EngineX/xc-llm-ascend
3
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

[bugfix]Qwen-Omni quantization bugfix (#7042)

Browse Source 
### What this PR does / why we need it?
[bugfix]Qwen-Omni quantization bugfix
fix Qwen-Omni quantization weight mapping to float weight
### Does this PR introduce _any_ user-facing change?

### How was this patch tested?

- vLLM version: v0.16.0
- vLLM main:
4034c3d32e

---------

Signed-off-by: tanhaoan333 <tanhaoan@huawei.com>
This commit is contained in:
tanhaoan333
2026-03-06 17:24:22 +08:00
committed by GitHub
parent a51d6366b9
commit 094eb0eff9
1 changed files with 26 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

34
vllm_ascend/quantization/modelslim_config.py
View File

@@ -64,6 +64,19 @@ QUANT_MODEL_PREFIX_MAPPINGS: dict[str, dict[str, str]] = {
"mm_projector.linear_1": "mm_projector.proj.0",
"mm_projector.linear_2": "mm_projector.proj.2",
},
"qwen3_omni_moe": {
"language_model.lm_head.": "thinker.lm_head.",
"language_model.model.": "thinker.model.",
"visual.": "thinker.visual.",
},
"qwen2_5_omni": {
"language_model.lm_head.": "thinker.lm_head.",
"language_model.model.": "thinker.model.",
"visual.": "thinker.visual.",
},
"qwen2_5_omni_text": {
"language_model.": "thinker.",
},
}
# key: model_type
@@ -192,6 +205,11 @@ packed_modules_model_mapping: dict[str, dict[str, list[str]]] = {
"k_proj",
"v_proj",
],
"attn_qkv_proj": [
"attn_q_proj",
"attn_k_proj",
"attn_v_proj",
],
"gate_up_proj": [
"gate_proj",
"up_proj",
@@ -204,20 +222,20 @@ packed_modules_model_mapping: dict[str, dict[str, list[str]]] = {
"k_proj",
"v_proj",
],
"attn.qkv": [
"attn.q",
"attn.k",
"attn.v",
],
"gate_up_proj": [
"gate_proj",
"up_proj",
"attn_qkv_proj": [
"attn_q_proj",
"attn_k_proj",
"attn_v_proj",
],
"qkv": [
"q",
"k",
"v",
],
"gate_up_proj": [
"gate_proj",
"up_proj",
],
},
}