ayo/ipex-llm
1
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 1

fix load low bit com dtype (#11832)

Browse source
This commit is contained in:
Zhao Changmin 2024-08-19 13:43:19 +08:00 committed by GitHub
parent cfc959defa
commit 6841a9ac8f
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
1 changed files with 1 additions and 1 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
python/llm/src/ipex_llm/transformers/convert.py
View file

@ -1845,7 +1845,7 @@ def _optimize_post(model, lightweight_bmm=False):
from ipex_llm.transformers.models.minicpmv import minicpmv_get_vision_embedding from ipex_llm.transformers.models.minicpmv import minicpmv_get_vision_embedding
convert_forward(model.vpm, vpm_module.Attention, vision_transformer_attention_forward) convert_forward(model.vpm, vpm_module.Attention, vision_transformer_attention_forward)
model.get_vision_embedding = MethodType(minicpmv_get_vision_embedding, model) model.get_vision_embedding = MethodType(minicpmv_get_vision_embedding, model)
elif model.vpm.config.model_type == "siglip": elif "siglip" in model.vpm.config.model_type:
# MiniCPM-V 2.6 # MiniCPM-V 2.6
from ipex_llm.transformers.models.minicpmv import siglip_attention_forward from ipex_llm.transformers.models.minicpmv import siglip_attention_forward
convert_forward(model.vpm, vpm_module.SiglipAttention, siglip_attention_forward) convert_forward(model.vpm, vpm_module.SiglipAttention, siglip_attention_forward)