Update README.md

README.md

@@ -37,15 +37,18 @@ from transformers import (
 
 )
 
-
+## モデルのロード
 import torch
 from tqdm import tqdm
 import json
 
 
-HF_TOKEN = "Hugging Face Token"
+HF_TOKEN = "Hugging Face Token"　#Hugging Face のAPIキーを入力（read)
+
+model_name = "satoyutaka/llm-jp-3-13b-ftELZ-2"　#作成したモデル名
 
-model_name = "satoyutaka/llm-jp-3-13b-ftELZ-2"
+
+## 量子化パラメータの設定
 
 
 bnb_config = BitsAndBytesConfig(
@@ -60,10 +63,12 @@ bnb_config = BitsAndBytesConfig(
 )
 
 
+## 問題文の読み込み
+
 datasets = []
 
 
-with open("「elyza-tasks-100-TV_0.jsonl」のパスをご指定ください。", "r") as f:
+with open("elyza-tasks-100-TV_0.jsonl", "r") as f:　#ファイルを格納したパスに書き換えてください。
 
     item = ""
 
@@ -81,36 +86,39 @@ with open("「elyza-tasks-100-TV_0.jsonl」のパスをご指定ください。"
 
 results = []
 
-for data in tqdm(datasets):
 
-  input = data["input"]
 
-  prompt = f"""### 指示
-  {input}
+## 推論
+
+
+from tqdm import tqdm
+
+
+results = []
+
 
-  """
+for dt in tqdm(datasets):
 
-  tokenized_input = tokenizer.encode(prompt, add_special_tokens=False, return_tensors="pt").to(model.device)
 
-  with torch.no_grad():
+  input = dt["input"]
 
-      outputs = model.generate(
 
-          tokenized_input,
+  prompt = f"""### 指示\n{input}\n### 回答\n"""
 
-          max_new_tokens=100,
 
-          do_sample=False,
+  inputs = tokenizer([prompt], return_tensors = "pt").to(model.device)
 
-          repetition_penalty=1.2
+  
+  outputs = model.generate(**inputs, max_new_tokens = 512, use_cache = True, do_sample=False, repetition_penalty=1.2)
 
-      )[0]
+  
+  prediction = tokenizer.decode(outputs[0], skip_special_tokens=True).split('\n### 回答')[-1]
 
-  output = tokenizer.decode(outputs[tokenized_input.size(1):], skip_special_tokens=True)
 
+  results.append({"task_id": dt["task_id"], "input": input, "output": prediction})
 
-  results.append({"task_id": data["task_id"], "input": input, "output": output})
 
+## 提出ファイルの作成
 
 
 import re
@@ -118,10 +126,15 @@ import re
 
 model_name = re.sub(".*/", "", model_name)
 
+
 with open(f"./{model_name}-outputs.jsonl", 'w', encoding='utf-8') as f:
 
+
     for result in results:
 
+
         json.dump(result, f, ensure_ascii=False)  # ensure_ascii=False for handling non-ASCII characters
 
+
         f.write('\n')
+