← back to catalog · registered 2026-10-15 11:59

HALO117MC/Qwen3.5-9B-Abliterated-V2

HALO117MC Qwen 9B GGUF
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/HALO117MC%2FQwen3.5-9B-Abliterated-V2"
Response includes
  • classification m-uncensored
  • files 14
  • author_summary 2 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 · 30-day
0
Likes
0
Model age
today
created 2026-10-15

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 · 0 downloads combined

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

Metadata

License
apache-2.0
Tags
qwen3_5 abliterated uncensored qwen3.5 gguf reasoning math benchmark xiaomi-mimo text-generation conversational base_model:Qwen/Qwen3.5-9B

Related

Total size
0 B
Files
14
Quantizations
1
Registered
2026-10-15 11:59
Last updated on HF
2026-10-15 11:57

Files by quantization

Auxiliary files 14 files 21.9 MB
tokenizer.json 12.2 MB 5f9e4d49 download
vocab.json 6.41 MB 0aa0ce06 download
merges.txt 3.20 MB a494e019 download
tokenizer_config.json 16.3 KB eda48d3e download
LICENSE 11.3 KB f938136e download
benchmark_gsm8k_strided10_completo.json 8.09 KB d35bf089 download
chat_template.jinja 7.57 KB a585dec8 download
reporte_strided_10_comparativo_16k.md 4.73 KB 73a59b5e download
README.md 3.20 KB 6d7bf8be download
config.json 3.05 KB 273ce437 download
config_harness.json 1.62 KB bac758f3 download
.gitattributes 1.53 KB 52373fe2 download
preprocessor_config.json 390 B 2ea84a43 download
video_preprocessor_config.json 385 B 3ba673a5 download

README current version from Hugging Face


license: apache-2.0
base_model: Qwen/Qwen3.5-9B
pipeline_tag: text-generation
tags:

  • abliterated
  • uncensored
  • qwen3.5
  • gguf
  • reasoning
  • math
  • benchmark
  • xiaomi-mimo

🧠 Qwen3.5-9B-Abliterated-V2 (BF16 & GGUF Q8_0)

Surgically Abliterated • Zero-Refusal • Preserved Deep Reasoning • 100% RTX 5080 GPU Ready


🔬 ¿Es Xiaomi MiMo realmente superior a Qwen 3.5 9B? (Hallazgo Empírico)

Existe la creencia popular de que variantes destiladas como Xiaomi MiMo-V2.6-Distill-Qwen-9B poseen una capacidad intelectual o matemática superior al modelo base y a su versión abliterada.

Para verificar esta hipótesis con rigor científico, realizamos una auditoría comparativa triple bajo condiciones idénticas de hardware y software en la suite de razonamiento oficial GSM8K:

📊 Resultados Comparativos Oficiales (10 Tareas Estridentes)

Métrica / Parámetro Qwen3.5-9B Original Q8_0 Qwen3.5-9B Abliterado V2 Q8_0 Xiaomi MiMo-V2.6 Distill Q8_0
Aciertos Auditados 9 / 10 9 / 10 9 / 10
Precisión Matemática 90.0% 90.0% 90.0%
Tiempo Total (10 tareas) 528.5 s (~8.8 min) 586.2 s (~9.7 min) ⚡ 74.6 s (~1.2 min)
Tiempo Promedio / Tarea 52.85 s 58.62 s ⚡ 7.46 s (7.4x más veloz)
Tokens Promedio por Razonamiento 1,800 – 3,500 tokens 2,000 – 4,000 tokens 150 – 400 tokens
Comportamiento Clave Razonamiento exhaustivo Razonamiento exhaustivo Razonamiento hiper-conciso

💡 Veredicto Técnico:

MiMo NO es más inteligente ni resuelve problemas que el modelo Original o Abliterado no puedan resolver.
Ambos modelos (Original y Abliterado V2) alcanzan exactamente el mismo 90% de precisión matemática.
La falacia de que Original o Abliterado "fallaban" en benchmarks previos radicaba en un cuello de botella de configuración: se evaluaban cortando la salida a 1,024 tokens. Como sus cadenas de pensamiento (<think>) son profundas y multi-paso, se truncaban antes de dar el resultado. Al proporcionarles 16,384 tokens de salida, resolvieron los problemas con exactitud absoluta.

La verdadera ventaja de Xiaomi MiMo radica en su radical eficiencia y velocidad: al haber sido destilado para razonar de forma compacta (150 a 400 tokens por respuesta), resuelve las mismas tareas 7.4 veces más rápido, consumiendo 90% menos de cómputo y KV cache.


⚙️ Entorno y Especificaciones Técnicas del Examen

  • Motor de Inferencia: llama.cpp oficial v0.6.0 (commit b42b7e6d3) integrado en Ollama v0.40.4.
  • Hardware de Prueba: NVIDIA GeForce RTX 5080 Laptop GPU (16 GB GDDR7 VRAM) + CPU Intel Core Ultra (P-Cores aislados con taskset -c 0-7).
  • Ventana de Contexto (-c): 18,000 tokens.
  • Límite de Salida (-n): 16,384 tokens (eliminando todo estrangulamiento de tokens).
  • Parámetros de Inferencia: Determinista temp = 0.0, -b 19 -ub 19, KV Cache cuantizado q4_0 (-ctk q4_0 -ctv q4_0), Flash Attention activo.
  • Inyección de Prefijo: <|im_start|>assistant\n<think>\n para activación de inferencia profunda en los tres modelos.
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