metadata

base_model:
  - llm-jp/llm-jp-3-13b
tags:
  - text-generation-inference
  - transformers
  - unsloth
  - llama
  - trl
license: apache-2.0
language:
  - en
  - ja

Uploaded model

Developed by: ikedachin
License: apache-2.0
Finetuned from model : llm-jp/llm-jp-3-13b

This llama model was trained 2x faster with Unsloth and Huggingface's TRL library.

学習データ

使用したSupervised fien-tune用dataset：下記からランダムに20000データを抽出
DeL-TaiseiOzaki/Tengentoppa-sft-v1.0
🌾 ランダムに20000データを取り出して学習

ベースモデル

SFTに用いた継続事前学習モデル（ベースとしたモデル） ikedachin/llm-jp-3-13b-october-news-e1-all-3-5 このベースモデルのLoRAパラメータの一部を追加学習

追加学習前のベースモデル

llm-jp/llm-jp-3-13b

環境構築

pip uninstall unsloth -y
pip install --upgrade --no-cache-dir "unsloth[colab-new] @ git+https://github.com/unslothai/unsloth.git"
pip install --upgrade torch
pip install --upgrade xformers

実行コード

# import libraries
import re
import json

import torch

from peft import PeftModel
from tqdm import tqdm
from unsloth import FastLanguageModel

# define base model_id and peft model_id
model_id = "llm-jp/llm-jp-3-13b"
adapter_id = "ikedachin/llm-jp-3-13b-october-news-e1-all-3-5-sft-Ozaki-Magpie-20000-sorted-params"

dtype = None
load_in_4bit = True
max_seq_length = 1024

# down load base model
model, tokenizer = FastLanguageModel.from_pretrained(
    model_name=model_id,
    dtype=dtype,
    load_in_4bit=load_in_4bit,
    trust_remote_code=True,
)

# adapt peft model
model = PeftModel.from_pretrained(model, adapter_id, token = HF_TOKEN)


# prepare dataset elyza-tasks-100-TV_0.jsonl
datasets_elyza = []

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_elyza.append(json.loads(item))
            item = ""


# change mode for inference
FastLanguageModel.for_inference(model)

# inferrence
results = []

for dt in tqdm(datasets):
    input = dt["input"]

    prompt = f"""### 指示\n{input}\n### 回答\n"""

    inputs = tokenizer([prompt], return_tensors = "pt").to(model.device)

    outputs = model.generate(**inputs, max_new_tokens = max_seq_length, use_cache = True, do_sample=False, repetition_penalty=1.2)
    prediction = tokenizer.decode(outputs[0], skip_special_tokens=True).split('\n### 回答')[-1]

    results.append({"task_id": dt["task_id"], "input": input, "output": prediction})

# create result file as jsonl type
json_file_id = re.sub(".*/", adapter_id)
with open(f"/content/{json_file_id}_output.jsonl", 'w', encoding='utf-8') as f:
    for result in results:
        json.dump(result, f, ensure_ascii=False)
        f.write('\n')