← back to catalog · registered 2026-11-16 20:03

raydollete/Qwen3.8-Flash-Next-HyperFlare-A16-Abliterated-1M

Your rig guess connected
? Why do I need an app?
Reading your rig…

This is a rough estimate. Install the free app - we'll show exact numbers.

Reading real hardware from your app right now. Numbers below are exact.

Below is the per-quantization compatibility for this model.

curl -H "Authorization: Bearer $ABL_KEY" \
     "https://abliteration.org/api/v1/models/raydollete%2FQwen3.8-Flash-Next-HyperFlare-A16-Abliterated-1M"
Response includes
  • classification m1
  • files 6
  • author_summary 1 models
  • readme_text full
10 credits · hourly refresh · ~4 KB payload Get an API key →
Abliteration classifier · v1.0.0
M1
Primary method

Direct removal

No other method signals detected in this model.
Confidence
MEDIUM
Why this label 3 signals
Method inferred from partial signals - repository name, related files, or tag patterns. Producer identity not confirmed; label may sharpen or shift as we gather more evidence.
  • 'abliterated' in name/tags
  • is_gguf=0 (base model)
  • no specific method indicators - defaulting to M1 (most common)
Refusal direction extracted via
Extraction technique

Difference-of-means

Confidence
MEDIUM
Why we say so
primary_method=M1; difference-of-means is the reference extraction for M1/M3 (Arditi 2024)
Downloads · 30-day
0
Likes
1
Model age
5d ago
created 2026-11-16
Downloads over time
Now 0 → from 0 ↑ 0%
0011 0 on Nov 16 0 on Nov 22 Nov
Nov 16 → Nov 22 · 7 snapshots · spans 6 days

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.

Metadata

License
other
Tags
safetensors flare hyperflare qwen3.8 nvfp4 dflash2 abliterated base_model:incoai/Qwen3.8-Flash-Next-DFlash2 base_model:finetune:incoai/Qwen3.8-Flash-Next-DFlash2 license:other 8-bit region:us
Total size
15.6 MB
Files
6
Quantizations
1
Registered
2026-11-16 20:03
Last updated on HF
2026-11-16 20:18

Files by quantization

Auxiliary files 6 files 15.6 MB
model.safetensors 15.6 MB 53ab9538 download
LICENSE 31.1 KB b3fc4cf4 download
README.md 3.33 KB f255d22b download
NOTICE 2.29 KB c2bdd548 download
.gitattributes 1.48 KB a6344aac download
publisher.json 458 B 4fdeb641 download

README current version from Hugging Face


license: other
license_name: lil-license-1.0
license_link: LICENSE
base_model:

  • local-inference-lab/Qwen3.8-Flash-Next-NVFP4-MXFP8-CSF-QAD
  • lflare/Qwen3.8-Flash-Next-NVFP4-MXFP8-CSF-QAD-Abliterated
    tags:
  • flare
  • hyperflare
  • qwen3.8
  • nvfp4
  • abliterated

Qwen3.8-Flash-Next HyperFlare A16 Abliterated 1M

This model is based on Qwen3.8-Flash-Next-NVFP4-MXFP8-CSF-QAD by Local Inference Lab, Inc., a non-profit organization, available at https://huggingface.co/local-inference-lab/Qwen3.8-Flash-Next-NVFP4-MXFP8-CSF-QAD. Qwen3.8-Flash-Next-NVFP4-MXFP8-CSF-QAD is licensed under the Local Inference Lab License, Version 1.0.

A prepared Flare runtime core of the abliterated Qwen3.8-Flash-Next checkpoint, for serving with the HyperFlare engine image ghcr.io/raydius/hyperflare-inference.

What this is

Weights lflare/Qwen3.8-Flash-Next-NVFP4-MXFP8-CSF-QAD-Abliterated tag v3, built on Local Inference Lab revision dc5b8ab2; weight bytes unchanged
Hardware One NVIDIA SM120 GPU with 96 GB (RTX PRO 6000 Blackwell); the core reserves 73.12 GiB of device memory
Context 1,048,576 tokens
Concurrency 16 requests, one rank
Speculation MTP, widths 1–8
Publisher raydollete; the verification key is publisher.json in this repository

Run

The core runs only in Flare. Use the HyperFlare image; it carries the serving options this core was prepared and tested with.

docker volume create flare-home
docker run --rm -it --name flare \
  --gpus 'device=0,capabilities=compute' \
  --security-opt seccomp=unconfined \
  --ulimit memlock=-1:-1 --ipc host --shm-size 8g \
  --stop-timeout 60 \
  -p 127.0.0.1:8000:8000 \
  -v flare-home:/home/flare \
  ghcr.io/raydius/hyperflare-inference:0.2.5 \
  run raydollete/Qwen3.8-Flash-Next-HyperFlare-A16-Abliterated-1M \
  --gpus 0 --host 0.0.0.0 --port 8000

The first run downloads the core (about 99 GB) into the volume and asks you to accept the raydollete publisher once, which also accepts the license named in the core. The volume keeps the download and the acceptance across container restarts. seccomp=unconfined is required: the core reads its embedding tables through io_uring, which Docker's default seccomp profile blocks (Engram io_uring: Operation not permitted). The server exposes an OpenAI-compatible API on port 8000.

License

This repository is a re-upload of a Substantially Similar Copy under the Local Inference Lab License, Version 1.0, made with the permission of Local Inference Lab, Inc. That permission does not extend to you: you may download and run this core, but you may not re-upload it or any Substantially Similar Copy. Read LICENSE and NOTICE before use.

The Qwen3.8-Flash-Next base model is governed by the Qwen Community License 1.0. The license name recorded inside the core, which Flare shows at acceptance, is nvidia-open-model-agreement, carried from the source checkpoint's model card; this repository's LICENSE and NOTICE state the binding terms.

The abliteration is lflare's work; see that repository for its method and measurements.

README history 4 versions

The author's README evolved over time. Click a version to see its content at that point.

  1. 2026-11-16 Card: draft-token default and range d30a7f0 4.5 KB
    Loading...
  2. 2026-11-16 Card: DFlash2 drafter and its Apache-2.0 license; serving settings ec4dc1e 4.5 KB
    Loading...
  3. 2026-11-16 Run command: allow io_uring with seccomp=unconfined 0e386b2 3.3 KB
    Loading...
  4. 2026-11-16 Add signed Flare core, model card, and license files fd3b86c 3.1 KB
    Loading...
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 Abliteration" button that hands the model directly to the first-party desktop client, at the quantization your rig can actually run. No API keys, no subscription, no prompt leakage.

Open in Abliteration