TL;DR: Huihui-Qwythos-9B-Claude-Mythos, quantized to NVFP4 (W4A4) for vLLM on NVIDIA Blackwell. 7.5 GB, wikitext-2 PPL 8.25, 1M-context creative writer, abliterated.

Huihui-Qwythos-9B-Claude-Mythos-5-1M-abliterated NVFP4

NVFP4 (W4A4) quantization of huihui-ai/Huihui-Qwythos-9B-Claude-Mythos-5-1M-abliterated, packed in the compressed-tensors nvfp4-pack-quantized format with llm-compressor. Weights are quantized with GPTQ (error-compensated rounding) and an MSE observer, on a domain-matched calibration blend.

Near-lossless. Fused layers (q/k/v, gate/up) share one NVFP4 global scale, so vLLM loads it cleanly with no per-layer-scale warning or fallback. wikitext-2 perplexity for this build: 8.25.

About 7.5 GB on disk versus about 19.3 GB for the bf16 source (about 39%).
Built for vLLM on NVIDIA Blackwell, where both the 4-bit weight and 4-bit activation paths are accelerated. On pre-Blackwell GPUs vLLM runs it weight-only.
Loading and generation verified in vLLM on an NVIDIA GB10 (Blackwell, sm_121).

Uncensored / abliterated model. It follows instructions without content guardrails, including NSFW. Behaviour and alignment are inherited entirely from the base model.

Fidelity

Near-lossless versus the bf16 source, 7.5 GB vs 19.3 GB bf16 (~39%), at wikitext-2 perplexity 8.25. GPTQ error compensation and an MSE observer keep the drop from bf16 minimal; the header lists the full characteristics and Quantization covers the recipe.

Quickstart

NVFP4 is auto-detected from config.json (compressed-tensors); no quantization flag needed. --reasoning-parser qwen3 splits the <think> block into reasoning_content.

bash

vllm serve maci0/Huihui-Qwythos-9B-Claude-Mythos-5-1M-abliterated-NVFP4 \
  --served-model-name qwythos-9b-nvfp4 \
  --max-model-len 131072 \
  --gpu-memory-utilization 0.90 \
  --kv-cache-dtype fp8 \
  --reasoning-parser qwen3 \
  --enable-auto-tool-choice --tool-call-parser qwen3_coder

Supports up to 1048576 tokens; keep at least 128K to preserve thinking quality. --max-model-len 131072 is a safe default; raise it if memory allows.
Add --language-model-only to skip the vision tower and free KV cache for text use.
The parser flags are not auto-detected; pass them explicitly. Drop the tool-call line if you do not need tool calling.

Python (OpenAI client)

python

from openai import OpenAI
client = OpenAI(base_url="http://localhost:8000/v1", api_key="x")
r = client.chat.completions.create(
    model="qwythos-9b-nvfp4",
    messages=[{"role": "user", "content": "Write a short mythic tale about a lantern that remembers the sea."}],
)
print(r.choices[0].message.content)

curl

bash

curl http://localhost:8000/v1/chat/completions -H "Content-Type: application/json" -d '{
  "model": "qwythos-9b-nvfp4",
  "messages": [{"role": "user", "content": "Write a short mythic tale about a lantern that remembers the sea."}]
}'

About the base model

An abliterated (uncensored) build of empero-ai/Qwythos-9B-Claude-Mythos-5-1M, a 9B Qwen3.5-family vision-language model tuned for creative / Mythos writing with a Claude distillation, and a 1M-token context window.

32 decoder layers: hybrid gated delta-net linear attention plus full attention, dense MLP, plus a vision tower for image and video input.
1M context (max_position_embeddings 1048576).
Thinking mode by default, with an instruct toggle. In non-thinking mode the base model adheres strictly to the first prompt's instructions (output only, no fluff).

Quantization


Scheme	NVFP4, W4A4
Weight rounding	GPTQ (Hessian-based error compensation), MSE observer
Weights	FP4 (E2M1), `group_size=16`, `tensor_group`, FP8 (E4M3) group scales, shared across fused layers
Activations	FP4, dynamic per-group, FP8 (E4M3) scales
Quantized	all language-model `Linear` layers
Kept in bf16	vision tower (`model.visual.*`), `lm_head`, MTP head
Untouched	gated delta-net `Conv1d` and SSM params (`A_log`, `dt_bias`), never `Linear`

GPTQ is a quantization-time cost only; inference speed and format are identical to plain round-to-nearest NVFP4, but it chooses better 4-bit values.

Calibration: 512 domain-matched samples (long reasoning + general chat + code), max_seq_len=2048, text-only path through the VL model.

Recommended sampling

Thinking mode is the default.

Thinking, general / creative: temperature=1.0, top_p=0.95, top_k=20
Instruct / non-thinking: temperature=0.7, top_p=0.80, top_k=20
To run non-thinking, set {%- set enable_thinking = false %} in the chat template, or pass extra_body={"chat_template_kwargs": {"enable_thinking": false}}.

Reproduction

Toolchain: llmcompressor==0.12.0, compressed-tensors==0.17.1, transformers==5.12.1, torch==2.11.0+cu130, on an NVIDIA GB10 (Blackwell, sm_121). llm-compressor 0.12 shares the NVFP4 global scale across fused layers automatically (q/k/v, gate/up).

Base model: huihui-ai/Huihui-Qwythos-9B-Claude-Mythos-5-1M-abliterated
Space: Rogue Quants
Collection: NVFP4 Quants
Sibling NVFP4 quants:

Notes

Needs NVIDIA Blackwell (sm_121, e.g. GB10) for accelerated W4A4; pre-Blackwell GPUs run it weight-only.
--reasoning-parser and --tool-call-parser are not auto-detected; pass them explicitly.
Thinking mode is on by default; toggle it via the chat template or chat_template_kwargs.
No refusal guardrails; you are responsible for how you use it.

License

Apache-2.0, following the base model. Intended use and all responsibility for use follow the base model.

Credits

Base model: huihui-ai (abliteration) / empero-ai (Qwythos)
Quantization tooling: llm-compressor / compressed-tensors

Part of Rogue Quants · NVFP4 component datasheets · collection. Fabricated on GB10 (Blackwell) with llm-compressor. Refusals shown per 100 harmful prompts; "n/a" = not separately measured (base-inherited).

maci0/Huihui-Qwythos-9B-Claude-Mythos-5-1M-abliterated-NVFP4