language:

• en
• zh
  license: other
  license_name: qwen-community-1.0
  license_link: https://huggingface.co/Qwen/Qwen3.8-Flash-Next/blob/main/LICENSE
  base_model: orcarouter/Qwen3.8-Flash-Next-Uncensored
  tags:
• qwen3.8
• qwen4_exp
• auto-round
• signroundv2
• mixed-precision
• agentic-coding
• mtp
• vision-language

Qwen3.8-Flash-Next-Uncensored AutoRound ~3 bpw

This is a coding-calibrated mixed-bit quantization of the pinned
OrcaRouter BF16 source
revision 8336e613ea508b13c2159bd0f68965d97a606b95. It retains the
original vision tower, native BF16 MTP head, tokenizer, chat template, and
BF16 PLE table. ~3 bpw applies to the routed experts, not the entire
checkpoint; the PLE table alone is approximately 102.4 GB.

Quantization

Intel AutoRound SignRoundV2 tuned all 48 backbone MoE blocks using 64
repository-distinct coding/tool windows of 2,048 tokens and 50 iterations.
Blocks 0–11 and 36–47 use W3A16G128; blocks 12–35 use W2A16G64. Dense
attention and DeltaNet projections use W8A16G128; routers, shared experts,
hyperconnections, vision, MTP, and PLE remain BF16. The calibration source
is NVIDIA Open-SWE-Traces
at f8fb5b3d2c787f85f8a00f5fe04fe3f1a11088ef (CC BY 4.0). See
RECIPE.md and recipe.json for reproducibility details.

Validation

• Audit: 35 indexed weight shards, 223,082 tensors, 156,770,420,728 tensor
  bytes; all 31 native BF16 MTP tensors were bit-identical to the source.
• Patched-vLLM generation and native MTP speculative decoding were tested.
  A 131,072-token, one-token generation smoke test peaked at 64,927 MiB
  (~63.4 GiB) GPU memory on an RTX PRO 6000 Blackwell with a 0.61
  utilization cap.
• Paired held-out coding-trace BF16→quantized fidelity, one window and 128
  scored positions per length: 8K context KL 0.03718 nats, top-1
  agreement 96.09%; 32K context KL 0.13202 nats, top-1 agreement
  92.19%.
• Four-task mini-SWE-agent/SWE-bench Verified check: 4/4 resolved
  by the official grader. Testing
  used thinking mode with temperature 1.0, top-p 0.95, top-k 20, min-p 0,
  presence penalty 0, and repetition penalty 1. The per-turn max_tokens
  limit was 8,192; observed completions stayed below it. All task
  repositories were absent from the calibration shard.

Serving

Use the pinned patched vLLM runtime and guide,
including its PLE SSD-offload and mixed-bit patch
against vLLM a5a30471ff2bb7f0824f2da10e358af98d304472. Set
QWEN_MODEL_DIR to this checkpoint and follow the guide's build steps; do
not download the previous checkpoint. The PLE table stays on SSD. For the
native BF16 MTP draft on the tested Blackwell host, use Triton MoE backend.
Qwen's recommended thinking-mode sampling was used for the agent validation.

This derivative retains the Qwen Community License 1.0 notice.