access vllm

This commit is contained in:
hmp
2024-04-11 22:00:07 +08:00
parent 02b6f26b05
commit 2406022c2a
3 changed files with 76 additions and 2 deletions

View File

@@ -784,6 +784,29 @@ for model in [m for m in AVAIL_LLM_MODELS if m.startswith("one-api-")]:
"token_cnt": get_token_num_gpt35,
},
})
# -=-=-=-=-=-=- vllm 对齐支持 -=-=-=-=-=-=-
for model in [m for m in AVAIL_LLM_MODELS if m.startswith("vllm-")]:
# 为了更灵活地接入vllm多模型管理界面设计了此接口例子AVAIL_LLM_MODELS = ["vllm-/home/hmp/llm/cache/Qwen1___5-32B-Chat(max_token=6666)"]
# 其中
# "vllm-" 是前缀(必要)
# "mixtral-8x7b" 是模型名(必要)
# "(max_token=6666)" 是配置(非必要)
try:
_, max_token_tmp = read_one_api_model_name(model)
except:
print(f"vllm模型 {model} 的 max_token 配置不是整数,请检查配置文件。")
continue
model_info.update({
model: {
"fn_with_ui": chatgpt_ui,
"fn_without_ui": chatgpt_noui,
"can_multi_thread": True,
"endpoint": openai_endpoint,
"max_token": max_token_tmp,
"tokenizer": tokenizer_gpt35,
"token_cnt": get_token_num_gpt35,
},
})
# -=-=-=-=-=-=- azure模型对齐支持 -=-=-=-=-=-=-