Skip to content

Commit d5728f5

Browse files
committed
Append perf-changelog entry for PR #1586
1 parent 03c20f7 commit d5728f5

1 file changed

Lines changed: 8 additions & 0 deletions

File tree

perf-changelog.yaml

Lines changed: 8 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -3200,3 +3200,11 @@
32003200
- "Bump image to lmsysorg/sglang-rocm:v0.5.12.post1-rocm720-mi35x-20260523, 1P1D TP8/EP1, dp-attn false, conc [8..512]"
32013201
- "MoRI conn.py overlay (48e459bd) via job.slurm; launcher qwen3.5_fp4_mi355x_sglang-disagg.sh"
32023202
pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1579
3203+
3204+
- config-keys:
3205+
- dsv4-fp4-gb300-dynamo-sglang
3206+
description:
3207+
- "Add wide-EP sweep configs (EP=12/16/24/32/40) matching srt-slurm PR#173 topology (18 nodes total)"
3208+
- "EP=12 15P+3D conc=12000, EP=16 14P+4D conc=8192, EP=24 12P+6D conc=3000, EP=32 10P+8D conc=2500, EP=40 8P+10D conc=2048"
3209+
- "Aligned decode params with Weiliang config: swa-full-tokens-ratio=0.20, max-running-requests=18432, moe-dense-tp-size=1; added prefill enable-dp-lm-head and cuda-graph-max-bs=512"
3210+
pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1586

0 commit comments

Comments
 (0)