mirror of
https://github.com/hiyouga/LLaMA-Factory.git
synced 2026-08-23 00:15:44 +08:00
Compare commits
3 Commits
9c0b4b3835
...
a48af5cc69
| Author | SHA1 | Date | |
|---|---|---|---|
|
|
a48af5cc69 | ||
|
|
c383c0d067 | ||
|
|
50ff45176a |
@@ -3,6 +3,7 @@ trust_remote_code: true
|
||||
model_class: llm
|
||||
|
||||
template: qwen3_nothink
|
||||
flash_attn: flash_attention_2
|
||||
|
||||
# FSDP Config
|
||||
dist_config:
|
||||
|
||||
@@ -655,8 +655,10 @@ def get_template_and_fix_tokenizer(tokenizer: "PreTrainedTokenizer", data_args:
|
||||
|
||||
if isinstance(template, ReasoningTemplate):
|
||||
logger.warning_rank0(
|
||||
"You are using reasoning template, "
|
||||
"please add `_nothink` suffix if the model is not a reasoning model. "
|
||||
"You are using reasoning template. "
|
||||
"If the base model is NOT a reasoning model (i.e., it has a separate Instruct variant), "
|
||||
"please add `_nothink` suffix to disable thinking. "
|
||||
"For reasoning-only model families (e.g., Qwen3.6), the suffix is not needed. "
|
||||
"e.g., qwen3_vl_nothink"
|
||||
)
|
||||
template.enable_thinking = data_args.enable_thinking
|
||||
|
||||
@@ -3372,6 +3372,19 @@ register_model_group(
|
||||
)
|
||||
|
||||
|
||||
# Qwen-AgentWorld: language world model based on Qwen3.5-35B-A3B (MoE).
|
||||
# Reference: https://github.com/QwenLM/Qwen-AgentWorld
|
||||
register_model_group(
|
||||
models={
|
||||
"Qwen-AgentWorld-35B-A3B-Thinking": {
|
||||
DownloadSource.DEFAULT: "Qwen/Qwen-AgentWorld-35B-A3B",
|
||||
DownloadSource.MODELSCOPE: "Qwen/Qwen-AgentWorld-35B-A3B",
|
||||
},
|
||||
},
|
||||
template="qwen3",
|
||||
)
|
||||
|
||||
|
||||
register_model_group(
|
||||
models={
|
||||
"Vicuna-v1.5-7B-Chat": {
|
||||
|
||||
Reference in New Issue
Block a user