RadixArk/Qwen3.8-27B-NVFP4-BF16-LMHead

closed

Model revision: Campaign ID:

Target GPUs

RTX5090

4× GPU per bench

Submissions

23 passing · 6 rejected

Scoring rule

median e2e speedup

RTX5090

p99 latency

2000 / 50 ms

TTFT / ITL ceiling

Objective

Priority metric

gpu hours

Win condition

>=10% GPU-hour reduction at SLA

Scoring rule

median e2e speedup

Correctness

Mean logprob ≥ -4
Min token logprob ≥ -16
Coverage ≥ 0.5

Model

Precision

bfloat16 · modelopt_mixed

Context window

262,144 tokens

Provenance

Baseline repo

github.com/sgl-project/sglang.git

Baseline commit

Patch visibility

Private for this campaign

Manifest hash

Base image

Baseline engine

Workload

Target GPUs

RTX5090

4× GPU per bench

Submissions

23 passing · 6 rejected

Scoring rule

median e2e speedup

RTX5090

p99 latency

2000 / 50 ms

TTFT / ITL ceiling

Objective

Priority metric

gpu hours

Win condition

>=10% GPU-hour reduction at SLA

Scoring rule

median e2e speedup

Correctness

Mean logprob ≥ -4
Min token logprob ≥ -16
Coverage ≥ 0.5

Model

Precision

bfloat16 · modelopt_mixed

Context window

262,144 tokens

Provenance

Baseline repo

github.com/sgl-project/sglang.git

Baseline commit

Patch visibility

Private for this campaign

Manifest hash

Base image

Baseline engine

Workload

Patch scope

Allowed paths

python/sglang/**rust/**

Denied paths

tests/**benchmarks/**.github/**docker/****/Dockerfile***/pyproject.toml**/setup.py**/setup.cfg**/requirements*.txttest/**benchmark/**python/sglang/test/**python/sglang/kernels/aot/tests/**python/sglang/kernels/aot/python/sgl_kernel/test_utils.pypython/sglang/kernels/aot/python/sgl_kernel/testing/**rust/**/tests/**rust/**/benches/**