← back to catalog · registered 2026-08-22 13:56

jpalmae/qwen35b-espanol-abliterated-gguf

jpalmae Qwen GGUF MoE second-order 262K 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/jpalmae%2Fqwen35b-espanol-abliterated-gguf"
Response includes
  • classification m8
  • files 4
  • hub_downloads_all_time 331
  • author_summary 1 models
  • readme_text full
10 credits · hourly refresh · ~4 KB payload Get an API key →
Abliteration classifier · v1.0.0
M8
Primary method

Repackaging (quantization)

Applied on top of direct removal inherited from the base model.
Confidence
MEDIUM
Inherited from base model
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=1
  • assume M1 (base ablation) + M8 (GGUF quant) - default when producer unknown
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
331
66 last 30d - stable
Likes
0
Model age
2mo ago
created 2026-07-18
Downloads over time
Now346→from10↑3,360%
012725338010 on Jul 15346 on Oct 11346 on Oct 10JulAugSepOct
Jul 15 → Oct 11 · 53 snapshots · spans 88 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
mit
Quantizations
F16 Q4_K
Tags
llama.cpp gguf f16 q4_k_m qwen3.5 qwen3.6 spanish espanol chile fine-tuned moe abliterated
Total size
84.3 GB
Files
4
Quantizations
3
Registered
2026-08-22 13:56
Last updated on HF
2026-07-18 02:50

Files by quantization

F16 1 file 64.6 GB
model_F16_patched.gguf 64.6 GB 43ede246 download
Q4_K 1 file 19.7 GB
model_Q4_K_M.gguf 19.7 GB 08c70ee1 download
Auxiliary files 2 files 3.22 KB
README.md 1.63 KB ea22f71f download
.gitattributes 1.59 KB de5c5121 download

README current version from Hugging Face


license: mit
library_name: llama.cpp
tags:

  • gguf
  • f16
  • q4_k_m
  • qwen3.5
  • qwen3.6
  • spanish
  • espanol
  • chile
  • fine-tuned
  • moe
  • abliterated
  • uncensored
    base_model: mlabonne/Huihui-Qwen3.6-35B-A3B-abliterated
    pipeline_tag: text-generation

Qwen3.6-35B-A3B Abliterated — Español Neutro/Chileno 🇨🇱 (GGUF)

Fine-tune de Huihui-Qwen3.6-35B-A3B-abliterated (MoE 35B/3B activos) en GGUF — combina uncensored (abliterated) + español neutro/chileno (sin sesgo rioplatense).

Archivos

Archivo Tamaño Uso
model_Q4_K_M.gguf 20GB Consumer: RTX 3090/4090 24GB (~169 tok/s)
model_F16_patched.gguf 65GB Precisión completa / re-cuantizar / A100·H100·RTX PRO 6000

⚠️ Parche MTP aplicado

Ambos GGUF llevan el parche binario que corrige la metadata MTP del conversor (block_count 41→40, nextn_predict_layers 1→0). Sin parche llama.cpp falla con missing tensor 'blk.39.nextn.eh_proj.weight'; con parche carga perfecto.

Uso con llama.cpp

# Consumer 24GB
llama-server -m model_Q4_K_M.gguf --alias qwen35b-espanol-abl -c 4096 -ngl 99 --port 8080

# GPU grande (F16)
llama-server -m model_F16_patched.gguf --alias qwen35b-espanol-abl -c 8192 -ngl 99 --port 8080

Entrenamiento

LoRA r=64 sobre Huihui-Qwen3.6-35B-A3B-abliterated, 1200 pasos, dataset 200k (C4 .cl + neutro, filtro anti-voseo). Detalles en jpalmae/qwen35b-espanol-abliterated.

Aviso

"Abliterated/uncensored": el modelo no rehúsa por defecto. Úsalo responsablemente.

Licencia

MIT.

README history 1 version

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

  1. 2026-07-18Upload README.md with huggingface_hub550b27b1.6 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