Create app.py

app.py

@@ -0,0 +1,26 @@
+import torch
+from fastapi import FastAPI, Request
+from transformers import AutoTokenizer, AutoModelForCausalLM
+from peft import PeftModel
+
+base_model_id = "google/gemma-1.1-2b-it"
+lora_model_id = "programci48/heytak-lora-v1"
+
+tokenizer = AutoTokenizer.from_pretrained(base_model_id)
+base_model = AutoModelForCausalLM.from_pretrained(base_model_id, torch_dtype=torch.float32, device_map=None)
+model = PeftModel.from_pretrained(base_model, lora_model_id)
+model.eval()
+
+app = FastAPI()
+
+@app.post("/run/predict")
+async def predict(request: Request):
+    data = await request.json()
+    prompt = data["data"][0]
+
+    inputs = tokenizer(prompt, return_tensors="pt")
+    with torch.no_grad():
+        outputs = model.generate(**inputs, max_new_tokens=100)
+    response = tokenizer.decode(outputs[0], skip_special_tokens=True)
+
+    return {"data": [response]}