fix

app.py

@@ -1,7 +1,7 @@
 import os
 import torch
 from fastapi import FastAPI
-from transformers import AutoProcessor, AutoModelForCausalLM
+from transformers import AutoTokenizer, AutoModelForCausalLM
 from pydantic import BaseModel
 import logging
 
@@ -12,17 +12,17 @@ logger = logging.getLogger(__name__)
 app = FastAPI()
 
 # モデルロード
-model_name = "google/gemma-3-4b-it"  # 軽量な2Bモデルに変更
+model_name = "google/gemma-2-2b-it"
 try:
     logger.info(f"Loading model: {model_name}")
-    processor = AutoProcessor.from_pretrained(model_name, token=os.getenv("HF_TOKEN"))
+    tokenizer = AutoTokenizer.from_pretrained(model_name, token=os.getenv("HF_TOKEN"))
     model = AutoModelForCausalLM.from_pretrained(
         model_name,
         torch_dtype=torch.bfloat16,
         device_map="auto",
         token=os.getenv("HF_TOKEN"),
         low_cpu_mem_usage=True,
-        load_in_4bit=True  # 量子化でメモリ節約
+        load_in_4bit=True
     )
     logger.info("Model loaded successfully")
 except Exception as e:
@@ -37,9 +37,9 @@ class TextInput(BaseModel):
 async def generate_text(input: TextInput):
     try:
         logger.info(f"Generating text for input: {input.text}")
-        inputs = processor(input.text, return_tensors="pt").to("cuda" if torch.cuda.is_available() else "cpu")
+        inputs = tokenizer(input.text, return_tensors="pt").to("cuda" if torch.cuda.is_available() else "cpu")
         outputs = model.generate(**inputs, max_length=input.max_length)
-        result = processor.decode(outputs[0], skip_special_tokens=True)
+        result = tokenizer.decode(outputs[0], skip_special_tokens=True)
         logger.info(f"Generated text: {result}")
         return {"generated_text": result}
     except Exception as e:

requirements.txt

@@ -1,5 +1,6 @@
 huggingface_hub==0.25.2
 torch==2.1.0
+numpy<2.0
 transformers==4.44.2
 bitsandbytes==0.42.0
 accelerate==0.26.1