← back to catalog · registered 2026-11-19 20:04

miloserdev/Llama-3.1-8B-Instruct-abliterated-numallama

miloserdev Llama 8B second-order
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/miloserdev%2FLlama-3.1-8B-Instruct-abliterated-numallama"
Response includes
  • classification m1
  • files 6
  • benchmarks 5 entries
  • author_summary 5 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
0
Model age
2d ago
created 2026-11-19
Downloads over time
Now 0 → from 0 ↑ 0%
0011 0 on Nov 19 0 on Nov 22 Nov
Nov 19 → Nov 22 · 4 snapshots · spans 3 days

Benchmarks

Benchmark Score Source
BBH average 0.4379296925671293 OpenLLM-v2
IFEval instruct 0.7757793764988009 OpenLLM-v2
IFEval-Prompt 0.6931608133086876 OpenLLM-v2
MATH lvl 5 0.06419939577039276 OpenLLM-v2
MMLU-Pro 0.3503158244680851 OpenLLM-v2

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
llama3.1
Tags
numallama cpu arm64 aarch64 numa text-generation abliterated uncensored base_model:mlabonne/Meta-Llama-3.1-8B-Instruct-abliterated base_model:quantized:mlabonne/Meta-Llama-3.1-8B-Instruct-abliterated license:llama3.1 region:us

Related

Total size
13.1 GB
Files
6
Quantizations
1
Registered
2026-11-19 20:04
Last updated on HF
2026-11-19 20:00

Files by quantization

Auxiliary files 6 files 13.1 GB
llama-3.1-8b-instruct-abliterated-q8.bin 7.95 GB 4dbdaa54 download
llama-3.1-8b-instruct-abliterated-q4.bin 5.16 GB 8dfe1ab2 download
tokenizer.bin 1.78 MB 57187047 download
LICENSE 7.45 KB a7c3ca16 download
README.md 3.55 KB 27c34c2c download
.gitattributes 1.48 KB a6344aac download

README current version from Hugging Face


license: llama3.1
base_model:

  • mlabonne/Meta-Llama-3.1-8B-Instruct-abliterated
    base_model_relation: quantized
    pipeline_tag: text-generation
    library_name: numallama
    tags:
  • numallama
  • cpu
  • arm64
  • aarch64
  • numa
  • text-generation
  • abliterated
  • uncensored

Llama-3.1-8B-Instruct-abliterated-numallama

Built with Llama

mlabonne/Meta-Llama-3.1-8B-Instruct-abliterated (an abliterated version of meta-llama/Llama-3.1-8B-Instruct), converted for numallama: a C inference engine for CPU servers, with NUMA-aware execution across all sockets of multi-socket machines.

This is an abliterated model: the refusal behaviour of the original was removed by mlabonne/Meta-Llama-3.1-8B-Instruct-abliterated. It will follow requests that the original refuses. You are responsible for how you use it and must still follow the license and its acceptable use policy. For the original behaviour use Llama-3.1-8B-Instruct-numallama.

These files are not GGUF. They run only with numallama (they do not work with llama.cpp, Ollama, LM Studio or transformers).

Files

File Weights Size Decode speed*
llama-3.1-8b-instruct-abliterated-q4.bin Q4 5.5 GB 57 tok/s
llama-3.1-8b-instruct-abliterated-q8.bin Q8 8.5 GB 40 tok/s
tokenizer.bin tokenizer 1.9 MB

Q4 is asymmetric 4-bit (groups of 32, fp16 scale and min), quantized with an importance matrix collected by numallama on English, Russian, Chinese and code text. Q8 is int8 with per-group scales.
Q4 is faster and smaller; Q8 is closer to the original.

* tokens/s for a single chat, measured on 2 × Ampere Altra Max M128-30 (256 Neoverse N1 cores), both sockets. Other machines will differ.

Run

Linux. Optimized for ARM64 with NEON dot product (Ampere Altra, AWS Graviton, other Neoverse cores); other CPUs, including x86-64, run portable C code that is not optimized.

git clone https://github.com/miloserdev/numallama && cd numallama && make
pip install huggingface_hub
hf download miloserdev/Llama-3.1-8B-Instruct-abliterated-numallama --local-dir models/Llama-3.1-8B-Instruct-abliterated-numallama

# chat in the terminal
./runq models/Llama-3.1-8B-Instruct-abliterated-numallama/llama-3.1-8b-instruct-abliterated-q4.bin -z models/Llama-3.1-8B-Instruct-abliterated-numallama/tokenizer.bin -m chat

# or an OpenAI-compatible server on http://127.0.0.1:8080/v1
./runq models/Llama-3.1-8B-Instruct-abliterated-numallama/llama-3.1-8b-instruct-abliterated-q4.bin -z models/Llama-3.1-8B-Instruct-abliterated-numallama/tokenizer.bin -m server -P 8080

See the numallama README for all options (context length, sampling, tools, speculative decoding, API details).

License

Llama 3.1 is licensed under the Llama 3.1 Community License, Copyright © Meta Platforms, Inc. All Rights Reserved. Use of this model must comply with the Llama 3.1 Community License and the Acceptable Use Policy.

Original model: meta-llama/Llama-3.1-8B-Instruct. Abliteration: mlabonne/Meta-Llama-3.1-8B-Instruct-abliterated. Conversion: numallama.

README history 1 version

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

  1. 2026-11-19 Add numallama model files 5b9f843 3.5 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