Naotaka/gemma-2-9b-r16-768-512-it_lora

Uploaded model

• Developed by: Naotaka
• License: apache-2.0
• Finetuned from model : google/gemma-2-9b

This gemma2 model was trained 2x faster with Unsloth and Huggingface's TRL library.

Usage

Execute following code in Google Colab

!pip install -U pip --quiet

######### to avoid using unsloth model
!pip uninstall unsloth -y --quiet
!pip install -q --upgrade --no-cache-dir "unsloth[colab-new] @ git+https://github.com/niryuu/unsloth.git@use-exact-model-name"
######### /to avoid using unsloth model

!pip install --upgrade torch --quiet
!pip install --upgrade xformers --quiet
!pip install -U peft --quiet
!pip install -U openai --quiet
!pip install -U transformers --quiet
!pip install -U bitsandbytes --quiet
!pip install -U accelerate --quiet
!pip install -U datasets --quiet
!pip install -U peft --quiet
!pip install -U trl --quiet

# Install Flash Attention 2 for softcapping support
import torch
if torch.cuda.get_device_capability()[0] >= 8:
    !pip install --no-deps packaging ninja einops "flash-attn>=2.6.3" --quiet

# Hugging Face Token
from google.colab import userdata

HF_TOKEN = userdata.get('HF_TOKEN')

import torch
from unsloth import FastLanguageModel

dtype = None # Noneにしておけば自動で設定
load_in_4bit = True # 今回は13Bモデルを扱うためTrue

# model_size = '27b'
# USE_OZAKI_DATA = True
# DATA_SAMPLING_RATE = 0.05


lr = 2e-4
per_device_train_batch_size = 8
lora_r = 16
lora_alpha = 32
max_seq_length = 768 # unslothではRoPEをサポートしているのでコンテキスト長は自由に設定可能
max_seq_length_output = 512

####################################### Gemma2
USE_OZAKI_DATA = True
DATA_SAMPLING_RATE = 0.2
model_size = '9b'
model_id = f"google/gemma-2-{model_size}"
new_model_id = f"gemma-2-{model_size}-r{lora_r}-{max_seq_length}-{max_seq_length_output}-it"
lora_model_id = new_model_id+"_lora"
adapter_id = f"Naotaka/{lora_model_id}"

# FastLanguageModel インスタンスを作成
model, tokenizer = FastLanguageModel.from_pretrained(
    model_name=model_id,
    dtype=dtype,
    load_in_4bit=load_in_4bit,
    trust_remote_code=True,
)

# LoRAアダプタを適用
# lora_model_idからLoRAアダプタを読み込みモデルにマージ
from peft import PeftModel

model = PeftModel.from_pretrained(
    model,
    adapter_id,
    token=HF_TOKEN
)

# ELYZA-tasks-100-TVの読み込み。事前にファイルをアップロードしてください
# データセットの読み込み。
# omnicampusの開発環境では、左にタスクのjsonlをドラッグアンドドロップしてから実行。
import json
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 = ""

# 学習したモデルを用いてタスクを実行
from tqdm import tqdm

# 推論するためにモデルのモードを変更
FastLanguageModel.for_inference(model)

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 = 368, 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})

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