docs: Update README.md #63
Conversation
| counter_prompt_tokens | ||
| # Number of generation tokens processed. | ||
| counter_generation_tokens | ||
| # Counter of prefill tokens processed. |
There was a problem hiding this comment.
Do we need both this section and the one right below it? Why not just use the below section instead?
VLLM stats are reported by the metrics endpoint in fields that are prefixed with
vllm:. For example, the metrics reported by Triton will look similar to the following:
# HELP vllm:prompt_tokens_total Number of prefill tokens processed.
# TYPE vllm:prompt_tokens_total counter
vllm:prompt_tokens_total{model="vllm_model",version="1"} 10
# HELP vllm:generation_tokens_total Number of generation tokens processed.
# TYPE vllm:generation_tokens_total counter
vllm:generation_tokens_total{model="vllm_model",version="1"} 16
...There was a problem hiding this comment.
I'd rather have another section listing all supported metrics. The sample metrics output is just hard to follow IMO.
There was a problem hiding this comment.
And for example, changing counter_prompt_tokens to vllm:prompt_tokens_total allows reader to easily locate the corresponding output from below section.
There was a problem hiding this comment.
@oandreeva-nv do you mind driving this PR if you have time?
There was a problem hiding this comment.
How about we create a dedicated doc for metrics? So that our front facing README is nice and non-cluttered ?
|
There will be changes to the vLLM backend metrics in the upcoming releases. Converting this PR to draft. @oandreeva-nv @rmccorm4 |
List metrics in
vllm:*instead of the variable name.