license: other
license_name: custom-license
license_link: https://huggingface.co/nicoboss/MedraN-E4B-Uncensored-EP7/blob/main/LICENSE.txt
base_model: nicoboss/MedraN-E4B-Uncensored-EP7
tags:
- mlx
- medical
- uncensored
- quantized
Medra3n-E4B-Uncensored-MLX-Quantized
This repository contains quantized MLX-optimized versions of nicoboss/MedraN-E4B-Uncensored-EP7, converted for use with Apple Silicon devices using the MLX framework.
Model Description
MedraN (Medical Reasoning and Analysis) is a specialized language model fine-tuned for medical applications. This E4B (Episode 4B) variant is an uncensored version that provides comprehensive medical information without content restrictions.
Available Quantizations
This repository includes two quantized versions optimized for different use cases:
Q6 Version (6-bit quantization)
- Size: ~5.2GB
- Quality: High quality with minimal degradation
- Use case: Best balance between size and performance
- Actual quantization: 6.501 bits per weight
Q4 Version (4-bit quantization)
- Size: ~3.6GB
- Quality: Good quality with some degradation
- Use case: Maximum speed and memory efficiency
- Actual quantization: 4.501 bits per weight
Usage
These models are optimized for use with the MLX framework on Apple Silicon devices. You can use them with:
Q6 Version:
from mlx_lm import load, generate
model, tokenizer = load("drwlf/MedraN-E4B-Uncensored-MLX-Quantized", model_path="q6")
response = generate(model, tokenizer, "What are the symptoms of...", max_tokens=512)
Q4 Version:
from mlx_lm import load, generate
model, tokenizer = load("drwlf/MedraN-E4B-Uncensored-MLX-Quantized", model_path="q4")
response = generate(model, tokenizer, "What are the symptoms of...", max_tokens=512)
Model Comparison
| Version | Size | Quality | Speed | Memory Usage |
|---|---|---|---|---|
| Q6 | 5.2GB | High | Good | Medium |
| Q4 | 3.6GB | Good | Fast | Low |
| Full | 13GB | Best | Slow | High |
Original Model
This is a conversion of the original model available at: https://huggingface.co/nicoboss/MedraN-E4B-Uncensored-EP7
Full precision MLX version: https://huggingface.co/drwlf/MedraN-E4B-Uncensored-MLX
Conversion Details
- Framework: MLX
- Base precision: float16
- Quantization: 4-bit and 6-bit
- Optimized for: Apple Silicon (M1/M2/M3/M4 chips)
License
This model follows the same licensing terms as the original model. Please refer to the original model's license for usage terms.