← back to catalog · registered 2026-09-16 06:56

lambsea/Qwen3.8-Flash-Next-Uncensored-NVFP4-AWQ-GPTQ

lambsea MoE second-order
Abliteration classifier · v1.0.0
M-U
Primary method

Uncensored (method unknown)

No other method signals detected in this model.
Confidence
LOW
Why this label 3 signals
Weak or ambiguous signals. Best guess based on catalog patterns; treat as tentative and check the evidence below.
  • 'uncensored' in name/tags but no 'abliterated' marker
  • method not identifiable from author declaration alone
  • may be DPO fine-tune, prompt engineering, or unknown technique
Refusal direction extraction

No specific extraction method could be identified for this model. The producer either did not document it or used a proprietary pipeline.

What is a refusal direction? →
Downloads · 30-day
7
Likes
0
Model age
today
created 2026-09-16

Genealogy 0 direct forks

Full fork graph →

This model's place in the market. Above: what it was derived from. Below: the tree of everything derived from it.

Variants by this author 2 formats · 7 downloads combined

The same weights this author released in different packaging. Pick the format that matches your runtime.

Metadata

License
apache-2.0
Tags
vLLM safetensors qwen4_exp nvfp4 awq gptq quantization moe text-generation conversational en base_model:orcarouter/Qwen3.8-Flash-Next-Uncensored

Related

Total size
171 GB
Files
28
Quantizations
1
Registered
2026-09-16 06:56
Last updated on HF
2026-09-16 07:59

Files by quantization

Auxiliary files 28 files 171 GB
model-00002-of-00017.safetensors 95.4 GB 64836044 download
model-mtp.safetensors 4.86 GB 5b134ebb download
model-00007-of-00017.safetensors 4.66 GB 04a72311 download
model-00003-of-00017.safetensors 4.66 GB 4cd8bb97 download
model-00009-of-00017.safetensors 4.66 GB 113c1830 download
model-00006-of-00017.safetensors 4.66 GB dc4d89c6 download
model-00013-of-00017.safetensors 4.66 GB 4b50b637 download
model-00008-of-00017.safetensors 4.66 GB cd52a3df download
model-00011-of-00017.safetensors 4.66 GB 1d775e13 download
model-00014-of-00017.safetensors 4.66 GB 651a5692 download
model-00015-of-00017.safetensors 4.66 GB 3e6e346f download
model-00010-of-00017.safetensors 4.66 GB 66f0d044 download
model-00005-of-00017.safetensors 4.66 GB e7d0bce9 download
model-00004-of-00017.safetensors 4.66 GB b285e874 download
model-00001-of-00017.safetensors 4.66 GB 8df33fa5 download
model-00012-of-00017.safetensors 4.63 GB e96fc471 download
model-00016-of-00017.safetensors 4.28 GB 03c4c2fe download
model-00017-of-00017.safetensors 1.18 GB 50be0ccd download
scales.safetensors 8.17 MB 966602c5 download
model.safetensors.index.json 23.5 MB b2f2963b download
tokenizer.json 19.1 MB 06b95093 download
config.json 73.4 KB c75c3372 download
chat_template.jinja 8.74 KB c0c686f9 download
README.md 3.59 KB c22004ab download
.gitattributes 1.60 KB aa7aacd0 download
tokenizer_config.json 1.10 KB d1a20cc3 download
preprocessor_config.json 390 B 2ea84a43 download
generation_config.json 205 B a382b041 download

README current version from Hugging Face


language: en
library_name: vLLM
base_model: orcarouter/Qwen3.8-Flash-Next-Uncensored
pipeline_tag: text-generation
license: apache-2.0
tags: [nvfp4, awq, gptq, quantization, moe]

Qwen3.8 Flash Next — NVFP4 (AWQ + GPTQ, calibrated)

Calibrated 4-bit weights of orcarouter/Qwen3.8-Flash-Next-Uncensored: routed experts in NVFP4
(W4A4, AWQ smoothing + GPTQ error compensation on ~1K calibration
sequences), attention / GDN / shared-expert in FP8 per-output-channel;
embeddings, lm_head, norms, router gates and the PLE table kept in BF16.
Includes the MTP head (model-mtp.safetensors).

Benchmarks

Measured against the BF16 base model on 328,800 positions of
self-distilled thinking-mode output (8 documents: 2 multi-turn agentic
coding sessions with tool loops, 6 plain reasoning/math/writing seeds,
all generated to natural stop), teacher-forced top-1/top-2 at every
position. 8xRTX Pro 6000, TP8 + expert parallel, vLLM 0.29.0, BF16 KV
(see Serving notes). Agreement = raw argmax match vs the BF16 base;
buckets split by the BF16 top1-top2 logprob margin. Disagreement % shown
(lower is better).

checkpoint top-1 near-tie moderate confident certain
orcarouter community NVFP4 93.90% 39.82 12.34 2.60 0.15
this model (NVFP4 + AWQ + GPTQ) 93.93% 39.67 12.31 2.59 0.12

Per-category disagreement vs BF16 (tokens scored):

category tokens this model community NVFP4
coding (agentic, tool loops) 200,682 7.76% 7.79%
math 53,166 3.98% 4.05%
reasoning 63,769 2.39% 2.38%
writing 11,175 6.67% 6.95%

Paired per-position logprob test on the BF16-preferred token: this model
carries +0.0019 nats over the community quant (SE 0.0005, t=+4.1,
321,736 paired positions). Residual disagreement concentrates in BF16
near-ties (whitespace width, table digits, synonym flips) — sampled
splits show no reasoning or code defects on either side.

Serving

Tested with vLLM >= 0.29 (native qwen4_exp support), TP8 with
--enable-expert-parallel on 8xRTX Pro 6000 (tree is ~180 GB):

vllm serve lambsea/Qwen3.8-Flash-Next-Uncensored-NVFP4-AWQ-GPTQ \
  --tensor-parallel-size 8 --trust-remote-code \
  --enable-expert-parallel --enable-auto-tool-choice \
  --tool-call-parser qwen3_coder

Notes:

  • KV cache must be BF16 for now: vLLM's QSA layers for this
    architecture raise Qwen4Exp QSA does not support KV quantization. The
    shipped config.json declares no kv_cache_scheme.
  • The quantization manifest targets use the model.language_model.layers.*
    prefix (VLM-wrapper naming); the checkpoint loads through the
    Qwen4ExpForConditionalGeneration path even for text-only serving, so
    preprocessor_config.json must stay in the repo.
  • FP8 tensors carry per-output-channel weight_scale of shape
    [out_features, 1]; NVFP4 experts carry block-16 weight_scale
    (F8_E4M3, in global-scaled space) plus a quant-side scalar
    weight_global_scale — i.e. the kernel dequant multiplier is
    1 / weight_global_scale, the standard checkpoint convention.

Provenance

Quantized 2026-09 on 8xH100 (8-GPU DDP calibration, compressed-tensors
disk-offload pipeline), exported to a serving tree, repaired scales
(FP8 per-channel re-quantized from the linearized BF16 source; NVFP4
global scales normalized to the quant-side convention), and validated
end-to-end by the benchmark above. All measurements, corpus, and the
BF16 ground-truth score files are reproducible from the published run
profile.

Catalog is the map. Apps are the tools.

Run models on your own machine, not in the cloud.

Every model page has an "Open in app" button that hands off directly to a local runtime of your choice - Infrahuman, LM Studio, or Ollama. No API keys, no subscription, no prompt leakage.