Qwen3.8-Flash-Next-NVFP4

모델 소개

Qwen3.8-Flash-Next is 360 GB in BF16 and needs two data-center GPUs in FP8. This build serves it on one 96 GB Blackwell : 88.8 GiB of VRAM, the 51B n-gram table in host RAM, and no patched runtime. Two flags are load-bearing on a single GPU. `--distributed-executor-backend mp`: the default single-GPU executor never starts the n-gram offload worker, and the first forward waits on it forever — the server looks healthy and hangs. `VLLM_PLE_OFFLOAD_READY_TIMEOUT=1800`: the worker loads a 95 GB table before serving and the 600 s default can expire first. Host needs about 100 GB of free RAM. The reasoning parser and tool-call parser in this block are validated on this exact checkpoint: thinking lands in `reasoning` with no markup in `content`, and tool calls arrive as structured `tool_calls` with valid JSON arguments and `finish_reason: tool_calls`. If you have seen other NVFP4 cards warn that the automatic MoE backend picks `FLASHINFER_TRTLLM` and corrupts output: on this image on an RTX PRO 6000 the automatic choice is `FLASHINFER_CUTLASS`, printed at boot, and every number on this card went through it. Measured on the mixed sibling (same expert kernel), `--moe-backend marlin` decodes 10% faster single-stream (93.2 vs 84.6 tok/s) and 1–3% slower at concurrency 32, with TTFT up 12%; tool-calling under Marlin 79.5 (n=1). Marlin does not combine with MTP on this image: the BF16 MTP draft inherits the backend and refuses it, and the draft-scoped override in `--speculative-config` is ignored by Model Runner V2 until vllm-project/vllm#54788 lands. Choose MTP for interactive use, Marlin only without speculation. On vLLM nightly builds from early September the SM120 selector picks Marlin by default anyway (vllm#54788); the flag matters on this image. Two more flags are in the block since 2026-09-06: without them the image's defaults do not boot on a 96 GB card. A field report found it first and we reproduced both failures on this box the same day. Without MTP the default `--max-num-seqs` (1024) exceeds the Mamba state blocks left at 0.92 utilization (598 here) and CUDA graph capture aborts with `Engine core initialization failed`; with MTP at the native 262,1

주요 특징

特点摘要待依据来源补齐，不凭模型名称推断能力。

한계 및 유의 사항

구체적 한계를 판단할 출처가 부족합니다. 실제 성능, 지연, 데이터 정책, 라이선스와 버전 수명을 확인하세요.

출처별 요금 및 비용 분석

요금은 채널, 지역, 설정에 따라 다릅니다. 각 채널 요금은 개별 표시되며 제공사의 통일 공식 요금이 아닙니다.

공개 평가 및 실행 설정

暂无已采集的可核验公开评分。模型未上榜不代表能力较低。

자주 묻는 질문

출처 및 추적 기록

许可证：other。模型介绍与第三方评测均以原来源的具体版本和方法为准。