deepgeek/CosyVoice
2
0
Fork 0
mirror of https://github.com/FunAudioLLM/CosyVoice.git synced 2026-02-05 09:59:23 +08:00
Code Issues Packages Projects Releases Wiki Activity

add vllm inference

Browse Source
This commit is contained in:
lyuxiang.lx
2025-05-30 07:22:35 +00:00
parent 9f55c5af8f
commit 6dd68b9d5e
6 changed files with 105 additions and 64 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
cosyvoice/utils/file_utils.py
View File

@@ -58,7 +58,7 @@ def convert_onnx_to_trt(trt_model, trt_kwargs, onnx_model, fp16):
network = builder.create_network(network_flags)
parser = trt.OnnxParser(network, logger)
config = builder.create_builder_config()
config.set_memory_pool_limit(trt.MemoryPoolType.WORKSPACE, 1 << 31) # 1GB
config.set_memory_pool_limit(trt.MemoryPoolType.WORKSPACE, 1 << 32) # 4GB
if fp16:
config.set_flag(trt.BuilderFlag.FP16)
profile = builder.create_optimization_profile()
@@ -122,6 +122,7 @@ def export_cosyvoice2_vllm(model, model_path, device):
model.llm.model.config.tie_word_embeddings = False
model.llm.model.config.use_bias = True
model.llm.model.save_pretrained(model_path)
os.system('sed -i s@Qwen2ForCausalLM@CosyVoice2ForCausalLM@g {}/config.json'.format(os.path.abspath(model_path)))
model.llm.model.config.vocab_size = tmp_vocab_size
model.llm.model.config.tie_word_embeddings = tmp_tie_embedding
model.llm.model.set_input_embeddings(embed_tokens)