whatsapp-twin-lora

LoRA adapter for mlx-community/Qwen3-4B-Instruct-2507-4bit trained to reply as "Miska" in a WhatsApp-style chat. SFT on ~992 context→reply examples from a private chat export (MLX LoRA, rank-8 on 16 layers, mask_prompt, seq 640, ~600 iters, M4).

Usage (MLX)

from mlx_lm import load, generate
model, tok = load("mlx-community/Qwen3-4B-Instruct-2507-4bit",
                  adapter_path="<this repo dir>")
prompt = tok.apply_chat_template(
    [{"role": "system", "content": "You are Miska replying in a WhatsApp chat..."},
     {"role": "user", "content": "Aatu: moi"}],
    tokenize=False, add_generation_prompt=True)
print(generate(model, tok, prompt=prompt, max_tokens=80))

Fuse into a standalone model:

python -m mlx_lm fuse --model mlx-community/Qwen3-4B-Instruct-2507-4bit \
    --adapter-path <this repo dir> --save-path miska-twin-4b

Private repo — chat training data is not included.