Qwen3.8-27B · SGLang · NVFP4 + DFlash2 · conc-8
status: blocked
Configuration
| Model | Qwen/Qwen3.8-27B |
|---|---|
| Company | Alibaba |
| Family | Qwen |
| Parameters | 27B (dense, hybrid linear-attn) |
| Engine | SGLang + DFlash2 (speculative decoding) |
| Quant / precision | NVFP4 |
| Why this quant | Same Spark-recipe NVFP4 target (RadixArk/Qwen3.8-27B-NVFP4) as the DSpark line, but DFlash2 block-diffusion draft (z-lab/Qwen3.8-27B-DFlash2 @50307d4) on the derived lmsysorg/sglang:qwen38-27b-dflash2 image. mem-fraction-static 0.90. conc-8 point of the ShareGPT sweep (client conc 1/2/4/8/16/32). |
| Download | RadixArk/Qwen3.8-27B-NVFP4 |
| Context window | 262144 |
| Input modalities | text, image, video (served text-only here) |
Measured results
| Prefill tok/s | — |
|---|---|
| Decode tok/s | — |
| Peak memory (GB) | — |
| Completed | 2026-08-20 05:25 +0800 |
Full run command
# Not run — memory ceiling. Last attempted boot used MAX_CONCURRENT_REQUESTS=5.
cd ~/Desktop/Qwen3.8-27B-SGLang-DGX-Spark
MAX_CONCURRENT_REQUESTS=5 ./start-dflash.sh
python3 ~/Desktop/autobench/scripts/bench-serving.py \
--base-url http://127.0.0.1:8888 --model qwen3.8-27b-sglang \
--dataset ~/Desktop/autobench/benchmark_data/ShareGPT_V3_unfiltered_cleaned_split.json \
--num-prompts 1000 --max-seconds 900 --concurrency 8 --max-tokens 256
Notes
BLOCKED — memory ceiling above DFlash2 conc-4. The 4-running boot (MAX_CONCURRENT_REQUESTS=5, 5 mamba slots/req → 4 running, --cuda-graph-max-bs-decode 4) SIGKILL’d 69 s into conc-4 (27 ok / 973 err, exit 137) after a scheduler GC freeze on top of the 0.90 reservation (~7 GB free at ready). The 8-running DFlash2 boot already died mid conc-2. Do not retry conc-8 at 0.90.