84basi/llm-jp-3-13b-it-4.0

Uploaded model

• Developed by: 84basi
• 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.

Readme

token = "" # token
model_id = "llm-jp-3-13b-it-4.0" # llm-jp-3-13b-it-4.17, gemma-2-27b-it-4.19
model_name = "84basi/" + model_id
answer_json_file = "./elyza-tasks-100-TV_0.jsonl"
output_json_file = "./" + model_id + "_output.jsonl"

%%capture
!pip install unsloth -q
!pip uninstall unsloth -y && pip install --upgrade --no-cache-dir "unsloth[colab-new] @ git+https://github.com/unslothai/unsloth.git" -q

from unsloth import FastLanguageModel
from peft import PeftModel
import torch
import json

max_seq_length = 2048
dtype = None
load_in_4bit = True

model, tokenizer = FastLanguageModel.from_pretrained(
    model_name = model_name,
    max_seq_length = max_seq_length,
    dtype = dtype,
    load_in_4bit = load_in_4bit,
    token = token,
    trust_remote_code=True,
)

# 推論モードに切り替え
FastLanguageModel.for_inference(model)

# 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。
datasets = []
with open(answer_json_file, "r") as f:
    item = ""
    for line in f:
      line = line.strip()
      item += line
      if item.endswith("}"):
        datasets.append(json.loads(item))
        item = ""

# 推論
from tqdm import tqdm

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=512, 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})

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