423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2025-08-22 22:02:51 +08:00
Code Issues Packages Projects Releases Wiki Activity

[hparam] add enable think argument (#7928)

Browse Source
This commit is contained in:
hoshi-hiyouga 2025-04-30 17:21:30 +08:00 committed by GitHub
parent 41ec928683
commit 6a584b4092
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
4 changed files with 10 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
src/llamafactory/chat/hf_engine.py
View File

@ -106,7 +106,8 @@ class HuggingfaceEngine(BaseEngine):
# add thought words to avoid skipping thinking # add thought words to avoid skipping thinking
paired_messages = messages + [{"role": "assistant", "content": template.add_thought("")}] paired_messages = messages + [{"role": "assistant", "content": template.add_thought("")}]
system = system or generating_args["default_system"] system = system or generating_args["default_system"]
enable_thinking = input_kwargs.pop("enable_thinking", True) enable_thinking = input_kwargs.pop("enable_thinking", None)
enable_thinking = enable_thinking if enable_thinking is not None else generating_args["enable_thinking"]
prompt_ids, _ = template.encode_oneturn(tokenizer, paired_messages, system, tools, enable_thinking) prompt_ids, _ = template.encode_oneturn(tokenizer, paired_messages, system, tools, enable_thinking)
prompt_ids, _ = template.mm_plugin.process_token_ids( prompt_ids, _ = template.mm_plugin.process_token_ids(
prompt_ids, prompt_ids,

3
src/llamafactory/chat/sglang_engine.py
View File

@ -149,7 +149,8 @@ class SGLangEngine(BaseEngine):
# add thought words to avoid skipping thinking # add thought words to avoid skipping thinking
paired_messages = messages + [{"role": "assistant", "content": self.template.add_thought("")}] paired_messages = messages + [{"role": "assistant", "content": self.template.add_thought("")}]
system = system or self.generating_args["default_system"] system = system or self.generating_args["default_system"]
enable_thinking = input_kwargs.pop("enable_thinking", True) enable_thinking = input_kwargs.pop("enable_thinking", None)
enable_thinking = enable_thinking if enable_thinking is not None else self.generating_args["enable_thinking"]
prompt_ids, _ = self.template.encode_oneturn(self.tokenizer, paired_messages, system, tools, enable_thinking) prompt_ids, _ = self.template.encode_oneturn(self.tokenizer, paired_messages, system, tools, enable_thinking)
prompt_length = len(prompt_ids) prompt_length = len(prompt_ids)

3
src/llamafactory/chat/vllm_engine.py
View File

@ -126,7 +126,8 @@ class VllmEngine(BaseEngine):
# add thought words to avoid skipping thinking # add thought words to avoid skipping thinking
paired_messages = messages + [{"role": "assistant", "content": self.template.add_thought("")}] paired_messages = messages + [{"role": "assistant", "content": self.template.add_thought("")}]
system = system or self.generating_args["default_system"] system = system or self.generating_args["default_system"]
enable_thinking = input_kwargs.pop("enable_thinking", True) enable_thinking = input_kwargs.pop("enable_thinking", None)
enable_thinking = enable_thinking if enable_thinking is not None else self.generating_args["enable_thinking"]
prompt_ids, _ = self.template.encode_oneturn(self.tokenizer, paired_messages, system, tools, enable_thinking) prompt_ids, _ = self.template.encode_oneturn(self.tokenizer, paired_messages, system, tools, enable_thinking)
prompt_length = len(prompt_ids) prompt_length = len(prompt_ids)

4
src/llamafactory/hparams/generating_args.py
View File

@ -70,6 +70,10 @@ class GeneratingArguments:
default=True, default=True,
metadata={"help": "Whether or not to remove special tokens in the decoding."}, metadata={"help": "Whether or not to remove special tokens in the decoding."},
) )
enable_thinking: bool = field(
default=True,
metadata={"help": "Whether or not to enable thinking mode for reasoning models."},
)
def to_dict(self, obey_generation_config: bool = False) -> dict[str, Any]: def to_dict(self, obey_generation_config: bool = False) -> dict[str, Any]:
args = asdict(self) args = asdict(self)