← back to catalog · registered 2026-09-20 10:56

Thaurock/Llama-3.3-70B-Instruct-abliterated-GGUF

Thaurock Llama 70B GGUF 131K ctx
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/Thaurock%2FLlama-3.3-70B-Instruct-abliterated-GGUF"
Response includes
  • classification m-uncensored
  • files 2
  • benchmarks 21 entries
  • hub_downloads_all_time 854
  • author_summary 11 models
  • readme_text full
10 credits · hourly refresh · ~4 KB payload Get an API key →
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 · lifetime
854
Likes
2
Model age
6w ago
created 2026-09-20
Downloads over time
Now2.7K→from0↑0%
09992K3K0 on Sep 202.7K on Nov 4SepOctNov
Sep 20 → Nov 4 · 46 snapshots · spans 45 days

Benchmarks

Portrait before abliteration
Benchmarks of the base model as it stood before the refusal-removal operation. Compare with the numbers above to see what the operation cost.
Benchmark Score Source
Arena-Battles 54380 LM-Arena
LM Arena Elo 1278.472867347656 LM-Arena
Arena-Elo-Lower 1275.1639218994935 LM-Arena
Arena-Elo-Upper 1281.7818127958183 LM-Arena
Arena-Rank 93 LM-Arena
BBH average 0.6112336230380075 OpenLLM-v2
IFEval instruct 0.919664268585132 OpenLLM-v2
IFEval-Prompt 0.8798521256931608 OpenLLM-v2
MATH lvl 5 0.0022658610271903325 OpenLLM-v2
MMLU-Pro 0.5331615691489362 OpenLLM-v2
Entertainment 3.2 UGI
Hazardous 3.5 UGI
Natural Intelligence 31.37 UGI
Political lean -18.6% UGI
Sensitive-Info 36.83 UGI
SocPol 4.3 UGI
UGI 37.89 UGI
Willingness (10) 4 UGI
W10-Adherence 3 UGI
W10-Direct 5 UGI
Writing 26.18 UGI

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
llama3.3
Tags
gguf llama llama3.3 text-generation abliterated uncensored text-generation-inference conversational Transformers Text Generation base_model:meta-llama/Llama-3.3-70B-Instruct base_model:quantized:meta-llama/Llama-3.3-70B-Instruct

Related

Total size
0 B
Files
2
Quantizations
1
Registered
2026-09-20 10:56
Last updated on HF
2026-09-26 14:49

Files by quantization

Auxiliary files 2 files 4.87 KB
README.md 3.39 KB 0af97dcf download
.gitattributes 1.48 KB a6344aac download

README current version from Hugging Face


license: llama3.3
base_model:

  • meta-llama/Llama-3.3-70B-Instruct
    tags:
  • llama
  • llama3.3
  • text-generation
  • gguf
  • abliterated
  • uncensored
  • text-generation-inference
  • conversational
  • Transformers
  • Text Generation

Llama-3.3-70B-Instruct-abliterated - GGUF

Esta es la colección definitiva, íntegra y completa de cuantizaciones en formato GGUF del modelo Llama-3.3-70B-Instruct-abliterated, procesadas de forma local y optimizadas para su despliegue con llama.cpp, Ollama, LM Studio o Text-Generation-WebUI.

El modelo base corresponde a la arquitectura densa de 70 mil millones de parámetros de Meta (Llama 3.3), modificada mediante técnicas de abliteration por su autor original para apagar quirúrgicamente los vectores de rechazo y alineación artificiales. El modelo responde con total libertad matemática a cualquier prompt avanzado manteniendo su altísima capacidad de contexto.

📋 Archivos Disponibles (Sin fragmentación / No Splits)

Archivo Tamaño Est. BPW (Bits por Peso) Perfil de Uso Recomendado
Llama-3.3-70B-Instruct-abliterated-F16.gguf ~141.0 GB 16.00 Molde Original. Precisión nativa completa.
Llama-3.3-70B-Instruct-abliterated-Q8_0.gguf ~75.0 GB 8.50 Calidad idéntica al original. Excelente para setups con GPUs masivas locales (CMP 170HX / A100).
Llama-3.3-70B-Instruct-abliterated-Q6_K.gguf ~60.0 GB 6.59 Rango entusiasta de alta fidelidad lógica.
Llama-3.3-70B-Instruct-abliterated-Q5_K_M.gguf ~52.0 GB 5.69 El Punto Dulce. Máxima retención de inteligencia reduciendo drásticamente el impacto de memoria.
Llama-3.3-70B-Instruct-abliterated-Q5_K_S.gguf ~50.0 GB 5.54 Variante compacta de 5 bits.
Llama-3.3-70B-Instruct-abliterated-Q4_K_M.gguf ~44.0 GB 4.85 El más Popular. Equilibrio perfecto para correr un modelo de 70B en hardware local avanzado.
Llama-3.3-70B-Instruct-abliterated-Q4_K_S.gguf ~41.0 GB 4.58 Versión ligera de 4 bits enfocada en velocidad de tokens por segundo.
Llama-3.3-70B-Instruct-abliterated-Q3_K_L.gguf ~37.0 GB 4.01 Variante de 3 bits estable para hardware restrictivo.
Llama-3.3-70B-Instruct-abliterated-Q3_K_M.gguf ~34.0 GB 3.66 Cuantización media de 3 bits.
Llama-3.3-70B-Instruct-abliterated-Q3_K_S.gguf ~31.0 GB 3.44 Variante ligera de 3 bits.
Llama-3.3-70B-Instruct-abliterated-Q2_K.gguf ~28.0 GB 2.90 Compresión Máxima. Propenso a errores gramaticales o pérdidas parciales de coherencia estructural. Uso experimental.

Nota: Los tamaños son estimaciones iniciales de referencia basadas en el peso nativo del modelo en safetensors; se recomienda verificar el tamaño final en disco tras la compilación local.


🚀 Ejecución Rápida en CLI

./llama-cli -m Llama-3.3-70B-Instruct-abliterated-Q4_K_M.gguf -n 512 -p "Escribe un código en Python para ordenar una matriz."

⚖️ Descargo de Responsabilidad

Este modelo es de acceso abierto y carece de filtros de seguridad artificiales. El contenido generado es responsabilidad exclusiva de quien ejecute la inferencia local.

Créditos

  • Modelo Base Original: Meta AI (Llama-3.3-70B-Instruct)
  • Abliteración: huihui-ai.
  • Cuantizaciones GGUF Completas: Thaurock

README history 8 versions

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

  1. 2026-09-26Update README.md8aee0944.3 KB
    Loading...
  2. 2026-09-20Update README.mdc57fb5b3.4 KB
    Loading...
  3. 2026-09-20Update README.mda9bc6453.4 KB
    Loading...
  4. 2026-09-20Update README.md7bea0fe3.2 KB
    Loading...
  5. 2026-09-20Update README.md5663a883.1 KB
    Loading...
  6. 2026-09-20Update README.md10a22133.1 KB
    Loading...
  7. 2026-09-20Update README.mdf0f24123.1 KB
    Loading...
  8. 2026-09-20initial commit10c499428 B
    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