mirror of
https://github.com/deepseek-ai/DeepSeek-V3.git
synced 2025-05-22 02:06:45 -04:00
d3de9e8d1f
This commit fixes the chat template to properly handle function calls by: 1. Using safe dictionary access with .get() 2. Converting function arguments to JSON with |tojson filter 3. Adding better empty content handling Fixes #860
35 lines
2.8 KiB
JSON
35 lines
2.8 KiB
JSON
{
|
||
"add_bos_token": false,
|
||
"add_eos_token": false,
|
||
"bos_token": {
|
||
"__type": "AddedToken",
|
||
"content": "<｜begin of sentence｜>",
|
||
"lstrip": false,
|
||
"normalized": true,
|
||
"rstrip": false,
|
||
"single_word": false
|
||
},
|
||
"chat_template": "{% set ns = namespace(found=false, is_first=false, is_tool=false) %}\n{%- for message in messages %}\n {%- if message['role'] == 'system' %}\n {{ '<｜begin of sentence｜><｜system｜>' + message['content'] + '<｜end of sentence｜>' }}\n {%- elif message['role'] == 'user' %}\n {%- if message.get('content') and message.get('tool_results') %}\n {{ '<｜begin of sentence｜><｜User｜>' + message['content'] }}\n {%- for result in message['tool_results'] %}\n {{ '\n<｜tool result begin｜>' + result['name'] + '\n```json\n' + result['result'] + '\n```<｜tool result end｜>' }}\n {%- endfor %}\n {{ '<｜end of sentence｜>' }}\n {%- elif message.get('content') %}\n {{ '<｜begin of sentence｜><｜User｜>' + message['content'] + '<｜end of sentence｜>' }}\n {%- endif %}\n {%- elif message['role'] == 'assistant' %}\n {%- if message.get('content') is none or message.get('content') == '' %}\n {%- set ns.is_tool = false -%}\n {%- for tool in message.get('tool_calls', []) %}\n {%- if not ns.is_first %}\n {{ '<｜Assistant｜><｜tool calls begin｜><｜tool call begin｜>' + tool['type'] + '<｜tool sep｜>' + tool['function']['name'] + '\n```json\n' + (tool['function']['arguments']|tojson) + '\n```<｜tool call end｜>' }}\n {%- set ns.is_first = true -%}\n {%- else %}\n {{ '\n<｜tool call begin｜>' + tool['type'] + '<｜tool sep｜>' + tool['function']['name'] + '\n```json\n' + (tool['function']['arguments']|tojson) + '\n```<｜tool call end｜>' }}\n {{ '<｜tool calls end｜><｜end of sentence｜>' }}\n {%- endif %}\n {%- endfor %}\n {%- else %}\n {{ '<｜begin of sentence｜><｜Assistant｜>' + message['content'] + '<｜end of sentence｜>' }}\n {%- endif %}\n {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n {{ '<｜begin of sentence｜><｜Assistant｜>' }}\n{%- endif %}\n",
|
||
"clean_up_tokenization_spaces": true,
|
||
"model_max_length": 131072,
|
||
"eos_token": {
|
||
"__type": "AddedToken",
|
||
"content": "<｜end of sentence｜>",
|
||
"lstrip": false,
|
||
"normalized": true,
|
||
"rstrip": false,
|
||
"single_word": false
|
||
},
|
||
"tokenizer_class": "CodeLlamaTokenizer",
|
||
"pad_token": null,
|
||
"special_tokens_map_file": null,
|
||
"tokenizer_file": "tokenizer.json",
|
||
"unk_token": {
|
||
"__type": "AddedToken",
|
||
"content": "<｜unknown｜>",
|
||
"lstrip": false,
|
||
"normalized": true,
|
||
"rstrip": false,
|
||
"single_word": false
|
||
}
|
||
} |