QFT-OLMO3-32B-epoch-3

This repository contains a mathematical reasoning language model in Hugging Face format.

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "qingyangzhang/QFT-OLMO3-32B-epoch-3"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype="auto",
    device_map="auto",
)

messages = [
    {
        "role": "user",
        "content": (
            "Solve the problem.\n"
            "Please reason step by step, and put your final answer within \\boxed{}."
        ),
    }
]

inputs = tokenizer.apply_chat_template(
    messages,
    add_generation_prompt=True,
    return_tensors="pt",
).to(model.device)

outputs = model.generate(
    inputs,
    max_new_tokens=32768,
    do_sample=True,
    temperature=0.6,
    top_p=0.95,
    top_k=20,
    eos_token_id=[151643, 151645],
)

print(tokenizer.decode(outputs[0][inputs.shape[-1]:], skip_special_tokens=False))

Model configuration

• Maximum context length: 49,152 tokens
• EOS token: <|endoftext|> (151643)
• PAD token: <|endoftext|> (151643)
• The bundled chat template should be used for prompt rendering.