Create app.py

app.py

@@ -0,0 +1,62 @@
+import gradio as gr
+from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
+import os
+import subprocess
+import torch
+
+# ---------- STEP 1: Fine-tuned 모델 Git에서 clone ----------
+repo_url = "https://github.com/yourusername/your-finetuned-model"
+local_dir = "./finetuned_model"
+
+if not os.path.exists(local_dir):
+    subprocess.run(["git", "clone", repo_url, local_dir])
+
+# ---------- STEP 2: Tokenizer와 모델 로드 ----------
+base_model = "mistralai/Mistral-7B-Instruct-v0.2"
+
+tokenizer = AutoTokenizer.from_pretrained(base_model)
+
+model = AutoModelForCausalLM.from_pretrained(
+    local_dir,
+    torch_dtype=torch.float32,
+)
+model = model.to("cpu")
+
+pipe = pipeline("text-generation", model=model, tokenizer=tokenizer, device=-1)
+
+# ---------- STEP 3: Gradio 함수 정의 ----------
+def generate_response(prompt, max_length=256, temperature=0.7):
+    # max_length를 제한하여 속도를 빠르게 함
+    outputs = pipe(
+        prompt,
+        max_length=max_length,
+        temperature=temperature,
+        do_sample=True,
+        top_p=0.9,
+        num_return_sequences=1,
+        pad_token_id=tokenizer.eos_token_id,
+    )
+    return outputs[0]["generated_text"]
+
+# ---------- STEP 4: Gradio UI ----------
+with gr.Blocks() as demo:
+    gr.Markdown("# 🚀 Fine-tuned Mistral-7B (CPU Optimized)")
+
+    with gr.Row():
+        prompt_input = gr.Textbox(label="Input Prompt", placeholder="Type your prompt here...", lines=4)
+
+    with gr.Row():
+        max_len_slider = gr.Slider(64, 512, value=256, step=16, label="Max Length (lower = faster)")
+        temp_slider = gr.Slider(0.1, 1.5, value=0.7, step=0.1, label="Temperature")
+
+    generate_button = gr.Button("Generate")
+    output_box = gr.Textbox(label="Generated Output", lines=10)
+
+    generate_button.click(
+        fn=generate_response,
+        inputs=[prompt_input, max_len_slider, temp_slider],
+        outputs=output_box,
+    )
+
+# ---------- STEP 5: Launch ----------
+demo.launch()