Ministral 3 Reasoning Heretic: LLMs without filters

The open-source community has received a new series of large language models (LLMs): Ministral 3 Reasoning Heretic. These models are distinguished by the absence of censorship, offering developers greater freedom in exploring their capabilities.

Coder3101 has curated a collection of these models, available on Hugging Face at https://huggingface.co/collections/coder3101/ministral-3-reasoning-heretic.

Key Features

  • Uncensored: The models have been modified to remove typical restrictions, allowing for broader responses.
  • Vision capabilities: Equipped with vision functionality thanks to the integration of MMPROJ.
  • Quantization: Available in different quantized versions (Q4, Q5, Q8, BF16) to optimize performance on different hardware.
  • Variants: Models with 4 billion, 8 billion and 14 billion parameters are available.

Quantization is a process that reduces the precision of a model's weights, allowing for reduced memory usage and improved inference speed. This makes the models more accessible and usable on devices with limited resources.