From b6f33d5c4d5c04f8c89f21521758ddb8cfdac4bb Mon Sep 17 00:00:00 2001 From: Yishuo Wang Date: Mon, 3 Mar 2025 09:21:15 +0800 Subject: [PATCH] optimize moonlight again (#12909) --- python/llm/src/ipex_llm/transformers/convert.py | 2 ++ 1 file changed, 2 insertions(+) diff --git a/python/llm/src/ipex_llm/transformers/convert.py b/python/llm/src/ipex_llm/transformers/convert.py index 2ea563cb..d61d804d 100644 --- a/python/llm/src/ipex_llm/transformers/convert.py +++ b/python/llm/src/ipex_llm/transformers/convert.py @@ -2029,10 +2029,12 @@ def _optimize_post(model): modeling_module_name = model.__class__.__module__ module = importlib.import_module(modeling_module_name) from ipex_llm.transformers.models.common import rms_norm_forward + from ipex_llm.transformers.models.common import mlp_silu_forward from ipex_llm.transformers.models.deepseek import deepseek_model_forward from ipex_llm.transformers.models.deepseek import deepseek_attention_forward from ipex_llm.transformers.models.deepseek import deepseek_moe_forward convert_forward(model, module.DeepseekV3RMSNorm, rms_norm_forward) + convert_forward(model, module.DeepseekV3MLP, mlp_silu_forward) convert_forward(model, module.DeepseekV3Model, deepseek_model_forward) convert_forward(model, module.DeepseekV3Attention, deepseek_attention_forward) convert_forward(model, module.DeepseekV3MoE, deepseek_moe_forward)