Update Dockerfile

Dockerfile

@@ -1,23 +1,15 @@
 FROM ghcr.io/ggml-org/llama.cpp:full
 
-WORKDIR /app
-
-# パッケージのインストール 
-RUN apt update && apt install -y python3-pip
-RUN pip install -U huggingface_hub
-
-# モデルをHugging Faceからダウンロード 
-RUN python3 -c 'from huggingface_hub import hf_hub_download; \
-    repo="AIencoder/Logic-Coder-7B-Q4_K_M-GGUF"; \
-    hf_hub_download(repo_id=repo, filename="logic-coder-7b-q4_k_m.gguf", local_dir="/app")'
-
-# サーバーの設定
-CMD ["--server", \
-     "-m", "/app/logic-coder-7b-q4_k_m.gguf", \
-     "--host", "0.0.0.0", \
-     "--port", "7860", \
-     "-t", "2", \
-     "--cache-type-k", "q8_0", \
-     "--cache-type-v", "iq4_nl", \
-     "-c", "32000", \
-     "-n", "8000"]
+RUN apt update && apt install wget -y
+
+# Download model
+RUN wget "https://huggingface.co/unsloth/gemma-3-1b-it-GGUF/resolve/main/gemma-3-1b-it-Q4_K_M.gguf" -O /model.gguf
+
+# API-optimized settings: larger context, GPU offload, flash attention
+CMD ["--server", "-m", "/model.gguf", "--port", "7860", "--host", "0.0.0.0", 
+     "-c", "4096",           # Context size: 4K tokens
+     "-n", "1024",           # Max generation: 1K tokens  
+     "-ngl", "999",          # Offload all layers to GPU (if available)
+     "-fa",                  # Flash Attention for speed
+     "--chat-template", "gemma",  # Proper chat formatting
+     "--timeout", "300"]     # 5 min timeout for long requests