update func export_codec_vllm

This commit is contained in:
雾聪
2025-02-26 16:48:21 +08:00
parent f6a18ee07a
commit f280558bcb
3 changed files with 5 additions and 2 deletions

View File

@@ -343,7 +343,7 @@ class Qwen2LM(TransformerLM):
max_tokens=max_len)
request_id = uuid.uuid4()
vllm_codec_engine.add_request(request_id,
{"prompt_embeds": lm_input.to(torch.bfloat16).to(device)},
{"prompt_embeds": lm_input.squeeze(0).to(torch.bfloat16).to(device)},
sampling_params)
## generator
out_token_ids = []