From 841845030024259441721516556d67476ffb71b6 Mon Sep 17 00:00:00 2001 From: Yishuo Wang Date: Mon, 17 Feb 2025 14:53:37 +0800 Subject: [PATCH] optimize minicpm-o's tts part (#12833) --- python/llm/src/ipex_llm/transformers/convert.py | 6 ++++++ 1 file changed, 6 insertions(+) diff --git a/python/llm/src/ipex_llm/transformers/convert.py b/python/llm/src/ipex_llm/transformers/convert.py index 168b5ab9..4466834b 100644 --- a/python/llm/src/ipex_llm/transformers/convert.py +++ b/python/llm/src/ipex_llm/transformers/convert.py @@ -1032,6 +1032,9 @@ def _optimize_pre(model, qtype=None): if hasattr(model, "vpm"): from ipex_llm.transformers.models.minicpmv import merge_qkv model.vpm.apply(merge_qkv) + # tts opt + if hasattr(model, "tts"): + _optimize_pre(model.tts.model, qtype=qtype) # llm opt model.llm.config.model_type = "qwen2" _optimize_pre(model.llm, qtype=qtype) @@ -1971,6 +1974,9 @@ def _optimize_post(model): from transformers.models.whisper.modeling_whisper import WhisperSdpaAttention from ipex_llm.transformers.models.whisper import whisper_attention_forward convert_forward(model.apm, WhisperSdpaAttention, whisper_attention_forward) + # tts opt + if hasattr(model, "tts"): + _optimize_post(model.tts.model) # llm opt model.llm.config.model_type = "qwen2" _optimize_post(model.llm)