cannot stream openai compatible response out

I have the below code for streaming the response, the generator is working but cannnot stream the response: llm = Llama(model_path="Phi-3-mini-4k-instruct-q4.gguf", n_gpu_layers=-1, n_ctx=4096, ) class JobInput: def init(self, job): self.openai_route = job.get("openai_route") self.openai_input = job.get("openai_input", {}) self.is_completion = "v1completions" in self.openai_route self.is_embedding = "embeddings" in self.openai_route self.embedding_format = self.openai_input.get('encoding_format', 'unknown') self.is_chatcompletion = "chat" in self.openai_route def infer(job_params): if 'n' in job_params.openai_input: del job_params.openai_input['n'] if job_params.openai_route and job_params.is_embedding: yield [ErrorResponse( message="The embedding endpoint is not supported on this URL.", type="unsupported_endpoint", code=501 # Not Implemented ).model_dump()] else: if job_params.openai_route and job_params.is_chatcompletion: llm_engine = llm.create_chat_completion else: llm_engine = llm.create_completion if not job_params.openai_input.get("stream", False): yield llm_engine(job_params.openai_input) elif job_params.openai_input.get("stream", False): llm_op = llm_engine(job_params.openai_input) yield llm_op async def handler(event): inp = event["input"] job_input = JobInput(inp) for line in infer(job_input): if isinstance(line, Generator): for l in line: yield l else: yield line if name == "main": runpod.serverless.start({"handler": handler, "return_aggregate_stream": True,}) Need help to fix!
5 Replies
digigoblin
digigoblin5mo ago
Use the OpenAI SDK if you are not already doing so.
ngagefreak05
ngagefreak05OP5mo ago
any directions how to use it?
digigoblin
digigoblin5mo ago
Oh looks like you are implementing it yourself or something, I suggest using RunPod's vllm worker instead, its also available as a template from the Explore section. https://github.com/runpod-workers/worker-vllm
GitHub
GitHub - runpod-workers/worker-vllm: The RunPod worker template for...
The RunPod worker template for serving our large language model endpoints. Powered by vLLM. - runpod-workers/worker-vllm
ngagefreak05
ngagefreak05OP5mo ago
but it is specifically for vllm, i am trying to use llama-cpp
nerdylive
nerdylive5mo ago
try to read the docs on how to stream for that And is it compatible with openai compatible stream?
Want results from more Discord servers?
Add your server