423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2025-12-16 11:50:35 +08:00
Code Issues Packages Projects Releases Wiki Activity

[model] add qwen2 omni 3b (#7945)

Browse Source
This commit is contained in:
hoshi-hiyouga
2025-05-03 16:36:51 +08:00
committed by GitHub
parent c94518718c
commit 634efb3ac9
7 changed files with 33 additions and 23 deletions
Download Patch File Download Diff File Expand all files Collapse all files

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

@@ -103,8 +103,7 @@ class HuggingfaceEngine(BaseEngine):
messages = template.mm_plugin.process_messages(
messages, mm_input_dict["images"], mm_input_dict["videos"], mm_input_dict["audios"], processor
)
# add thought words to avoid skipping thinking
paired_messages = messages + [{"role": "assistant", "content": template.add_thought("")}]
paired_messages = messages + [{"role": "assistant", "content": ""}]
system = system or generating_args["default_system"]
enable_thinking = input_kwargs.pop("enable_thinking", None)
enable_thinking = enable_thinking if enable_thinking is not None else generating_args["enable_thinking"]

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

@@ -146,8 +146,7 @@ class SGLangEngine(BaseEngine):
messages = self.template.mm_plugin.process_messages(
messages, images or [], videos or [], audios or [], self.processor
)
# add thought words to avoid skipping thinking
paired_messages = messages + [{"role": "assistant", "content": self.template.add_thought("")}]
paired_messages = messages + [{"role": "assistant", "content": ""}]
system = system or self.generating_args["default_system"]
enable_thinking = input_kwargs.pop("enable_thinking", None)
enable_thinking = enable_thinking if enable_thinking is not None else self.generating_args["enable_thinking"]

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

@@ -123,8 +123,7 @@ class VllmEngine(BaseEngine):
messages = self.template.mm_plugin.process_messages(
messages, images or [], videos or [], audios or [], self.processor
)
# add thought words to avoid skipping thinking
paired_messages = messages + [{"role": "assistant", "content": self.template.add_thought("")}]
paired_messages = messages + [{"role": "assistant", "content": ""}]
system = system or self.generating_args["default_system"]
enable_thinking = input_kwargs.pop("enable_thinking", None)
enable_thinking = enable_thinking if enable_thinking is not None else self.generating_args["enable_thinking"]

6
src/llamafactory/data/template.py
View File

@@ -60,7 +60,7 @@ class Template:
messages: list[dict[str, str]],
system: Optional[str] = None,
tools: Optional[str] = None,
enable_thinking: bool = True,
enable_thinking: bool = False,
) -> tuple[list[int], list[int]]:
r"""Return a single pair of token ids representing prompt and response respectively."""
encoded_messages = self._encode(tokenizer, messages, system, tools)
@@ -406,7 +406,7 @@ class ReasoningTemplate(Template):
messages: list[dict[str, str]],
system: Optional[str] = None,
tools: Optional[str] = None,
enable_thinking: bool = True,
enable_thinking: bool = False,
) -> tuple[list[int], list[int]]:
messages = deepcopy(messages)
for i in range(len(messages)):
@@ -418,7 +418,7 @@ class ReasoningTemplate(Template):
for encoded_ids in encoded_messages[:-1]:
prompt_ids += encoded_ids
if not enable_thinking or (
if not enable_thinking and (
messages[-1]["role"] == Role.ASSISTANT
and self.thought_words[0] not in messages[-1]["content"]
and self.thought_words[1] not in messages[-1]["content"]

14
src/llamafactory/extras/constants.py
View File

@@ -2479,6 +2479,14 @@ register_model_group(
DownloadSource.DEFAULT: "Qwen/Qwen3-235B-A22B",
DownloadSource.MODELSCOPE: "Qwen/Qwen3-235B-A22B",
},
"Qwen3-14B-Instruct-AWQ": {
DownloadSource.DEFAULT: "Qwen/Qwen3-14B-AWQ",
DownloadSource.MODELSCOPE: "Qwen/Qwen3-14B-AWQ",
},
"Qwen3-32B-Instruct-AWQ": {
DownloadSource.DEFAULT: "Qwen/Qwen3-32B-AWQ",
DownloadSource.MODELSCOPE: "Qwen/Qwen3-32B-AWQ",
},
},
template="qwen3",
)
@@ -2502,10 +2510,14 @@ register_model_group(
register_model_group(
models={
"Qwen2.5-Omni-3B": {
DownloadSource.DEFAULT: "Qwen/Qwen2.5-Omni-3B",
DownloadSource.MODELSCOPE: "Qwen/Qwen2.5-Omni-3B",
},
"Qwen2.5-Omni-7B": {
DownloadSource.DEFAULT: "Qwen/Qwen2.5-Omni-7B",
DownloadSource.MODELSCOPE: "Qwen/Qwen2.5-Omni-7B",
}
},
},
template="qwen2_omni",
multimodal=True,