Update DocSum README and environment configuration

[daniel-de-leon-user293]

Description

Updates to DocSum documentation and environment configurations

Issues

n/a

Type of change

List the type of change like below. Please delete options that are not relevant.

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds new functionality)
• Breaking change (fix or feature that would break existing design and interface)
• Others (enhancement, documentation, validation, etc.)

Dependencies

n/a

Tests

Manual walkthrough of the README and DocSum Gaudi unit test

[github-actions]

Dependency Review

✅ No vulnerabilities or license issues found.

Scanned Files

None

[MSCetin37]

LGTM

[ashahba]

LGTM!

[eero-t]

Why this disables vLLM warmup? On Gaudi warmup is too long for autoscaling to be responsive, and it's rather annoying also in development, but do you have some numbers on what's the runtime performance impact of disabling the warmup for "normal" usage?

[eero-t]

 + RESPONSE_BODY='Internal Server Error'
+ docker logs docsum-xeon-backend-server
...
 requests.exceptions.ReadTimeout: HTTPConnectionPool(host='10.244.159.134', port=9000): Read timed out. (read timeout=1000)
+ '[' 500 -ne 200 ']'
+ echo '[ docsum-xeon-backend-server ] HTTP status is not 200. Received status was 500'
+ exit 1
[ docsum-xeon-backend-server ] HTTP status is not 200. Received status was 500

=> PR to double the timeout time was merged today, it may be enough to get CI passing for this: opea-project/GenAIComps#1687

[chensuyue]

Know issue, please wait for fix #1936

[daniel-de-leon-user293]

Why this disables vLLM warmup? On Gaudi warmup is too long for autoscaling to be responsive, and it's rather annoying also in development, but do you have some numbers on what's the runtime performance impact of disabling the warmup for "normal" usage?

Thanks for the review @eero-t! vllm warmup was set to skip by default to reduce that developer friction that you mention. You bring up a valid point, however, that reminds me that the final state of the code should be ready for "normal" usage. And, since I don't currently have any numbers showing a difference in performance with/without vllm warmup, I went ahead and set the default to not skip vllm warmup. I made a note in the README to remind developers they can manually set VLLM_SKIP_WARMUP to true.

[eero-t]

Approved, but note that this overlaps with changes in #1942.