qwen-test-ppo

This is a fail expriment, I can't obtain a reasoning model with qwen2.5_3B or my dataset.

How to use( Although it is a fail model)

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "jiangchengchengNLP/qwen-test-ppo"

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

prompt = "How many integers  $n>1$  are there such that  $n$  divides  $x^{13}-x$  for every positive integer  $x$ ? Show your work in <think> </think> tags. And return the final answer in <answer> </answer> tags."
messages = [
    {"role": "system", "content": "You are Qwen, created by Alibaba Cloud. You are a helpful assistant."},
    {"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)

generated_ids = model.generate(
    **model_inputs,
    max_new_tokens=1024,
    temperature=0.2
)
generated_ids = [
    output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]

response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]

more detail:

I mix countdown,K_and_K,S1 and the dataset which from Open resoner zero, and find hardship samples from them, to create a 9K train data.But, train this model after 620 step, I find the model's solving ability betbeen countdown and other category of promblem is much confict. It make model can't find a best solution to solve those problem.