12 Commits

Author SHA1 Message Date
Root User
5e95cee1a0 bench: add adaptive concurrency search 2026-07-11 06:11:40 +00:00
Root User
d783143ddb vllm tp_dp_matrix: dockerize deployment and fix missing DOCKER_CLIENT_IMAGE export
- Run vLLM server inside vllm/vllm-openai:latest container on H200.
- Run benchmark client inside lmsysorg/sglang:latest using
  sglang.bench_serving --backend vllm, matching the sglang reference.
- Use RUNTIME_BASE for self-contained logs/pids/tmp instead of hardcoded
  /data/user1/yy paths.
- Update model path to /data3/hf_models/DeepSeek-V4-Flash.
- Sample only low/high concurrency endpoints to match sglang matrix.
- Fix docker 'invalid reference format' caused by DOCKER_CLIENT_IMAGE not
  being exported to bash -c subshells used by run_bench_serving.
- Add offline dummy ShareGPT dataset for random benchmark mode.
2026-07-11 06:11:40 +00:00
yy-fighting
666c583844 vllm tp_dp_matrix: add final results (report, results.json, comparison, scenarios)
Source code was already committed; this adds only the structured final
artifacts from run 20260709-152419_matrix. Raw jsonl/logs/gpu_logs remain
ignored per .gitignore.
2026-07-10 02:20:36 +00:00
yy-fighting
31731ebb75 dsv4_h200_vllm_tp_dp_matrix: export CONCURRENCY_SAMPLES so generate_scenarios.py sees it 2026-07-09 15:23:57 +00:00
yy-fighting
0ae1ea0a9f dsv4_h200_vllm_tp_dp_matrix: fix output naming, parser/compat, add Docker startup option 2026-07-09 12:33:52 +00:00
yy-fighting
0aee540c67 dsv4_h200_vllm_tp_dp_matrix: single server per config, OOM/restart handling, DRY_RUN/GRID_LIMIT, include 1M 2026-07-09 10:50:59 +00:00
yy-fighting
329d4c3975 dsv4_h200_vllm_tp_dp_matrix: use vLLM internal DP LB, drop proxy/supervisor 2026-07-09 10:28:50 +00:00
yy-fighting
59dbadb110 add dp_proxy for vLLM DP, write kv_cache_analysis.md, fix DP bench port 2026-07-09 10:19:30 +00:00
yy-fighting
5898d578ed dsv4_h200_vllm_tp_dp_matrix: allow MATRIX_FILE env override 2026-07-09 10:02:45 +00:00
yy-fighting
39326eaf4a remove partial smoke test artifacts from killed run 2026-07-09 10:02:03 +00:00
yy-fighting
15b5489535 dsv4_h200_vllm_tp_dp_matrix: use minimal official vLLM args, no EP, Y-only, exclude 1M 2026-07-09 10:01:51 +00:00
yy-fighting
04a384b265 add dsv4_h200_vllm_tp_dp_matrix experiment (TP×DP matrix with Y-only mode, 1M excluded) 2026-07-09 09:55:18 +00:00