Repos made with Qwen3.8-Flash-Next

Qwen3.8-Flash-Next on ONE DGX Spark (TP=1)

The fastest way to run Qwen3.8-Flash-Next on Strix Halo (gfx1151)

Qwen3.8 Flash Next on one DGX Spark (TensorFold)

recipe for running Qwen3.8-Flash-Next on a single DGX Spark

Qwen3.8-Flash-Next-NVFP4-vLLM-Dual-DGX-Spark

qwen38-flash-next recipe for solo GB10

qwen38-flash-next recipe for tp=2 GB10

Strata fork: Qwen3.8-Flash-Next 125B MoE in NVFP4 on one RTX 20-50 card (12 GB+) and 64 GB of RAM or more. Setup installs our GPTQ quants (huihui abliterated, OrcaRouter) from Hugging Face. CPU experts on AVX-512 or AVX2 (+AVX-VNNI), all experts in RAM from 92 GB, a low-RAM mode…

An inference framework customized for AI MAX 395, compatible with both Linux and Windows, and supports deploying only qwen3.8-flash-next. This project is for reference and learning purposes only. 为AI MAX 395定制的推理框架，Linux和Windows通用，只能部署qwen3.8-flash-next。本项目仅供参考学习使用。

Qwen3.8-Flash-Next (NVFP4) on DGX Spark in vLLM: one Spark 43.9 tok/s with our disk-backed n-gram table patch, staged gather and reduced-vocab MTP draft; TP2 SPEED 53.7; TP4 CONTEXT 9M KV pool. Launchers, patch, 40-prompt harness, KV ledger, credits. SGLang TP2 day-0 lane kept.

Run Qwen3.8-Flash-Next (180B) on a single DGX Spark by keeping its 51B n-gram embedding table on NVMe

Qwen3.8-Flash-Next on 2× RTX 3090: with 128 GB RAM up to 4,191 tok/s prefill · 111.5 tok/s decode (131K prompt), full 256K window at 2,865 tok/s; with 64 GB RAM 3,410 tok/s prefill · 84 tok/s decode. New: opt-in uncensored mode (runtime abliteration, no new weights).

Portable optimized ROCm and Qwen3.8-27B setup for AMD Strix Halo

Serve Qwen3.8-Flash-Next (turboderp ExLlamaV3 pack) on one NVIDIA DGX Spark with exllamav3

Recipe for the fastest 4bit qwen3.8FlashNext

Qwen3.8-Flash-Next on one Intel Arc Pro B70 with experts on NVMe and 32 GB host RAM: SGLang + exl3xpu tier, staged NVMe prefill, SYCL sparse attention, measured results

A custom inference engine tuned for Qwen4 architecture. Currently runs Qwen3.8-Flash-Next at 15+ tok/s on 16GB of VRAM and 30 GB of RAM

Shape-specialized ROCm inference for Qwen3.8 Flash Next on RDNA4

Run Qwen3.8-Flash-Next (180B, 6B active) on a single AMD Strix Halo via the halogen-flash-server container

Verified Qwen3.8 Flash-Next GGUF recipe for one NVIDIA DGX Spark or ASUS Ascent GX10

Async PLE SSD offload for Qwen3.8 Flash Next on a 64 GiB SM80 GPU: vLLM patch, setup guide, and measured PP/TG results.

The published EXL3 quant — r0b0tlab/Qwen3.8-Flash-Next-EXL3-2.50bpw
— tuned for a 24 GB RTX 3090, running at the model's full 262,144-token native context
with the MTP head (NEXTN) and the vision tower intact, using MoE expert CPU
offload to fit the 125B model (6B active…

A script to launch Qwen3.8-Flash-Next in a single DGX-Spark (or its variants) at 40+ tok/s inference speed.

双 SM80 / CMP 170HX 的 Qwen3.8-Flash-Next 社区 vLLM：NVIDIA NVFP4、GDS、TEP2、MTP6＋历史草稿、内存与 PCIe 通信优化。中英文部署与可构建源码。