Quantization made by Richard Erkhov.

Llama-3.1-Tulu-3-8B-mlx - GGUF

Model creator: https://huggingface.co/cmcmaster/
Original model: https://huggingface.co/cmcmaster/Llama-3.1-Tulu-3-8B-mlx/

Name	Quant method	Size
Llama-3.1-Tulu-3-8B-mlx.Q2_K.gguf	Q2_K	2.96GB
Llama-3.1-Tulu-3-8B-mlx.IQ3_XS.gguf	IQ3_XS	3.28GB
Llama-3.1-Tulu-3-8B-mlx.IQ3_S.gguf	IQ3_S	3.43GB
Llama-3.1-Tulu-3-8B-mlx.Q3_K_S.gguf	Q3_K_S	3.41GB
Llama-3.1-Tulu-3-8B-mlx.IQ3_M.gguf	IQ3_M	3.52GB
Llama-3.1-Tulu-3-8B-mlx.Q3_K.gguf	Q3_K	3.74GB
Llama-3.1-Tulu-3-8B-mlx.Q3_K_M.gguf	Q3_K_M	3.74GB
Llama-3.1-Tulu-3-8B-mlx.Q3_K_L.gguf	Q3_K_L	4.03GB
Llama-3.1-Tulu-3-8B-mlx.IQ4_XS.gguf	IQ4_XS	4.18GB
Llama-3.1-Tulu-3-8B-mlx.Q4_0.gguf	Q4_0	4.34GB
Llama-3.1-Tulu-3-8B-mlx.IQ4_NL.gguf	IQ4_NL	4.38GB
Llama-3.1-Tulu-3-8B-mlx.Q4_K_S.gguf	Q4_K_S	4.37GB
Llama-3.1-Tulu-3-8B-mlx.Q4_K.gguf	Q4_K	4.58GB
Llama-3.1-Tulu-3-8B-mlx.Q4_K_M.gguf	Q4_K_M	4.58GB
Llama-3.1-Tulu-3-8B-mlx.Q4_1.gguf	Q4_1	4.78GB
Llama-3.1-Tulu-3-8B-mlx.Q5_0.gguf	Q5_0	5.21GB
Llama-3.1-Tulu-3-8B-mlx.Q5_K_S.gguf	Q5_K_S	5.21GB
Llama-3.1-Tulu-3-8B-mlx.Q5_K.gguf	Q5_K	5.34GB
Llama-3.1-Tulu-3-8B-mlx.Q5_K_M.gguf	Q5_K_M	5.34GB
Llama-3.1-Tulu-3-8B-mlx.Q5_1.gguf	Q5_1	5.65GB
Llama-3.1-Tulu-3-8B-mlx.Q6_K.gguf	Q6_K	6.14GB
Llama-3.1-Tulu-3-8B-mlx.Q8_0.gguf	Q8_0	7.95GB

Original model description:

license: llama3.1 language: - en pipeline_tag: text-generation datasets: - allenai/RLVR-GSM-MATH-IF-Mixed-Constraints base_model: allenai/Llama-3.1-Tulu-3-8B library_name: transformers tags: - mlx

cmcmaster/Llama-3.1-Tulu-3-8B-mlx

The Model cmcmaster/Llama-3.1-Tulu-3-8B-mlx was converted to MLX format from allenai/Llama-3.1-Tulu-3-8B using mlx-lm version 0.19.2.

Use with mlx

pip install mlx-lm

from mlx_lm import load, generate

model, tokenizer = load("cmcmaster/Llama-3.1-Tulu-3-8B-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)