From 6841a9ac8fc8b3f4eb06e41fa3944f7877fd8f94 Mon Sep 17 00:00:00 2001 From: Zhao Changmin Date: Mon, 19 Aug 2024 13:43:19 +0800 Subject: [PATCH] fix load low bit com dtype (#11832) --- python/llm/src/ipex_llm/transformers/convert.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/python/llm/src/ipex_llm/transformers/convert.py b/python/llm/src/ipex_llm/transformers/convert.py index 65944f05..d43043d3 100644 --- a/python/llm/src/ipex_llm/transformers/convert.py +++ b/python/llm/src/ipex_llm/transformers/convert.py @@ -1845,7 +1845,7 @@ def _optimize_post(model, lightweight_bmm=False): from ipex_llm.transformers.models.minicpmv import minicpmv_get_vision_embedding convert_forward(model.vpm, vpm_module.Attention, vision_transformer_attention_forward) 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 from ipex_llm.transformers.models.minicpmv import siglip_attention_forward convert_forward(model.vpm, vpm_module.SiglipAttention, siglip_attention_forward)