Add full tools support to the chat template
Browse filesThis PR is still in progress! It should work, but it needs testing to verify it exactly matches the outputs of `mistral-common`.
- tokenizer_config.json +1 -1
tokenizer_config.json
CHANGED
|
@@ -6173,7 +6173,7 @@
|
|
| 6173 |
}
|
| 6174 |
},
|
| 6175 |
"bos_token": "<s>",
|
| 6176 |
-
"chat_template": "{%- if messages[0]['role'] == 'system' %}\n {%- set system_message = messages[0]['content'] %}\n {%- set loop_messages = messages[1:] %}\n{%- else %}\n {%- set loop_messages = messages %}\n{%- endif %}\n\n{{
|
| 6177 |
"clean_up_tokenization_spaces": false,
|
| 6178 |
"eos_token": "</s>",
|
| 6179 |
"legacy": false,
|
|
|
|
| 6173 |
}
|
| 6174 |
},
|
| 6175 |
"bos_token": "<s>",
|
| 6176 |
+
"chat_template": "{%- if messages[0]['role'] == 'system' %}\n {%- set system_message = messages[0]['content'] %}\n {%- set loop_messages = messages[1:] %}\n{%- else %}\n {%- set loop_messages = messages %}\n{%- endif %}\n{%- if not tools is defined %}\n {%- set tools = none %}\n{%- endif %}\n{%- set user_messages = loop_messages | selectattr('role', 'equalto', 'user') | list %}\n\n{%- for message in loop_messages | rejectattr(\"role\", \"equalto\", \"tool\") | rejectattr(\"role\", \"equalto\", \"tool_results\") | selectattr(\"tool_calls\", \"undefined\") %}\n {%- if (message['role'] == 'user') != (loop.index0 % 2 == 0) %}\n {{- raise_exception('After the optional system message, conversation roles must alternate user/assistant/user/assistant/...') }}\n {%- endif %}\n{%- endfor %}\n\n{{- bos_token }}\n{%- for message in loop_messages %}\n {%- if message['role'] == 'user' %}\n {%- if tools is not none and (message == user_messages[-1]) %}\n {{- '[AVAILABLE_TOOLS] [' }}\n {%- for tool in tools %}\n\t\t{%- set tool = tool.function %}\n\t\t{{- \"{'type': 'function', 'function': {\" }}\n\t\t{%- for key, val in tool.items() if key != \"return\" %}\n\t\t {%- if val is string %}\n\t\t\t{{- \"'\" + key + \"': '\" + val + \"'\" }}\n\t\t {%- else %}\n\t\t\t{{- \"'\" + key + \"': \" + val|string }}\n\t\t {%- endif %}\n\t\t {%- if not loop.last %}\n\t\t\t{{- \", \" }}\n\t\t {%- endif %}\n\t\t{%- endfor %}\n\t\t{{- \"}}\" }}\n {%- if not loop.last %}\n {{- \", \" }}\n {%- else %}\n {{- \"]\" }}\n {%- endif %}\n {%- endfor %}\n {{- '[/AVAILABLE_TOOLS]' }}\n {%- endif %}\n {%- if loop.last and system_message is defined %}\n {{- '[INST] ' + system_message + '\\n\\n' + message['content'] + '[/INST]' }}\n {%- else %}\n {{- '[INST] ' + message['content'] + '[/INST]' }}\n {%- endif %}\n {%- elif message['role'] == 'tool_calls' or message.tool_calls is defined %}\n {%- if message.tool_calls is defined %}\n {%- set tool_calls = message.tool_calls %}\n {%- else %}\n {%- set tool_calls = message.content %}\n {%- endif %}\n {{- '[TOOL_CALLS]' + tool_calls|string + eos_token }}\n {%- elif message['role'] == 'assistant' %}\n {{- ' ' + message['content'] + eos_token}}\n {%- elif message['role'] == 'tool_results' or message['role'] == 'tool' %}\n \t{%- if message.content is defined and message.content.content is defined %}\n \t {%- set content = message.content.content %}\n \t{%- else %}\n \t {%- set content = message.content %}\n \t{%- endif %}\n {{- '[TOOL_RESULTS]{\"content\": ' + content|string + '}[/TOOL_RESULTS]' }}\n {%- else %}\n {{- raise_exception('Only user and assistant roles are supported, with the exception of an initial optional system message!') }}\n {%- endif %}\n{%- endfor %}\n",
|
| 6177 |
"clean_up_tokenization_spaces": false,
|
| 6178 |
"eos_token": "</s>",
|
| 6179 |
"legacy": false,
|