gemma2-9b-sft

base_model: gemma2_9b gemma2 27bにより生成したself instructデータ10kによりinstrution tuningを実施

Eval

elyza-task-100

Use

model = AutoModelForCausalLM.from_pretrained(
            model_name,
            torch_dtype=torch.bfloat16,
            device_map="auto"  # GPU自動割り当て
        )

tokenizer = AutoTokenizer.from_pretrained(model_name)

messages_list = [
  [
    {"role": "user", "content": "仕事の熱意を取り戻すためのアイデアを5つ挙げてください。"}
  ]
]

prompts = [self.tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True) for messages in messages_list]
        
inputs = self.tokenizer(prompts, return_tensors="pt", padding=True).to(self.model.device)

outputs = self.model.generate(
    **inputs,
    temperature=self.generate_configs["temperature"],
    max_new_tokens=self.generate_configs["max_new_tokens"],
    top_p=self.generate_configs["top_p"],
    top_k=self.generate_configs["top_k"],
    repetition_penalty=self.generate_configs["repetition_penalty"],
    pad_token_id=self.tokenizer.pad_token_id,
    eos_token_id=self.tokenizer.eos_token_id,
)

print(outputs)