Skip to content

Conversation

the-david-oy
Copy link
Contributor

Allow direct payloads to be passed into GenAI-Perf as a JSONL file.

If we proceed with this feature, we'll need to update the parser descriptions and docs. We'll want to add unit testing and a CI test.

Example:

Server run command: docker run -it --net=host --gpus=all vllm/vllm-openai:latest --model gpt2 --dtype float16 --max-model-len 1024

Direct payload file (JSONL):

{"model": "gpt2", "prompt": ["Generate an example of a user profile."], "stream": false, "temperature": 0.0}

Example command: genai-perf profile -v -m gpt2 --service-kind openai --endpoint-type completions --input-file payload:direct_payload_example.jsonl

Output:
image

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant