[NVIDIA] chore: B300 single node DeepSeek v4 SGLang LOW LATENCY ONLY#1143

Merged

cquil11 merged 21 commits intomainfrom

chore/dsv4-sgl-b300-low-latency

Apr 24, 2026

Commits on Apr 24, 2026

feat: add DeepSeek-V4-Flash FP4 B300 SGLang benchmark

cquil11
and
claude
committed
fix: switch dsv4-fp4-b300-sglang to Pro + Max-Throughput recipe

cquil11
and
claude
committed
chore: sync launch_b200-dgxc-slurm.sh cache mount from claude/add-dsv4-fp4-b200-vllm

cquil11
and
claude
committed
fix: restore trailing whitespace stripped from glm5.1 changelog entry

cquil11
and
claude
committed
chore: add flock-guarded squash import to B300 runner

cquil11
and
claude
committed
fix: drop ENROOT_CACHE_PATH override from B300 runner

cquil11
and
claude
committed
chore: point B300 runner at shared gharunners/{squash,hf-hub-cache}

cquil11
and
claude
committed
fix: move enroot import out of srun to avoid pyxis namespace collision

cquil11
and
claude
committed
fix: wipe stale pyxis scratch dirs for this JOB_ID before benchmark srun

cquil11
and
claude
committed
Revert: drop all B300 runner changes, mirror #1128's approach

cquil11
and
claude
committed
runner: add head-node flock-guarded squash import on B300

cquil11
and
claude
committed
fix: mount at /ix and clear baked-in CUDA_VISIBLE_DEVICES

cquil11
and
claude
committed
Merge branch 'main' into chore/dsv4-sgl-b300
cquil11
authored
runner: use /data/models pre-staged path for dsv4 on B300

cquil11
and
claude
committed
fix: switch B300 dsv4 sglang to bw-ultra-compiled image

cquil11
and
claude
committed
fix: switch B300 dsv4 sglang image to yhyang201/sglang-b300:v3

cquil11
and
claude
committed
update b300
cquil11
committed
feat(dsv4-fp4-b300-sglang): pick recipe by CONC; split search-space

cquil11
and
claude
committed
update b300

cquil11
and
claude
committed
feat(dsv4-fp4-b300-sglang): low-latency recipe at every CONC (fallback)

cquil11
and
claude
committed
fix: align perf-changelog and config comments with low-latency fallback

committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[NVIDIA] chore: B300 single node DeepSeek v4 SGLang LOW LATENCY ONLY#1143

[NVIDIA] chore: B300 single node DeepSeek v4 SGLang LOW LATENCY ONLY#1143
cquil11 merged 21 commits intomainfrom
chore/dsv4-sgl-b300-low-latency

Commits on Apr 24, 2026

feat: add DeepSeek-V4-Flash FP4 B300 SGLang benchmark

fix: switch dsv4-fp4-b300-sglang to Pro + Max-Throughput recipe

chore: sync launch_b200-dgxc-slurm.sh cache mount from claude/add-dsv4-fp4-b200-vllm

fix: restore trailing whitespace stripped from glm5.1 changelog entry

chore: add flock-guarded squash import to B300 runner

fix: drop ENROOT_CACHE_PATH override from B300 runner

chore: point B300 runner at shared gharunners/{squash,hf-hub-cache}

fix: move enroot import out of srun to avoid pyxis namespace collision

fix: wipe stale pyxis scratch dirs for this JOB_ID before benchmark srun

Revert: drop all B300 runner changes, mirror #1128's approach

runner: add head-node flock-guarded squash import on B300

fix: mount at /ix and clear baked-in CUDA_VISIBLE_DEVICES

Merge branch 'main' into chore/dsv4-sgl-b300

runner: use /data/models pre-staged path for dsv4 on B300

fix: switch B300 dsv4 sglang to bw-ultra-compiled image

fix: switch B300 dsv4 sglang image to yhyang201/sglang-b300:v3

update b300

feat(dsv4-fp4-b300-sglang): pick recipe by CONC; split search-space

update b300

feat(dsv4-fp4-b300-sglang): low-latency recipe at every CONC (fallback)

fix: align perf-changelog and config comments with low-latency fallback