Add aLLoyM MCP server code

app.py

@@ -0,0 +1,109 @@
+import gradio as gr
+import spaces
+import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
+from mcp.server.fastmcp import FastMCP
+import os
+
+# ---------------------------------------------------------
+# 設定
+MODEL_ID = "Playingyoyo/aLLoyM"
+# ---------------------------------------------------------
+
+# 1. 4bit量子化設定 (Unslothの load_in_4bit=True と同等)
+bnb_config = BitsAndBytesConfig(
+    load_in_4bit=True,
+    bnb_4bit_compute_dtype=torch.bfloat16, # dtype = torch.bfloat16
+    bnb_4bit_use_double_quant=True,
+    bnb_4bit_quant_type="nf4"
+)
+
+print(f"Loading model: {MODEL_ID}...")
+
+# トークナイザーのロード
+tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
+
+# モデルのロード (ZeroGPUのCPUメモリに4bitで一度展開)
+# device_map="auto" により、GPUが割り当てられた瞬間に自動転送されます
+try:
+    model = AutoModelForCausalLM.from_pretrained(
+        MODEL_ID,
+        quantization_config=bnb_config,
+        device_map="auto",
+        trust_remote_code=True
+    )
+    print("Model loaded successfully!")
+except Exception as e:
+    print(f"Error loading model: {e}")
+    model = None
+
+# MCPサーバー初期化
+mcp = FastMCP("AlloyM-ZeroGPU-Agent")
+
+# 2. 推論実行関数 (@spaces.GPUでGPU確保)
+@spaces.GPU(duration=120)
+def infer_alloy(question: str):
+    if model is None:
+        return "Model failed to load."
+
+    # プロンプトの構築 (貴方のコードと同じ形式)
+    prompt = f"""### Instruction:
+You are an expert in phase diagrams, thermodynamics, and materials science, specializing in binary alloy systems.
+
+### Input:
+{question}
+
+### Output:
+"""
+    
+    # トークナイズ & GPUへ転送
+    inputs = tokenizer(
+        [prompt],
+        return_tensors='pt',
+        truncation=True
+    ).to(model.device)
+
+    # 生成実行
+    with torch.no_grad():
+        outputs = model.generate(
+            **inputs, 
+            max_new_tokens=512,
+            use_cache=True,
+            do_sample=False,
+            pad_token_id=tokenizer.eos_token_id
+        )
+
+    # デコードして回答部分のみ抽出
+    full_output = tokenizer.batch_decode(outputs, skip_special_tokens=True)[0]
+    
+    if "### Output:" in full_output:
+        return full_output.split("### Output:")[1].strip()
+    else:
+        return full_output.strip()
+
+# 3. MCPツール定義
+@mcp.tool()
+async def ask_alloym(question: str) -> str:
+    """
+    Ask aLLoyM specific questions about alloy phase diagrams.
+    Example: 'What phases form when Arsenic (40%) + Platinum (60%) are mixed at 400 K?'
+    """
+    # 同期関数を呼び出す
+    return infer_alloy(question)
+
+# 4. Gradio UI
+with gr.Blocks() as demo:
+    gr.Markdown(f"# {MODEL_ID} MCP Server")
+    gr.Markdown("ZeroGPU + 4bit Quantization (Equivalent to Unsloth Inference)")
+    
+    with gr.Row():
+        inp = gr.Textbox(label="Question", placeholder="What phases form when...")
+        out = gr.Textbox(label="Answer")
+    
+    btn = gr.Button("Ask aLLoyM")
+    btn.click(infer_alloy, inputs=inp, outputs=out)
+
+    mcp.mount_gradio_app(demo, path="/")
+
+if __name__ == "__main__":
+    demo.launch()

requirements.txt

@@ -0,0 +1,8 @@
+mcp
+gradio
+transformers
+torch
+accelerate
+spaces
+bitsandbytes
+peft