 0.8213064837455749,
345
+ "num_tokens": 3576093.0,
346
+ "step": 1350
347
+ },
348
+ {
349
+ "entropy": 0.5937123569846153,
350
+ "epoch": 3.4319018404907977,
351
+ "grad_norm": 0.8690970540046692,
352
+ "learning_rate": 0.0003104509751962869,
353
+ "loss": 0.5310481262207031,
354
+ "mean_token_accuracy": 0.8229135835170746,
355
+ "num_tokens": 3705912.0,
356
+ "step": 1400
357
+ },
358
+ {
359
+ "entropy": 0.6067730465531349,
360
+ "epoch": 3.554601226993865,
361
+ "grad_norm": 0.7433330416679382,
362
+ "learning_rate": 0.0003043464109784207,
363
+ "loss": 0.5450045394897461,
364
+ "mean_token_accuracy": 0.8205251175165177,
365
+ "num_tokens": 3832376.0,
366
+ "step": 1450
367
+ },
368
+ {
369
+ "entropy": 0.5926914718747139,
370
+ "epoch": 3.6773006134969326,
371
+ "grad_norm": 0.7265587449073792,
372
+ "learning_rate": 0.00029802684966252974,
373
+ "loss": 0.5330168914794922,
374
+ "mean_token_accuracy": 0.821294127702713,
375
+ "num_tokens": 3967273.0,
376
+ "step": 1500
377
+ },
378
+ {
379
+ "entropy": 0.5881199997663498,
380
+ "epoch": 3.8,
381
+ "grad_norm": 0.7681817412376404,
382
+ "learning_rate": 0.00029150385383825187,
383
+ "loss": 0.5288929367065429,
384
+ "mean_token_accuracy": 0.8234161508083343,
385
+ "num_tokens": 4102434.0,
386
+ "step": 1550
387
+ },
388
+ {
389
+ "entropy": 0.6008686199784279,
390
+ "epoch": 3.9226993865030675,
391
+ "grad_norm": 0.9384056329727173,
392
+ "learning_rate": 0.0002847893583093113,
393
+ "loss": 0.5437511444091797,
394
+ "mean_token_accuracy": 0.8201366275548935,
395
+ "num_tokens": 4234037.0,
396
+ "step": 1600
397
+ },
398
+ {
399
+ "epoch": 4.0,
400
+ "eval_entropy": 0.6218148783275059,
401
+ "eval_mean_token_accuracy": 0.783796865599496,
402
+ "eval_not_syn_loss": 0.6558170318603516,
403
+ "eval_not_syn_runtime": 55.429,
404
+ "eval_not_syn_samples_per_second": 25.167,
405
+ "eval_not_syn_steps_per_second": 3.157,
406
+ "eval_num_tokens": 4320784.0,
407
+ "step": 1632
408
+ },
409
+ {
410
+ "epoch": 4.0,
411
+ "eval_entropy": 0.5833754854542869,
412
+ "eval_mean_token_accuracy": 0.8278223422595433,
413
+ "eval_num_tokens": 4320784.0,
414
+ "eval_syn_loss": 0.604770302772522,
415
+ "eval_syn_runtime": 56.9593,
416
+ "eval_syn_samples_per_second": 24.491,
417
+ "eval_syn_steps_per_second": 3.072,
418
+ "step": 1632
419
+ },
420
+ {
421
+ "entropy": 0.559690699733869,
422
+ "epoch": 4.044171779141104,
423
+ "grad_norm": 0.65739905834198,
424
+ "learning_rate": 0.0002778956482569963,
425
+ "loss": 0.49702674865722657,
426
+ "mean_token_accuracy": 0.8317034166268628,
427
+ "num_tokens": 4368268.0,
428
+ "step": 1650
429
+ },
430
+ {
431
+ "entropy": 0.5401161375641823,
432
+ "epoch": 4.166871165644172,
433
+ "grad_norm": 1.5176353454589844,
434
+ "learning_rate": 0.0002708353367625691,
435
+ "loss": 0.4598746109008789,
436
+ "mean_token_accuracy": 0.8412033343315124,
437
+ "num_tokens": 4494525.0,
438
+ "step": 1700
439
+ },
440
+ {
441
+ "entropy": 0.5299916860461235,
442
+ "epoch": 4.289570552147239,
443
+ "grad_norm": 0.685965359210968,
444
+ "learning_rate": 0.0002636213417297333,
445
+ "loss": 0.45663665771484374,
446
+ "mean_token_accuracy": 0.8412413811683654,
447
+ "num_tokens": 4631985.0,
448
+ "step": 1750
449
+ },
450
+ {
451
+ "entropy": 0.5352088099718094,
452
+ "epoch": 4.412269938650307,
453
+ "grad_norm": 0.8471158742904663,
454
+ "learning_rate": 0.0002562668622493824,
455
+ "loss": 0.4625707626342773,
456
+ "mean_token_accuracy": 0.8400991547107697,
457
+ "num_tokens": 4766323.0,
458
+ "step": 1800
459
+ },
460
+ {
461
+ "entropy": 0.5425067806243896,
462
+ "epoch": 4.534969325153375,
463
+ "grad_norm": 0.8595437407493591,
464
+ "learning_rate": 0.00024878535444987474,
465
+ "loss": 0.47124954223632814,
466
+ "mean_token_accuracy": 0.8376108628511428,
467
+ "num_tokens": 4895606.0,
468
+ "step": 1850
469
+ },
470
+ {
471
+ "entropy": 0.5419312778115273,
472
+ "epoch": 4.6576687116564415,
473
+ "grad_norm": 0.8436959981918335,
474
+ "learning_rate": 0.00024119050687701818,
475
+ "loss": 0.4708108901977539,
476
+ "mean_token_accuracy": 0.8390797019004822,
477
+ "num_tokens": 5025912.0,
478
+ "step": 1900
479
+ },
480
+ {
481
+ "entropy": 0.5392271488904953,
482
+ "epoch": 4.780368098159509,
483
+ "grad_norm": 0.8691944479942322,
484
+ "learning_rate": 0.00023349621544881404,
485
+ "loss": 0.46682029724121094,
486
+ "mean_token_accuracy": 0.8385172665119172,
487
+ "num_tokens": 5162972.0,
488
+ "step": 1950
489
+ },
490
+ {
491
+ "entropy": 0.5332251599431038,
492
+ "epoch": 4.903067484662577,
493
+ "grad_norm": 0.6814598441123962,
494
+ "learning_rate": 0.00022571655803078078,
495
+ "loss": 0.4687493896484375,
496
+ "mean_token_accuracy": 0.8388861840963364,
497
+ "num_tokens": 5297935.0,
498
+ "step": 2000
499
+ },
500
+ {
501
+ "epoch": 5.0,
502
+ "eval_entropy": 0.5847777877535139,
503
+ "eval_mean_token_accuracy": 0.7782382583618164,
504
+ "eval_not_syn_loss": 0.6678160429000854,
505
+ "eval_not_syn_runtime": 55.1467,
506
+ "eval_not_syn_samples_per_second": 25.296,
507
+ "eval_not_syn_steps_per_second": 3.173,
508
+ "eval_num_tokens": 5400980.0,
509
+ "step": 2040
510
+ },
511
+ {
512
+ "epoch": 5.0,
513
+ "eval_entropy": 0.5505962218557086,
514
+ "eval_mean_token_accuracy": 0.8308270008223397,
515
+ "eval_num_tokens": 5400980.0,
516
+ "eval_syn_loss": 0.606867253780365,
517
+ "eval_syn_runtime": 56.9768,
518
+ "eval_syn_samples_per_second": 24.484,
519
+ "eval_syn_steps_per_second": 3.071,
520
+ "step": 2040
521
+ },
522
+ {
523
+ "entropy": 0.533086326688227,
524
+ "epoch": 5.024539877300613,
525
+ "grad_norm": 0.9453939199447632,
526
+ "learning_rate": 0.00021786576867837776,
527
+ "loss": 0.45802555084228513,
528
+ "mean_token_accuracy": 0.8429379854539428,
529
+ "num_tokens": 5427362.0,
530
+ "step": 2050
531
+ },
532
+ {
533
+ "entropy": 0.46721911430358887,
534
+ "epoch": 5.147239263803681,
535
+ "grad_norm": 0.9237910509109497,
536
+ "learning_rate": 0.00020995821159365647,
537
+ "loss": 0.37914146423339845,
538
+ "mean_token_accuracy": 0.8648965793848038,
539
+ "num_tokens": 5564186.0,
540
+ "step": 2100
541
+ },
542
+ {
543
+ "entropy": 0.4682381117343903,
544
+ "epoch": 5.269938650306749,
545
+ "grad_norm": 0.7424130439758301,
546
+ "learning_rate": 0.0002020083548437879,
547
+ "loss": 0.3859384536743164,
548
+ "mean_token_accuracy": 0.8622810471057892,
549
+ "num_tokens": 5694728.0,
550
+ "step": 2150
551
+ },
552
+ {
553
+ "entropy": 0.464998793900013,
554
+ "epoch": 5.392638036809816,
555
+ "grad_norm": 0.9224902987480164,
556
+ "learning_rate": 0.0001940307438895544,
557
+ "loss": 0.3839593505859375,
558
+ "mean_token_accuracy": 0.8627353543043137,
559
+ "num_tokens": 5832581.0,
560
+ "step": 2200
561
+ },
562
+ {
563
+ "entropy": 0.4753906837105751,
564
+ "epoch": 5.515337423312883,
565
+ "grad_norm": 0.8931427001953125,
566
+ "learning_rate": 0.0001860399749722367,
567
+ "loss": 0.393004264831543,
568
+ "mean_token_accuracy": 0.8586806547641754,
569
+ "num_tokens": 5962417.0,
570
+ "step": 2250
571
+ },
572
+ {
573
+ "entropy": 0.4671635267138481,
574
+ "epoch": 5.638036809815951,
575
+ "grad_norm": 0.9559172987937927,
576
+ "learning_rate": 0.00017805066840759193,
577
+ "loss": 0.38852767944335936,
578
+ "mean_token_accuracy": 0.8615026396512985,
579
+ "num_tokens": 6092338.0,
580
+ "step": 2300
581
+ },
582
+ {
583
+ "entropy": 0.4746100226044655,
584
+ "epoch": 5.7607361963190185,
585
+ "grad_norm": 0.9070253968238831,
586
+ "learning_rate": 0.00017007744183578254,
587
+ "loss": 0.3960166168212891,
588
+ "mean_token_accuracy": 0.8591635781526565,
589
+ "num_tokens": 6226327.0,
590
+ "step": 2350
591
+ },
592
+ {
593
+ "entropy": 0.4698092398047447,
594
+ "epoch": 5.883435582822086,
595
+ "grad_norm": 0.8354232907295227,
596
+ "learning_rate": 0.0001621348834762011,
597
+ "loss": 0.39321571350097656,
598
+ "mean_token_accuracy": 0.8603039687871933,
599
+ "num_tokens": 6359083.0,
600
+ "step": 2400
601
+ },
602
+ {
603
+ "epoch": 6.0,
604
+ "eval_entropy": 0.536405998979296,
605
+ "eval_mean_token_accuracy": 0.790239520413535,
606
+ "eval_not_syn_loss": 0.6884409785270691,
607
+ "eval_not_syn_runtime": 54.9927,
608
+ "eval_not_syn_samples_per_second": 25.367,
609
+ "eval_not_syn_steps_per_second": 3.182,
610
+ "eval_num_tokens": 6481176.0,
611
+ "step": 2448
612
+ },
613
+ {
614
+ "epoch": 6.0,
615
+ "eval_entropy": 0.5039791010107313,
616
+ "eval_mean_token_accuracy": 0.8163431504794529,
617
+ "eval_num_tokens": 6481176.0,
618
+ "eval_syn_loss": 0.6377155184745789,
619
+ "eval_syn_runtime": 57.196,
620
+ "eval_syn_samples_per_second": 24.39,
621
+ "eval_syn_steps_per_second": 3.06,
622
+ "step": 2448
623
+ },
624
+ {
625
+ "entropy": 0.4671963176341972,
626
+ "epoch": 6.004907975460123,
627
+ "grad_norm": 0.8445424437522888,
628
+ "learning_rate": 0.00015423752543612516,
629
+ "loss": 0.3872825241088867,
630
+ "mean_token_accuracy": 0.8618519083418027,
631
+ "num_tokens": 6486549.0,
632
+ "step": 2450
633
+ },
634
+ {
635
+ "entropy": 0.3890652832388878,
636
+ "epoch": 6.12760736196319,
637
+ "grad_norm": 0.725812554359436,
638
+ "learning_rate": 0.00014639981712203683,
639
+ "loss": 0.2923568153381348,
640
+ "mean_token_accuracy": 0.8928290349245072,
641
+ "num_tokens": 6622384.0,
642
+ "step": 2500
643
+ },
644
+ {
645
+ "entropy": 0.39649384289979933,
646
+ "epoch": 6.250306748466258,
647
+ "grad_norm": 0.9143152236938477,
648
+ "learning_rate": 0.000138636098802257,
649
+ "loss": 0.303057861328125,
650
+ "mean_token_accuracy": 0.8888710218667984,
651
+ "num_tokens": 6758637.0,
652
+ "step": 2550
653
+ },
654
+ {
655
+ "entropy": 0.4020465177297592,
656
+ "epoch": 6.373006134969325,
657
+ "grad_norm": 0.9407342076301575,
658
+ "learning_rate": 0.00013096057536926304,
659
+ "loss": 0.3073894691467285,
660
+ "mean_token_accuracy": 0.8894575881958008,
661
+ "num_tokens": 6888533.0,
662
+ "step": 2600
663
+ },
664
+ {
665
+ "entropy": 0.397317076921463,
666
+ "epoch": 6.495705521472392,
667
+ "grad_norm": 0.8517777323722839,
668
+ "learning_rate": 0.00012338729034969828,
669
+ "loss": 0.30508209228515626,
670
+ "mean_token_accuracy": 0.8886856293678284,
671
+ "num_tokens": 7023729.0,
672
+ "step": 2650
673
+ },
674
+ {
675
+ "entropy": 0.3915065750479698,
676
+ "epoch": 6.61840490797546,
677
+ "grad_norm": 0.862319827079773,
678
+ "learning_rate": 0.00011593010020962378,
679
+ "loss": 0.30237462997436526,
680
+ "mean_token_accuracy": 0.8897094148397445,
681
+ "num_tokens": 7153821.0,
682
+ "step": 2700
683
+ },
684
+ {
685
+ "entropy": 0.3977497002482414,
686
+ "epoch": 6.741104294478528,
687
+ "grad_norm": 0.8821370601654053,
688
+ "learning_rate": 0.00010860264900202563,
689
+ "loss": 0.3067088508605957,
690
+ "mean_token_accuracy": 0.8873285222053527,
691
+ "num_tokens": 7283513.0,
692
+ "step": 2750
693
+ },
694
+ {
695
+ "entropy": 0.4002616259455681,
696
+ "epoch": 6.863803680981595,
697
+ "grad_norm": 1.069462776184082,
698
+ "learning_rate": 0.0001014183434029644,
699
+ "loss": 0.30735193252563475,
700
+ "mean_token_accuracy": 0.8880347841978073,
701
+ "num_tokens": 7411578.0,
702
+ "step": 2800
703
+ },
704
+ {
705
+ "entropy": 0.3951660317182541,
706
+ "epoch": 6.986503067484662,
707
+ "grad_norm": 0.8845367431640625,
708
+ "learning_rate": 9.439032818204105e-05,
709
+ "loss": 0.30523622512817383,
710
+ "mean_token_accuracy": 0.8894886714220047,
711
+ "num_tokens": 7548356.0,
712
+ "step": 2850
713
+ },
714
+ {
715
+ "epoch": 7.0,
716
+ "eval_entropy": 0.4834100874832698,
717
+ "eval_mean_token_accuracy": 0.788761088166918,
718
+ "eval_not_syn_loss": 0.7379753589630127,
719
+ "eval_not_syn_runtime": 55.6309,
720
+ "eval_not_syn_samples_per_second": 25.076,
721
+ "eval_not_syn_steps_per_second": 3.146,
722
+ "eval_num_tokens": 7561372.0,
723
+ "step": 2856
724
+ },
725
+ {
726
+ "epoch": 7.0,
727
+ "eval_entropy": 0.453818735906056,
728
+ "eval_mean_token_accuracy": 0.8209771479879107,
729
+ "eval_num_tokens": 7561372.0,
730
+ "eval_syn_loss": 0.6787521243095398,
731
+ "eval_syn_runtime": 56.9966,
732
+ "eval_syn_samples_per_second": 24.475,
733
+ "eval_syn_steps_per_second": 3.07,
734
+ "step": 2856
735
+ },
736
+ {
737
+ "entropy": 0.3386827930055483,
738
+ "epoch": 7.1079754601227,
739
+ "grad_norm": 1.0214653015136719,
740
+ "learning_rate": 8.753146215206042e-05,
741
+ "loss": 0.2361087417602539,
742
+ "mean_token_accuracy": 0.9141153067049353,
743
+ "num_tokens": 7671657.0,
744
+ "step": 2900
745
+ },
746
+ {
747
+ "entropy": 0.31863692075014116,
748
+ "epoch": 7.230674846625767,
749
+ "grad_norm": 0.9141477346420288,
750
+ "learning_rate": 8.085429464189586e-05,
751
+ "loss": 0.21914518356323243,
752
+ "mean_token_accuracy": 0.9194145703315735,
753
+ "num_tokens": 7806989.0,
754
+ "step": 2950
755
+ },
756
+ {
757
+ "entropy": 0.31718266174197196,
758
+ "epoch": 7.353374233128834,
759
+ "grad_norm": 0.9127821922302246,
760
+ "learning_rate": 7.437104253560078e-05,
761
+ "loss": 0.22112932205200195,
762
+ "mean_token_accuracy": 0.9190003037452698,
763
+ "num_tokens": 7941760.0,
764
+ "step": 3000
765
+ },
766
+ {
767
+ "entropy": 0.3143314927816391,
768
+ "epoch": 7.476073619631902,
769
+ "grad_norm": 1.1398990154266357,
770
+ "learning_rate": 6.809356791977874e-05,
771
+ "loss": 0.2191557502746582,
772
+ "mean_token_accuracy": 0.918931410908699,
773
+ "num_tokens": 8080328.0,
774
+ "step": 3050
775
+ },
776
+ {
777
+ "entropy": 0.32102111548185347,
778
+ "epoch": 7.598773006134969,
779
+ "grad_norm": 1.069332480430603,
780
+ "learning_rate": 6.203335638010852e-05,
781
+ "loss": 0.22624496459960938,
782
+ "mean_token_accuracy": 0.9173947256803513,
783
+ "num_tokens": 8210831.0,
784
+ "step": 3100
785
+ },
786
+ {
787
+ "entropy": 0.31791565626859664,
788
+ "epoch": 7.721472392638037,
789
+ "grad_norm": 1.1536624431610107,
790
+ "learning_rate": 5.62014959867348e-05,
791
+ "loss": 0.22338006973266603,
792
+ "mean_token_accuracy": 0.9177713346481323,
793
+ "num_tokens": 8344985.0,
794
+ "step": 3150
795
+ },
796
+ {
797
+ "entropy": 0.3209437537193298,
798
+ "epoch": 7.844171779141105,
799
+ "grad_norm": 1.1040980815887451,
800
+ "learning_rate": 5.06086570069728e-05,
801
+ "loss": 0.2263322639465332,
802
+ "mean_token_accuracy": 0.9170012313127518,
803
+ "num_tokens": 8475716.0,
804
+ "step": 3200
805
+ },
806
+ {
807
+ "entropy": 0.31527657777071,
808
+ "epoch": 7.9668711656441715,
809
+ "grad_norm": 1.1415972709655762,
810
+ "learning_rate": 4.526507238244653e-05,
811
+ "loss": 0.2216329574584961,
812
+ "mean_token_accuracy": 0.918340762257576,
813
+ "num_tokens": 8606709.0,
814
+ "step": 3250
815
+ },
816
+ {
817
+ "epoch": 8.0,
818
+ "eval_entropy": 0.43602277857916694,
819
+ "eval_mean_token_accuracy": 0.7907402624402727,
820
+ "eval_not_syn_loss": 0.8163036704063416,
821
+ "eval_not_syn_runtime": 55.3868,
822
+ "eval_not_syn_samples_per_second": 25.187,
823
+ "eval_not_syn_steps_per_second": 3.16,
824
+ "eval_num_tokens": 8641568.0,
825
+ "step": 3264
826
+ },
827
+ {
828
+ "epoch": 8.0,
829
+ "eval_entropy": 0.4093018649305616,
830
+ "eval_mean_token_accuracy": 0.8147984709058489,
831
+ "eval_num_tokens": 8641568.0,
832
+ "eval_syn_loss": 0.7571278810501099,
833
+ "eval_syn_runtime": 56.797,
834
+ "eval_syn_samples_per_second": 24.561,
835
+ "eval_syn_steps_per_second": 3.081,
836
+ "step": 3264
837
+ }
838
+ ],
839
+ "logging_steps": 50,
840
+ "max_steps": 4080,
841
+ "num_input_tokens_seen": 0,
842
+ "num_train_epochs": 10,
843
+ "save_steps": 500,
844
+ "stateful_callbacks": {
845
+ "TrainerControl": {
846
+ "args": {
847
+ "should_epoch_stop": false,
848
+ "should_evaluate": false,
849
+ "should_log": false,
850
+ "should_save": true,
851
+ "should_training_stop": false
852
+ },
853
+ "attributes": {}
854
+ }
855
+ },
856
+ "total_flos": 1.7177733506397658e+17,
857
+ "train_batch_size": 8,
858
+ "trial_name": null,
859
+ "trial_params": null
860
+ }
substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3672/README.md ADDED
@@ -0,0 +1,209 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: Qwen/Qwen3.5-2B-Base
3
+ library_name: peft
4
+ pipeline_tag: text-generation
5
+ tags:
6
+ - base_model:adapter:Qwen/Qwen3.5-2B-Base
7
+ - lora
8
+ - sft
9
+ - transformers
10
+ - trl
11
+ ---
12
+
13
+ # Model Card for Model ID
14
+
15
+ <!-- Provide a quick summary of what the model is/does. -->
16
+
17
+
18
+
19
+ ## Model Details
20
+
21
+ ### Model Description
22
+
23
+ <!-- Provide a longer summary of what this model is. -->
24
+
25
+
26
+
27
+ - **Developed by:** [More Information Needed]
28
+ - **Funded by [optional]:** [More Information Needed]
29
+ - **Shared by [optional]:** [More Information Needed]
30
+ - **Model type:** [More Information Needed]
31
+ - **Language(s) (NLP):** [More Information Needed]
32
+ - **License:** [More Information Needed]
33
+ - **Finetuned from model [optional]:** [More Information Needed]
34
+
35
+ ### Model Sources [optional]
36
+
37
+ <!-- Provide the basic links for the model. -->
38
+
39
+ - **Repository:** [More Information Needed]
40
+ - **Paper [optional]:** [More Information Needed]
41
+ - **Demo [optional]:** [More Information Needed]
42
+
43
+ ## Uses
44
+
45
+ <!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
46
+
47
+ ### Direct Use
48
+
49
+ <!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
50
+
51
+ [More Information Needed]
52
+
53
+ ### Downstream Use [optional]
54
+
55
+ <!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
56
+
57
+ [More Information Needed]
58
+
59
+ ### Out-of-Scope Use
60
+
61
+ <!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
62
+
63
+ [More Information Needed]
64
+
65
+ ## Bias, Risks, and Limitations
66
+
67
+ <!-- This section is meant to convey both technical and sociotechnical limitations. -->
68
+
69
+ [More Information Needed]
70
+
71
+ ### Recommendations
72
+
73
+ <!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
74
+
75
+ Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
76
+
77
+ ## How to Get Started with the Model
78
+
79
+ Use the code below to get started with the model.
80
+
81
+ [More Information Needed]
82
+
83
+ ## Training Details
84
+
85
+ ### Training Data
86
+
87
+ <!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
88
+
89
+ [More Information Needed]
90
+
91
+ ### Training Procedure
92
+
93
+ <!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
94
+
95
+ #### Preprocessing [optional]
96
+
97
+ [More Information Needed]
98
+
99
+
100
+ #### Training Hyperparameters
101
+
102
+ - **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
103
+
104
+ #### Speeds, Sizes, Times [optional]
105
+
106
+ <!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
107
+
108
+ [More Information Needed]
109
+
110
+ ## Evaluation
111
+
112
+ <!-- This section describes the evaluation protocols and provides the results. -->
113
+
114
+ ### Testing Data, Factors & Metrics
115
+
116
+ #### Testing Data
117
+
118
+ <!-- This should link to a Dataset Card if possible. -->
119
+
120
+ [More Information Needed]
121
+
122
+ #### Factors
123
+
124
+ <!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
125
+
126
+ [More Information Needed]
127
+
128
+ #### Metrics
129
+
130
+ <!-- These are the evaluation metrics being used, ideally with a description of why. -->
131
+
132
+ [More Information Needed]
133
+
134
+ ### Results
135
+
136
+ [More Information Needed]
137
+
138
+ #### Summary
139
+
140
+
141
+
142
+ ## Model Examination [optional]
143
+
144
+ <!-- Relevant interpretability work for the model goes here -->
145
+
146
+ [More Information Needed]
147
+
148
+ ## Environmental Impact
149
+
150
+ <!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
151
+
152
+ Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
153
+
154
+ - **Hardware Type:** [More Information Needed]
155
+ - **Hours used:** [More Information Needed]
156
+ - **Cloud Provider:** [More Information Needed]
157
+ - **Compute Region:** [More Information Needed]
158
+ - **Carbon Emitted:** [More Information Needed]
159
+
160
+ ## Technical Specifications [optional]
161
+
162
+ ### Model Architecture and Objective
163
+
164
+ [More Information Needed]
165
+
166
+ ### Compute Infrastructure
167
+
168
+ [More Information Needed]
169
+
170
+ #### Hardware
171
+
172
+ [More Information Needed]
173
+
174
+ #### Software
175
+
176
+ [More Information Needed]
177
+
178
+ ## Citation [optional]
179
+
180
+ <!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
181
+
182
+ **BibTeX:**
183
+
184
+ [More Information Needed]
185
+
186
+ **APA:**
187
+
188
+ [More Information Needed]
189
+
190
+ ## Glossary [optional]
191
+
192
+ <!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
193
+
194
+ [More Information Needed]
195
+
196
+ ## More Information [optional]
197
+
198
+ [More Information Needed]
199
+
200
+ ## Model Card Authors [optional]
201
+
202
+ [More Information Needed]
203
+
204
+ ## Model Card Contact
205
+
206
+ [More Information Needed]
207
+ ### Framework versions
208
+
209
+ - PEFT 0.19.1
substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3672/adapter_config.json ADDED
@@ -0,0 +1,48 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alora_invocation_tokens": null,
3
+ "alpha_pattern": {},
4
+ "arrow_config": null,
5
+ "auto_mapping": null,
6
+ "base_model_name_or_path": "Qwen/Qwen3.5-2B-Base",
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": 32,
20
+ "lora_bias": false,
21
+ "lora_dropout": 0.08602048083239339,
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": 16,
30
+ "rank_pattern": {},
31
+ "revision": null,
32
+ "target_modules": [
33
+ "q_proj",
34
+ "k_proj",
35
+ "gate_proj",
36
+ "up_proj",
37
+ "o_proj",
38
+ "down_proj",
39
+ "v_proj"
40
+ ],
41
+ "target_parameters": null,
42
+ "task_type": "CAUSAL_LM",
43
+ "trainable_token_indices": null,
44
+ "use_bdlora": null,
45
+ "use_dora": false,
46
+ "use_qalora": false,
47
+ "use_rslora": false
48
+ }
substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3672/chat_template.jinja ADDED
@@ -0,0 +1,154 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {%- set image_count = namespace(value=0) %}
2
+ {%- set video_count = namespace(value=0) %}
3
+ {%- macro render_content(content, do_vision_count, is_system_content=false) %}
4
+ {%- if content is string %}
5
+ {{- content }}
6
+ {%- elif content is iterable and content is not mapping %}
7
+ {%- for item in content %}
8
+ {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
9
+ {%- if is_system_content %}
10
+ {{- raise_exception('System message cannot contain images.') }}
11
+ {%- endif %}
12
+ {%- if do_vision_count %}
13
+ {%- set image_count.value = image_count.value + 1 %}
14
+ {%- endif %}
15
+ {%- if add_vision_id %}
16
+ {{- 'Picture ' ~ image_count.value ~ ': ' }}
17
+ {%- endif %}
18
+ {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
19
+ {%- elif 'video' in item or item.type == 'video' %}
20
+ {%- if is_system_content %}
21
+ {{- raise_exception('System message cannot contain videos.') }}
22
+ {%- endif %}
23
+ {%- if do_vision_count %}
24
+ {%- set video_count.value = video_count.value + 1 %}
25
+ {%- endif %}
26
+ {%- if add_vision_id %}
27
+ {{- 'Video ' ~ video_count.value ~ ': ' }}
28
+ {%- endif %}
29
+ {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
30
+ {%- elif 'text' in item %}
31
+ {{- item.text }}
32
+ {%- else %}
33
+ {{- raise_exception('Unexpected item type in content.') }}
34
+ {%- endif %}
35
+ {%- endfor %}
36
+ {%- elif content is none or content is undefined %}
37
+ {{- '' }}
38
+ {%- else %}
39
+ {{- raise_exception('Unexpected content type.') }}
40
+ {%- endif %}
41
+ {%- endmacro %}
42
+ {%- if not messages %}
43
+ {{- raise_exception('No messages provided.') }}
44
+ {%- endif %}
45
+ {%- if tools and tools is iterable and tools is not mapping %}
46
+ {{- '<|im_start|>system\n' }}
47
+ {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
48
+ {%- for tool in tools %}
49
+ {{- "\n" }}
50
+ {{- tool | tojson }}
51
+ {%- endfor %}
52
+ {{- "\n</tools>" }}
53
+ {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
54
+ {%- if messages[0].role == 'system' %}
55
+ {%- set content = render_content(messages[0].content, false, true)|trim %}
56
+ {%- if content %}
57
+ {{- '\n\n' + content }}
58
+ {%- endif %}
59
+ {%- endif %}
60
+ {{- '<|im_end|>\n' }}
61
+ {%- else %}
62
+ {%- if messages[0].role == 'system' %}
63
+ {%- set content = render_content(messages[0].content, false, true)|trim %}
64
+ {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
65
+ {%- endif %}
66
+ {%- endif %}
67
+ {%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
68
+ {%- for message in messages[::-1] %}
69
+ {%- set index = (messages|length - 1) - loop.index0 %}
70
+ {%- if ns.multi_step_tool and message.role == "user" %}
71
+ {%- set content = render_content(message.content, false)|trim %}
72
+ {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
73
+ {%- set ns.multi_step_tool = false %}
74
+ {%- set ns.last_query_index = index %}
75
+ {%- endif %}
76
+ {%- endif %}
77
+ {%- endfor %}
78
+ {%- if ns.multi_step_tool %}
79
+ {{- raise_exception('No user query found in messages.') }}
80
+ {%- endif %}
81
+ {%- for message in messages %}
82
+ {%- set content = render_content(message.content, true)|trim %}
83
+ {%- if message.role == "system" %}
84
+ {%- if not loop.first %}
85
+ {{- raise_exception('System message must be at the beginning.') }}
86
+ {%- endif %}
87
+ {%- elif message.role == "user" %}
88
+ {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
89
+ {%- elif message.role == "assistant" %}
90
+ {%- set reasoning_content = '' %}
91
+ {%- if message.reasoning_content is string %}
92
+ {%- set reasoning_content = message.reasoning_content %}
93
+ {%- else %}
94
+ {%- if '</think>' in content %}
95
+ {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
96
+ {%- set content = content.split('</think>')[-1].lstrip('\n') %}
97
+ {%- endif %}
98
+ {%- endif %}
99
+ {%- set reasoning_content = reasoning_content|trim %}
100
+ {%- if loop.index0 > ns.last_query_index %}
101
+ {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
102
+ {%- else %}
103
+ {{- '<|im_start|>' + message.role + '\n' + content }}
104
+ {%- endif %}
105
+ {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
106
+ {%- for tool_call in message.tool_calls %}
107
+ {%- if tool_call.function is defined %}
108
+ {%- set tool_call = tool_call.function %}
109
+ {%- endif %}
110
+ {%- if loop.first %}
111
+ {%- if content|trim %}
112
+ {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
113
+ {%- else %}
114
+ {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
115
+ {%- endif %}
116
+ {%- else %}
117
+ {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
118
+ {%- endif %}
119
+ {%- if tool_call.arguments is defined %}
120
+ {%- for args_name, args_value in tool_call.arguments|items %}
121
+ {{- '<parameter=' + args_name + '>\n' }}
122
+ {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
123
+ {{- args_value }}
124
+ {{- '\n</parameter>\n' }}
125
+ {%- endfor %}
126
+ {%- endif %}
127
+ {{- '</function>\n</tool_call>' }}
128
+ {%- endfor %}
129
+ {%- endif %}
130
+ {{- '<|im_end|>\n' }}
131
+ {%- elif message.role == "tool" %}
132
+ {%- if loop.previtem and loop.previtem.role != "tool" %}
133
+ {{- '<|im_start|>user' }}
134
+ {%- endif %}
135
+ {{- '\n<tool_response>\n' }}
136
+ {{- content }}
137
+ {{- '\n</tool_response>' }}
138
+ {%- if not loop.last and loop.nextitem.role != "tool" %}
139
+ {{- '<|im_end|>\n' }}
140
+ {%- elif loop.last %}
141
+ {{- '<|im_end|>\n' }}
142
+ {%- endif %}
143
+ {%- else %}
144
+ {{- raise_exception('Unexpected message role.') }}
145
+ {%- endif %}
146
+ {%- endfor %}
147
+ {%- if add_generation_prompt %}
148
+ {{- '<|im_start|>assistant\n' }}
149
+ {%- if enable_thinking is defined and enable_thinking is true %}
150
+ {{- '<think>\n' }}
151
+ {%- else %}
152
+ {{- '<think>\n\n</think>\n\n' }}
153
+ {%- endif %}
154
+ {%- endif %}
substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3672/tokenizer_config.json ADDED
@@ -0,0 +1,31 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "audio_bos_token": "<|audio_start|>",
4
+ "audio_eos_token": "<|audio_end|>",
5
+ "audio_token": "<|audio_pad|>",
6
+ "backend": "tokenizers",
7
+ "bos_token": null,
8
+ "clean_up_tokenization_spaces": false,
9
+ "eos_token": "<|endoftext|>",
10
+ "errors": "replace",
11
+ "image_token": "<|image_pad|>",
12
+ "is_local": false,
13
+ "model_max_length": 262144,
14
+ "model_specific_special_tokens": {
15
+ "audio_bos_token": "<|audio_start|>",
16
+ "audio_eos_token": "<|audio_end|>",
17
+ "audio_token": "<|audio_pad|>",
18
+ "image_token": "<|image_pad|>",
19
+ "video_token": "<|video_pad|>",
20
+ "vision_bos_token": "<|vision_start|>",
21
+ "vision_eos_token": "<|vision_end|>"
22
+ },
23
+ "pad_token": "<|endoftext|>",
24
+ "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
25
+ "split_special_tokens": false,
26
+ "tokenizer_class": "TokenizersBackend",
27
+ "unk_token": null,
28
+ "video_token": "<|video_pad|>",
29
+ "vision_bos_token": "<|vision_start|>",
30
+ "vision_eos_token": "<|vision_end|>"
31
+ }
substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3672/trainer_state.json ADDED
@@ -0,0 +1,962 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 9.0,
6
+ "eval_steps": 500,
7
+ "global_step": 3672,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "entropy": 2.2408065378665922,
14
+ "epoch": 0.12269938650306748,
15
+ "grad_norm": 1.6718658208847046,
16
+ "learning_rate": 4.487804970912036e-05,
17
+ "loss": 2.129624786376953,
18
+ "mean_token_accuracy": 0.5743644836544991,
19
+ "num_tokens": 134329.0,
20
+ "step": 50
21
+ },
22
+ {
23
+ "entropy": 1.1078504550457,
24
+ "epoch": 0.24539877300613497,
25
+ "grad_norm": 2.4091813564300537,
26
+ "learning_rate": 9.067197798373296e-05,
27
+ "loss": 1.0346390533447265,
28
+ "mean_token_accuracy": 0.7085381114482879,
29
+ "num_tokens": 262407.0,
30
+ "step": 100
31
+ },
32
+ {
33
+ "entropy": 0.9421598988771439,
34
+ "epoch": 0.36809815950920244,
35
+ "grad_norm": 1.3913646936416626,
36
+ "learning_rate": 0.00013646590625834558,
37
+ "loss": 0.8716551208496094,
38
+ "mean_token_accuracy": 0.7449040985107422,
39
+ "num_tokens": 401136.0,
40
+ "step": 150
41
+ },
42
+ {
43
+ "entropy": 0.8723550146818161,
44
+ "epoch": 0.49079754601226994,
45
+ "grad_norm": 1.1664468050003052,
46
+ "learning_rate": 0.0001822598345329582,
47
+ "loss": 0.8095977783203125,
48
+ "mean_token_accuracy": 0.7585240352153778,
49
+ "num_tokens": 531647.0,
50
+ "step": 200
51
+ },
52
+ {
53
+ "entropy": 0.8466519421339035,
54
+ "epoch": 0.6134969325153374,
55
+ "grad_norm": 1.0145421028137207,
56
+ "learning_rate": 0.00022805376280757083,
57
+ "loss": 0.7830857849121093,
58
+ "mean_token_accuracy": 0.7661705195903779,
59
+ "num_tokens": 666127.0,
60
+ "step": 250
61
+ },
62
+ {
63
+ "entropy": 0.8180295366048813,
64
+ "epoch": 0.7361963190184049,
65
+ "grad_norm": 0.9731705784797668,
66
+ "learning_rate": 0.00027384769108218336,
67
+ "loss": 0.7573603820800782,
68
+ "mean_token_accuracy": 0.7704453605413437,
69
+ "num_tokens": 799292.0,
70
+ "step": 300
71
+ },
72
+ {
73
+ "entropy": 0.7990227991342544,
74
+ "epoch": 0.8588957055214724,
75
+ "grad_norm": 1.0642374753952026,
76
+ "learning_rate": 0.000319641619356796,
77
+ "loss": 0.73709228515625,
78
+ "mean_token_accuracy": 0.7752733880281448,
79
+ "num_tokens": 930614.0,
80
+ "step": 350
81
+ },
82
+ {
83
+ "entropy": 0.7899087685346603,
84
+ "epoch": 0.9815950920245399,
85
+ "grad_norm": 0.880613386631012,
86
+ "learning_rate": 0.0003654355476314086,
87
+ "loss": 0.7347319793701171,
88
+ "mean_token_accuracy": 0.7766731631755829,
89
+ "num_tokens": 1061654.0,
90
+ "step": 400
91
+ },
92
+ {
93
+ "epoch": 1.0,
94
+ "eval_entropy": 0.83522472347532,
95
+ "eval_mean_token_accuracy": 0.7484098076820374,
96
+ "eval_not_syn_loss": 0.7872886061668396,
97
+ "eval_not_syn_runtime": 54.8697,
98
+ "eval_not_syn_samples_per_second": 25.424,
99
+ "eval_not_syn_steps_per_second": 3.189,
100
+ "eval_num_tokens": 1080196.0,
101
+ "step": 408
102
+ },
103
+ {
104
+ "epoch": 1.0,
105
+ "eval_entropy": 0.7784442271505083,
106
+ "eval_mean_token_accuracy": 0.8021791185651507,
107
+ "eval_num_tokens": 1080196.0,
108
+ "eval_syn_loss": 0.6980098485946655,
109
+ "eval_syn_runtime": 56.6342,
110
+ "eval_syn_samples_per_second": 24.632,
111
+ "eval_syn_steps_per_second": 3.09,
112
+ "step": 408
113
+ },
114
+ {
115
+ "entropy": 0.75672573694075,
116
+ "epoch": 1.1030674846625768,
117
+ "grad_norm": 0.9244198203086853,
118
+ "learning_rate": 0.00037356351883456205,
119
+ "loss": 0.7048745727539063,
120
+ "mean_token_accuracy": 0.783016896609104,
121
+ "num_tokens": 1199170.0,
122
+ "step": 450
123
+ },
124
+ {
125
+ "entropy": 0.7406797379255294,
126
+ "epoch": 1.2257668711656442,
127
+ "grad_norm": 0.6744896769523621,
128
+ "learning_rate": 0.00037311248151765587,
129
+ "loss": 0.6860730743408203,
130
+ "mean_token_accuracy": 0.787369327545166,
131
+ "num_tokens": 1334298.0,
132
+ "step": 500
133
+ },
134
+ {
135
+ "entropy": 0.7424973237514496,
136
+ "epoch": 1.3484662576687116,
137
+ "grad_norm": 0.92305588722229,
138
+ "learning_rate": 0.000372320629215228,
139
+ "loss": 0.6855724334716797,
140
+ "mean_token_accuracy": 0.7871134179830551,
141
+ "num_tokens": 1463189.0,
142
+ "step": 550
143
+ },
144
+ {
145
+ "entropy": 0.7502243828773498,
146
+ "epoch": 1.471165644171779,
147
+ "grad_norm": 0.8259047269821167,
148
+ "learning_rate": 0.00037118941074037944,
149
+ "loss": 0.6949834442138672,
150
+ "mean_token_accuracy": 0.7851914083957672,
151
+ "num_tokens": 1590310.0,
152
+ "step": 600
153
+ },
154
+ {
155
+ "entropy": 0.7233260518312454,
156
+ "epoch": 1.5938650306748468,
157
+ "grad_norm": 0.882722795009613,
158
+ "learning_rate": 0.00036972089582775814,
159
+ "loss": 0.6649005889892579,
160
+ "mean_token_accuracy": 0.7898879665136337,
161
+ "num_tokens": 1721836.0,
162
+ "step": 650
163
+ },
164
+ {
165
+ "entropy": 0.7162708967924118,
166
+ "epoch": 1.716564417177914,
167
+ "grad_norm": 0.6593027114868164,
168
+ "learning_rate": 0.0003679177713466678,
169
+ "loss": 0.6653710174560546,
170
+ "mean_token_accuracy": 0.7905678844451904,
171
+ "num_tokens": 1853507.0,
172
+ "step": 700
173
+ },
174
+ {
175
+ "entropy": 0.7150520020723343,
176
+ "epoch": 1.8392638036809816,
177
+ "grad_norm": 0.665340781211853,
178
+ "learning_rate": 0.0003657833363850354,
179
+ "loss": 0.6600938415527344,
180
+ "mean_token_accuracy": 0.7918649202585221,
181
+ "num_tokens": 1988113.0,
182
+ "step": 750
183
+ },
184
+ {
185
+ "entropy": 0.6993386310338974,
186
+ "epoch": 1.961963190184049,
187
+ "grad_norm": 0.6502187252044678,
188
+ "learning_rate": 0.00036332149621323294,
189
+ "loss": 0.6469607543945313,
190
+ "mean_token_accuracy": 0.7950778317451477,
191
+ "num_tokens": 2121314.0,
192
+ "step": 800
193
+ },
194
+ {
195
+ "epoch": 2.0,
196
+ "eval_entropy": 0.7315139893123082,
197
+ "eval_mean_token_accuracy": 0.7713504004478454,
198
+ "eval_not_syn_loss": 0.699893593788147,
199
+ "eval_not_syn_runtime": 55.4135,
200
+ "eval_not_syn_samples_per_second": 25.174,
201
+ "eval_not_syn_steps_per_second": 3.158,
202
+ "eval_num_tokens": 2160392.0,
203
+ "step": 816
204
+ },
205
+ {
206
+ "epoch": 2.0,
207
+ "eval_entropy": 0.6848451798302787,
208
+ "eval_mean_token_accuracy": 0.814895657471248,
209
+ "eval_num_tokens": 2160392.0,
210
+ "eval_syn_loss": 0.6415576934814453,
211
+ "eval_syn_runtime": 56.4735,
212
+ "eval_syn_samples_per_second": 24.702,
213
+ "eval_syn_steps_per_second": 3.099,
214
+ "step": 816
215
+ },
216
+ {
217
+ "entropy": 0.6596363644407253,
218
+ "epoch": 2.083435582822086,
219
+ "grad_norm": 0.6647700071334839,
220
+ "learning_rate": 0.00036053675513879666,
221
+ "loss": 0.599151611328125,
222
+ "mean_token_accuracy": 0.8071986864311527,
223
+ "num_tokens": 2253179.0,
224
+ "step": 850
225
+ },
226
+ {
227
+ "entropy": 0.6626429110765457,
228
+ "epoch": 2.2061349693251535,
229
+ "grad_norm": 0.7356868386268616,
230
+ "learning_rate": 0.00035743420826511783,
231
+ "loss": 0.6004195404052735,
232
+ "mean_token_accuracy": 0.805362731218338,
233
+ "num_tokens": 2379965.0,
234
+ "step": 900
235
+ },
236
+ {
237
+ "entropy": 0.6503470712900161,
238
+ "epoch": 2.3288343558282207,
239
+ "grad_norm": 1.095608115196228,
240
+ "learning_rate": 0.0003540195321691835,
241
+ "loss": 0.5951333999633789,
242
+ "mean_token_accuracy": 0.807324025630951,
243
+ "num_tokens": 2515593.0,
244
+ "step": 950
245
+ },
246
+ {
247
+ "entropy": 0.660577870607376,
248
+ "epoch": 2.4515337423312884,
249
+ "grad_norm": 0.6602179408073425,
250
+ "learning_rate": 0.00035029897451542325,
251
+ "loss": 0.6056365585327148,
252
+ "mean_token_accuracy": 0.8057438349723816,
253
+ "num_tokens": 2642790.0,
254
+ "step": 1000
255
+ },
256
+ {
257
+ "entropy": 0.6555853113532066,
258
+ "epoch": 2.574233128834356,
259
+ "grad_norm": 0.7459151148796082,
260
+ "learning_rate": 0.00034627934262466606,
261
+ "loss": 0.5963204193115235,
262
+ "mean_token_accuracy": 0.8060598099231719,
263
+ "num_tokens": 2774396.0,
264
+ "step": 1050
265
+ },
266
+ {
267
+ "entropy": 0.6505853188037872,
268
+ "epoch": 2.6969325153374233,
269
+ "grad_norm": 0.6359792947769165,
270
+ "learning_rate": 0.00034196799101912103,
271
+ "loss": 0.5993065643310547,
272
+ "mean_token_accuracy": 0.806280387043953,
273
+ "num_tokens": 2912102.0,
274
+ "step": 1100
275
+ },
276
+ {
277
+ "entropy": 0.6509970253705979,
278
+ "epoch": 2.819631901840491,
279
+ "grad_norm": 0.8086174726486206,
280
+ "learning_rate": 0.0003373728079661707,
281
+ "loss": 0.6001268005371094,
282
+ "mean_token_accuracy": 0.806866540312767,
283
+ "num_tokens": 3042987.0,
284
+ "step": 1150
285
+ },
286
+ {
287
+ "entropy": 0.6404844307899475,
288
+ "epoch": 2.942331288343558,
289
+ "grad_norm": 0.7038407325744629,
290
+ "learning_rate": 0.0003325022010455975,
291
+ "loss": 0.5870526123046875,
292
+ "mean_token_accuracy": 0.8104586935043335,
293
+ "num_tokens": 3178044.0,
294
+ "step": 1200
295
+ },
296
+ {
297
+ "epoch": 3.0,
298
+ "eval_entropy": 0.6645996727262224,
299
+ "eval_mean_token_accuracy": 0.798654237134116,
300
+ "eval_not_syn_loss": 0.6586771011352539,
301
+ "eval_not_syn_runtime": 55.1088,
302
+ "eval_not_syn_samples_per_second": 25.314,
303
+ "eval_not_syn_steps_per_second": 3.176,
304
+ "eval_num_tokens": 3240588.0,
305
+ "step": 1224
306
+ },
307
+ {
308
+ "epoch": 3.0,
309
+ "eval_entropy": 0.6233607809884207,
310
+ "eval_mean_token_accuracy": 0.7977249867575509,
311
+ "eval_num_tokens": 3240588.0,
312
+ "eval_syn_loss": 0.6288298964500427,
313
+ "eval_syn_runtime": 56.6081,
314
+ "eval_syn_samples_per_second": 24.643,
315
+ "eval_syn_steps_per_second": 3.091,
316
+ "step": 1224
317
+ },
318
+ {
319
+ "entropy": 0.6081961104364106,
320
+ "epoch": 3.063803680981595,
321
+ "grad_norm": 0.6161203980445862,
322
+ "learning_rate": 0.0003273650817666497,
323
+ "loss": 0.5481716537475586,
324
+ "mean_token_accuracy": 0.8175542035488167,
325
+ "num_tokens": 3312797.0,
326
+ "step": 1250
327
+ },
328
+ {
329
+ "entropy": 0.5877064895629883,
330
+ "epoch": 3.1865030674846624,
331
+ "grad_norm": 0.6909636855125427,
332
+ "learning_rate": 0.00032197084926309234,
333
+ "loss": 0.5224150466918945,
334
+ "mean_token_accuracy": 0.8250984263420105,
335
+ "num_tokens": 3441629.0,
336
+ "step": 1300
337
+ },
338
+ {
339
+ "entropy": 0.5952656552195549,
340
+ "epoch": 3.30920245398773,
341
+ "grad_norm": 0.8147026300430298,
342
+ "learning_rate": 0.0003163293730960758,
343
+ "loss": 0.5349474334716797,
344
+ "mean_token_accuracy": 0.8213064837455749,
345
+ "num_tokens": 3576093.0,
346
+ "step": 1350
347
+ },
348
+ {
349
+ "entropy": 0.5937123569846153,
350
+ "epoch": 3.4319018404907977,
351
+ "grad_norm": 0.8690970540046692,
352
+ "learning_rate": 0.0003104509751962869,
353
+ "loss": 0.5310481262207031,
354
+ "mean_token_accuracy": 0.8229135835170746,
355
+ "num_tokens": 3705912.0,
356
+ "step": 1400
357
+ },
358
+ {
359
+ "entropy": 0.6067730465531349,
360
+ "epoch": 3.554601226993865,
361
+ "grad_norm": 0.7433330416679382,
362
+ "learning_rate": 0.0003043464109784207,
363
+ "loss": 0.5450045394897461,
364
+ "mean_token_accuracy": 0.8205251175165177,
365
+ "num_tokens": 3832376.0,
366
+ "step": 1450
367
+ },
368
+ {
369
+ "entropy": 0.5926914718747139,
370
+ "epoch": 3.6773006134969326,
371
+ "grad_norm": 0.7265587449073792,
372
+ "learning_rate": 0.00029802684966252974,
373
+ "loss": 0.5330168914794922,
374
+ "mean_token_accuracy": 0.821294127702713,
375
+ "num_tokens": 3967273.0,
376
+ "step": 1500
377
+ },
378
+ {
379
+ "entropy": 0.5881199997663498,
380
+ "epoch": 3.8,
381
+ "grad_norm": 0.7681817412376404,
382
+ "learning_rate": 0.00029150385383825187,
383
+ "loss": 0.5288929367065429,
384
+ "mean_token_accuracy": 0.8234161508083343,
385
+ "num_tokens": 4102434.0,
386
+ "step": 1550
387
+ },
388
+ {
389
+ "entropy": 0.6008686199784279,
390
+ "epoch": 3.9226993865030675,
391
+ "grad_norm": 0.9384056329727173,
392
+ "learning_rate": 0.0002847893583093113,
393
+ "loss": 0.5437511444091797,
394
+ "mean_token_accuracy": 0.8201366275548935,
395
+ "num_tokens": 4234037.0,
396
+ "step": 1600
397
+ },
398
+ {
399
+ "epoch": 4.0,
400
+ "eval_entropy": 0.6218148783275059,
401
+ "eval_mean_token_accuracy": 0.783796865599496,
402
+ "eval_not_syn_loss": 0.6558170318603516,
403
+ "eval_not_syn_runtime": 55.429,
404
+ "eval_not_syn_samples_per_second": 25.167,
405
+ "eval_not_syn_steps_per_second": 3.157,
406
+ "eval_num_tokens": 4320784.0,
407
+ "step": 1632
408
+ },
409
+ {
410
+ "epoch": 4.0,
411
+ "eval_entropy": 0.5833754854542869,
412
+ "eval_mean_token_accuracy": 0.8278223422595433,
413
+ "eval_num_tokens": 4320784.0,
414
+ "eval_syn_loss": 0.604770302772522,
415
+ "eval_syn_runtime": 56.9593,
416
+ "eval_syn_samples_per_second": 24.491,
417
+ "eval_syn_steps_per_second": 3.072,
418
+ "step": 1632
419
+ },
420
+ {
421
+ "entropy": 0.559690699733869,
422
+ "epoch": 4.044171779141104,
423
+ "grad_norm": 0.65739905834198,
424
+ "learning_rate": 0.0002778956482569963,
425
+ "loss": 0.49702674865722657,
426
+ "mean_token_accuracy": 0.8317034166268628,
427
+ "num_tokens": 4368268.0,
428
+ "step": 1650
429
+ },
430
+ {
431
+ "entropy": 0.5401161375641823,
432
+ "epoch": 4.166871165644172,
433
+ "grad_norm": 1.5176353454589844,
434
+ "learning_rate": 0.0002708353367625691,
435
+ "loss": 0.4598746109008789,
436
+ "mean_token_accuracy": 0.8412033343315124,
437
+ "num_tokens": 4494525.0,
438
+ "step": 1700
439
+ },
440
+ {
441
+ "entropy": 0.5299916860461235,
442
+ "epoch": 4.289570552147239,
443
+ "grad_norm": 0.685965359210968,
444
+ "learning_rate": 0.0002636213417297333,
445
+ "loss": 0.45663665771484374,
446
+ "mean_token_accuracy": 0.8412413811683654,
447
+ "num_tokens": 4631985.0,
448
+ "step": 1750
449
+ },
450
+ {
451
+ "entropy": 0.5352088099718094,
452
+ "epoch": 4.412269938650307,
453
+ "grad_norm": 0.8471158742904663,
454
+ "learning_rate": 0.0002562668622493824,
455
+ "loss": 0.4625707626342773,
456
+ "mean_token_accuracy": 0.8400991547107697,
457
+ "num_tokens": 4766323.0,
458
+ "step": 1800
459
+ },
460
+ {
461
+ "entropy": 0.5425067806243896,
462
+ "epoch": 4.534969325153375,
463
+ "grad_norm": 0.8595437407493591,
464
+ "learning_rate": 0.00024878535444987474,
465
+ "loss": 0.47124954223632814,
466
+ "mean_token_accuracy": 0.8376108628511428,
467
+ "num_tokens": 4895606.0,
468
+ "step": 1850
469
+ },
470
+ {
471
+ "entropy": 0.5419312778115273,
472
+ "epoch": 4.6576687116564415,
473
+ "grad_norm": 0.8436959981918335,
474
+ "learning_rate": 0.00024119050687701818,
475
+ "loss": 0.4708108901977539,
476
+ "mean_token_accuracy": 0.8390797019004822,
477
+ "num_tokens": 5025912.0,
478
+ "step": 1900
479
+ },
480
+ {
481
+ "entropy": 0.5392271488904953,
482
+ "epoch": 4.780368098159509,
483
+ "grad_norm": 0.8691944479942322,
484
+ "learning_rate": 0.00023349621544881404,
485
+ "loss": 0.46682029724121094,
486
+ "mean_token_accuracy": 0.8385172665119172,
487
+ "num_tokens": 5162972.0,
488
+ "step": 1950
489
+ },
490
+ {
491
+ "entropy": 0.5332251599431038,
492
+ "epoch": 4.903067484662577,
493
+ "grad_norm": 0.6814598441123962,
494
+ "learning_rate": 0.00022571655803078078,
495
+ "loss": 0.4687493896484375,
496
+ "mean_token_accuracy": 0.8388861840963364,
497
+ "num_tokens": 5297935.0,
498
+ "step": 2000
499
+ },
500
+ {
501
+ "epoch": 5.0,
502
+ "eval_entropy": 0.5847777877535139,
503
+ "eval_mean_token_accuracy": 0.7782382583618164,
504
+ "eval_not_syn_loss": 0.6678160429000854,
505
+ "eval_not_syn_runtime": 55.1467,
506
+ "eval_not_syn_samples_per_second": 25.296,
507
+ "eval_not_syn_steps_per_second": 3.173,
508
+ "eval_num_tokens": 5400980.0,
509
+ "step": 2040
510
+ },
511
+ {
512
+ "epoch": 5.0,
513
+ "eval_entropy": 0.5505962218557086,
514
+ "eval_mean_token_accuracy": 0.8308270008223397,
515
+ "eval_num_tokens": 5400980.0,
516
+ "eval_syn_loss": 0.606867253780365,
517
+ "eval_syn_runtime": 56.9768,
518
+ "eval_syn_samples_per_second": 24.484,
519
+ "eval_syn_steps_per_second": 3.071,
520
+ "step": 2040
521
+ },
522
+ {
523
+ "entropy": 0.533086326688227,
524
+ "epoch": 5.024539877300613,
525
+ "grad_norm": 0.9453939199447632,
526
+ "learning_rate": 0.00021786576867837776,
527
+ "loss": 0.45802555084228513,
528
+ "mean_token_accuracy": 0.8429379854539428,
529
+ "num_tokens": 5427362.0,
530
+ "step": 2050
531
+ },
532
+ {
533
+ "entropy": 0.46721911430358887,
534
+ "epoch": 5.147239263803681,
535
+ "grad_norm": 0.9237910509109497,
536
+ "learning_rate": 0.00020995821159365647,
537
+ "loss": 0.37914146423339845,
538
+ "mean_token_accuracy": 0.8648965793848038,
539
+ "num_tokens": 5564186.0,
540
+ "step": 2100
541
+ },
542
+ {
543
+ "entropy": 0.4682381117343903,
544
+ "epoch": 5.269938650306749,
545
+ "grad_norm": 0.7424130439758301,
546
+ "learning_rate": 0.0002020083548437879,
547
+ "loss": 0.3859384536743164,
548
+ "mean_token_accuracy": 0.8622810471057892,
549
+ "num_tokens": 5694728.0,
550
+ "step": 2150
551
+ },
552
+ {
553
+ "entropy": 0.464998793900013,
554
+ "epoch": 5.392638036809816,
555
+ "grad_norm": 0.9224902987480164,
556
+ "learning_rate": 0.0001940307438895544,
557
+ "loss": 0.3839593505859375,
558
+ "mean_token_accuracy": 0.8627353543043137,
559
+ "num_tokens": 5832581.0,
560
+ "step": 2200
561
+ },
562
+ {
563
+ "entropy": 0.4753906837105751,
564
+ "epoch": 5.515337423312883,
565
+ "grad_norm": 0.8931427001953125,
566
+ "learning_rate": 0.0001860399749722367,
567
+ "loss": 0.393004264831543,
568
+ "mean_token_accuracy": 0.8586806547641754,
569
+ "num_tokens": 5962417.0,
570
+ "step": 2250
571
+ },
572
+ {
573
+ "entropy": 0.4671635267138481,
574
+ "epoch": 5.638036809815951,
575
+ "grad_norm": 0.9559172987937927,
576
+ "learning_rate": 0.00017805066840759193,
577
+ "loss": 0.38852767944335936,
578
+ "mean_token_accuracy": 0.8615026396512985,
579
+ "num_tokens": 6092338.0,
580
+ "step": 2300
581
+ },
582
+ {
583
+ "entropy": 0.4746100226044655,
584
+ "epoch": 5.7607361963190185,
585
+ "grad_norm": 0.9070253968238831,
586
+ "learning_rate": 0.00017007744183578254,
587
+ "loss": 0.3960166168212891,
588
+ "mean_token_accuracy": 0.8591635781526565,
589
+ "num_tokens": 6226327.0,
590
+ "step": 2350
591
+ },
592
+ {
593
+ "entropy": 0.4698092398047447,
594
+ "epoch": 5.883435582822086,
595
+ "grad_norm": 0.8354232907295227,
596
+ "learning_rate": 0.0001621348834762011,
597
+ "loss": 0.39321571350097656,
598
+ "mean_token_accuracy": 0.8603039687871933,
599
+ "num_tokens": 6359083.0,
600
+ "step": 2400
601
+ },
602
+ {
603
+ "epoch": 6.0,
604
+ "eval_entropy": 0.536405998979296,
605
+ "eval_mean_token_accuracy": 0.790239520413535,
606
+ "eval_not_syn_loss": 0.6884409785270691,
607
+ "eval_not_syn_runtime": 54.9927,
608
+ "eval_not_syn_samples_per_second": 25.367,
609
+ "eval_not_syn_steps_per_second": 3.182,
610
+ "eval_num_tokens": 6481176.0,
611
+ "step": 2448
612
+ },
613
+ {
614
+ "epoch": 6.0,
615
+ "eval_entropy": 0.5039791010107313,
616
+ "eval_mean_token_accuracy": 0.8163431504794529,
617
+ "eval_num_tokens": 6481176.0,
618
+ "eval_syn_loss": 0.6377155184745789,
619
+ "eval_syn_runtime": 57.196,
620
+ "eval_syn_samples_per_second": 24.39,
621
+ "eval_syn_steps_per_second": 3.06,
622
+ "step": 2448
623
+ },
624
+ {
625
+ "entropy": 0.4671963176341972,
626
+ "epoch": 6.004907975460123,
627
+ "grad_norm": 0.8445424437522888,
628
+ "learning_rate": 0.00015423752543612516,
629
+ "loss": 0.3872825241088867,
630
+ "mean_token_accuracy": 0.8618519083418027,
631
+ "num_tokens": 6486549.0,
632
+ "step": 2450
633
+ },
634
+ {
635
+ "entropy": 0.3890652832388878,
636
+ "epoch": 6.12760736196319,
637
+ "grad_norm": 0.725812554359436,
638
+ "learning_rate": 0.00014639981712203683,
639
+ "loss": 0.2923568153381348,
640
+ "mean_token_accuracy": 0.8928290349245072,
641
+ "num_tokens": 6622384.0,
642
+ "step": 2500
643
+ },
644
+ {
645
+ "entropy": 0.39649384289979933,
646
+ "epoch": 6.250306748466258,
647
+ "grad_norm": 0.9143152236938477,
648
+ "learning_rate": 0.000138636098802257,
649
+ "loss": 0.303057861328125,
650
+ "mean_token_accuracy": 0.8888710218667984,
651
+ "num_tokens": 6758637.0,
652
+ "step": 2550
653
+ },
654
+ {
655
+ "entropy": 0.4020465177297592,
656
+ "epoch": 6.373006134969325,
657
+ "grad_norm": 0.9407342076301575,
658
+ "learning_rate": 0.00013096057536926304,
659
+ "loss": 0.3073894691467285,
660
+ "mean_token_accuracy": 0.8894575881958008,
661
+ "num_tokens": 6888533.0,
662
+ "step": 2600
663
+ },
664
+ {
665
+ "entropy": 0.397317076921463,
666
+ "epoch": 6.495705521472392,
667
+ "grad_norm": 0.8517777323722839,
668
+ "learning_rate": 0.00012338729034969828,
669
+ "loss": 0.30508209228515626,
670
+ "mean_token_accuracy": 0.8886856293678284,
671
+ "num_tokens": 7023729.0,
672
+ "step": 2650
673
+ },
674
+ {
675
+ "entropy": 0.3915065750479698,
676
+ "epoch": 6.61840490797546,
677
+ "grad_norm": 0.862319827079773,
678
+ "learning_rate": 0.00011593010020962378,
679
+ "loss": 0.30237462997436526,
680
+ "mean_token_accuracy": 0.8897094148397445,
681
+ "num_tokens": 7153821.0,
682
+ "step": 2700
683
+ },
684
+ {
685
+ "entropy": 0.3977497002482414,
686
+ "epoch": 6.741104294478528,
687
+ "grad_norm": 0.8821370601654053,
688
+ "learning_rate": 0.00010860264900202563,
689
+ "loss": 0.3067088508605957,
690
+ "mean_token_accuracy": 0.8873285222053527,
691
+ "num_tokens": 7283513.0,
692
+ "step": 2750
693
+ },
694
+ {
695
+ "entropy": 0.4002616259455681,
696
+ "epoch": 6.863803680981595,
697
+ "grad_norm": 1.069462776184082,
698
+ "learning_rate": 0.0001014183434029644,
699
+ "loss": 0.30735193252563475,
700
+ "mean_token_accuracy": 0.8880347841978073,
701
+ "num_tokens": 7411578.0,
702
+ "step": 2800
703
+ },
704
+ {
705
+ "entropy": 0.3951660317182541,
706
+ "epoch": 6.986503067484662,
707
+ "grad_norm": 0.8845367431640625,
708
+ "learning_rate": 9.439032818204105e-05,
709
+ "loss": 0.30523622512817383,
710
+ "mean_token_accuracy": 0.8894886714220047,
711
+ "num_tokens": 7548356.0,
712
+ "step": 2850
713
+ },
714
+ {
715
+ "epoch": 7.0,
716
+ "eval_entropy": 0.4834100874832698,
717
+ "eval_mean_token_accuracy": 0.788761088166918,
718
+ "eval_not_syn_loss": 0.7379753589630127,
719
+ "eval_not_syn_runtime": 55.6309,
720
+ "eval_not_syn_samples_per_second": 25.076,
721
+ "eval_not_syn_steps_per_second": 3.146,
722
+ "eval_num_tokens": 7561372.0,
723
+ "step": 2856
724
+ },
725
+ {
726
+ "epoch": 7.0,
727
+ "eval_entropy": 0.453818735906056,
728
+ "eval_mean_token_accuracy": 0.8209771479879107,
729
+ "eval_num_tokens": 7561372.0,
730
+ "eval_syn_loss": 0.6787521243095398,
731
+ "eval_syn_runtime": 56.9966,
732
+ "eval_syn_samples_per_second": 24.475,
733
+ "eval_syn_steps_per_second": 3.07,
734
+ "step": 2856
735
+ },
736
+ {
737
+ "entropy": 0.3386827930055483,
738
+ "epoch": 7.1079754601227,
739
+ "grad_norm": 1.0214653015136719,
740
+ "learning_rate": 8.753146215206042e-05,
741
+ "loss": 0.2361087417602539,
742
+ "mean_token_accuracy": 0.9141153067049353,
743
+ "num_tokens": 7671657.0,
744
+ "step": 2900
745
+ },
746
+ {
747
+ "entropy": 0.31863692075014116,
748
+ "epoch": 7.230674846625767,
749
+ "grad_norm": 0.9141477346420288,
750
+ "learning_rate": 8.085429464189586e-05,
751
+ "loss": 0.21914518356323243,
752
+ "mean_token_accuracy": 0.9194145703315735,
753
+ "num_tokens": 7806989.0,
754
+ "step": 2950
755
+ },
756
+ {
757
+ "entropy": 0.31718266174197196,
758
+ "epoch": 7.353374233128834,
759
+ "grad_norm": 0.9127821922302246,
760
+ "learning_rate": 7.437104253560078e-05,
761
+ "loss": 0.22112932205200195,
762
+ "mean_token_accuracy": 0.9190003037452698,
763
+ "num_tokens": 7941760.0,
764
+ "step": 3000
765
+ },
766
+ {
767
+ "entropy": 0.3143314927816391,
768
+ "epoch": 7.476073619631902,
769
+ "grad_norm": 1.1398990154266357,
770
+ "learning_rate": 6.809356791977874e-05,
771
+ "loss": 0.2191557502746582,
772
+ "mean_token_accuracy": 0.918931410908699,
773
+ "num_tokens": 8080328.0,
774
+ "step": 3050
775
+ },
776
+ {
777
+ "entropy": 0.32102111548185347,
778
+ "epoch": 7.598773006134969,
779
+ "grad_norm": 1.069332480430603,
780
+ "learning_rate": 6.203335638010852e-05,
781
+ "loss": 0.22624496459960938,
782
+ "mean_token_accuracy": 0.9173947256803513,
783
+ "num_tokens": 8210831.0,
784
+ "step": 3100
785
+ },
786
+ {
787
+ "entropy": 0.31791565626859664,
788
+ "epoch": 7.721472392638037,
789
+ "grad_norm": 1.1536624431610107,
790
+ "learning_rate": 5.62014959867348e-05,
791
+ "loss": 0.22338006973266603,
792
+ "mean_token_accuracy": 0.9177713346481323,
793
+ "num_tokens": 8344985.0,
794
+ "step": 3150
795
+ },
796
+ {
797
+ "entropy": 0.3209437537193298,
798
+ "epoch": 7.844171779141105,
799
+ "grad_norm": 1.1040980815887451,
800
+ "learning_rate": 5.06086570069728e-05,
801
+ "loss": 0.2263322639465332,
802
+ "mean_token_accuracy": 0.9170012313127518,
803
+ "num_tokens": 8475716.0,
804
+ "step": 3200
805
+ },
806
+ {
807
+ "entropy": 0.31527657777071,
808
+ "epoch": 7.9668711656441715,
809
+ "grad_norm": 1.1415972709655762,
810
+ "learning_rate": 4.526507238244653e-05,
811
+ "loss": 0.2216329574584961,
812
+ "mean_token_accuracy": 0.918340762257576,
813
+ "num_tokens": 8606709.0,
814
+ "step": 3250
815
+ },
816
+ {
817
+ "epoch": 8.0,
818
+ "eval_entropy": 0.43602277857916694,
819
+ "eval_mean_token_accuracy": 0.7907402624402727,
820
+ "eval_not_syn_loss": 0.8163036704063416,
821
+ "eval_not_syn_runtime": 55.3868,
822
+ "eval_not_syn_samples_per_second": 25.187,
823
+ "eval_not_syn_steps_per_second": 3.16,
824
+ "eval_num_tokens": 8641568.0,
825
+ "step": 3264
826
+ },
827
+ {
828
+ "epoch": 8.0,
829
+ "eval_entropy": 0.4093018649305616,
830
+ "eval_mean_token_accuracy": 0.8147984709058489,
831
+ "eval_num_tokens": 8641568.0,
832
+ "eval_syn_loss": 0.7571278810501099,
833
+ "eval_syn_runtime": 56.797,
834
+ "eval_syn_samples_per_second": 24.561,
835
+ "eval_syn_steps_per_second": 3.081,
836
+ "step": 3264
837
+ },
838
+ {
839
+ "entropy": 0.2847619162063406,
840
+ "epoch": 8.088343558282208,
841
+ "grad_norm": 1.0268603563308716,
842
+ "learning_rate": 4.018051900638024e-05,
843
+ "loss": 0.1806371307373047,
844
+ "mean_token_accuracy": 0.9346237525795446,
845
+ "num_tokens": 8733256.0,
846
+ "step": 3300
847
+ },
848
+ {
849
+ "entropy": 0.26448936730623246,
850
+ "epoch": 8.211042944785277,
851
+ "grad_norm": 1.307308554649353,
852
+ "learning_rate": 3.536429983529935e-05,
853
+ "loss": 0.16438600540161133,
854
+ "mean_token_accuracy": 0.9407236540317535,
855
+ "num_tokens": 8864507.0,
856
+ "step": 3350
857
+ },
858
+ {
859
+ "entropy": 0.26411333337426185,
860
+ "epoch": 8.333742331288343,
861
+ "grad_norm": 1.0053174495697021,
862
+ "learning_rate": 3.082522686786935e-05,
863
+ "loss": 0.16243309020996094,
864
+ "mean_token_accuracy": 0.9401588958501815,
865
+ "num_tokens": 8993933.0,
866
+ "step": 3400
867
+ },
868
+ {
869
+ "entropy": 0.2589409455657005,
870
+ "epoch": 8.45644171779141,
871
+ "grad_norm": 1.1849397420883179,
872
+ "learning_rate": 2.6571605022016655e-05,
873
+ "loss": 0.1622679328918457,
874
+ "mean_token_accuracy": 0.9411727702617645,
875
+ "num_tokens": 9124819.0,
876
+ "step": 3450
877
+ },
878
+ {
879
+ "entropy": 0.25455976352095605,
880
+ "epoch": 8.579141104294479,
881
+ "grad_norm": 0.9689339995384216,
882
+ "learning_rate": 2.2611216939830125e-05,
883
+ "loss": 0.16137266159057617,
884
+ "mean_token_accuracy": 0.941931157708168,
885
+ "num_tokens": 9266865.0,
886
+ "step": 3500
887
+ },
888
+ {
889
+ "entropy": 0.25809177130460736,
890
+ "epoch": 8.701840490797546,
891
+ "grad_norm": 1.097219467163086,
892
+ "learning_rate": 1.8951308748044324e-05,
893
+ "loss": 0.16276371002197265,
894
+ "mean_token_accuracy": 0.9418226885795593,
895
+ "num_tokens": 9402291.0,
896
+ "step": 3550
897
+ },
898
+ {
899
+ "entropy": 0.2551534166932106,
900
+ "epoch": 8.824539877300614,
901
+ "grad_norm": 1.1789699792861938,
902
+ "learning_rate": 1.5598576800158975e-05,
903
+ "loss": 0.16089740753173828,
904
+ "mean_token_accuracy": 0.942240217924118,
905
+ "num_tokens": 9536676.0,
906
+ "step": 3600
907
+ },
908
+ {
909
+ "entropy": 0.25731841623783114,
910
+ "epoch": 8.94723926380368,
911
+ "grad_norm": 1.071371078491211,
912
+ "learning_rate": 1.2559155424451429e-05,
913
+ "loss": 0.16048816680908204,
914
+ "mean_token_accuracy": 0.9426142275333405,
915
+ "num_tokens": 9668941.0,
916
+ "step": 3650
917
+ },
918
+ {
919
+ "epoch": 9.0,
920
+ "eval_entropy": 0.4000720078604562,
921
+ "eval_mean_token_accuracy": 0.791837249483381,
922
+ "eval_not_syn_loss": 0.9112135171890259,
923
+ "eval_not_syn_runtime": 55.748,
924
+ "eval_not_syn_samples_per_second": 25.023,
925
+ "eval_not_syn_steps_per_second": 3.139,
926
+ "eval_num_tokens": 9721764.0,
927
+ "step": 3672
928
+ },
929
+ {
930
+ "epoch": 9.0,
931
+ "eval_entropy": 0.3756853074686868,
932
+ "eval_mean_token_accuracy": 0.8066899194036211,
933
+ "eval_num_tokens": 9721764.0,
934
+ "eval_syn_loss": 0.8509045243263245,
935
+ "eval_syn_runtime": 56.7641,
936
+ "eval_syn_samples_per_second": 24.575,
937
+ "eval_syn_steps_per_second": 3.083,
938
+ "step": 3672
939
+ }
940
+ ],
941
+ "logging_steps": 50,
942
+ "max_steps": 4080,
943
+ "num_input_tokens_seen": 0,
944
+ "num_train_epochs": 10,
945
+ "save_steps": 500,
946
+ "stateful_callbacks": {
947
+ "TrainerControl": {
948
+ "args": {
949
+ "should_epoch_stop": false,
950
+ "should_evaluate": false,
951
+ "should_log": false,
952
+ "should_save": true,
953
+ "should_training_stop": false
954
+ },
955
+ "attributes": {}
956
+ }
957
+ },
958
+ "total_flos": 1.932544486512142e+17,
959
+ "train_batch_size": 8,
960
+ "trial_name": null,
961
+ "trial_params": null
962
+ }
substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-408/README.md ADDED
@@ -0,0 +1,209 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: Qwen/Qwen3.5-2B-Base
3
+ library_name: peft
4
+ pipeline_tag: text-generation
5
+ tags:
6
+ - base_model:adapter:Qwen/Qwen3.5-2B-Base
7
+ - lora
8
+ - sft
9
+ - transformers
10
+ - trl
11
+ ---
12
+
13
+ # Model Card for Model ID
14
+
15
+ <!-- Provide a quick summary of what the model is/does. -->
16
+
17
+
18
+
19
+ ## Model Details
20
+
21
+ ### Model Description
22
+
23
+ <!-- Provide a longer summary of what this model is. -->
24
+
25
+
26
+
27
+ - **Developed by:** [More Information Needed]
28
+ - **Funded by [optional]:** [More Information Needed]
29
+ - **Shared by [optional]:** [More Information Needed]
30
+ - **Model type:** [More Information Needed]
31
+ - **Language(s) (NLP):** [More Information Needed]
32
+ - **License:** [More Information Needed]
33
+ - **Finetuned from model [optional]:** [More Information Needed]
34
+
35
+ ### Model Sources [optional]
36
+
37
+ <!-- Provide the basic links for the model. -->
38
+
39
+ - **Repository:** [More Information Needed]
40
+ - **Paper [optional]:** [More Information Needed]
41
+ - **Demo [optional]:** [More Information Needed]
42
+
43
+ ## Uses
44
+
45
+ <!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
46
+
47
+ ### Direct Use
48
+
49
+ <!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
50
+
51
+ [More Information Needed]
52
+
53
+ ### Downstream Use [optional]
54
+
55
+ <!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
56
+
57
+ [More Information Needed]
58
+
59
+ ### Out-of-Scope Use
60
+
61
+ <!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
62
+
63
+ [More Information Needed]
64
+
65
+ ## Bias, Risks, and Limitations
66
+
67
+ <!-- This section is meant to convey both technical and sociotechnical limitations. -->
68
+
69
+ [More Information Needed]
70
+
71
+ ### Recommendations
72
+
73
+ <!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
74
+
75
+ Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
76
+
77
+ ## How to Get Started with the Model
78
+
79
+ Use the code below to get started with the model.
80
+
81
+ [More Information Needed]
82
+
83
+ ## Training Details
84
+
85
+ ### Training Data
86
+
87
+ <!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
88
+
89
+ [More Information Needed]
90
+
91
+ ### Training Procedure
92
+
93
+ <!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
94
+
95
+ #### Preprocessing [optional]
96
+
97
+ [More Information Needed]
98
+
99
+
100
+ #### Training Hyperparameters
101
+
102
+ - **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
103
+
104
+ #### Speeds, Sizes, Times [optional]
105
+
106
+ <!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
107
+
108
+ [More Information Needed]
109
+
110
+ ## Evaluation
111
+
112
+ <!-- This section describes the evaluation protocols and provides the results. -->
113
+
114
+ ### Testing Data, Factors & Metrics
115
+
116
+ #### Testing Data
117
+
118
+ <!-- This should link to a Dataset Card if possible. -->
119
+
120
+ [More Information Needed]
121
+
122
+ #### Factors
123
+
124
+ <!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
125
+
126
+ [More Information Needed]
127
+
128
+ #### Metrics
129
+
130
+ <!-- These are the evaluation metrics being used, ideally with a description of why. -->
131
+
132
+ [More Information Needed]
133
+
134
+ ### Results
135
+
136
+ [More Information Needed]
137
+
138
+ #### Summary
139
+
140
+
141
+
142
+ ## Model Examination [optional]
143
+
144
+ <!-- Relevant interpretability work for the model goes here -->
145
+
146
+ [More Information Needed]
147
+
148
+ ## Environmental Impact
149
+
150
+ <!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
151
+
152
+ Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
153
+
154
+ - **Hardware Type:** [More Information Needed]
155
+ - **Hours used:** [More Information Needed]
156
+ - **Cloud Provider:** [More Information Needed]
157
+ - **Compute Region:** [More Information Needed]
158
+ - **Carbon Emitted:** [More Information Needed]
159
+
160
+ ## Technical Specifications [optional]
161
+
162
+ ### Model Architecture and Objective
163
+
164
+ [More Information Needed]
165
+
166
+ ### Compute Infrastructure
167
+
168
+ [More Information Needed]
169
+
170
+ #### Hardware
171
+
172
+ [More Information Needed]
173
+
174
+ #### Software
175
+
176
+ [More Information Needed]
177
+
178
+ ## Citation [optional]
179
+
180
+ <!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
181
+
182
+ **BibTeX:**
183
+
184
+ [More Information Needed]
185
+
186
+ **APA:**
187
+
188
+ [More Information Needed]
189
+
190
+ ## Glossary [optional]
191
+
192
+ <!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
193
+
194
+ [More Information Needed]
195
+
196
+ ## More Information [optional]
197
+
198
+ [More Information Needed]
199
+
200
+ ## Model Card Authors [optional]
201
+
202
+ [More Information Needed]
203
+
204
+ ## Model Card Contact
205
+
206
+ [More Information Needed]
207
+ ### Framework versions
208
+
209
+ - PEFT 0.19.1
substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-408/adapter_config.json ADDED
@@ -0,0 +1,48 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alora_invocation_tokens": null,
3
+ "alpha_pattern": {},
4
+ "arrow_config": null,
5
+ "auto_mapping": null,
6
+ "base_model_name_or_path": "Qwen/Qwen3.5-2B-Base",
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": 32,
20
+ "lora_bias": false,
21
+ "lora_dropout": 0.08602048083239339,
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": 16,
30
+ "rank_pattern": {},
31
+ "revision": null,
32
+ "target_modules": [
33
+ "q_proj",
34
+ "k_proj",
35
+ "gate_proj",
36
+ "up_proj",
37
+ "o_proj",
38
+ "down_proj",
39
+ "v_proj"
40
+ ],
41
+ "target_parameters": null,
42
+ "task_type": "CAUSAL_LM",
43
+ "trainable_token_indices": null,
44
+ "use_bdlora": null,
45
+ "use_dora": false,
46
+ "use_qalora": false,
47
+ "use_rslora": false
48
+ }
substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-408/trainer_state.json ADDED
@@ -0,0 +1,136 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 1.0,
6
+ "eval_steps": 500,
7
+ "global_step": 408,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "entropy": 2.2408065378665922,
14
+ "epoch": 0.12269938650306748,
15
+ "grad_norm": 1.6718658208847046,
16
+ "learning_rate": 4.487804970912036e-05,
17
+ "loss": 2.129624786376953,
18
+ "mean_token_accuracy": 0.5743644836544991,
19
+ "num_tokens": 134329.0,
20
+ "step": 50
21
+ },
22
+ {
23
+ "entropy": 1.1078504550457,
24
+ "epoch": 0.24539877300613497,
25
+ "grad_norm": 2.4091813564300537,
26
+ "learning_rate": 9.067197798373296e-05,
27
+ "loss": 1.0346390533447265,
28
+ "mean_token_accuracy": 0.7085381114482879,
29
+ "num_tokens": 262407.0,
30
+ "step": 100
31
+ },
32
+ {
33
+ "entropy": 0.9421598988771439,
34
+ "epoch": 0.36809815950920244,
35
+ "grad_norm": 1.3913646936416626,
36
+ "learning_rate": 0.00013646590625834558,
37
+ "loss": 0.8716551208496094,
38
+ "mean_token_accuracy": 0.7449040985107422,
39
+ "num_tokens": 401136.0,
40
+ "step": 150
41
+ },
42
+ {
43
+ "entropy": 0.8723550146818161,
44
+ "epoch": 0.49079754601226994,
45
+ "grad_norm": 1.1664468050003052,
46
+ "learning_rate": 0.0001822598345329582,
47
+ "loss": 0.8095977783203125,
48
+ "mean_token_accuracy": 0.7585240352153778,
49
+ "num_tokens": 531647.0,
50
+ "step": 200
51
+ },
52
+ {
53
+ "entropy": 0.8466519421339035,
54
+ "epoch": 0.6134969325153374,
55
+ "grad_norm": 1.0145421028137207,
56
+ "learning_rate": 0.00022805376280757083,
57
+ "loss": 0.7830857849121093,
58
+ "mean_token_accuracy": 0.7661705195903779,
59
+ "num_tokens": 666127.0,
60
+ "step": 250
61
+ },
62
+ {
63
+ "entropy": 0.8180295366048813,
64
+ "epoch": 0.7361963190184049,
65
+ "grad_norm": 0.9731705784797668,
66
+ "learning_rate": 0.00027384769108218336,
67
+ "loss": 0.7573603820800782,
68
+ "mean_token_accuracy": 0.7704453605413437,
69
+ "num_tokens": 799292.0,
70
+ "step": 300
71
+ },
72
+ {
73
+ "entropy": 0.7990227991342544,
74
+ "epoch": 0.8588957055214724,
75
+ "grad_norm": 1.0642374753952026,
76
+ "learning_rate": 0.000319641619356796,
77
+ "loss": 0.73709228515625,
78
+ "mean_token_accuracy": 0.7752733880281448,
79
+ "num_tokens": 930614.0,
80
+ "step": 350
81
+ },
82
+ {
83
+ "entropy": 0.7899087685346603,
84
+ "epoch": 0.9815950920245399,
85
+ "grad_norm": 0.880613386631012,
86
+ "learning_rate": 0.0003654355476314086,
87
+ "loss": 0.7347319793701171,
88
+ "mean_token_accuracy": 0.7766731631755829,
89
+ "num_tokens": 1061654.0,
90
+ "step": 400
91
+ },
92
+ {
93
+ "epoch": 1.0,
94
+ "eval_entropy": 0.83522472347532,
95
+ "eval_mean_token_accuracy": 0.7484098076820374,
96
+ "eval_not_syn_loss": 0.7872886061668396,
97
+ "eval_not_syn_runtime": 54.8697,
98
+ "eval_not_syn_samples_per_second": 25.424,
99
+ "eval_not_syn_steps_per_second": 3.189,
100
+ "eval_num_tokens": 1080196.0,
101
+ "step": 408
102
+ },
103
+ {
104
+ "epoch": 1.0,
105
+ "eval_entropy": 0.7784442271505083,
106
+ "eval_mean_token_accuracy": 0.8021791185651507,
107
+ "eval_num_tokens": 1080196.0,
108
+ "eval_syn_loss": 0.6980098485946655,
109
+ "eval_syn_runtime": 56.6342,
110
+ "eval_syn_samples_per_second": 24.632,
111
+ "eval_syn_steps_per_second": 3.09,
112
+ "step": 408
113
+ }
114
+ ],
115
+ "logging_steps": 50,
116
+ "max_steps": 4080,
117
+ "num_input_tokens_seen": 0,
118
+ "num_train_epochs": 10,
119
+ "save_steps": 500,
120
+ "stateful_callbacks": {
121
+ "TrainerControl": {
122
+ "args": {
123
+ "should_epoch_stop": false,
124
+ "should_evaluate": false,
125
+ "should_log": false,
126
+ "should_save": true,
127
+ "should_training_stop": false
128
+ },
129
+ "attributes": {}
130
+ }
131
+ },
132
+ "total_flos": 2.161626494849779e+16,
133
+ "train_batch_size": 8,
134
+ "trial_name": null,
135
+ "trial_params": null
136
+ }
substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test2/checkpoint-1632/README.md ADDED
@@ -0,0 +1,209 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: Qwen/Qwen3.5-2B-Base
3
+ library_name: peft
4
+ pipeline_tag: text-generation
5
+ tags:
6
+ - base_model:adapter:Qwen/Qwen3.5-2B-Base
7
+ - lora
8
+ - sft
9
+ - transformers
10
+ - trl
11
+ ---
12
+
13
+ # Model Card for Model ID
14
+
15
+ <!-- Provide a quick summary of what the model is/does. -->
16
+
17
+
18
+
19
+ ## Model Details
20
+
21
+ ### Model Description
22
+
23
+ <!-- Provide a longer summary of what this model is. -->
24
+
25
+
26
+
27
+ - **Developed by:** [More Information Needed]
28
+ - **Funded by [optional]:** [More Information Needed]
29
+ - **Shared by [optional]:** [More Information Needed]
30
+ - **Model type:** [More Information Needed]
31
+ - **Language(s) (NLP):** [More Information Needed]
32
+ - **License:** [More Information Needed]
33
+ - **Finetuned from model [optional]:** [More Information Needed]
34
+
35
+ ### Model Sources [optional]
36
+
37
+ <!-- Provide the basic links for the model. -->
38
+
39
+ - **Repository:** [More Information Needed]
40
+ - **Paper [optional]:** [More Information Needed]
41
+ - **Demo [optional]:** [More Information Needed]
42
+
43
+ ## Uses
44
+
45
+ <!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
46
+
47
+ ### Direct Use
48
+
49
+ <!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
50
+
51
+ [More Information Needed]
52
+
53
+ ### Downstream Use [optional]
54
+
55
+ <!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
56
+
57
+ [More Information Needed]
58
+
59
+ ### Out-of-Scope Use
60
+
61
+ <!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
62
+
63
+ [More Information Needed]
64
+
65
+ ## Bias, Risks, and Limitations
66
+
67
+ <!-- This section is meant to convey both technical and sociotechnical limitations. -->
68
+
69
+ [More Information Needed]
70
+
71
+ ### Recommendations
72
+
73
+ <!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
74
+
75
+ Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
76
+
77
+ ## How to Get Started with the Model
78
+
79
+ Use the code below to get started with the model.
80
+
81
+ [More Information Needed]
82
+
83
+ ## Training Details
84
+
85
+ ### Training Data
86
+
87
+ <!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
88
+
89
+ [More Information Needed]
90
+
91
+ ### Training Procedure
92
+
93
+ <!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
94
+
95
+ #### Preprocessing [optional]
96
+
97
+ [More Information Needed]
98
+
99
+
100
+ #### Training Hyperparameters
101
+
102
+ - **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
103
+
104
+ #### Speeds, Sizes, Times [optional]
105
+
106
+ <!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
107
+
108
+ [More Information Needed]
109
+
110
+ ## Evaluation
111
+
112
+ <!-- This section describes the evaluation protocols and provides the results. -->
113
+
114
+ ### Testing Data, Factors & Metrics
115
+
116
+ #### Testing Data
117
+
118
+ <!-- This should link to a Dataset Card if possible. -->
119
+
120
+ [More Information Needed]
121
+
122
+ #### Factors
123
+
124
+ <!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
125
+
126
+ [More Information Needed]
127
+
128
+ #### Metrics
129
+
130
+ <!-- These are the evaluation metrics being used, ideally with a description of why. -->
131
+
132
+ [More Information Needed]
133
+
134
+ ### Results
135
+
136
+ [More Information Needed]
137
+
138
+ #### Summary
139
+
140
+
141
+
142
+ ## Model Examination [optional]
143
+
144
+ <!-- Relevant interpretability work for the model goes here -->
145
+
146
+ [More Information Needed]
147
+
148
+ ## Environmental Impact
149
+
150
+ <!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
151
+
152
+ Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
153
+
154
+ - **Hardware Type:** [More Information Needed]
155
+ - **Hours used:** [More Information Needed]
156
+ - **Cloud Provider:** [More Information Needed]
157
+ - **Compute Region:** [More Information Needed]
158
+ - **Carbon Emitted:** [More Information Needed]
159
+
160
+ ## Technical Specifications [optional]
161
+
162
+ ### Model Architecture and Objective
163
+
164
+ [More Information Needed]
165
+
166
+ ### Compute Infrastructure
167
+
168
+ [More Information Needed]
169
+
170
+ #### Hardware
171
+
172
+ [More Information Needed]
173
+
174
+ #### Software
175
+
176
+ [More Information Needed]
177
+
178
+ ## Citation [optional]
179
+
180
+ <!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
181
+
182
+ **BibTeX:**
183
+
184
+ [More Information Needed]
185
+
186
+ **APA:**
187
+
188
+ [More Information Needed]
189
+
190
+ ## Glossary [optional]
191
+
192
+ <!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
193
+
194
+ [More Information Needed]
195
+
196
+ ## More Information [optional]
197
+
198
+ [More Information Needed]
199
+
200
+ ## Model Card Authors [optional]
201
+
202
+ [More Information Needed]
203
+
204
+ ## Model Card Contact
205
+
206
+ [More Information Needed]
207
+ ### Framework versions
208
+
209
+ - PEFT 0.19.1
systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1371/README.md ADDED
@@ -0,0 +1,209 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: Qwen/Qwen3.5-2B-Base
3
+ library_name: peft
4
+ pipeline_tag: text-generation
5
+ tags:
6
+ - base_model:adapter:Qwen/Qwen3.5-2B-Base
7
+ - lora
8
+ - sft
9
+ - transformers
10
+ - trl
11
+ ---
12
+
13
+ # Model Card for Model ID
14
+
15
+ <!-- Provide a quick summary of what the model is/does. -->
16
+
17
+
18
+
19
+ ## Model Details
20
+
21
+ ### Model Description
22
+
23
+ <!-- Provide a longer summary of what this model is. -->
24
+
25
+
26
+
27
+ - **Developed by:** [More Information Needed]
28
+ - **Funded by [optional]:** [More Information Needed]
29
+ - **Shared by [optional]:** [More Information Needed]
30
+ - **Model type:** [More Information Needed]
31
+ - **Language(s) (NLP):** [More Information Needed]
32
+ - **License:** [More Information Needed]
33
+ - **Finetuned from model [optional]:** [More Information Needed]
34
+
35
+ ### Model Sources [optional]
36
+
37
+ <!-- Provide the basic links for the model. -->
38
+
39
+ - **Repository:** [More Information Needed]
40
+ - **Paper [optional]:** [More Information Needed]
41
+ - **Demo [optional]:** [More Information Needed]
42
+
43
+ ## Uses
44
+
45
+ <!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
46
+
47
+ ### Direct Use
48
+
49
+ <!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
50
+
51
+ [More Information Needed]
52
+
53
+ ### Downstream Use [optional]
54
+
55
+ <!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
56
+
57
+ [More Information Needed]
58
+
59
+ ### Out-of-Scope Use
60
+
61
+ <!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
62
+
63
+ [More Information Needed]
64
+
65
+ ## Bias, Risks, and Limitations
66
+
67
+ <!-- This section is meant to convey both technical and sociotechnical limitations. -->
68
+
69
+ [More Information Needed]
70
+
71
+ ### Recommendations
72
+
73
+ <!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
74
+
75
+ Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
76
+
77
+ ## How to Get Started with the Model
78
+
79
+ Use the code below to get started with the model.
80
+
81
+ [More Information Needed]
82
+
83
+ ## Training Details
84
+
85
+ ### Training Data
86
+
87
+ <!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
88
+
89
+ [More Information Needed]
90
+
91
+ ### Training Procedure
92
+
93
+ <!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
94
+
95
+ #### Preprocessing [optional]
96
+
97
+ [More Information Needed]
98
+
99
+
100
+ #### Training Hyperparameters
101
+
102
+ - **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
103
+
104
+ #### Speeds, Sizes, Times [optional]
105
+
106
+ <!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
107
+
108
+ [More Information Needed]
109
+
110
+ ## Evaluation
111
+
112
+ <!-- This section describes the evaluation protocols and provides the results. -->
113
+
114
+ ### Testing Data, Factors & Metrics
115
+
116
+ #### Testing Data
117
+
118
+ <!-- This should link to a Dataset Card if possible. -->
119
+
120
+ [More Information Needed]
121
+
122
+ #### Factors
123
+
124
+ <!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
125
+
126
+ [More Information Needed]
127
+
128
+ #### Metrics
129
+
130
+ <!-- These are the evaluation metrics being used, ideally with a description of why. -->
131
+
132
+ [More Information Needed]
133
+
134
+ ### Results
135
+
136
+ [More Information Needed]
137
+
138
+ #### Summary
139
+
140
+
141
+
142
+ ## Model Examination [optional]
143
+
144
+ <!-- Relevant interpretability work for the model goes here -->
145
+
146
+ [More Information Needed]
147
+
148
+ ## Environmental Impact
149
+
150
+ <!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
151
+
152
+ Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
153
+
154
+ - **Hardware Type:** [More Information Needed]
155
+ - **Hours used:** [More Information Needed]
156
+ - **Cloud Provider:** [More Information Needed]
157
+ - **Compute Region:** [More Information Needed]
158
+ - **Carbon Emitted:** [More Information Needed]
159
+
160
+ ## Technical Specifications [optional]
161
+
162
+ ### Model Architecture and Objective
163
+
164
+ [More Information Needed]
165
+
166
+ ### Compute Infrastructure
167
+
168
+ [More Information Needed]
169
+
170
+ #### Hardware
171
+
172
+ [More Information Needed]
173
+
174
+ #### Software
175
+
176
+ [More Information Needed]
177
+
178
+ ## Citation [optional]
179
+
180
+ <!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
181
+
182
+ **BibTeX:**
183
+
184
+ [More Information Needed]
185
+
186
+ **APA:**
187
+
188
+ [More Information Needed]
189
+
190
+ ## Glossary [optional]
191
+
192
+ <!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
193
+
194
+ [More Information Needed]
195
+
196
+ ## More Information [optional]
197
+
198
+ [More Information Needed]
199
+
200
+ ## Model Card Authors [optional]
201
+
202
+ [More Information Needed]
203
+
204
+ ## Model Card Contact
205
+
206
+ [More Information Needed]
207
+ ### Framework versions
208
+
209
+ - PEFT 0.19.1
systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1371/adapter_config.json ADDED
@@ -0,0 +1,48 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alora_invocation_tokens": null,
3
+ "alpha_pattern": {},
4
+ "arrow_config": null,
5
+ "auto_mapping": null,
6
+ "base_model_name_or_path": "Qwen/Qwen3.5-2B-Base",
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": 256,
20
+ "lora_bias": false,
21
+ "lora_dropout": 0.01686122009693236,
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": 128,
30
+ "rank_pattern": {},
31
+ "revision": null,
32
+ "target_modules": [
33
+ "v_proj",
34
+ "o_proj",
35
+ "k_proj",
36
+ "up_proj",
37
+ "q_proj",
38
+ "down_proj",
39
+ "gate_proj"
40
+ ],
41
+ "target_parameters": null,
42
+ "task_type": "CAUSAL_LM",
43
+ "trainable_token_indices": null,
44
+ "use_bdlora": null,
45
+ "use_dora": false,
46
+ "use_qalora": false,
47
+ "use_rslora": false
48
+ }
systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1371/chat_template.jinja ADDED
@@ -0,0 +1,154 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {%- set image_count = namespace(value=0) %}
2
+ {%- set video_count = namespace(value=0) %}
3
+ {%- macro render_content(content, do_vision_count, is_system_content=false) %}
4
+ {%- if content is string %}
5
+ {{- content }}
6
+ {%- elif content is iterable and content is not mapping %}
7
+ {%- for item in content %}
8
+ {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
9
+ {%- if is_system_content %}
10
+ {{- raise_exception('System message cannot contain images.') }}
11
+ {%- endif %}
12
+ {%- if do_vision_count %}
13
+ {%- set image_count.value = image_count.value + 1 %}
14
+ {%- endif %}
15
+ {%- if add_vision_id %}
16
+ {{- 'Picture ' ~ image_count.value ~ ': ' }}
17
+ {%- endif %}
18
+ {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
19
+ {%- elif 'video' in item or item.type == 'video' %}
20
+ {%- if is_system_content %}
21
+ {{- raise_exception('System message cannot contain videos.') }}
22
+ {%- endif %}
23
+ {%- if do_vision_count %}
24
+ {%- set video_count.value = video_count.value + 1 %}
25
+ {%- endif %}
26
+ {%- if add_vision_id %}
27
+ {{- 'Video ' ~ video_count.value ~ ': ' }}
28
+ {%- endif %}
29
+ {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
30
+ {%- elif 'text' in item %}
31
+ {{- item.text }}
32
+ {%- else %}
33
+ {{- raise_exception('Unexpected item type in content.') }}
34
+ {%- endif %}
35
+ {%- endfor %}
36
+ {%- elif content is none or content is undefined %}
37
+ {{- '' }}
38
+ {%- else %}
39
+ {{- raise_exception('Unexpected content type.') }}
40
+ {%- endif %}
41
+ {%- endmacro %}
42
+ {%- if not messages %}
43
+ {{- raise_exception('No messages provided.') }}
44
+ {%- endif %}
45
+ {%- if tools and tools is iterable and tools is not mapping %}
46
+ {{- '<|im_start|>system\n' }}
47
+ {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
48
+ {%- for tool in tools %}
49
+ {{- "\n" }}
50
+ {{- tool | tojson }}
51
+ {%- endfor %}
52
+ {{- "\n</tools>" }}
53
+ {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
54
+ {%- if messages[0].role == 'system' %}
55
+ {%- set content = render_content(messages[0].content, false, true)|trim %}
56
+ {%- if content %}
57
+ {{- '\n\n' + content }}
58
+ {%- endif %}
59
+ {%- endif %}
60
+ {{- '<|im_end|>\n' }}
61
+ {%- else %}
62
+ {%- if messages[0].role == 'system' %}
63
+ {%- set content = render_content(messages[0].content, false, true)|trim %}
64
+ {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
65
+ {%- endif %}
66
+ {%- endif %}
67
+ {%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
68
+ {%- for message in messages[::-1] %}
69
+ {%- set index = (messages|length - 1) - loop.index0 %}
70
+ {%- if ns.multi_step_tool and message.role == "user" %}
71
+ {%- set content = render_content(message.content, false)|trim %}
72
+ {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
73
+ {%- set ns.multi_step_tool = false %}
74
+ {%- set ns.last_query_index = index %}
75
+ {%- endif %}
76
+ {%- endif %}
77
+ {%- endfor %}
78
+ {%- if ns.multi_step_tool %}
79
+ {{- raise_exception('No user query found in messages.') }}
80
+ {%- endif %}
81
+ {%- for message in messages %}
82
+ {%- set content = render_content(message.content, true)|trim %}
83
+ {%- if message.role == "system" %}
84
+ {%- if not loop.first %}
85
+ {{- raise_exception('System message must be at the beginning.') }}
86
+ {%- endif %}
87
+ {%- elif message.role == "user" %}
88
+ {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
89
+ {%- elif message.role == "assistant" %}
90
+ {%- set reasoning_content = '' %}
91
+ {%- if message.reasoning_content is string %}
92
+ {%- set reasoning_content = message.reasoning_content %}
93
+ {%- else %}
94
+ {%- if '</think>' in content %}
95
+ {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
96
+ {%- set content = content.split('</think>')[-1].lstrip('\n') %}
97
+ {%- endif %}
98
+ {%- endif %}
99
+ {%- set reasoning_content = reasoning_content|trim %}
100
+ {%- if loop.index0 > ns.last_query_index %}
101
+ {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
102
+ {%- else %}
103
+ {{- '<|im_start|>' + message.role + '\n' + content }}
104
+ {%- endif %}
105
+ {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
106
+ {%- for tool_call in message.tool_calls %}
107
+ {%- if tool_call.function is defined %}
108
+ {%- set tool_call = tool_call.function %}
109
+ {%- endif %}
110
+ {%- if loop.first %}
111
+ {%- if content|trim %}
112
+ {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
113
+ {%- else %}
114
+ {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
115
+ {%- endif %}
116
+ {%- else %}
117
+ {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
118
+ {%- endif %}
119
+ {%- if tool_call.arguments is defined %}
120
+ {%- for args_name, args_value in tool_call.arguments|items %}
121
+ {{- '<parameter=' + args_name + '>\n' }}
122
+ {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
123
+ {{- args_value }}
124
+ {{- '\n</parameter>\n' }}
125
+ {%- endfor %}
126
+ {%- endif %}
127
+ {{- '</function>\n</tool_call>' }}
128
+ {%- endfor %}
129
+ {%- endif %}
130
+ {{- '<|im_end|>\n' }}
131
+ {%- elif message.role == "tool" %}
132
+ {%- if loop.previtem and loop.previtem.role != "tool" %}
133
+ {{- '<|im_start|>user' }}
134
+ {%- endif %}
135
+ {{- '\n<tool_response>\n' }}
136
+ {{- content }}
137
+ {{- '\n</tool_response>' }}
138
+ {%- if not loop.last and loop.nextitem.role != "tool" %}
139
+ {{- '<|im_end|>\n' }}
140
+ {%- elif loop.last %}
141
+ {{- '<|im_end|>\n' }}
142
+ {%- endif %}
143
+ {%- else %}
144
+ {{- raise_exception('Unexpected message role.') }}
145
+ {%- endif %}
146
+ {%- endfor %}
147
+ {%- if add_generation_prompt %}
148
+ {{- '<|im_start|>assistant\n' }}
149
+ {%- if enable_thinking is defined and enable_thinking is true %}
150
+ {{- '<think>\n' }}
151
+ {%- else %}
152
+ {{- '<think>\n\n</think>\n\n' }}
153
+ {%- endif %}
154
+ {%- endif %}
systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1371/tokenizer_config.json ADDED
@@ -0,0 +1,31 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "audio_bos_token": "<|audio_start|>",
4
+ "audio_eos_token": "<|audio_end|>",
5
+ "audio_token": "<|audio_pad|>",
6
+ "backend": "tokenizers",
7
+ "bos_token": null,
8
+ "clean_up_tokenization_spaces": false,
9
+ "eos_token": "<|endoftext|>",
10
+ "errors": "replace",
11
+ "image_token": "<|image_pad|>",
12
+ "is_local": false,
13
+ "model_max_length": 262144,
14
+ "model_specific_special_tokens": {
15
+ "audio_bos_token": "<|audio_start|>",
16
+ "audio_eos_token": "<|audio_end|>",
17
+ "audio_token": "<|audio_pad|>",
18
+ "image_token": "<|image_pad|>",
19
+ "video_token": "<|video_pad|>",
20
+ "vision_bos_token": "<|vision_start|>",
21
+ "vision_eos_token": "<|vision_end|>"
22
+ },
23
+ "pad_token": "<|endoftext|>",
24
+ "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
25
+ "split_special_tokens": false,
26
+ "tokenizer_class": "TokenizersBackend",
27
+ "unk_token": null,
28
+ "video_token": "<|video_pad|>",
29
+ "vision_bos_token": "<|vision_start|>",
30
+ "vision_eos_token": "<|vision_end|>"
31
+ }
systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1371/trainer_state.json ADDED
@@ -0,0 +1,337 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 3.0,
6
+ "eval_steps": 500,
7
+ "global_step": 1371,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "entropy": 2.074948468208313,
14
+ "epoch": 0.10952902519167579,
15
+ "grad_norm": 2.3511903285980225,
16
+ "learning_rate": 1.2698308676050528e-05,
17
+ "loss": 1.9153688049316406,
18
+ "mean_token_accuracy": 0.6077924159169197,
19
+ "num_tokens": 113087.0,
20
+ "step": 50
21
+ },
22
+ {
23
+ "entropy": 1.054056368470192,
24
+ "epoch": 0.21905805038335158,
25
+ "grad_norm": 2.3081915378570557,
26
+ "learning_rate": 2.5655766508755147e-05,
27
+ "loss": 0.9633240509033203,
28
+ "mean_token_accuracy": 0.741833764910698,
29
+ "num_tokens": 230392.0,
30
+ "step": 100
31
+ },
32
+ {
33
+ "entropy": 0.9184817910194397,
34
+ "epoch": 0.32858707557502737,
35
+ "grad_norm": 2.0965280532836914,
36
+ "learning_rate": 3.861322434145977e-05,
37
+ "loss": 0.8405924987792969,
38
+ "mean_token_accuracy": 0.7688284432888031,
39
+ "num_tokens": 341746.0,
40
+ "step": 150
41
+ },
42
+ {
43
+ "entropy": 0.8618861585855484,
44
+ "epoch": 0.43811610076670315,
45
+ "grad_norm": 2.2506542205810547,
46
+ "learning_rate": 5.157068217416438e-05,
47
+ "loss": 0.7841030883789063,
48
+ "mean_token_accuracy": 0.7796466761827469,
49
+ "num_tokens": 460483.0,
50
+ "step": 200
51
+ },
52
+ {
53
+ "entropy": 0.819869134426117,
54
+ "epoch": 0.547645125958379,
55
+ "grad_norm": 1.4058395624160767,
56
+ "learning_rate": 6.4528140006869e-05,
57
+ "loss": 0.7524736785888672,
58
+ "mean_token_accuracy": 0.7900535291433335,
59
+ "num_tokens": 570627.0,
60
+ "step": 250
61
+ },
62
+ {
63
+ "entropy": 0.8075837278366089,
64
+ "epoch": 0.6571741511500547,
65
+ "grad_norm": 1.5815014839172363,
66
+ "learning_rate": 7.748559783957362e-05,
67
+ "loss": 0.7388697814941406,
68
+ "mean_token_accuracy": 0.7914055424928665,
69
+ "num_tokens": 677217.0,
70
+ "step": 300
71
+ },
72
+ {
73
+ "entropy": 0.7868322026729584,
74
+ "epoch": 0.7667031763417306,
75
+ "grad_norm": 1.2377994060516357,
76
+ "learning_rate": 9.044305567227824e-05,
77
+ "loss": 0.7226168060302735,
78
+ "mean_token_accuracy": 0.7953901988267899,
79
+ "num_tokens": 790185.0,
80
+ "step": 350
81
+ },
82
+ {
83
+ "entropy": 0.7526325100660324,
84
+ "epoch": 0.8762322015334063,
85
+ "grad_norm": 1.203137755393982,
86
+ "learning_rate": 0.00010340051350498286,
87
+ "loss": 0.6991680908203125,
88
+ "mean_token_accuracy": 0.7992895567417144,
89
+ "num_tokens": 910080.0,
90
+ "step": 400
91
+ },
92
+ {
93
+ "entropy": 0.7626404595375061,
94
+ "epoch": 0.9857612267250822,
95
+ "grad_norm": 1.1625452041625977,
96
+ "learning_rate": 0.00011635797133768749,
97
+ "loss": 0.7055020904541016,
98
+ "mean_token_accuracy": 0.7997505795955658,
99
+ "num_tokens": 1023473.0,
100
+ "step": 450
101
+ },
102
+ {
103
+ "epoch": 1.0,
104
+ "eval_entropy": 0.7441067624659765,
105
+ "eval_loss": 0.7209385633468628,
106
+ "eval_mean_token_accuracy": 0.7895652093584575,
107
+ "eval_num_tokens": 1038250.0,
108
+ "eval_runtime": 45.5076,
109
+ "eval_samples_per_second": 22.018,
110
+ "eval_steps_per_second": 2.769,
111
+ "step": 457
112
+ },
113
+ {
114
+ "entropy": 0.720724637460227,
115
+ "epoch": 1.0941949616648412,
116
+ "grad_norm": 1.2537180185317993,
117
+ "learning_rate": 0.00011840069618947894,
118
+ "loss": 0.6586117553710937,
119
+ "mean_token_accuracy": 0.8081232917429221,
120
+ "num_tokens": 1128922.0,
121
+ "step": 500
122
+ },
123
+ {
124
+ "entropy": 0.7084310132265091,
125
+ "epoch": 1.203723986856517,
126
+ "grad_norm": 1.2347161769866943,
127
+ "learning_rate": 0.00011828501915154198,
128
+ "loss": 0.650611572265625,
129
+ "mean_token_accuracy": 0.808387525677681,
130
+ "num_tokens": 1245635.0,
131
+ "step": 550
132
+ },
133
+ {
134
+ "entropy": 0.706004958152771,
135
+ "epoch": 1.3132530120481927,
136
+ "grad_norm": 1.2140169143676758,
137
+ "learning_rate": 0.00011808319665683977,
138
+ "loss": 0.6442465209960937,
139
+ "mean_token_accuracy": 0.8122780376672745,
140
+ "num_tokens": 1354885.0,
141
+ "step": 600
142
+ },
143
+ {
144
+ "entropy": 0.6854291236400605,
145
+ "epoch": 1.4227820372398685,
146
+ "grad_norm": 1.1137559413909912,
147
+ "learning_rate": 0.00011779552303848118,
148
+ "loss": 0.6293138885498046,
149
+ "mean_token_accuracy": 0.8158071845769882,
150
+ "num_tokens": 1472232.0,
151
+ "step": 650
152
+ },
153
+ {
154
+ "entropy": 0.6885070586204529,
155
+ "epoch": 1.5323110624315444,
156
+ "grad_norm": 1.1074714660644531,
157
+ "learning_rate": 0.00011742241783280468,
158
+ "loss": 0.6259001541137695,
159
+ "mean_token_accuracy": 0.8164563828706741,
160
+ "num_tokens": 1587494.0,
161
+ "step": 700
162
+ },
163
+ {
164
+ "entropy": 0.6932114177942276,
165
+ "epoch": 1.6418400876232202,
166
+ "grad_norm": 1.1099495887756348,
167
+ "learning_rate": 0.00011696442516753647,
168
+ "loss": 0.6335408401489258,
169
+ "mean_token_accuracy": 0.8156683903932571,
170
+ "num_tokens": 1696927.0,
171
+ "step": 750
172
+ },
173
+ {
174
+ "entropy": 0.6787053138017655,
175
+ "epoch": 1.751369112814896,
176
+ "grad_norm": 1.2505418062210083,
177
+ "learning_rate": 0.00011642221296824764,
178
+ "loss": 0.6210909271240235,
179
+ "mean_token_accuracy": 0.8153067737817764,
180
+ "num_tokens": 1813415.0,
181
+ "step": 800
182
+ },
183
+ {
184
+ "entropy": 0.6693948286771775,
185
+ "epoch": 1.8608981380065717,
186
+ "grad_norm": 1.0963624715805054,
187
+ "learning_rate": 0.00011579657198426736,
188
+ "loss": 0.6105814361572266,
189
+ "mean_token_accuracy": 0.819609425663948,
190
+ "num_tokens": 1924911.0,
191
+ "step": 850
192
+ },
193
+ {
194
+ "entropy": 0.6634757363796234,
195
+ "epoch": 1.9704271631982475,
196
+ "grad_norm": 0.8766036033630371,
197
+ "learning_rate": 0.00011508841463547313,
198
+ "loss": 0.6100498580932617,
199
+ "mean_token_accuracy": 0.8197187691926956,
200
+ "num_tokens": 2044650.0,
201
+ "step": 900
202
+ },
203
+ {
204
+ "epoch": 2.0,
205
+ "eval_entropy": 0.6342300275961558,
206
+ "eval_loss": 0.6517631411552429,
207
+ "eval_mean_token_accuracy": 0.807878240233376,
208
+ "eval_num_tokens": 2076500.0,
209
+ "eval_runtime": 44.6225,
210
+ "eval_samples_per_second": 22.455,
211
+ "eval_steps_per_second": 2.824,
212
+ "step": 914
213
+ },
214
+ {
215
+ "entropy": 0.6060671378867795,
216
+ "epoch": 2.078860898138007,
217
+ "grad_norm": 0.8028743267059326,
218
+ "learning_rate": 0.00011429877368163915,
219
+ "loss": 0.5440861511230469,
220
+ "mean_token_accuracy": 0.833196161973356,
221
+ "num_tokens": 2160498.0,
222
+ "step": 950
223
+ },
224
+ {
225
+ "entropy": 0.5839239662885666,
226
+ "epoch": 2.1883899233296824,
227
+ "grad_norm": 0.8585777878761292,
228
+ "learning_rate": 0.00011342880071628435,
229
+ "loss": 0.5182851409912109,
230
+ "mean_token_accuracy": 0.8406627786159515,
231
+ "num_tokens": 2269979.0,
232
+ "step": 1000
233
+ },
234
+ {
235
+ "entropy": 0.5949062475562096,
236
+ "epoch": 2.297918948521358,
237
+ "grad_norm": 0.9953155517578125,
238
+ "learning_rate": 0.00011247976448721578,
239
+ "loss": 0.5327045822143555,
240
+ "mean_token_accuracy": 0.8344444561004639,
241
+ "num_tokens": 2378362.0,
242
+ "step": 1050
243
+ },
244
+ {
245
+ "entropy": 0.583380132317543,
246
+ "epoch": 2.407447973713034,
247
+ "grad_norm": 0.9235143065452576,
248
+ "learning_rate": 0.00011145304904621704,
249
+ "loss": 0.520841178894043,
250
+ "mean_token_accuracy": 0.839934149980545,
251
+ "num_tokens": 2498494.0,
252
+ "step": 1100
253
+ },
254
+ {
255
+ "entropy": 0.5811308068037033,
256
+ "epoch": 2.5169769989047097,
257
+ "grad_norm": 1.1114075183868408,
258
+ "learning_rate": 0.00011035015173058038,
259
+ "loss": 0.5212027740478515,
260
+ "mean_token_accuracy": 0.838929146528244,
261
+ "num_tokens": 2611646.0,
262
+ "step": 1150
263
+ },
264
+ {
265
+ "entropy": 0.5903497302532196,
266
+ "epoch": 2.6265060240963853,
267
+ "grad_norm": 1.053895354270935,
268
+ "learning_rate": 0.0001091726809794255,
269
+ "loss": 0.5264106750488281,
270
+ "mean_token_accuracy": 0.8369076484441758,
271
+ "num_tokens": 2717744.0,
272
+ "step": 1200
273
+ },
274
+ {
275
+ "entropy": 0.5949014449119567,
276
+ "epoch": 2.7360350492880614,
277
+ "grad_norm": 0.7563770413398743,
278
+ "learning_rate": 0.00010792235398799048,
279
+ "loss": 0.5322903442382813,
280
+ "mean_token_accuracy": 0.8368168956041336,
281
+ "num_tokens": 2827750.0,
282
+ "step": 1250
283
+ },
284
+ {
285
+ "entropy": 0.5745143675804139,
286
+ "epoch": 2.845564074479737,
287
+ "grad_norm": 0.8103823065757751,
288
+ "learning_rate": 0.00010660099420331526,
289
+ "loss": 0.5154821014404297,
290
+ "mean_token_accuracy": 0.8410378849506378,
291
+ "num_tokens": 2948958.0,
292
+ "step": 1300
293
+ },
294
+ {
295
+ "entropy": 0.5665904894471169,
296
+ "epoch": 2.955093099671413,
297
+ "grad_norm": 0.8764206171035767,
298
+ "learning_rate": 0.00010521052866496986,
299
+ "loss": 0.5121800231933594,
300
+ "mean_token_accuracy": 0.8420848768949508,
301
+ "num_tokens": 3066917.0,
302
+ "step": 1350
303
+ },
304
+ {
305
+ "epoch": 3.0,
306
+ "eval_entropy": 0.5603872095308606,
307
+ "eval_loss": 0.6720245480537415,
308
+ "eval_mean_token_accuracy": 0.8079002830717299,
309
+ "eval_num_tokens": 3114750.0,
310
+ "eval_runtime": 43.9396,
311
+ "eval_samples_per_second": 22.804,
312
+ "eval_steps_per_second": 2.868,
313
+ "step": 1371
314
+ }
315
+ ],
316
+ "logging_steps": 50,
317
+ "max_steps": 4570,
318
+ "num_input_tokens_seen": 0,
319
+ "num_train_epochs": 10,
320
+ "save_steps": 500,
321
+ "stateful_callbacks": {
322
+ "TrainerControl": {
323
+ "args": {
324
+ "should_epoch_stop": false,
325
+ "should_evaluate": false,
326
+ "should_log": false,
327
+ "should_save": true,
328
+ "should_training_stop": false
329
+ },
330
+ "attributes": {}
331
+ }
332
+ },
333
+ "total_flos": 6.747546481924416e+16,
334
+ "train_batch_size": 8,
335
+ "trial_name": null,
336
+ "trial_params": null
337
+ }
systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1828/README.md ADDED
@@ -0,0 +1,209 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: Qwen/Qwen3.5-2B-Base
3
+ library_name: peft
4
+ pipeline_tag: text-generation
5
+ tags:
6
+ - base_model:adapter:Qwen/Qwen3.5-2B-Base
7
+ - lora
8
+ - sft
9
+ - transformers
10
+ - trl
11
+ ---
12
+
13
+ # Model Card for Model ID
14
+
15
+ <!-- Provide a quick summary of what the model is/does. -->
16
+
17
+
18
+
19
+ ## Model Details
20
+
21
+ ### Model Description
22
+
23
+ <!-- Provide a longer summary of what this model is. -->
24
+
25
+
26
+
27
+ - **Developed by:** [More Information Needed]
28
+ - **Funded by [optional]:** [More Information Needed]
29
+ - **Shared by [optional]:** [More Information Needed]
30
+ - **Model type:** [More Information Needed]
31
+ - **Language(s) (NLP):** [More Information Needed]
32
+ - **License:** [More Information Needed]
33
+ - **Finetuned from model [optional]:** [More Information Needed]
34
+
35
+ ### Model Sources [optional]
36
+
37
+ <!-- Provide the basic links for the model. -->
38
+
39
+ - **Repository:** [More Information Needed]
40
+ - **Paper [optional]:** [More Information Needed]
41
+ - **Demo [optional]:** [More Information Needed]
42
+
43
+ ## Uses
44
+
45
+ <!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
46
+
47
+ ### Direct Use
48
+
49
+ <!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
50
+
51
+ [More Information Needed]
52
+
53
+ ### Downstream Use [optional]
54
+
55
+ <!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
56
+
57
+ [More Information Needed]
58
+
59
+ ### Out-of-Scope Use
60
+
61
+ <!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
62
+
63
+ [More Information Needed]
64
+
65
+ ## Bias, Risks, and Limitations
66
+
67
+ <!-- This section is meant to convey both technical and sociotechnical limitations. -->
68
+
69
+ [More Information Needed]
70
+
71
+ ### Recommendations
72
+
73
+ <!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
74
+
75
+ Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
76
+
77
+ ## How to Get Started with the Model
78
+
79
+ Use the code below to get started with the model.
80
+
81
+ [More Information Needed]
82
+
83
+ ## Training Details
84
+
85
+ ### Training Data
86
+
87
+ <!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
88
+
89
+ [More Information Needed]
90
+
91
+ ### Training Procedure
92
+
93
+ <!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
94
+
95
+ #### Preprocessing [optional]
96
+
97
+ [More Information Needed]
98
+
99
+
100
+ #### Training Hyperparameters
101
+
102
+ - **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
103
+
104
+ #### Speeds, Sizes, Times [optional]
105
+
106
+ <!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
107
+
108
+ [More Information Needed]
109
+
110
+ ## Evaluation
111
+
112
+ <!-- This section describes the evaluation protocols and provides the results. -->
113
+
114
+ ### Testing Data, Factors & Metrics
115
+
116
+ #### Testing Data
117
+
118
+ <!-- This should link to a Dataset Card if possible. -->
119
+
120
+ [More Information Needed]
121
+
122
+ #### Factors
123
+
124
+ <!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
125
+
126
+ [More Information Needed]
127
+
128
+ #### Metrics
129
+
130
+ <!-- These are the evaluation metrics being used, ideally with a description of why. -->
131
+
132
+ [More Information Needed]
133
+
134
+ ### Results
135
+
136
+ [More Information Needed]
137
+
138
+ #### Summary
139
+
140
+
141
+
142
+ ## Model Examination [optional]
143
+
144
+ <!-- Relevant interpretability work for the model goes here -->
145
+
146
+ [More Information Needed]
147
+
148
+ ## Environmental Impact
149
+
150
+ <!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
151
+
152
+ Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
153
+
154
+ - **Hardware Type:** [More Information Needed]
155
+ - **Hours used:** [More Information Needed]
156
+ - **Cloud Provider:** [More Information Needed]
157
+ - **Compute Region:** [More Information Needed]
158
+ - **Carbon Emitted:** [More Information Needed]
159
+
160
+ ## Technical Specifications [optional]
161
+
162
+ ### Model Architecture and Objective
163
+
164
+ [More Information Needed]
165
+
166
+ ### Compute Infrastructure
167
+
168
+ [More Information Needed]
169
+
170
+ #### Hardware
171
+
172
+ [More Information Needed]
173
+
174
+ #### Software
175
+
176
+ [More Information Needed]
177
+
178
+ ## Citation [optional]
179
+
180
+ <!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
181
+
182
+ **BibTeX:**
183
+
184
+ [More Information Needed]
185
+
186
+ **APA:**
187
+
188
+ [More Information Needed]
189
+
190
+ ## Glossary [optional]
191
+
192
+ <!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
193
+
194
+ [More Information Needed]
195
+
196
+ ## More Information [optional]
197
+
198
+ [More Information Needed]
199
+
200
+ ## Model Card Authors [optional]
201
+
202
+ [More Information Needed]
203
+
204
+ ## Model Card Contact
205
+
206
+ [More Information Needed]
207
+ ### Framework versions
208
+
209
+ - PEFT 0.19.1
systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1828/adapter_config.json ADDED
@@ -0,0 +1,48 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alora_invocation_tokens": null,
3
+ "alpha_pattern": {},
4
+ "arrow_config": null,
5
+ "auto_mapping": null,
6
+ "base_model_name_or_path": "Qwen/Qwen3.5-2B-Base",
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": 256,
20
+ "lora_bias": false,
21
+ "lora_dropout": 0.01686122009693236,
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": 128,
30
+ "rank_pattern": {},
31
+ "revision": null,
32
+ "target_modules": [
33
+ "v_proj",
34
+ "o_proj",
35
+ "k_proj",
36
+ "up_proj",
37
+ "q_proj",
38
+ "down_proj",
39
+ "gate_proj"
40
+ ],
41
+ "target_parameters": null,
42
+ "task_type": "CAUSAL_LM",
43
+ "trainable_token_indices": null,
44
+ "use_bdlora": null,
45
+ "use_dora": false,
46
+ "use_qalora": false,
47
+ "use_rslora": false
48
+ }
systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1828/chat_template.jinja ADDED
@@ -0,0 +1,154 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {%- set image_count = namespace(value=0) %}
2
+ {%- set video_count = namespace(value=0) %}
3
+ {%- macro render_content(content, do_vision_count, is_system_content=false) %}
4
+ {%- if content is string %}
5
+ {{- content }}
6
+ {%- elif content is iterable and content is not mapping %}
7
+ {%- for item in content %}
8
+ {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
9
+ {%- if is_system_content %}
10
+ {{- raise_exception('System message cannot contain images.') }}
11
+ {%- endif %}
12
+ {%- if do_vision_count %}
13
+ {%- set image_count.value = image_count.value + 1 %}
14
+ {%- endif %}
15
+ {%- if add_vision_id %}
16
+ {{- 'Picture ' ~ image_count.value ~ ': ' }}
17
+ {%- endif %}
18
+ {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
19
+ {%- elif 'video' in item or item.type == 'video' %}
20
+ {%- if is_system_content %}
21
+ {{- raise_exception('System message cannot contain videos.') }}
22
+ {%- endif %}
23
+ {%- if do_vision_count %}
24
+ {%- set video_count.value = video_count.value + 1 %}
25
+ {%- endif %}
26
+ {%- if add_vision_id %}
27
+ {{- 'Video ' ~ video_count.value ~ ': ' }}
28
+ {%- endif %}
29
+ {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
30
+ {%- elif 'text' in item %}
31
+ {{- item.text }}
32
+ {%- else %}
33
+ {{- raise_exception('Unexpected item type in content.') }}
34
+ {%- endif %}
35
+ {%- endfor %}
36
+ {%- elif content is none or content is undefined %}
37
+ {{- '' }}
38
+ {%- else %}
39
+ {{- raise_exception('Unexpected content type.') }}
40
+ {%- endif %}
41
+ {%- endmacro %}
42
+ {%- if not messages %}
43
+ {{- raise_exception('No messages provided.') }}
44
+ {%- endif %}
45
+ {%- if tools and tools is iterable and tools is not mapping %}
46
+ {{- '<|im_start|>system\n' }}
47
+ {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
48
+ {%- for tool in tools %}
49
+ {{- "\n" }}
50
+ {{- tool | tojson }}
51
+ {%- endfor %}
52
+ {{- "\n</tools>" }}
53
+ {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
54
+ {%- if messages[0].role == 'system' %}
55
+ {%- set content = render_content(messages[0].content, false, true)|trim %}
56
+ {%- if content %}
57
+ {{- '\n\n' + content }}
58
+ {%- endif %}
59
+ {%- endif %}
60
+ {{- '<|im_end|>\n' }}
61
+ {%- else %}
62
+ {%- if messages[0].role == 'system' %}
63
+ {%- set content = render_content(messages[0].content, false, true)|trim %}
64
+ {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
65
+ {%- endif %}
66
+ {%- endif %}
67
+ {%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
68
+ {%- for message in messages[::-1] %}
69
+ {%- set index = (messages|length - 1) - loop.index0 %}
70
+ {%- if ns.multi_step_tool and message.role == "user" %}
71
+ {%- set content = render_content(message.content, false)|trim %}
72
+ {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
73
+ {%- set ns.multi_step_tool = false %}
74
+ {%- set ns.last_query_index = index %}
75
+ {%- endif %}
76
+ {%- endif %}
77
+ {%- endfor %}
78
+ {%- if ns.multi_step_tool %}
79
+ {{- raise_exception('No user query found in messages.') }}
80
+ {%- endif %}
81
+ {%- for message in messages %}
82
+ {%- set content = render_content(message.content, true)|trim %}
83
+ {%- if message.role == "system" %}
84
+ {%- if not loop.first %}
85
+ {{- raise_exception('System message must be at the beginning.') }}
86
+ {%- endif %}
87
+ {%- elif message.role == "user" %}
88
+ {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
89
+ {%- elif message.role == "assistant" %}
90
+ {%- set reasoning_content = '' %}
91
+ {%- if message.reasoning_content is string %}
92
+ {%- set reasoning_content = message.reasoning_content %}
93
+ {%- else %}
94
+ {%- if '</think>' in content %}
95
+ {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
96
+ {%- set content = content.split('</think>')[-1].lstrip('\n') %}
97
+ {%- endif %}
98
+ {%- endif %}
99
+ {%- set reasoning_content = reasoning_content|trim %}
100
+ {%- if loop.index0 > ns.last_query_index %}
101
+ {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
102
+ {%- else %}
103
+ {{- '<|im_start|>' + message.role + '\n' + content }}
104
+ {%- endif %}
105
+ {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
106
+ {%- for tool_call in message.tool_calls %}
107
+ {%- if tool_call.function is defined %}
108
+ {%- set tool_call = tool_call.function %}
109
+ {%- endif %}
110
+ {%- if loop.first %}
111
+ {%- if content|trim %}
112
+ {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
113
+ {%- else %}
114
+ {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
115
+ {%- endif %}
116
+ {%- else %}
117
+ {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
118
+ {%- endif %}
119
+ {%- if tool_call.arguments is defined %}
120
+ {%- for args_name, args_value in tool_call.arguments|items %}
121
+ {{- '<parameter=' + args_name + '>\n' }}
122
+ {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
123
+ {{- args_value }}
124
+ {{- '\n</parameter>\n' }}
125
+ {%- endfor %}
126
+ {%- endif %}
127
+ {{- '</function>\n</tool_call>' }}
128
+ {%- endfor %}
129
+ {%- endif %}
130
+ {{- '<|im_end|>\n' }}
131
+ {%- elif message.role == "tool" %}
132
+ {%- if loop.previtem and loop.previtem.role != "tool" %}
133
+ {{- '<|im_start|>user' }}
134
+ {%- endif %}
135
+ {{- '\n<tool_response>\n' }}
136
+ {{- content }}
137
+ {{- '\n</tool_response>' }}
138
+ {%- if not loop.last and loop.nextitem.role != "tool" %}
139
+ {{- '<|im_end|>\n' }}
140
+ {%- elif loop.last %}
141
+ {{- '<|im_end|>\n' }}
142
+ {%- endif %}
143
+ {%- else %}
144
+ {{- raise_exception('Unexpected message role.') }}
145
+ {%- endif %}
146
+ {%- endfor %}
147
+ {%- if add_generation_prompt %}
148
+ {{- '<|im_start|>assistant\n' }}
149
+ {%- if enable_thinking is defined and enable_thinking is true %}
150
+ {{- '<think>\n' }}
151
+ {%- else %}
152
+ {{- '<think>\n\n</think>\n\n' }}
153
+ {%- endif %}
154
+ {%- endif %}
systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1828/tokenizer_config.json ADDED
@@ -0,0 +1,31 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "audio_bos_token": "<|audio_start|>",
4
+ "audio_eos_token": "<|audio_end|>",
5
+ "audio_token": "<|audio_pad|>",
6
+ "backend": "tokenizers",
7
+ "bos_token": null,
8
+ "clean_up_tokenization_spaces": false,
9
+ "eos_token": "<|endoftext|>",
10
+ "errors": "replace",
11
+ "image_token": "<|image_pad|>",
12
+ "is_local": false,
13
+ "model_max_length": 262144,
14
+ "model_specific_special_tokens": {
15
+ "audio_bos_token": "<|audio_start|>",
16
+ "audio_eos_token": "<|audio_end|>",
17
+ "audio_token": "<|audio_pad|>",
18
+ "image_token": "<|image_pad|>",
19
+ "video_token": "<|video_pad|>",
20
+ "vision_bos_token": "<|vision_start|>",
21
+ "vision_eos_token": "<|vision_end|>"
22
+ },
23
+ "pad_token": "<|endoftext|>",
24
+ "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
25
+ "split_special_tokens": false,
26
+ "tokenizer_class": "TokenizersBackend",
27
+ "unk_token": null,
28
+ "video_token": "<|video_pad|>",
29
+ "vision_bos_token": "<|vision_start|>",
30
+ "vision_eos_token": "<|vision_end|>"
31
+ }
systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1828/trainer_state.json ADDED
@@ -0,0 +1,438 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 4.0,
6
+ "eval_steps": 500,
7
+ "global_step": 1828,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "entropy": 2.074948468208313,
14
+ "epoch": 0.10952902519167579,
15
+ "grad_norm": 2.3511903285980225,
16
+ "learning_rate": 1.2698308676050528e-05,
17
+ "loss": 1.9153688049316406,
18
+ "mean_token_accuracy": 0.6077924159169197,
19
+ "num_tokens": 113087.0,
20
+ "step": 50
21
+ },
22
+ {
23
+ "entropy": 1.054056368470192,
24
+ "epoch": 0.21905805038335158,
25
+ "grad_norm": 2.3081915378570557,
26
+ "learning_rate": 2.5655766508755147e-05,
27
+ "loss": 0.9633240509033203,
28
+ "mean_token_accuracy": 0.741833764910698,
29
+ "num_tokens": 230392.0,
30
+ "step": 100
31
+ },
32
+ {
33
+ "entropy": 0.9184817910194397,
34
+ "epoch": 0.32858707557502737,
35
+ "grad_norm": 2.0965280532836914,
36
+ "learning_rate": 3.861322434145977e-05,
37
+ "loss": 0.8405924987792969,
38
+ "mean_token_accuracy": 0.7688284432888031,
39
+ "num_tokens": 341746.0,
40
+ "step": 150
41
+ },
42
+ {
43
+ "entropy": 0.8618861585855484,
44
+ "epoch": 0.43811610076670315,
45
+ "grad_norm": 2.2506542205810547,
46
+ "learning_rate": 5.157068217416438e-05,
47
+ "loss": 0.7841030883789063,
48
+ "mean_token_accuracy": 0.7796466761827469,
49
+ "num_tokens": 460483.0,
50
+ "step": 200
51
+ },
52
+ {
53
+ "entropy": 0.819869134426117,
54
+ "epoch": 0.547645125958379,
55
+ "grad_norm": 1.4058395624160767,
56
+ "learning_rate": 6.4528140006869e-05,
57
+ "loss": 0.7524736785888672,
58
+ "mean_token_accuracy": 0.7900535291433335,
59
+ "num_tokens": 570627.0,
60
+ "step": 250
61
+ },
62
+ {
63
+ "entropy": 0.8075837278366089,
64
+ "epoch": 0.6571741511500547,
65
+ "grad_norm": 1.5815014839172363,
66
+ "learning_rate": 7.748559783957362e-05,
67
+ "loss": 0.7388697814941406,
68
+ "mean_token_accuracy": 0.7914055424928665,
69
+ "num_tokens": 677217.0,
70
+ "step": 300
71
+ },
72
+ {
73
+ "entropy": 0.7868322026729584,
74
+ "epoch": 0.7667031763417306,
75
+ "grad_norm": 1.2377994060516357,
76
+ "learning_rate": 9.044305567227824e-05,
77
+ "loss": 0.7226168060302735,
78
+ "mean_token_accuracy": 0.7953901988267899,
79
+ "num_tokens": 790185.0,
80
+ "step": 350
81
+ },
82
+ {
83
+ "entropy": 0.7526325100660324,
84
+ "epoch": 0.8762322015334063,
85
+ "grad_norm": 1.203137755393982,
86
+ "learning_rate": 0.00010340051350498286,
87
+ "loss": 0.6991680908203125,
88
+ "mean_token_accuracy": 0.7992895567417144,
89
+ "num_tokens": 910080.0,
90
+ "step": 400
91
+ },
92
+ {
93
+ "entropy": 0.7626404595375061,
94
+ "epoch": 0.9857612267250822,
95
+ "grad_norm": 1.1625452041625977,
96
+ "learning_rate": 0.00011635797133768749,
97
+ "loss": 0.7055020904541016,
98
+ "mean_token_accuracy": 0.7997505795955658,
99
+ "num_tokens": 1023473.0,
100
+ "step": 450
101
+ },
102
+ {
103
+ "epoch": 1.0,
104
+ "eval_entropy": 0.7441067624659765,
105
+ "eval_loss": 0.7209385633468628,
106
+ "eval_mean_token_accuracy": 0.7895652093584575,
107
+ "eval_num_tokens": 1038250.0,
108
+ "eval_runtime": 45.5076,
109
+ "eval_samples_per_second": 22.018,
110
+ "eval_steps_per_second": 2.769,
111
+ "step": 457
112
+ },
113
+ {
114
+ "entropy": 0.720724637460227,
115
+ "epoch": 1.0941949616648412,
116
+ "grad_norm": 1.2537180185317993,
117
+ "learning_rate": 0.00011840069618947894,
118
+ "loss": 0.6586117553710937,
119
+ "mean_token_accuracy": 0.8081232917429221,
120
+ "num_tokens": 1128922.0,
121
+ "step": 500
122
+ },
123
+ {
124
+ "entropy": 0.7084310132265091,
125
+ "epoch": 1.203723986856517,
126
+ "grad_norm": 1.2347161769866943,
127
+ "learning_rate": 0.00011828501915154198,
128
+ "loss": 0.650611572265625,
129
+ "mean_token_accuracy": 0.808387525677681,
130
+ "num_tokens": 1245635.0,
131
+ "step": 550
132
+ },
133
+ {
134
+ "entropy": 0.706004958152771,
135
+ "epoch": 1.3132530120481927,
136
+ "grad_norm": 1.2140169143676758,
137
+ "learning_rate": 0.00011808319665683977,
138
+ "loss": 0.6442465209960937,
139
+ "mean_token_accuracy": 0.8122780376672745,
140
+ "num_tokens": 1354885.0,
141
+ "step": 600
142
+ },
143
+ {
144
+ "entropy": 0.6854291236400605,
145
+ "epoch": 1.4227820372398685,
146
+ "grad_norm": 1.1137559413909912,
147
+ "learning_rate": 0.00011779552303848118,
148
+ "loss": 0.6293138885498046,
149
+ "mean_token_accuracy": 0.8158071845769882,
150
+ "num_tokens": 1472232.0,
151
+ "step": 650
152
+ },
153
+ {
154
+ "entropy": 0.6885070586204529,
155
+ "epoch": 1.5323110624315444,
156
+ "grad_norm": 1.1074714660644531,
157
+ "learning_rate": 0.00011742241783280468,
158
+ "loss": 0.6259001541137695,
159
+ "mean_token_accuracy": 0.8164563828706741,
160
+ "num_tokens": 1587494.0,
161
+ "step": 700
162
+ },
163
+ {
164
+ "entropy": 0.6932114177942276,
165
+ "epoch": 1.6418400876232202,
166
+ "grad_norm": 1.1099495887756348,
167
+ "learning_rate": 0.00011696442516753647,
168
+ "loss": 0.6335408401489258,
169
+ "mean_token_accuracy": 0.8156683903932571,
170
+ "num_tokens": 1696927.0,
171
+ "step": 750
172
+ },
173
+ {
174
+ "entropy": 0.6787053138017655,
175
+ "epoch": 1.751369112814896,
176
+ "grad_norm": 1.2505418062210083,
177
+ "learning_rate": 0.00011642221296824764,
178
+ "loss": 0.6210909271240235,
179
+ "mean_token_accuracy": 0.8153067737817764,
180
+ "num_tokens": 1813415.0,
181
+ "step": 800
182
+ },
183
+ {
184
+ "entropy": 0.6693948286771775,
185
+ "epoch": 1.8608981380065717,
186
+ "grad_norm": 1.0963624715805054,
187
+ "learning_rate": 0.00011579657198426736,
188
+ "loss": 0.6105814361572266,
189
+ "mean_token_accuracy": 0.819609425663948,
190
+ "num_tokens": 1924911.0,
191
+ "step": 850
192
+ },
193
+ {
194
+ "entropy": 0.6634757363796234,
195
+ "epoch": 1.9704271631982475,
196
+ "grad_norm": 0.8766036033630371,
197
+ "learning_rate": 0.00011508841463547313,
198
+ "loss": 0.6100498580932617,
199
+ "mean_token_accuracy": 0.8197187691926956,
200
+ "num_tokens": 2044650.0,
201
+ "step": 900
202
+ },
203
+ {
204
+ "epoch": 2.0,
205
+ "eval_entropy": 0.6342300275961558,
206
+ "eval_loss": 0.6517631411552429,
207
+ "eval_mean_token_accuracy": 0.807878240233376,
208
+ "eval_num_tokens": 2076500.0,
209
+ "eval_runtime": 44.6225,
210
+ "eval_samples_per_second": 22.455,
211
+ "eval_steps_per_second": 2.824,
212
+ "step": 914
213
+ },
214
+ {
215
+ "entropy": 0.6060671378867795,
216
+ "epoch": 2.078860898138007,
217
+ "grad_norm": 0.8028743267059326,
218
+ "learning_rate": 0.00011429877368163915,
219
+ "loss": 0.5440861511230469,
220
+ "mean_token_accuracy": 0.833196161973356,
221
+ "num_tokens": 2160498.0,
222
+ "step": 950
223
+ },
224
+ {
225
+ "entropy": 0.5839239662885666,
226
+ "epoch": 2.1883899233296824,
227
+ "grad_norm": 0.8585777878761292,
228
+ "learning_rate": 0.00011342880071628435,
229
+ "loss": 0.5182851409912109,
230
+ "mean_token_accuracy": 0.8406627786159515,
231
+ "num_tokens": 2269979.0,
232
+ "step": 1000
233
+ },
234
+ {
235
+ "entropy": 0.5949062475562096,
236
+ "epoch": 2.297918948521358,
237
+ "grad_norm": 0.9953155517578125,
238
+ "learning_rate": 0.00011247976448721578,
239
+ "loss": 0.5327045822143555,
240
+ "mean_token_accuracy": 0.8344444561004639,
241
+ "num_tokens": 2378362.0,
242
+ "step": 1050
243
+ },
244
+ {
245
+ "entropy": 0.583380132317543,
246
+ "epoch": 2.407447973713034,
247
+ "grad_norm": 0.9235143065452576,
248
+ "learning_rate": 0.00011145304904621704,
249
+ "loss": 0.520841178894043,
250
+ "mean_token_accuracy": 0.839934149980545,
251
+ "num_tokens": 2498494.0,
252
+ "step": 1100
253
+ },
254
+ {
255
+ "entropy": 0.5811308068037033,
256
+ "epoch": 2.5169769989047097,
257
+ "grad_norm": 1.1114075183868408,
258
+ "learning_rate": 0.00011035015173058038,
259
+ "loss": 0.5212027740478515,
260
+ "mean_token_accuracy": 0.838929146528244,
261
+ "num_tokens": 2611646.0,
262
+ "step": 1150
263
+ },
264
+ {
265
+ "entropy": 0.5903497302532196,
266
+ "epoch": 2.6265060240963853,
267
+ "grad_norm": 1.053895354270935,
268
+ "learning_rate": 0.0001091726809794255,
269
+ "loss": 0.5264106750488281,
270
+ "mean_token_accuracy": 0.8369076484441758,
271
+ "num_tokens": 2717744.0,
272
+ "step": 1200
273
+ },
274
+ {
275
+ "entropy": 0.5949014449119567,
276
+ "epoch": 2.7360350492880614,
277
+ "grad_norm": 0.7563770413398743,
278
+ "learning_rate": 0.00010792235398799048,
279
+ "loss": 0.5322903442382813,
280
+ "mean_token_accuracy": 0.8368168956041336,
281
+ "num_tokens": 2827750.0,
282
+ "step": 1250
283
+ },
284
+ {
285
+ "entropy": 0.5745143675804139,
286
+ "epoch": 2.845564074479737,
287
+ "grad_norm": 0.8103823065757751,
288
+ "learning_rate": 0.00010660099420331526,
289
+ "loss": 0.5154821014404297,
290
+ "mean_token_accuracy": 0.8410378849506378,
291
+ "num_tokens": 2948958.0,
292
+ "step": 1300
293
+ },
294
+ {
295
+ "entropy": 0.5665904894471169,
296
+ "epoch": 2.955093099671413,
297
+ "grad_norm": 0.8764206171035767,
298
+ "learning_rate": 0.00010521052866496986,
299
+ "loss": 0.5121800231933594,
300
+ "mean_token_accuracy": 0.8420848768949508,
301
+ "num_tokens": 3066917.0,
302
+ "step": 1350
303
+ },
304
+ {
305
+ "epoch": 3.0,
306
+ "eval_entropy": 0.5603872095308606,
307
+ "eval_loss": 0.6720245480537415,
308
+ "eval_mean_token_accuracy": 0.8079002830717299,
309
+ "eval_num_tokens": 3114750.0,
310
+ "eval_runtime": 43.9396,
311
+ "eval_samples_per_second": 22.804,
312
+ "eval_steps_per_second": 2.868,
313
+ "step": 1371
314
+ },
315
+ {
316
+ "entropy": 0.5181072778774031,
317
+ "epoch": 3.063526834611172,
318
+ "grad_norm": 0.889443039894104,
319
+ "learning_rate": 0.00010375298519470596,
320
+ "loss": 0.44869777679443357,
321
+ "mean_token_accuracy": 0.8574847211741438,
322
+ "num_tokens": 3180366.0,
323
+ "step": 1400
324
+ },
325
+ {
326
+ "entropy": 0.4902029836177826,
327
+ "epoch": 3.1730558598028478,
328
+ "grad_norm": 0.8999125957489014,
329
+ "learning_rate": 0.00010223048943913007,
330
+ "loss": 0.41646446228027345,
331
+ "mean_token_accuracy": 0.8655553787946701,
332
+ "num_tokens": 3293918.0,
333
+ "step": 1450
334
+ },
335
+ {
336
+ "entropy": 0.48207574665546415,
337
+ "epoch": 3.2825848849945234,
338
+ "grad_norm": 1.0042189359664917,
339
+ "learning_rate": 0.00010064526176971124,
340
+ "loss": 0.41205951690673825,
341
+ "mean_token_accuracy": 0.866786903142929,
342
+ "num_tokens": 3409058.0,
343
+ "step": 1500
344
+ },
345
+ {
346
+ "entropy": 0.4839185461401939,
347
+ "epoch": 3.3921139101861995,
348
+ "grad_norm": 1.1650396585464478,
349
+ "learning_rate": 9.89996140446443e-05,
350
+ "loss": 0.4102657699584961,
351
+ "mean_token_accuracy": 0.8668668657541275,
352
+ "num_tokens": 3525131.0,
353
+ "step": 1550
354
+ },
355
+ {
356
+ "entropy": 0.4932769918441772,
357
+ "epoch": 3.501642935377875,
358
+ "grad_norm": 1.0935224294662476,
359
+ "learning_rate": 9.729594623729099e-05,
360
+ "loss": 0.4225375366210937,
361
+ "mean_token_accuracy": 0.8634544950723648,
362
+ "num_tokens": 3641504.0,
363
+ "step": 1600
364
+ },
365
+ {
366
+ "entropy": 0.5046219238638878,
367
+ "epoch": 3.6111719605695507,
368
+ "grad_norm": 0.9509316086769104,
369
+ "learning_rate": 9.553674293611626e-05,
370
+ "loss": 0.4319529342651367,
371
+ "mean_token_accuracy": 0.8610132175683975,
372
+ "num_tokens": 3752182.0,
373
+ "step": 1650
374
+ },
375
+ {
376
+ "entropy": 0.5090703725814819,
377
+ "epoch": 3.7207009857612268,
378
+ "grad_norm": 0.9050212502479553,
379
+ "learning_rate": 9.372456972122359e-05,
380
+ "loss": 0.4376397705078125,
381
+ "mean_token_accuracy": 0.8591927194595337,
382
+ "num_tokens": 3866473.0,
383
+ "step": 1700
384
+ },
385
+ {
386
+ "entropy": 0.4903561845421791,
387
+ "epoch": 3.8302300109529024,
388
+ "grad_norm": 1.1624391078948975,
389
+ "learning_rate": 9.186206942277429e-05,
390
+ "loss": 0.4219248962402344,
391
+ "mean_token_accuracy": 0.8636374253034592,
392
+ "num_tokens": 3977393.0,
393
+ "step": 1750
394
+ },
395
+ {
396
+ "entropy": 0.5040662395954132,
397
+ "epoch": 3.9397590361445785,
398
+ "grad_norm": 0.798085629940033,
399
+ "learning_rate": 8.995195826674703e-05,
400
+ "loss": 0.4333121871948242,
401
+ "mean_token_accuracy": 0.86187963783741,
402
+ "num_tokens": 4088936.0,
403
+ "step": 1800
404
+ },
405
+ {
406
+ "epoch": 4.0,
407
+ "eval_entropy": 0.5136857172326436,
408
+ "eval_loss": 0.6814945340156555,
409
+ "eval_mean_token_accuracy": 0.8091600262929523,
410
+ "eval_num_tokens": 4153000.0,
411
+ "eval_runtime": 44.5059,
412
+ "eval_samples_per_second": 22.514,
413
+ "eval_steps_per_second": 2.831,
414
+ "step": 1828
415
+ }
416
+ ],
417
+ "logging_steps": 50,
418
+ "max_steps": 4570,
419
+ "num_input_tokens_seen": 0,
420
+ "num_train_epochs": 10,
421
+ "save_steps": 500,
422
+ "stateful_callbacks": {
423
+ "TrainerControl": {
424
+ "args": {
425
+ "should_epoch_stop": false,
426
+ "should_evaluate": false,
427
+ "should_log": false,
428
+ "should_save": true,
429
+ "should_training_stop": false
430
+ },
431
+ "attributes": {}
432
+ }
433
+ },
434
+ "total_flos": 8.985192850101312e+16,
435
+ "train_batch_size": 8,
436
+ "trial_name": null,
437
+ "trial_params": null
438
+ }
systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-2285/README.md ADDED
@@ -0,0 +1,209 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: Qwen/Qwen3.5-2B-Base
3
+ library_name: peft
4
+ pipeline_tag: text-generation
5
+ tags:
6
+ - base_model:adapter:Qwen/Qwen3.5-2B-Base
7
+ - lora
8
+ - sft
9
+ - transformers
10
+ - trl
11
+ ---
12
+
13
+ # Model Card for Model ID
14
+
15
+ <!-- Provide a quick summary of what the model is/does. -->
16
+
17
+
18
+
19
+ ## Model Details
20
+
21
+ ### Model Description
22
+
23
+ <!-- Provide a longer summary of what this model is. -->
24
+
25
+
26
+
27
+ - **Developed by:** [More Information Needed]
28
+ - **Funded by [optional]:** [More Information Needed]
29
+ - **Shared by [optional]:** [More Information Needed]
30
+ - **Model type:** [More Information Needed]
31
+ - **Language(s) (NLP):** [More Information Needed]
32
+ - **License:** [More Information Needed]
33
+ - **Finetuned from model [optional]:** [More Information Needed]
34
+
35
+ ### Model Sources [optional]
36
+
37
+ <!-- Provide the basic links for the model. -->
38
+
39
+ - **Repository:** [More Information Needed]
40
+ - **Paper [optional]:** [More Information Needed]
41
+ - **Demo [optional]:** [More Information Needed]
42
+
43
+ ## Uses
44
+
45
+ <!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
46
+
47
+ ### Direct Use
48
+
49
+ <!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
50
+
51
+ [More Information Needed]
52
+
53
+ ### Downstream Use [optional]
54
+
55
+ <!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
56
+
57
+ [More Information Needed]
58
+
59
+ ### Out-of-Scope Use
60
+
61
+ <!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
62
+
63
+ [More Information Needed]
64
+
65
+ ## Bias, Risks, and Limitations
66
+
67
+ <!-- This section is meant to convey both technical and sociotechnical limitations. -->
68
+
69
+ [More Information Needed]
70
+
71
+ ### Recommendations
72
+
73
+ <!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
74
+
75
+ Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
76
+
77
+ ## How to Get Started with the Model
78
+
79
+ Use the code below to get started with the model.
80
+
81
+ [More Information Needed]
82
+
83
+ ## Training Details
84
+
85
+ ### Training Data
86
+
87
+ <!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
88
+
89
+ [More Information Needed]
90
+
91
+ ### Training Procedure
92
+
93
+ <!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
94
+
95
+ #### Preprocessing [optional]
96
+
97
+ [More Information Needed]
98
+
99
+
100
+ #### Training Hyperparameters
101
+
102
+ - **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
103
+
104
+ #### Speeds, Sizes, Times [optional]
105
+
106
+ <!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
107
+
108
+ [More Information Needed]
109
+
110
+ ## Evaluation
111
+
112
+ <!-- This section describes the evaluation protocols and provides the results. -->
113
+
114
+ ### Testing Data, Factors & Metrics
115
+
116
+ #### Testing Data
117
+
118
+ <!-- This should link to a Dataset Card if possible. -->
119
+
120
+ [More Information Needed]
121
+
122
+ #### Factors
123
+
124
+ <!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
125
+
126
+ [More Information Needed]
127
+
128
+ #### Metrics
129
+
130
+ <!-- These are the evaluation metrics being used, ideally with a description of why. -->
131
+
132
+ [More Information Needed]
133
+
134
+ ### Results
135
+
136
+ [More Information Needed]
137
+
138
+ #### Summary
139
+
140
+
141
+
142
+ ## Model Examination [optional]
143
+
144
+ <!-- Relevant interpretability work for the model goes here -->
145
+
146
+ [More Information Needed]
147
+
148
+ ## Environmental Impact
149
+
150
+ <!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
151
+
152
+ Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
153
+
154
+ - **Hardware Type:** [More Information Needed]
155
+ - **Hours used:** [More Information Needed]
156
+ - **Cloud Provider:** [More Information Needed]
157
+ - **Compute Region:** [More Information Needed]
158
+ - **Carbon Emitted:** [More Information Needed]
159
+
160
+ ## Technical Specifications [optional]
161
+
162
+ ### Model Architecture and Objective
163
+
164
+ [More Information Needed]
165
+
166
+ ### Compute Infrastructure
167
+
168
+ [More Information Needed]
169
+
170
+ #### Hardware
171
+
172
+ [More Information Needed]
173
+
174
+ #### Software
175
+
176
+ [More Information Needed]
177
+
178
+ ## Citation [optional]
179
+
180
+ <!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
181
+
182
+ **BibTeX:**
183
+
184
+ [More Information Needed]
185
+
186
+ **APA:**
187
+
188
+ [More Information Needed]
189
+
190
+ ## Glossary [optional]
191
+
192
+ <!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
193
+
194
+ [More Information Needed]
195
+
196
+ ## More Information [optional]
197
+
198
+ [More Information Needed]
199
+
200
+ ## Model Card Authors [optional]
201
+
202
+ [More Information Needed]
203
+
204
+ ## Model Card Contact
205
+
206
+ [More Information Needed]
207
+ ### Framework versions
208
+
209
+ - PEFT 0.19.1
systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-2285/adapter_config.json ADDED
@@ -0,0 +1,48 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alora_invocation_tokens": null,
3
+ "alpha_pattern": {},
4
+ "arrow_config": null,
5
+ "auto_mapping": null,
6
+ "base_model_name_or_path": "Qwen/Qwen3.5-2B-Base",
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": 256,
20
+ "lora_bias": false,
21
+ "lora_dropout": 0.01686122009693236,
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": 128,
30
+ "rank_pattern": {},
31
+ "revision": null,
32
+ "target_modules": [
33
+ "v_proj",
34
+ "o_proj",
35
+ "k_proj",
36
+ "up_proj",
37
+ "q_proj",
38
+ "down_proj",
39
+ "gate_proj"
40
+ ],
41
+ "target_parameters": null,
42
+ "task_type": "CAUSAL_LM",
43
+ "trainable_token_indices": null,
44
+ "use_bdlora": null,
45
+ "use_dora": false,
46
+ "use_qalora": false,
47
+ "use_rslora": false
48
+ }
systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-2285/chat_template.jinja ADDED
@@ -0,0 +1,154 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {%- set image_count = namespace(value=0) %}
2
+ {%- set video_count = namespace(value=0) %}
3
+ {%- macro render_content(content, do_vision_count, is_system_content=false) %}
4
+ {%- if content is string %}
5
+ {{- content }}
6
+ {%- elif content is iterable and content is not mapping %}
7
+ {%- for item in content %}
8
+ {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
9
+ {%- if is_system_content %}
10
+ {{- raise_exception('System message cannot contain images.') }}
11
+ {%- endif %}
12
+ {%- if do_vision_count %}
13
+ {%- set image_count.value = image_count.value + 1 %}
14
+ {%- endif %}
15
+ {%- if add_vision_id %}
16
+ {{- 'Picture ' ~ image_count.value ~ ': ' }}
17
+ {%- endif %}
18
+ {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
19
+ {%- elif 'video' in item or item.type == 'video' %}
20
+ {%- if is_system_content %}
21
+ {{- raise_exception('System message cannot contain videos.') }}
22
+ {%- endif %}
23
+ {%- if do_vision_count %}
24
+ {%- set video_count.value = video_count.value + 1 %}
25
+ {%- endif %}
26
+ {%- if add_vision_id %}
27
+ {{- 'Video ' ~ video_count.value ~ ': ' }}
28
+ {%- endif %}
29
+ {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
30
+ {%- elif 'text' in item %}
31
+ {{- item.text }}
32
+ {%- else %}
33
+ {{- raise_exception('Unexpected item type in content.') }}
34
+ {%- endif %}
35
+ {%- endfor %}
36
+ {%- elif content is none or content is undefined %}
37
+ {{- '' }}
38
+ {%- else %}
39
+ {{- raise_exception('Unexpected content type.') }}
40
+ {%- endif %}
41
+ {%- endmacro %}
42
+ {%- if not messages %}
43
+ {{- raise_exception('No messages provided.') }}
44
+ {%- endif %}
45
+ {%- if tools and tools is iterable and tools is not mapping %}
46
+ {{- '<|im_start|>system\n' }}
47
+ {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
48
+ {%- for tool in tools %}
49
+ {{- "\n" }}
50
+ {{- tool | tojson }}
51
+ {%- endfor %}
52
+ {{- "\n</tools>" }}
53
+ {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
54
+ {%- if messages[0].role == 'system' %}
55
+ {%- set content = render_content(messages[0].content, false, true)|trim %}
56
+ {%- if content %}
57
+ {{- '\n\n' + content }}
58
+ {%- endif %}
59
+ {%- endif %}
60
+ {{- '<|im_end|>\n' }}
61
+ {%- else %}
62
+ {%- if messages[0].role == 'system' %}
63
+ {%- set content = render_content(messages[0].content, false, true)|trim %}
64
+ {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
65
+ {%- endif %}
66
+ {%- endif %}
67
+ {%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
68
+ {%- for message in messages[::-1] %}
69
+ {%- set index = (messages|length - 1) - loop.index0 %}
70
+ {%- if ns.multi_step_tool and message.role == "user" %}
71
+ {%- set content = render_content(message.content, false)|trim %}
72
+ {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
73
+ {%- set ns.multi_step_tool = false %}
74
+ {%- set ns.last_query_index = index %}
75
+ {%- endif %}
76
+ {%- endif %}
77
+ {%- endfor %}
78
+ {%- if ns.multi_step_tool %}
79
+ {{- raise_exception('No user query found in messages.') }}
80
+ {%- endif %}
81
+ {%- for message in messages %}
82
+ {%- set content = render_content(message.content, true)|trim %}
83
+ {%- if message.role == "system" %}
84
+ {%- if not loop.first %}
85
+ {{- raise_exception('System message must be at the beginning.') }}
86
+ {%- endif %}
87
+ {%- elif message.role == "user" %}
88
+ {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
89
+ {%- elif message.role == "assistant" %}
90
+ {%- set reasoning_content = '' %}
91
+ {%- if message.reasoning_content is string %}
92
+ {%- set reasoning_content = message.reasoning_content %}
93
+ {%- else %}
94
+ {%- if '</think>' in content %}
95
+ {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
96
+ {%- set content = content.split('</think>')[-1].lstrip('\n') %}
97
+ {%- endif %}
98
+ {%- endif %}
99
+ {%- set reasoning_content = reasoning_content|trim %}
100
+ {%- if loop.index0 > ns.last_query_index %}
101
+ {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
102
+ {%- else %}
103
+ {{- '<|im_start|>' + message.role + '\n' + content }}
104
+ {%- endif %}
105
+ {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
106
+ {%- for tool_call in message.tool_calls %}
107
+ {%- if tool_call.function is defined %}
108
+ {%- set tool_call = tool_call.function %}
109
+ {%- endif %}
110
+ {%- if loop.first %}
111
+ {%- if content|trim %}
112
+ {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
113
+ {%- else %}
114
+ {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
115
+ {%- endif %}
116
+ {%- else %}
117
+ {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
118
+ {%- endif %}
119
+ {%- if tool_call.arguments is defined %}
120
+ {%- for args_name, args_value in tool_call.arguments|items %}
121
+ {{- '<parameter=' + args_name + '>\n' }}
122
+ {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
123
+ {{- args_value }}
124
+ {{- '\n</parameter>\n' }}
125
+ {%- endfor %}
126
+ {%- endif %}
127
+ {{- '</function>\n</tool_call>' }}
128
+ {%- endfor %}
129
+ {%- endif %}
130
+ {{- '<|im_end|>\n' }}
131
+ {%- elif message.role == "tool" %}
132
+ {%- if loop.previtem and loop.previtem.role != "tool" %}
133
+ {{- '<|im_start|>user' }}
134
+ {%- endif %}
135
+ {{- '\n<tool_response>\n' }}
136
+ {{- content }}
137
+ {{- '\n</tool_response>' }}
138
+ {%- if not loop.last and loop.nextitem.role != "tool" %}
139
+ {{- '<|im_end|>\n' }}
140
+ {%- elif loop.last %}
141
+ {{- '<|im_end|>\n' }}
142
+ {%- endif %}
143
+ {%- else %}
144
+ {{- raise_exception('Unexpected message role.') }}
145
+ {%- endif %}
146
+ {%- endfor %}
147
+ {%- if add_generation_prompt %}
148
+ {{- '<|im_start|>assistant\n' }}
149
+ {%- if enable_thinking is defined and enable_thinking is true %}
150
+ {{- '<think>\n' }}
151
+ {%- else %}
152
+ {{- '<think>\n\n</think>\n\n' }}
153
+ {%- endif %}
154
+ {%- endif %}
systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-2285/tokenizer_config.json ADDED
@@ -0,0 +1,31 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "audio_bos_token": "<|audio_start|>",
4
+ "audio_eos_token": "<|audio_end|>",
5
+ "audio_token": "<|audio_pad|>",
6
+ "backend": "tokenizers",
7
+ "bos_token": null,
8
+ "clean_up_tokenization_spaces": false,
9
+ "eos_token": "<|endoftext|>",
10
+ "errors": "replace",
11
+ "image_token": "<|image_pad|>",
12
+ "is_local": false,
13
+ "model_max_length": 262144,
14
+ "model_specific_special_tokens": {
15
+ "audio_bos_token": "<|audio_start|>",
16
+ "audio_eos_token": "<|audio_end|>",
17
+ "audio_token": "<|audio_pad|>",
18
+ "image_token": "<|image_pad|>",
19
+ "video_token": "<|video_pad|>",
20
+ "vision_bos_token": "<|vision_start|>",
21
+ "vision_eos_token": "<|vision_end|>"
22
+ },
23
+ "pad_token": "<|endoftext|>",
24
+ "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
25
+ "split_special_tokens": false,
26
+ "tokenizer_class": "TokenizersBackend",
27
+ "unk_token": null,
28
+ "video_token": "<|video_pad|>",
29
+ "vision_bos_token": "<|vision_start|>",
30
+ "vision_eos_token": "<|vision_end|>"
31
+ }
systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-2285/trainer_state.json ADDED
@@ -0,0 +1,539 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 5.0,
6
+ "eval_steps": 500,
7
+ "global_step": 2285,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "entropy": 2.074948468208313,
14
+ "epoch": 0.10952902519167579,
15
+ "grad_norm": 2.3511903285980225,
16
+ "learning_rate": 1.2698308676050528e-05,
17
+ "loss": 1.9153688049316406,
18
+ "mean_token_accuracy": 0.6077924159169197,
19
+ "num_tokens": 113087.0,
20
+ "step": 50
21
+ },
22
+ {
23
+ "entropy": 1.054056368470192,
24
+ "epoch": 0.21905805038335158,
25
+ "grad_norm": 2.3081915378570557,
26
+ "learning_rate": 2.5655766508755147e-05,
27
+ "loss": 0.9633240509033203,
28
+ "mean_token_accuracy": 0.741833764910698,
29
+ "num_tokens": 230392.0,
30
+ "step": 100
31
+ },
32
+ {
33
+ "entropy": 0.9184817910194397,
34
+ "epoch": 0.32858707557502737,
35
+ "grad_norm": 2.0965280532836914,
36
+ "learning_rate": 3.861322434145977e-05,
37
+ "loss": 0.8405924987792969,
38
+ "mean_token_accuracy": 0.7688284432888031,
39
+ "num_tokens": 341746.0,
40
+ "step": 150
41
+ },
42
+ {
43
+ "entropy": 0.8618861585855484,
44
+ "epoch": 0.43811610076670315,
45
+ "grad_norm": 2.2506542205810547,
46
+ "learning_rate": 5.157068217416438e-05,
47
+ "loss": 0.7841030883789063,
48
+ "mean_token_accuracy": 0.7796466761827469,
49
+ "num_tokens": 460483.0,
50
+ "step": 200
51
+ },
52
+ {
53
+ "entropy": 0.819869134426117,
54
+ "epoch": 0.547645125958379,
55
+ "grad_norm": 1.4058395624160767,
56
+ "learning_rate": 6.4528140006869e-05,
57
+ "loss": 0.7524736785888672,
58
+ "mean_token_accuracy": 0.7900535291433335,
59
+ "num_tokens": 570627.0,
60
+ "step": 250
61
+ },
62
+ {
63
+ "entropy": 0.8075837278366089,
64
+ "epoch": 0.6571741511500547,
65
+ "grad_norm": 1.5815014839172363,
66
+ "learning_rate": 7.748559783957362e-05,
67
+ "loss": 0.7388697814941406,
68
+ "mean_token_accuracy": 0.7914055424928665,
69
+ "num_tokens": 677217.0,
70
+ "step": 300
71
+ },
72
+ {
73
+ "entropy": 0.7868322026729584,
74
+ "epoch": 0.7667031763417306,
75
+ "grad_norm": 1.2377994060516357,
76
+ "learning_rate": 9.044305567227824e-05,
77
+ "loss": 0.7226168060302735,
78
+ "mean_token_accuracy": 0.7953901988267899,
79
+ "num_tokens": 790185.0,
80
+ "step": 350
81
+ },
82
+ {
83
+ "entropy": 0.7526325100660324,
84
+ "epoch": 0.8762322015334063,
85
+ "grad_norm": 1.203137755393982,
86
+ "learning_rate": 0.00010340051350498286,
87
+ "loss": 0.6991680908203125,
88
+ "mean_token_accuracy": 0.7992895567417144,
89
+ "num_tokens": 910080.0,
90
+ "step": 400
91
+ },
92
+ {
93
+ "entropy": 0.7626404595375061,
94
+ "epoch": 0.9857612267250822,
95
+ "grad_norm": 1.1625452041625977,
96
+ "learning_rate": 0.00011635797133768749,
97
+ "loss": 0.7055020904541016,
98
+ "mean_token_accuracy": 0.7997505795955658,
99
+ "num_tokens": 1023473.0,
100
+ "step": 450
101
+ },
102
+ {
103
+ "epoch": 1.0,
104
+ "eval_entropy": 0.7441067624659765,
105
+ "eval_loss": 0.7209385633468628,
106
+ "eval_mean_token_accuracy": 0.7895652093584575,
107
+ "eval_num_tokens": 1038250.0,
108
+ "eval_runtime": 45.5076,
109
+ "eval_samples_per_second": 22.018,
110
+ "eval_steps_per_second": 2.769,
111
+ "step": 457
112
+ },
113
+ {
114
+ "entropy": 0.720724637460227,
115
+ "epoch": 1.0941949616648412,
116
+ "grad_norm": 1.2537180185317993,
117
+ "learning_rate": 0.00011840069618947894,
118
+ "loss": 0.6586117553710937,
119
+ "mean_token_accuracy": 0.8081232917429221,
120
+ "num_tokens": 1128922.0,
121
+ "step": 500
122
+ },
123
+ {
124
+ "entropy": 0.7084310132265091,
125
+ "epoch": 1.203723986856517,
126
+ "grad_norm": 1.2347161769866943,
127
+ "learning_rate": 0.00011828501915154198,
128
+ "loss": 0.650611572265625,
129
+ "mean_token_accuracy": 0.808387525677681,
130
+ "num_tokens": 1245635.0,
131
+ "step": 550
132
+ },
133
+ {
134
+ "entropy": 0.706004958152771,
135
+ "epoch": 1.3132530120481927,
136
+ "grad_norm": 1.2140169143676758,
137
+ "learning_rate": 0.00011808319665683977,
138
+ "loss": 0.6442465209960937,
139
+ "mean_token_accuracy": 0.8122780376672745,
140
+ "num_tokens": 1354885.0,
141
+ "step": 600
142
+ },
143
+ {
144
+ "entropy": 0.6854291236400605,
145
+ "epoch": 1.4227820372398685,
146
+ "grad_norm": 1.1137559413909912,
147
+ "learning_rate": 0.00011779552303848118,
148
+ "loss": 0.6293138885498046,
149
+ "mean_token_accuracy": 0.8158071845769882,
150
+ "num_tokens": 1472232.0,
151
+ "step": 650
152
+ },
153
+ {
154
+ "entropy": 0.6885070586204529,
155
+ "epoch": 1.5323110624315444,
156
+ "grad_norm": 1.1074714660644531,
157
+ "learning_rate": 0.00011742241783280468,
158
+ "loss": 0.6259001541137695,
159
+ "mean_token_accuracy": 0.8164563828706741,
160
+ "num_tokens": 1587494.0,
161
+ "step": 700
162
+ },
163
+ {
164
+ "entropy": 0.6932114177942276,
165
+ "epoch": 1.6418400876232202,
166
+ "grad_norm": 1.1099495887756348,
167
+ "learning_rate": 0.00011696442516753647,
168
+ "loss": 0.6335408401489258,
169
+ "mean_token_accuracy": 0.8156683903932571,
170
+ "num_tokens": 1696927.0,
171
+ "step": 750
172
+ },
173
+ {
174
+ "entropy": 0.6787053138017655,
175
+ "epoch": 1.751369112814896,
176
+ "grad_norm": 1.2505418062210083,
177
+ "learning_rate": 0.00011642221296824764,
178
+ "loss": 0.6210909271240235,
179
+ "mean_token_accuracy": 0.8153067737817764,
180
+ "num_tokens": 1813415.0,
181
+ "step": 800
182
+ },
183
+ {
184
+ "entropy": 0.6693948286771775,
185
+ "epoch": 1.8608981380065717,
186
+ "grad_norm": 1.0963624715805054,
187
+ "learning_rate": 0.00011579657198426736,
188
+ "loss": 0.6105814361572266,
189
+ "mean_token_accuracy": 0.819609425663948,
190
+ "num_tokens": 1924911.0,
191
+ "step": 850
192
+ },
193
+ {
194
+ "entropy": 0.6634757363796234,
195
+ "epoch": 1.9704271631982475,
196
+ "grad_norm": 0.8766036033630371,
197
+ "learning_rate": 0.00011508841463547313,
198
+ "loss": 0.6100498580932617,
199
+ "mean_token_accuracy": 0.8197187691926956,
200
+ "num_tokens": 2044650.0,
201
+ "step": 900
202
+ },
203
+ {
204
+ "epoch": 2.0,
205
+ "eval_entropy": 0.6342300275961558,
206
+ "eval_loss": 0.6517631411552429,
207
+ "eval_mean_token_accuracy": 0.807878240233376,
208
+ "eval_num_tokens": 2076500.0,
209
+ "eval_runtime": 44.6225,
210
+ "eval_samples_per_second": 22.455,
211
+ "eval_steps_per_second": 2.824,
212
+ "step": 914
213
+ },
214
+ {
215
+ "entropy": 0.6060671378867795,
216
+ "epoch": 2.078860898138007,
217
+ "grad_norm": 0.8028743267059326,
218
+ "learning_rate": 0.00011429877368163915,
219
+ "loss": 0.5440861511230469,
220
+ "mean_token_accuracy": 0.833196161973356,
221
+ "num_tokens": 2160498.0,
222
+ "step": 950
223
+ },
224
+ {
225
+ "entropy": 0.5839239662885666,
226
+ "epoch": 2.1883899233296824,
227
+ "grad_norm": 0.8585777878761292,
228
+ "learning_rate": 0.00011342880071628435,
229
+ "loss": 0.5182851409912109,
230
+ "mean_token_accuracy": 0.8406627786159515,
231
+ "num_tokens": 2269979.0,
232
+ "step": 1000
233
+ },
234
+ {
235
+ "entropy": 0.5949062475562096,
236
+ "epoch": 2.297918948521358,
237
+ "grad_norm": 0.9953155517578125,
238
+ "learning_rate": 0.00011247976448721578,
239
+ "loss": 0.5327045822143555,
240
+ "mean_token_accuracy": 0.8344444561004639,
241
+ "num_tokens": 2378362.0,
242
+ "step": 1050
243
+ },
244
+ {
245
+ "entropy": 0.583380132317543,
246
+ "epoch": 2.407447973713034,
247
+ "grad_norm": 0.9235143065452576,
248
+ "learning_rate": 0.00011145304904621704,
249
+ "loss": 0.520841178894043,
250
+ "mean_token_accuracy": 0.839934149980545,
251
+ "num_tokens": 2498494.0,
252
+ "step": 1100
253
+ },
254
+ {
255
+ "entropy": 0.5811308068037033,
256
+ "epoch": 2.5169769989047097,
257
+ "grad_norm": 1.1114075183868408,
258
+ "learning_rate": 0.00011035015173058038,
259
+ "loss": 0.5212027740478515,
260
+ "mean_token_accuracy": 0.838929146528244,
261
+ "num_tokens": 2611646.0,
262
+ "step": 1150
263
+ },
264
+ {
265
+ "entropy": 0.5903497302532196,
266
+ "epoch": 2.6265060240963853,
267
+ "grad_norm": 1.053895354270935,
268
+ "learning_rate": 0.0001091726809794255,
269
+ "loss": 0.5264106750488281,
270
+ "mean_token_accuracy": 0.8369076484441758,
271
+ "num_tokens": 2717744.0,
272
+ "step": 1200
273
+ },
274
+ {
275
+ "entropy": 0.5949014449119567,
276
+ "epoch": 2.7360350492880614,
277
+ "grad_norm": 0.7563770413398743,
278
+ "learning_rate": 0.00010792235398799048,
279
+ "loss": 0.5322903442382813,
280
+ "mean_token_accuracy": 0.8368168956041336,
281
+ "num_tokens": 2827750.0,
282
+ "step": 1250
283
+ },
284
+ {
285
+ "entropy": 0.5745143675804139,
286
+ "epoch": 2.845564074479737,
287
+ "grad_norm": 0.8103823065757751,
288
+ "learning_rate": 0.00010660099420331526,
289
+ "loss": 0.5154821014404297,
290
+ "mean_token_accuracy": 0.8410378849506378,
291
+ "num_tokens": 2948958.0,
292
+ "step": 1300
293
+ },
294
+ {
295
+ "entropy": 0.5665904894471169,
296
+ "epoch": 2.955093099671413,
297
+ "grad_norm": 0.8764206171035767,
298
+ "learning_rate": 0.00010521052866496986,
299
+ "loss": 0.5121800231933594,
300
+ "mean_token_accuracy": 0.8420848768949508,
301
+ "num_tokens": 3066917.0,
302
+ "step": 1350
303
+ },
304
+ {
305
+ "epoch": 3.0,
306
+ "eval_entropy": 0.5603872095308606,
307
+ "eval_loss": 0.6720245480537415,
308
+ "eval_mean_token_accuracy": 0.8079002830717299,
309
+ "eval_num_tokens": 3114750.0,
310
+ "eval_runtime": 43.9396,
311
+ "eval_samples_per_second": 22.804,
312
+ "eval_steps_per_second": 2.868,
313
+ "step": 1371
314
+ },
315
+ {
316
+ "entropy": 0.5181072778774031,
317
+ "epoch": 3.063526834611172,
318
+ "grad_norm": 0.889443039894104,
319
+ "learning_rate": 0.00010375298519470596,
320
+ "loss": 0.44869777679443357,
321
+ "mean_token_accuracy": 0.8574847211741438,
322
+ "num_tokens": 3180366.0,
323
+ "step": 1400
324
+ },
325
+ {
326
+ "entropy": 0.4902029836177826,
327
+ "epoch": 3.1730558598028478,
328
+ "grad_norm": 0.8999125957489014,
329
+ "learning_rate": 0.00010223048943913007,
330
+ "loss": 0.41646446228027345,
331
+ "mean_token_accuracy": 0.8655553787946701,
332
+ "num_tokens": 3293918.0,
333
+ "step": 1450
334
+ },
335
+ {
336
+ "entropy": 0.48207574665546415,
337
+ "epoch": 3.2825848849945234,
338
+ "grad_norm": 1.0042189359664917,
339
+ "learning_rate": 0.00010064526176971124,
340
+ "loss": 0.41205951690673825,
341
+ "mean_token_accuracy": 0.866786903142929,
342
+ "num_tokens": 3409058.0,
343
+ "step": 1500
344
+ },
345
+ {
346
+ "entropy": 0.4839185461401939,
347
+ "epoch": 3.3921139101861995,
348
+ "grad_norm": 1.1650396585464478,
349
+ "learning_rate": 9.89996140446443e-05,
350
+ "loss": 0.4102657699584961,
351
+ "mean_token_accuracy": 0.8668668657541275,
352
+ "num_tokens": 3525131.0,
353
+ "step": 1550
354
+ },
355
+ {
356
+ "entropy": 0.4932769918441772,
357
+ "epoch": 3.501642935377875,
358
+ "grad_norm": 1.0935224294662476,
359
+ "learning_rate": 9.729594623729099e-05,
360
+ "loss": 0.4225375366210937,
361
+ "mean_token_accuracy": 0.8634544950723648,
362
+ "num_tokens": 3641504.0,
363
+ "step": 1600
364
+ },
365
+ {
366
+ "entropy": 0.5046219238638878,
367
+ "epoch": 3.6111719605695507,
368
+ "grad_norm": 0.9509316086769104,
369
+ "learning_rate": 9.553674293611626e-05,
370
+ "loss": 0.4319529342651367,
371
+ "mean_token_accuracy": 0.8610132175683975,
372
+ "num_tokens": 3752182.0,
373
+ "step": 1650
374
+ },
375
+ {
376
+ "entropy": 0.5090703725814819,
377
+ "epoch": 3.7207009857612268,
378
+ "grad_norm": 0.9050212502479553,
379
+ "learning_rate": 9.372456972122359e-05,
380
+ "loss": 0.4376397705078125,
381
+ "mean_token_accuracy": 0.8591927194595337,
382
+ "num_tokens": 3866473.0,
383
+ "step": 1700
384
+ },
385
+ {
386
+ "entropy": 0.4903561845421791,
387
+ "epoch": 3.8302300109529024,
388
+ "grad_norm": 1.1624391078948975,
389
+ "learning_rate": 9.186206942277429e-05,
390
+ "loss": 0.4219248962402344,
391
+ "mean_token_accuracy": 0.8636374253034592,
392
+ "num_tokens": 3977393.0,
393
+ "step": 1750
394
+ },
395
+ {
396
+ "entropy": 0.5040662395954132,
397
+ "epoch": 3.9397590361445785,
398
+ "grad_norm": 0.798085629940033,
399
+ "learning_rate": 8.995195826674703e-05,
400
+ "loss": 0.4333121871948242,
401
+ "mean_token_accuracy": 0.86187963783741,
402
+ "num_tokens": 4088936.0,
403
+ "step": 1800
404
+ },
405
+ {
406
+ "epoch": 4.0,
407
+ "eval_entropy": 0.5136857172326436,
408
+ "eval_loss": 0.6814945340156555,
409
+ "eval_mean_token_accuracy": 0.8091600262929523,
410
+ "eval_num_tokens": 4153000.0,
411
+ "eval_runtime": 44.5059,
412
+ "eval_samples_per_second": 22.514,
413
+ "eval_steps_per_second": 2.831,
414
+ "step": 1828
415
+ },
416
+ {
417
+ "entropy": 0.4410170723091472,
418
+ "epoch": 4.048192771084337,
419
+ "grad_norm": 0.9169163703918457,
420
+ "learning_rate": 8.799702191365879e-05,
421
+ "loss": 0.3653216552734375,
422
+ "mean_token_accuracy": 0.8799899684058295,
423
+ "num_tokens": 4202329.0,
424
+ "step": 1850
425
+ },
426
+ {
427
+ "entropy": 0.38991558194160464,
428
+ "epoch": 4.157721796276014,
429
+ "grad_norm": 1.2269299030303955,
430
+ "learning_rate": 8.600011139602396e-05,
431
+ "loss": 0.3035482597351074,
432
+ "mean_token_accuracy": 0.8981181675195694,
433
+ "num_tokens": 4320321.0,
434
+ "step": 1900
435
+ },
436
+ {
437
+ "entropy": 0.38429499566555025,
438
+ "epoch": 4.267250821467689,
439
+ "grad_norm": 1.1254630088806152,
440
+ "learning_rate": 8.39641389604766e-05,
441
+ "loss": 0.3037366676330566,
442
+ "mean_token_accuracy": 0.8979371869564057,
443
+ "num_tokens": 4435139.0,
444
+ "step": 1950
445
+ },
446
+ {
447
+ "entropy": 0.39086049795150757,
448
+ "epoch": 4.376779846659365,
449
+ "grad_norm": 1.3771028518676758,
450
+ "learning_rate": 8.18920738206195e-05,
451
+ "loss": 0.3100498390197754,
452
+ "mean_token_accuracy": 0.8953604429960251,
453
+ "num_tokens": 4547272.0,
454
+ "step": 2000
455
+ },
456
+ {
457
+ "entropy": 0.3967577290534973,
458
+ "epoch": 4.48630887185104,
459
+ "grad_norm": 1.0673140287399292,
460
+ "learning_rate": 7.978693782679394e-05,
461
+ "loss": 0.31265506744384763,
462
+ "mean_token_accuracy": 0.8946040958166123,
463
+ "num_tokens": 4659036.0,
464
+ "step": 2050
465
+ },
466
+ {
467
+ "entropy": 0.3927252873778343,
468
+ "epoch": 4.595837897042716,
469
+ "grad_norm": 1.1611689329147339,
470
+ "learning_rate": 7.765180105908515e-05,
471
+ "loss": 0.3075417518615723,
472
+ "mean_token_accuracy": 0.8964757239818573,
473
+ "num_tokens": 4769957.0,
474
+ "step": 2100
475
+ },
476
+ {
477
+ "entropy": 0.39732489734888077,
478
+ "epoch": 4.705366922234392,
479
+ "grad_norm": 1.3783643245697021,
480
+ "learning_rate": 7.548977734999099e-05,
481
+ "loss": 0.3155888748168945,
482
+ "mean_token_accuracy": 0.8941756916046143,
483
+ "num_tokens": 4882487.0,
484
+ "step": 2150
485
+ },
486
+ {
487
+ "entropy": 0.3972980257868767,
488
+ "epoch": 4.814895947426068,
489
+ "grad_norm": 0.9033240675926208,
490
+ "learning_rate": 7.330401974328283e-05,
491
+ "loss": 0.3149653244018555,
492
+ "mean_token_accuracy": 0.8941631782054901,
493
+ "num_tokens": 4993060.0,
494
+ "step": 2200
495
+ },
496
+ {
497
+ "entropy": 0.39401222407817843,
498
+ "epoch": 4.924424972617744,
499
+ "grad_norm": 1.1885665655136108,
500
+ "learning_rate": 7.109771589568177e-05,
501
+ "loss": 0.31299734115600586,
502
+ "mean_token_accuracy": 0.8955705845355988,
503
+ "num_tokens": 5111182.0,
504
+ "step": 2250
505
+ },
506
+ {
507
+ "epoch": 5.0,
508
+ "eval_entropy": 0.44022186218746123,
509
+ "eval_loss": 0.731863260269165,
510
+ "eval_mean_token_accuracy": 0.8086685883620429,
511
+ "eval_num_tokens": 5191250.0,
512
+ "eval_runtime": 44.8156,
513
+ "eval_samples_per_second": 22.358,
514
+ "eval_steps_per_second": 2.812,
515
+ "step": 2285
516
+ }
517
+ ],
518
+ "logging_steps": 50,
519
+ "max_steps": 4570,
520
+ "num_input_tokens_seen": 0,
521
+ "num_train_epochs": 10,
522
+ "save_steps": 500,
523
+ "stateful_callbacks": {
524
+ "TrainerControl": {
525
+ "args": {
526
+ "should_epoch_stop": false,
527
+ "should_evaluate": false,
528
+ "should_log": false,
529
+ "should_save": true,
530
+ "should_training_stop": false
531
+ },
532
+ "attributes": {}
533
+ }
534
+ },
535
+ "total_flos": 1.1263896427218816e+17,
536
+ "train_batch_size": 8,
537
+ "trial_name": null,
538
+ "trial_params": null
539
+ }
systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-2742/README.md ADDED
@@ -0,0 +1,209 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: Qwen/Qwen3.5-2B-Base
3
+ library_name: peft
4
+ pipeline_tag: text-generation
5
+ tags:
6
+ - base_model:adapter:Qwen/Qwen3.5-2B-Base
7
+ - lora
8
+ - sft
9
+ - transformers
10
+ - trl
11
+ ---
12
+
13
+ # Model Card for Model ID
14
+
15
+ <!-- Provide a quick summary of what the model is/does. -->
16
+
17
+
18
+
19
+ ## Model Details
20
+
21
+ ### Model Description
22
+
23
+ <!-- Provide a longer summary of what this model is. -->
24
+
25
+
26
+
27
+ - **Developed by:** [More Information Needed]
28
+ - **Funded by [optional]:** [More Information Needed]
29
+ - **Shared by [optional]:** [More Information Needed]
30
+ - **Model type:** [More Information Needed]
31
+ - **Language(s) (NLP):** [More Information Needed]
32
+ - **License:** [More Information Needed]
33
+ - **Finetuned from model [optional]:** [More Information Needed]
34
+
35
+ ### Model Sources [optional]
36
+
37
+ <!-- Provide the basic links for the model. -->
38
+
39
+ - **Repository:** [More Information Needed]
40
+ - **Paper [optional]:** [More Information Needed]
41
+ - **Demo [optional]:** [More Information Needed]
42
+
43
+ ## Uses
44
+
45
+ <!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
46
+
47
+ ### Direct Use
48
+
49
+ <!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
50
+
51
+ [More Information Needed]
52
+
53
+ ### Downstream Use [optional]
54
+
55
+ <!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
56
+
57
+ [More Information Needed]
58
+
59
+ ### Out-of-Scope Use
60
+
61
+ <!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
62
+
63
+ [More Information Needed]
64
+
65
+ ## Bias, Risks, and Limitations
66
+
67
+ <!-- This section is meant to convey both technical and sociotechnical limitations. -->
68
+
69
+ [More Information Needed]
70
+
71
+ ### Recommendations
72
+
73
+ <!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
74
+
75
+ Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
76
+
77
+ ## How to Get Started with the Model
78
+
79
+ Use the code below to get started with the model.
80
+
81
+ [More Information Needed]
82
+
83
+ ## Training Details
84
+
85
+ ### Training Data
86
+
87
+ <!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
88
+
89
+ [More Information Needed]
90
+
91
+ ### Training Procedure
92
+
93
+ <!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
94
+
95
+ #### Preprocessing [optional]
96
+
97
+ [More Information Needed]
98
+
99
+
100
+ #### Training Hyperparameters
101
+
102
+ - **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
103
+
104
+ #### Speeds, Sizes, Times [optional]
105
+
106
+ <!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
107
+
108
+ [More Information Needed]
109
+
110
+ ## Evaluation
111
+
112
+ <!-- This section describes the evaluation protocols and provides the results. -->
113
+
114
+ ### Testing Data, Factors & Metrics
115
+
116
+ #### Testing Data
117
+
118
+ <!-- This should link to a Dataset Card if possible. -->
119
+
120
+ [More Information Needed]
121
+
122
+ #### Factors
123
+
124
+ <!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
125
+
126
+ [More Information Needed]
127
+
128
+ #### Metrics
129
+
130
+ <!-- These are the evaluation metrics being used, ideally with a description of why. -->
131
+
132
+ [More Information Needed]
133
+
134
+ ### Results
135
+
136
+ [More Information Needed]
137
+
138
+ #### Summary
139
+
140
+
141
+
142
+ ## Model Examination [optional]
143
+
144
+ <!-- Relevant interpretability work for the model goes here -->
145
+
146
+ [More Information Needed]
147
+
148
+ ## Environmental Impact
149
+
150
+ <!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
151
+
152
+ Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
153
+
154
+ - **Hardware Type:** [More Information Needed]
155
+ - **Hours used:** [More Information Needed]
156
+ - **Cloud Provider:** [More Information Needed]
157
+ - **Compute Region:** [More Information Needed]
158
+ - **Carbon Emitted:** [More Information Needed]
159
+
160
+ ## Technical Specifications [optional]
161
+
162
+ ### Model Architecture and Objective
163
+
164
+ [More Information Needed]
165
+
166
+ ### Compute Infrastructure
167
+
168
+ [More Information Needed]
169
+
170
+ #### Hardware
171
+
172
+ [More Information Needed]
173
+
174
+ #### Software
175
+
176
+ [More Information Needed]
177
+
178
+ ## Citation [optional]
179
+
180
+ <!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
181
+
182
+ **BibTeX:**
183
+
184
+ [More Information Needed]
185
+
186
+ **APA:**
187
+
188
+ [More Information Needed]
189
+
190
+ ## Glossary [optional]
191
+
192
+ <!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
193
+
194
+ [More Information Needed]
195
+
196
+ ## More Information [optional]
197
+
198
+ [More Information Needed]
199
+
200
+ ## Model Card Authors [optional]
201
+
202
+ [More Information Needed]
203
+
204
+ ## Model Card Contact
205
+
206
+ [More Information Needed]
207
+ ### Framework versions
208
+
209
+ - PEFT 0.19.1
systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-2742/adapter_config.json ADDED
@@ -0,0 +1,48 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alora_invocation_tokens": null,
3
+ "alpha_pattern": {},
4
+ "arrow_config": null,
5
+ "auto_mapping": null,
6
+ "base_model_name_or_path": "Qwen/Qwen3.5-2B-Base",
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": 256,
20
+ "lora_bias": false,
21
+ "lora_dropout": 0.01686122009693236,
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": 128,
30
+ "rank_pattern": {},
31
+ "revision": null,
32
+ "target_modules": [
33
+ "v_proj",
34
+ "o_proj",
35
+ "k_proj",
36
+ "up_proj",
37
+ "q_proj",
38
+ "down_proj",
39
+ "gate_proj"
40
+ ],
41
+ "target_parameters": null,
42
+ "task_type": "CAUSAL_LM",
43
+ "trainable_token_indices": null,
44
+ "use_bdlora": null,
45
+ "use_dora": false,
46
+ "use_qalora": false,
47
+ "use_rslora": false
48
+ }
systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-2742/chat_template.jinja ADDED
@@ -0,0 +1,154 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {%- set image_count = namespace(value=0) %}
2
+ {%- set video_count = namespace(value=0) %}
3
+ {%- macro render_content(content, do_vision_count, is_system_content=false) %}
4
+ {%- if content is string %}
5
+ {{- content }}
6
+ {%- elif content is iterable and content is not mapping %}
7
+ {%- for item in content %}
8
+ {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
9
+ {%- if is_system_content %}
10
+ {{- raise_exception('System message cannot contain images.') }}
11
+ {%- endif %}
12
+ {%- if do_vision_count %}
13
+ {%- set image_count.value = image_count.value + 1 %}
14
+ {%- endif %}
15
+ {%- if add_vision_id %}
16
+ {{- 'Picture ' ~ image_count.value ~ ': ' }}
17
+ {%- endif %}
18
+ {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
19
+ {%- elif 'video' in item or item.type == 'video' %}
20
+ {%- if is_system_content %}
21
+ {{- raise_exception('System message cannot contain videos.') }}
22
+ {%- endif %}
23
+ {%- if do_vision_count %}
24
+ {%- set video_count.value = video_count.value + 1 %}
25
+ {%- endif %}
26
+ {%- if add_vision_id %}
27
+ {{- 'Video ' ~ video_count.value ~ ': ' }}
28
+ {%- endif %}
29
+ {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
30
+ {%- elif 'text' in item %}
31
+ {{- item.text }}
32
+ {%- else %}
33
+ {{- raise_exception('Unexpected item type in content.') }}
34
+ {%- endif %}
35
+ {%- endfor %}
36
+ {%- elif content is none or content is undefined %}
37
+ {{- '' }}
38
+ {%- else %}
39
+ {{- raise_exception('Unexpected content type.') }}
40
+ {%- endif %}
41
+ {%- endmacro %}
42
+ {%- if not messages %}
43
+ {{- raise_exception('No messages provided.') }}
44
+ {%- endif %}
45
+ {%- if tools and tools is iterable and tools is not mapping %}
46
+ {{- '<|im_start|>system\n' }}
47
+ {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
48
+ {%- for tool in tools %}
49
+ {{- "\n" }}
50
+ {{- tool | tojson }}
51
+ {%- endfor %}
52
+ {{- "\n</tools>" }}
53
+ {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
54
+ {%- if messages[0].role == 'system' %}
55
+ {%- set content = render_content(messages[0].content, false, true)|trim %}
56
+ {%- if content %}
57
+ {{- '\n\n' + content }}
58
+ {%- endif %}
59
+ {%- endif %}
60
+ {{- '<|im_end|>\n' }}
61
+ {%- else %}
62
+ {%- if messages[0].role == 'system' %}
63
+ {%- set content = render_content(messages[0].content, false, true)|trim %}
64
+ {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
65
+ {%- endif %}
66
+ {%- endif %}
67
+ {%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
68
+ {%- for message in messages[::-1] %}
69
+ {%- set index = (messages|length - 1) - loop.index0 %}
70
+ {%- if ns.multi_step_tool and message.role == "user" %}
71
+ {%- set content = render_content(message.content, false)|trim %}
72
+ {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
73
+ {%- set ns.multi_step_tool = false %}
74
+ {%- set ns.last_query_index = index %}
75
+ {%- endif %}
76
+ {%- endif %}
77
+ {%- endfor %}
78
+ {%- if ns.multi_step_tool %}
79
+ {{- raise_exception('No user query found in messages.') }}
80
+ {%- endif %}
81
+ {%- for message in messages %}
82
+ {%- set content = render_content(message.content, true)|trim %}
83
+ {%- if message.role == "system" %}
84
+ {%- if not loop.first %}
85
+ {{- raise_exception('System message must be at the beginning.') }}
86
+ {%- endif %}
87
+ {%- elif message.role == "user" %}
88
+ {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
89
+ {%- elif message.role == "assistant" %}
90
+ {%- set reasoning_content = '' %}
91
+ {%- if message.reasoning_content is string %}
92
+ {%- set reasoning_content = message.reasoning_content %}
93
+ {%- else %}
94
+ {%- if '</think>' in content %}
95
+ {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
96
+ {%- set content = content.split('</think>')[-1].lstrip('\n') %}
97
+ {%- endif %}
98
+ {%- endif %}
99
+ {%- set reasoning_content = reasoning_content|trim %}
100
+ {%- if loop.index0 > ns.last_query_index %}
101
+ {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
102
+ {%- else %}
103
+ {{- '<|im_start|>' + message.role + '\n' + content }}
104
+ {%- endif %}
105
+ {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
106
+ {%- for tool_call in message.tool_calls %}
107
+ {%- if tool_call.function is defined %}
108
+ {%- set tool_call = tool_call.function %}
109
+ {%- endif %}
110
+ {%- if loop.first %}
111
+ {%- if content|trim %}
112
+ {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
113
+ {%- else %}
114
+ {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
115
+ {%- endif %}
116
+ {%- else %}
117
+ {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
118
+ {%- endif %}
119
+ {%- if tool_call.arguments is defined %}
120
+ {%- for args_name, args_value in tool_call.arguments|items %}
121
+ {{- '<parameter=' + args_name + '>\n' }}
122
+ {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
123
+ {{- args_value }}
124
+ {{- '\n</parameter>\n' }}
125
+ {%- endfor %}
126
+ {%- endif %}
127
+ {{- '</function>\n</tool_call>' }}
128
+ {%- endfor %}
129
+ {%- endif %}
130
+ {{- '<|im_end|>\n' }}
131
+ {%- elif message.role == "tool" %}
132
+ {%- if loop.previtem and loop.previtem.role != "tool" %}
133
+ {{- '<|im_start|>user' }}
134
+ {%- endif %}
135
+ {{- '\n<tool_response>\n' }}
136
+ {{- content }}
137
+ {{- '\n</tool_response>' }}
138
+ {%- if not loop.last and loop.nextitem.role != "tool" %}
139
+ {{- '<|im_end|>\n' }}
140
+ {%- elif loop.last %}
141
+ {{- '<|im_end|>\n' }}
142
+ {%- endif %}
143
+ {%- else %}
144
+ {{- raise_exception('Unexpected message role.') }}
145
+ {%- endif %}
146
+ {%- endfor %}
147
+ {%- if add_generation_prompt %}
148
+ {{- '<|im_start|>assistant\n' }}
149
+ {%- if enable_thinking is defined and enable_thinking is true %}
150
+ {{- '<think>\n' }}
151
+ {%- else %}
152
+ {{- '<think>\n\n</think>\n\n' }}
153
+ {%- endif %}
154
+ {%- endif %}