patrickleenyc commited on
Commit
a7e9b28
·
verified ·
1 Parent(s): 3330429

(Trained with Unsloth)

Browse files
chat_template.jinja CHANGED
@@ -15,29 +15,38 @@
15
  {%- endif %}
16
  {%- endif %}
17
  {%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
18
- {%- for message in messages[::-1] %}
19
  {%- set index = (messages|length - 1) - loop.index0 %}
20
- {%- if ns.multi_step_tool and message.role == "user" and message.content is string and not(message.content.startswith('<tool_response>') and message.content.endswith('</tool_response>')) %}
 
 
 
 
 
 
 
 
 
 
 
 
21
  {%- set ns.multi_step_tool = false %}
22
  {%- set ns.last_query_index = index %}
23
  {%- endif %}
24
  {%- endfor %}
25
  {%- for message in messages %}
26
- {%- if message.content is string %}
27
- {%- set content = message.content %}
28
- {%- else %}
29
- {%- set content = '' %}
30
- {%- endif %}
31
  {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
32
- {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
33
  {%- elif message.role == "assistant" %}
 
34
  {%- set reasoning_content = '' %}
35
- {%- if message.reasoning_content is string %}
36
  {%- set reasoning_content = message.reasoning_content %}
37
  {%- else %}
38
- {%- if '</think>' in content %}
39
- {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
40
- {%- set content = content.split('</think>')[-1].lstrip('\n') %}
 
41
  {%- endif %}
42
  {%- endif %}
43
  {%- if loop.index0 > ns.last_query_index %}
@@ -74,7 +83,7 @@
74
  {{- '<|im_start|>user' }}
75
  {%- endif %}
76
  {{- '\n<tool_response>\n' }}
77
- {{- content }}
78
  {{- '\n</tool_response>' }}
79
  {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
80
  {{- '<|im_end|>\n' }}
 
15
  {%- endif %}
16
  {%- endif %}
17
  {%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
18
+ {%- for forward_message in messages %}
19
  {%- set index = (messages|length - 1) - loop.index0 %}
20
+ {%- set message = messages[index] %}
21
+ {%- set current_content = message.content if message.content is not none else '' %}
22
+ {%- set tool_start = '<tool_response>' %}
23
+ {%- set tool_start_length = tool_start|length %}
24
+ {%- set start_of_message = current_content[:tool_start_length] %}
25
+ {%- set tool_end = '</tool_response>' %}
26
+ {%- set tool_end_length = tool_end|length %}
27
+ {%- set start_pos = (current_content|length) - tool_end_length %}
28
+ {%- if start_pos < 0 %}
29
+ {%- set start_pos = 0 %}
30
+ {%- endif %}
31
+ {%- set end_of_message = current_content[start_pos:] %}
32
+ {%- if ns.multi_step_tool and message.role == "user" and not(start_of_message == tool_start and end_of_message == tool_end) %}
33
  {%- set ns.multi_step_tool = false %}
34
  {%- set ns.last_query_index = index %}
35
  {%- endif %}
36
  {%- endfor %}
37
  {%- for message in messages %}
 
 
 
 
 
38
  {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
39
+ {{- '<|im_start|>' + message.role + '\n' + message.content + '<|im_end|>' + '\n' }}
40
  {%- elif message.role == "assistant" %}
41
+ {%- set content = message.content %}
42
  {%- set reasoning_content = '' %}
43
+ {%- if message.reasoning_content is defined and message.reasoning_content is not none %}
44
  {%- set reasoning_content = message.reasoning_content %}
45
  {%- else %}
46
+ {%- if '</think>' in message.content %}
47
+ {%- set content = (message.content.split('</think>')|last).lstrip('\n') %}
48
+ {%- set reasoning_content = (message.content.split('</think>')|first).rstrip('\n') %}
49
+ {%- set reasoning_content = (reasoning_content.split('<think>')|last).lstrip('\n') %}
50
  {%- endif %}
51
  {%- endif %}
52
  {%- if loop.index0 > ns.last_query_index %}
 
83
  {{- '<|im_start|>user' }}
84
  {%- endif %}
85
  {{- '\n<tool_response>\n' }}
86
+ {{- message.content }}
87
  {{- '\n</tool_response>' }}
88
  {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
89
  {{- '<|im_end|>\n' }}
config.json CHANGED
@@ -4,7 +4,6 @@
4
  ],
5
  "attention_bias": false,
6
  "attention_dropout": 0.0,
7
- "bos_token_id": 151643,
8
  "eos_token_id": 151645,
9
  "head_dim": 128,
10
  "hidden_act": "silu",
@@ -55,13 +54,16 @@
55
  "num_attention_heads": 32,
56
  "num_hidden_layers": 36,
57
  "num_key_value_heads": 8,
 
58
  "rms_norm_eps": 1e-06,
59
  "rope_scaling": null,
60
  "rope_theta": 1000000,
61
  "sliding_window": null,
62
  "tie_word_embeddings": true,
63
- "torch_dtype": "float16",
64
- "transformers_version": "4.53.3",
 
 
65
  "use_cache": true,
66
  "use_sliding_window": false,
67
  "vocab_size": 151936
 
4
  ],
5
  "attention_bias": false,
6
  "attention_dropout": 0.0,
 
7
  "eos_token_id": 151645,
8
  "head_dim": 128,
9
  "hidden_act": "silu",
 
54
  "num_attention_heads": 32,
55
  "num_hidden_layers": 36,
56
  "num_key_value_heads": 8,
57
+ "pad_token_id": 151654,
58
  "rms_norm_eps": 1e-06,
59
  "rope_scaling": null,
60
  "rope_theta": 1000000,
61
  "sliding_window": null,
62
  "tie_word_embeddings": true,
63
+ "torch_dtype": "bfloat16",
64
+ "transformers_version": "4.54.0",
65
+ "unsloth_fixed": true,
66
+ "unsloth_version": "2025.7.8",
67
  "use_cache": true,
68
  "use_sliding_window": false,
69
  "vocab_size": 151936
generation_config.json CHANGED
@@ -5,9 +5,10 @@
5
  151645,
6
  151643
7
  ],
8
- "pad_token_id": 151643,
 
9
  "temperature": 0.6,
10
  "top_k": 20,
11
  "top_p": 0.95,
12
- "transformers_version": "4.53.3"
13
  }
 
5
  151645,
6
  151643
7
  ],
8
+ "max_length": 40960,
9
+ "pad_token_id": 151654,
10
  "temperature": 0.6,
11
  "top_k": 20,
12
  "top_p": 0.95,
13
+ "transformers_version": "4.54.0"
14
  }
special_tokens_map.json CHANGED
@@ -22,7 +22,7 @@
22
  "single_word": false
23
  },
24
  "pad_token": {
25
- "content": "<|endoftext|>",
26
  "lstrip": false,
27
  "normalized": false,
28
  "rstrip": false,
 
22
  "single_word": false
23
  },
24
  "pad_token": {
25
+ "content": "<|vision_pad|>",
26
  "lstrip": false,
27
  "normalized": false,
28
  "rstrip": false,
tokenizer_config.json CHANGED
@@ -231,8 +231,9 @@
231
  "eos_token": "<|im_end|>",
232
  "errors": "replace",
233
  "extra_special_tokens": {},
234
- "model_max_length": 131072,
235
- "pad_token": "<|endoftext|>",
 
236
  "split_special_tokens": false,
237
  "tokenizer_class": "Qwen2Tokenizer",
238
  "unk_token": null
 
231
  "eos_token": "<|im_end|>",
232
  "errors": "replace",
233
  "extra_special_tokens": {},
234
+ "model_max_length": 40960,
235
+ "pad_token": "<|vision_pad|>",
236
+ "padding_side": "right",
237
  "split_special_tokens": false,
238
  "tokenizer_class": "Qwen2Tokenizer",
239
  "unk_token": null