ceos/unsloth
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

apply suggestions

Browse source
This commit is contained in:
AlanPonnachan 2026-01-02 08:19:38 +00:00 • committed by Daniel Han
commit 63aeedbfcb
1 changed files with 10 additions and 7 deletions
Download patch file Download diff file Expand all files Collapse all files

17
unsloth/models/qwen3_moe.py
View file

@ -273,9 +273,7 @@ class FastQwen3OmniMoeModel(FastQwen3MoeModel):
trust_remote_code = False,
**kwargs,
):
model, tokenizer = super(
FastQwen3OmniMoeModel, FastQwen3OmniMoeModel
).from_pretrained(
model, tokenizer = FastQwen3MoeModel.from_pretrained(
model_name = model_name,
max_seq_length = max_seq_length,
dtype = dtype,
@ -305,7 +303,12 @@ class FastQwen3OmniMoeModel(FastQwen3MoeModel):
and hasattr(module, "q_norm")
and hasattr(module, "k_norm")
):
if hasattr(module, "q_proj") and hasattr(module, "o_proj"):
if (
hasattr(module, "q_proj")
and hasattr(module, "k_proj")
and hasattr(module, "v_proj")
and hasattr(module, "o_proj")
):
module.__class__.apply_qkv = qwen3_omni_apply_qkv
module.__class__.apply_o = qwen3_omni_apply_o
@ -326,9 +329,9 @@ class FastQwen3OmniMoeModel(FastQwen3MoeModel):
if past_key_values is not None:
return self._original_forward(
hidden_states,
position_embeddings,
attention_mask,
past_key_values,
position_embeddings = position_embeddings,
attention_mask = attention_mask,
past_key_values = past_key_values,
**kwargs,
)