Qwen2.5-3B-R1-Finance

This is a toy model using CoT-sft with GRPO.

Usage

tokenizer = AutoTokenizer.from_pretrained("yixuantt/Qwen2.5-3B-R1-Finance")

model = AutoModelForCausalLM.from_pretrained("yixuantt/Qwen2.5-3B-R1-Finance",
                                             torch_dtype = torch.bfloat16,
                                             device_map = "auto"
                                             )
model.eval()

print(model)
def generate(text):
    conv = [{"role": "system",
             "content": "You are a helpful AI Assistant that provides well-reasoned and detailed responses. You first think about the reasoning process as an internal monologue and then provide the user with the answer."},{"role": "user", "content": text}]
    prompt = tokenizer.apply_chat_template(conversation=conv, tokenize=False, add_generation_prompt=True)
    encoded = tokenizer(prompt, return_tensors="pt")
    generate_params = dict(
        max_new_tokens=1024,
        do_sample=True,
        top_k=20,
    )
    with torch.no_grad():
        generation_output = model.generate(input_ids=encoded.input_ids.cuda(),
                                           attention_mask=encoded.attention_mask.cuda(),
                                           tokenizer=tokenizer,
                                           **generate_params)

    generation_output = generation_output[:, encoded.input_ids.shape[1]:]
    out = tokenizer.decode(generation_output[0], skip_special_tokens=True)
    # print(out)
    return out