Mistral text-generation base
The captured configuration identifies MistralForCausalLM with model type mistral and Transformers support.
Open Source Model Profile · saidutta69
Mistral-Nemo-Instruct-heretic
Mistral-Nemo-Instruct-heretic is a 12.25B-parameter Mistral text-generation variant from saidutta69. According to the model card, it applies Heretic refusal ablation to Mistral-Nemo-Instruct-2407.
- Publisher
- saidutta69
- Task
- text-generation
- Model type
- mistral
- License
- apache-2.0
- Library
- transformers
- Publication status
- Accepted · not indexed
Model overview
Mistral-Nemo-Instruct-heretic is published by saidutta69 as a text-generation model. The captured configuration identifies MistralForCausalLM, and Safetensors metadata reports about 12.25B parameters. According to the model card, it is a decensored Mistral-Nemo-Instruct-2407 variant made with Heretic ablation.
Recorded capabilities
Abliteration editing
According to the model card, Heretic v1.4.0 edits refusal-related weight directions in attention and MLP projections instead of fine-tuning.
Full GGUF ladder
According to the model card, the release includes BF16 weights and GGUF Q4_K_M, Q5_K_M, Q6_K, and Q8_0 files produced with llama.cpp.
Local runtime documentation
According to the model card, the publisher documents Transformers chat-template use and lists Ollama, LM Studio, Jan, vLLM, and SGLang as runtimes.
Explicit safety caveat
According to the model card, refusal suppression is deliberate with no added safety filtering, so deployment responsibility rests with the user.
Use cases in the source record
- Direct instruction-following experiments where the publisher describes answering without base-model refusal.
- Local deployment tests using the documented GGUF quants and Transformers, llama.cpp, Ollama, LM Studio, Jan, vLLM, or SGLang paths.
Limitations and unknowns
- No evaluation results were extracted from this record.
- No context-window value was extracted from this record.
- According to the model card, the model may comply with requests the base model would refuse, so unmoderated public serving is discouraged by the publisher.
- Provider state is historical snapshot data and should be refreshed before being presented as current.
Source and provenance
Source: saidutta69/Mistral-Nemo-Instruct-heretic
Captured: Unknown. Processed: 2026-09-07T19:36:10.728648+00:00.
Mistral-Nemo-Instruct-heretic A decensored variant of mistralai/Mistral-Nemo-Instruct-2407 , produced with Heretic v1.4.0 (directional ablation / "abliteration"). a 12B model co-developed by Mistral AI and NVIDIA — strong reasoning and long-context now uncensored. Refusal behavior is suppressed via targeted weight edits to the attention output and MLP down-projections rather than fine-tuning, so the base model's knowledge and capabilities are left largely intact. Who this is for: developers who want a 12B multilingual model that answers directly instead of refusing — for complex reasoning tasks, long-context applications, or any use…
F001F002F003F004F005F006F007F008F009F010F011F012F013F015F016F017F018