AlejandroOlmedo/zeta-8bit-mlx

About:

Tuned from Qwen2.5 coder for coding tasks

• Its a fine-tuned version of Qwen2.5-Coder-7B to support edit prediction in Zed. Fine-tuned using zeta dataset.

Special thanks to the folks at Zed Industries for fine-tuning this version of Qwen2.5-Coder-7B. More information about the model can be found here:

https://huggingface.co/zed-industries/zeta (Base Model)

https://huggingface.co/lmstudio-community/zeta-GGUF (GGUF Version)

• Converted it to MLX format (using mlx-lm version 0.20.5.) with a quantization of 8-bit for better performance on Apple Silicon Macs (M1,M2,M3,M4 Chips).
• If looking for a larger or smaller (quantized) mlx model, see the models below.

Other Types:

Link	Type	Size	Notes
[MLX] (https://huggingface.co/AlejandroOlmedo/zeta-mlx)	Full	15.2 GB	Best Quality
[MLX] (https://huggingface.co/AlejandroOlmedo/zeta-8bit-mlx)	8-bit	8.10 GB	Better Quality
[MLX] (https://huggingface.co/AlejandroOlmedo/zeta-4bit-mlx)	4-bit	4.30 GB	Good Quality

AlejandroOlmedo/zeta-8bit-mlx

The Model AlejandroOlmedo/zeta-8bit-mlx was converted to MLX format from zed-industries/zeta using mlx-lm version 0.20.5.

Use with mlx

pip install mlx-lm

from mlx_lm import load, generate

model, tokenizer = load("AlejandroOlmedo/zeta-8bit-mlx")

prompt="hello"

if hasattr(tokenizer, "apply_chat_template") and tokenizer.chat_template is not None:
    messages = [{"role": "user", "content": prompt}]
    prompt = tokenizer.apply_chat_template(
        messages, tokenize=False, add_generation_prompt=True
    )

response = generate(model, tokenizer, prompt=prompt, verbose=True)