qqtang/sskj
1
2
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Commit Graph

  • 31731ebb75 dsv4_h200_vllm_tp_dp_matrix: export CONCURRENCY_SAMPLES so generate_scenarios.py sees it yy-fighting 2026-07-09 15:23:57 +00:00
  • 8d61e682cd add dsv4_h200_sglang_tp_dp_matrix experiment (Docker-based SGLang TP×DP matrix) Root User 2026-07-09 14:49:31 +00:00
  • 0ae1ea0a9f dsv4_h200_vllm_tp_dp_matrix: fix output naming, parser/compat, add Docker startup option yy-fighting 2026-07-09 12:33:52 +00:00
  • 0aee540c67 dsv4_h200_vllm_tp_dp_matrix: single server per config, OOM/restart handling, DRY_RUN/GRID_LIMIT, include 1M yy-fighting 2026-07-09 10:50:59 +00:00
  • 329d4c3975 dsv4_h200_vllm_tp_dp_matrix: use vLLM internal DP LB, drop proxy/supervisor yy-fighting 2026-07-09 10:28:50 +00:00
  • 59dbadb110 add dp_proxy for vLLM DP, write kv_cache_analysis.md, fix DP bench port yy-fighting 2026-07-09 10:19:30 +00:00
  • 5898d578ed dsv4_h200_vllm_tp_dp_matrix: allow MATRIX_FILE env override yy-fighting 2026-07-09 10:02:45 +00:00
  • 39326eaf4a remove partial smoke test artifacts from killed run yy-fighting 2026-07-09 10:02:03 +00:00
  • 15b5489535 dsv4_h200_vllm_tp_dp_matrix: use minimal official vLLM args, no EP, Y-only, exclude 1M yy-fighting 2026-07-09 10:01:51 +00:00
  • 04a384b265 add dsv4_h200_vllm_tp_dp_matrix experiment (TP×DP matrix with Y-only mode, 1M excluded) yy-fighting 2026-07-09 09:55:18 +00:00
  • 1482601ae5 Add num_prompts = 5×concurrency convention and update all experiments yy-fighting 2026-07-09 09:26:24 +00:00
  • a1da95ef44 Scale P800 long-context matrix to concurrency=1 for stability Quantong Qiu 2026-07-09 05:59:28 +00:00
  • 46e05a57a4 Re-enable chunked prefill for P800 long-context matrix stability Quantong Qiu 2026-07-09 05:13:04 +00:00
  • 59f480b1bc Disable chunked prefill and speculative decoding in P800 long-context matrix Quantong Qiu 2026-07-09 04:47:59 +00:00
  • 90eecc4ede Add P800 long-context matrix experiment (64k/128k) for DeepSeek-V4-Flash-INT8 Quantong Qiu 2026-07-09 04:45:04 +00:00
  • b1b5f50f56 results(p800): max context length exploration run 20260709-033134 Quantong Qiu 2026-07-09 04:08:40 +00:00
  • 06549d8faf cleanup(p800): remove local debugging artifacts from dsv4_p800_sglang/results Quantong Qiu 2026-07-09 03:30:58 +00:00
  • 3694614dd9 feat(p800): add max context length exploration experiment Quantong Qiu 2026-07-09 03:30:09 +00:00
  • 8324d944f2 fix(p800): define RUN_ID/RESULT_ROOT/LOG_DIR in start_server.sh Quantong Qiu 2026-07-09 02:53:38 +00:00
  • 7e614fb702 Add DSpark backend to long-context matrix experiment yy-fighting 2026-07-09 02:36:57 +00:00
  • 278653b6a6 Add three-way comparison: MTP vs DSpark vs default yy-fighting 2026-07-09 01:42:46 +00:00
  • 7d2ed7f05d Add vLLM+MTP speculative decoding benchmark vs default on H200 yy-fighting 2026-07-08 16:13:01 +00:00
  • a56c995091 results(max-context): complete H200 max context length exploration run 20260708-143534 yy-fighting 2026-07-08 15:33:48 +00:00
  • 4658e36b1e Fix P800 baseline server startup issues yy-fighting 2026-07-08 15:27:31 +00:00
  • f6a45f1ed8 Add P800 long-context matrix and tune benchmark/health settings yy-fighting 2026-07-08 15:02:22 +00:00
  • 539b3bc539 Merge branch 'main' of https://gitee.com/yy-fighting/sskj yy-fighting 2026-07-08 14:45:00 +00:00
  • cd0b083a10 results(vllm-dspark-vs-default): complete TTFT comparison run 20260708-142121 yy-fighting 2026-07-08 14:35:14 +00:00
  • 27ceaac47f results(long-context-matrix): complete 64k/128k/256k matrix run 20260708-100016; use common scripts yy-fighting 2026-07-08 11:21:15 +00:00
  • 9b135aa4ff revert: remove intermediate matrix results committed by mistake (again) yy-fighting 2026-07-08 11:06:54 +00:00
  • 80959cf765 docs: move SLO_STANDARDS.md from scripts/ to docs/; update references yy-fighting 2026-07-08 11:06:49 +00:00
  • ccf65e745b revert: remove intermediate matrix results committed by mistake yy-fighting 2026-07-08 11:04:23 +00:00
  • 9b485c05f7 cleanup(scripts): remove unused legacy scripts; update README and workflow docs yy-fighting 2026-07-08 11:04:16 +00:00
  • f51697aed6 refactor(common): remove duplicated parse/compare/warmup from existing experiments yy-fighting 2026-07-08 10:49:27 +00:00
  • a127cf7ef0 refactor(common): centralize parse/compare/warmup; add TEMPLATE and EXPERIMENT_GUIDE yy-fighting 2026-07-08 10:48:43 +00:00
  • be94b8e7ef Add P800 256k/4k long-context probe and baseline w8a8_int8 server mode yy-fighting 2026-07-08 10:30:05 +00:00
  • 868ebc6cdf feat(long-context-matrix): add SGLang vs vLLM long-context matrix experiment yy-fighting 2026-07-08 10:09:32 +00:00
  • cd2ba29d49 feat(256k-4k-probe): verify SGLang/vLLM can serve 256k+4k without OOM yy-fighting 2026-07-08 09:59:05 +00:00
  • 0f9399baf6 feat(64k): add SGLang vs vLLM 64k-context comparison and results run 20260708-092556 yy-fighting 2026-07-08 09:44:13 +00:00
  • ee954d7409 fix(sglang-vs-vllm): use jq for metadata update; add results run 20260708-083346 yy-fighting 2026-07-08 09:21:49 +00:00
  • 5eb605b7a2 feat: support resuming interrupted SGLang vs vLLM runs yy-fighting 2026-07-08 08:42:21 +00:00
  • fd0474099a chore: do not commit partial SGLang vs vLLM results while run is in progress yy-fighting 2026-07-08 08:36:13 +00:00
  • 19d336de08 feat: add max context length exploration experiment yy-fighting 2026-07-08 08:35:56 +00:00
  • b12039599d chore: remove partial SGLang vs vLLM run that failed mid-way yy-fighting 2026-07-08 08:33:37 +00:00
  • 0bb8adf2d9 feat: add SLO status field to per-backend reports and results.json yy-fighting 2026-07-08 08:28:37 +00:00
  • b5b636236d docs: require server launch args to be recorded in results.json yy-fighting 2026-07-08 08:22:33 +00:00
  • 162168da3c feat: record exact server launch args in results.json for reproducibility yy-fighting 2026-07-08 08:21:55 +00:00
  • 69756c598c fix: ignore nested backend raw_outputs/logs and remove accidentally committed intermediate files yy-fighting 2026-07-08 08:15:39 +00:00
  • 5d1cd9c266 feat: add vLLM+DSpark vs vLLM-default TTFT comparison experiment yy-fighting 2026-07-08 08:15:02 +00:00
  • 3cf8b8f0d7 fix: use --context-length instead of --max-model-len for SGLang server yy-fighting 2026-07-08 07:57:15 +00:00
  • 06b3398ad0 feat: add ShareGPT dataset scenarios to SGLang vs vLLM comparison yy-fighting 2026-07-08 07:42:05 +00:00
  • e6d9e99b6f feat: add SGLang vs vLLM controlled comparison experiment yy-fighting 2026-07-08 07:34:34 +00:00
  • 8b82f700e4 feat: add H200 SGLang baseline experiment and fix health check yy-fighting 2026-07-08 06:56:45 +00:00
  • 20d04a31ae refactor: move final legacy result dspark_grid into experiments/legacy_bench_results/ Quantong Qiu 2026-07-08 06:40:39 +00:00
  • 0397b4a2b5 refactor: move legacy bench_results into experiments/legacy_bench_results/ Quantong Qiu 2026-07-08 06:36:19 +00:00
  • b742187498 fix(dsv4_h200_vllm): scenario array parsing and sglang output format Quantong Qiu 2026-07-08 06:31:03 +00:00
  • 8e15ffe1c7 refactor: relocate old docs and add dsv4_h200_vllm experiment Quantong Qiu 2026-07-08 06:19:40 +00:00
  • 07835b9fed chore: remove obsolete scripts/test_block_sparse_attn.py after refactor Quantong Qiu 2026-07-08 06:01:31 +00:00
  • a5d69dfd9a docs(h200): H200 quick-start guide, porting workflow, and example experiment Quantong Qiu 2026-07-08 05:55:11 +00:00
  • f0d4c4753b fix(common): robust XPU platform auto-detection Quantong Qiu 2026-07-08 05:50:42 +00:00
  • e72e34c2f7 fix(p800): managed-mode server startup and smoke-test defaults Quantong Qiu 2026-07-08 05:49:55 +00:00
  • 227ef603a6 refactor: experiments/ + platforms/ layout and P800 SGLang support Quantong Qiu 2026-07-08 05:15:59 +00:00
  • 2c332ff712 docs: record chip/accelerator and inference engine in benchmark workflow Quantong Qiu 2026-07-08 03:52:23 +00:00
  • 7de88b995a docs: 规范 benchmark 结果存储格式，要求每个实验同时产出 .md 报告和结构化 results.json Quantong Qiu 2026-07-08 03:31:47 +00:00
  • eb431d3925 Initial commit: benchmark scripts and final reports Benchmark Bot 2026-07-08 02:17:13 +00:00