chen, suyue 3d8009aa91 Fix benchmark scripts (#1517) ...

- Align benchmark default config:  
1. Update default helm charts version. 
2. Add `# mandatory` comment. 
3. Update default model ID for LLM. 
- Fix deploy issue:  
1. Support different `replicaCount` for w/ w/o rerank test. 
2. Add `max_num_seqs` for vllm. 
3. Add resource setting for tune mode. 

- Fix Benchmark issue: 
1. Update `user_queries` and `concurrency` setting. 
2. Remove invalid parameters. 
3. Fix `dataset` and `prompt` setting. And dataset ingest into db. 
5. Fix the benchmark hang issue with large user queries. Update `"processes": 16` will fix this issue. 
6. Update the eval_path setting logical. 
- Optimize benchmark readme. 
- Optimize the log path to make the logs more readable. 

Signed-off-by: chensuyue <suyue.chen@intel.com>
Signed-off-by: Cathy Zhang <cathy.zhang@intel.com>
Signed-off-by: letonghan <letong.han@intel.com>

2025-02-28 10:30:54 +08:00

28 KiB

Raw Permalink Blame History

View Raw