nkmry/llmjp-13b-comp

Model Card

Description

• Developed by: nkmry
• Languages: Japanese, English
• License: Apache-2.0
• Finetuned from model: llm-jp/llm-jp-3-13b

Uses

Use the code below to get started with the model.

from peft import PeftModel
from unsloth import FastLanguageModel

BASE_MODEL_ID = "llm-jp/llm-jp-3-13b"
ADAPTER_ID = "nkmry/llmjp-13b-comp"

# Get the base model

base_model, tokenizer = FastLanguageModel.from_pretrained(
    model_name=BASE_MODEL_ID,
    dtype=None,
    load_in_4bit=True,
    trust_remote_code=True,
)

# Apply the LoRA adapter

model = PeftModel.from_pretrained(
    base_model,
    model_id=ADAPTER_ID,
    is_trainable=False
)

# Generate outputs

datasets = []
with open("./elyza-tasks-100-TV_0.jsonl", "r") as f:
    item = ""
    for line in f:
      line = line.strip()
      item += line
      if item.endswith("}"):
        datasets.append(json.loads(item))
        item = ""

PROMPT = """### 指示
{input}
### 回答
"""

results = []
FastLanguageModel.for_inference(model)
for data in tqdm(datasets):
    input = data["input"]
    tokenized_input = tokenizer.encode(
        PROMPT.format(input=input), add_special_tokens=False, return_tensors="pt").to(model.device)
    attention_mask = torch.ones_like(tokenized_input)
    with torch.no_grad():
        outputs = model.generate(
            tokenized_input,
            attention_mask=attention_mask,
            max_new_tokens=512,
            do_sample=False,
            repetition_penalty=1.2,
            pad_token_id=tokenizer.eos_token_id
        )[0]
    output = tokenizer.decode(outputs[tokenized_input.size(1):], skip_special_tokens=True)
    results.append({"task_id": data["task_id"], "input": input, "output": output})

with open(f"./outputs.jsonl", 'w', encoding='utf-8') as f:
    for result in results:
        json.dump(result, f, ensure_ascii=False)
        f.write('\n')

Details

Training Data

• DeL-TaiseiOzaki/Tengentoppa-sft-v1.0
• elyza/ELYZA-tasks-100

Framework versions

• PyTorch 2.5.1
• Transformers 4.46.3
• Unsloth 2024.12.4
• PEFT 0.14.0
• xformers 0.0.2