biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Model] Gemma3: Fix GGUF loading and quantization (#26189)

Browse Source 
Signed-off-by: Luciano Martins <lucianommartins@users.noreply.github.com>
Signed-off-by: Isotr0py <mozf@mail2.sysu.edu.cn>
Co-authored-by: Luciano Martins <lucianommartins@users.noreply.github.com>
Co-authored-by: Isotr0py <mozf@mail2.sysu.edu.cn>
This commit is contained in:
Luciano Martins
2025-10-09 04:00:53 -03:00
committed by GitHub
parent 5e49c3e777
commit 1317028aa8
2 changed files with 14 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
vllm/model_executor/model_loader/gguf_loader.py
View File

@@ -72,6 +72,10 @@ class GGUFModelLoader(BaseModelLoader):
# hack: ggufs have a different name than transformers
if model_type == "cohere":
model_type = "command-r"
if model_type == "gemma3_text":
# Gemma3 models use "gemma3_text" in HuggingFace but
# "gemma3" in GGUF architecture naming
model_type = "gemma3"
if model_type in ("deepseek_v3", "deepseek_v2"):
model_type = "deepseek2"
# GGUF layer map assumes that we will have a merged expert weights