Page values for "Qwen3.8 Flash Next on 3x 8 GB (clee PP3)"
From CMP Wiki
More actions
"Recipe" values
1 row is stored for this page| Field | Field type | Allowed values | Value |
|---|---|---|---|
| page | Page | Qwen3.8 Flash Next on 3x 8 GB (clee PP3) | |
| model | String | Qwen3.8-Flash-Next | |
| quant | String | W4A16 AutoRound: Intel/Qwen3.8-Flash-Next-W4A16-AutoRound; 95.4 GiB BF16 PLE table streamed from NVMe; KV fp8 e4m3 (software-decoded on SM80) | |
| engine | String | vLLM nightly 7d0b4e57 plus the author's pure-Python fork clee/vllm-sm80 (branch q38fn/stack) | |
| engine_version | String | q38fn/stack on nightly-7d0b4e57 | |
| cards | String | 8 GB · 10 GB · both | 8 GB |
| card_count | Integer | 3 | |
| topology | String | PP3, no P2P (works at PP2 on two cards) | |
| context | String | 1,048,576 (YaRN x4) | |
| decode_tps | Float | 74.4 | |
| prefill_tps | Float | 4,831.0 | |
| vram_per_card | String | 64 GiB, gpu-memory-utilization 0.92 | |
| power_cap | String | ||
| driver | String | ||
| unlock_build | String | ||
| flags | Text | pipeline-parallel-size 3, block-size 1728, kv-cache-dtype fp8_e4m3, max-model-len 1048576, max-num-seqs 8, max-num-batched-tokens 8192, FULL_AND_PIECEWISE; NCCL_P2P_DISABLE=1, VLLM_PP_LAYER_PARTITION=16,16,16, VLLM_PLE_STAGED=1 | |
| status | String | Proven · Reported · Theory · Contested · Warning | Proven |
| tested_by | String | c133 | |
| tested_on | Date | 2026-10-10 | |
| source | String | https://github.com/clee/qwen3.8-flash-next-3x170hx | |
| files | Wikitext |