add server, config, and scripts

.gitignore

@@ -0,0 +1,5 @@
+# Model files (large, downloaded separately)
+models/
+
+# log
+*.log

install.sh

@@ -0,0 +1,7 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+pip install torch==2.10.0 torchvision==0.25.0 \
+  --index-url https://download.pytorch.org/whl/cu128
+pip install vllm==0.18.0 flashinfer-python==0.6.6
+pip install transformers==5.12.0

opencode.json

@@ -0,0 +1,24 @@
+{
+  "$schema": "https://opencode.ai/config.json",
+  "provider": {
+    "Agent": {
+      "api": "openai",
+      "options": {
+        "baseURL": "http://localhost:8000/v1/",
+        "apiKey": "not-needed"
+      },
+      "models": {
+        "GLM-4.6V-Flash": {
+          "id": "GLM-4.6V-Flash",
+          "name": "GLM-4.6V-Flash",
+          "limit": { "context": 32768, "output": 8192 },
+          "cost": { "input": 0, "output": 0 },
+          "modalities": { "input": ["text", "image"], "output": ["text"] }
+        }
+      }
+    }
+  },
+  "experimental": {
+    "mcp_timeout": 60000
+  }
+}

server.py

@@ -0,0 +1,42 @@
+import os, subprocess
+from pathlib import Path
+
+def _pick_free_gpu():
+    if "CUDA_VISIBLE_DEVICES" in os.environ:
+        return os.environ["CUDA_VISIBLE_DEVICES"]
+    try:
+        out = subprocess.check_output(
+            ["nvidia-smi", "--query-gpu=index,memory.free", "--format=csv,noheader"],
+            text=True,
+        )
+        lines = [l.strip() for l in out.strip().split("\n") if l.strip()]
+        best = max(lines, key=lambda x: int(x.split(",")[1].strip().split()[0]))
+        gpu = best.split(",")[0].strip()
+    except Exception:
+        gpu = "0"
+    os.environ["CUDA_VISIBLE_DEVICES"] = gpu
+    return gpu
+
+if __name__ == "__main__":
+    _pick_free_gpu()
+
+    model = str(Path(__file__).resolve().parent / "models/GLM-4.6V-Flash")
+
+    os.environ["FLASHINFER_DISABLE_VERSION_CHECK"] = "1"
+
+    os.execvp("vllm", [
+        "vllm", "serve",
+        model,
+        "--host", "0.0.0.0",
+        "--port", "8000",
+        "--served-model-name", "GLM-4.6V-Flash",
+        "--trust-remote-code",
+        "--enable-auto-tool-choice",
+        "--tool-call-parser", "glm47",
+        "--max-model-len", "32768",
+        "--dtype", "bfloat16",
+        "--gpu-memory-utilization", "0.85",
+        "--enforce-eager",
+        "--generation-config", "vllm",
+        "--override-generation-config", '{"temperature": 0.8, "top_k": 50, "top_p": 0.9}',
+    ])

start.sh

@@ -0,0 +1,46 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+ACTION="${1:-all}"
+GPU="${2:-2}"
+PROJECT_DIR="$(cd "$(dirname "$0")" && pwd)"
+
+start_backend() {
+  echo "=== Starting Backend (vLLM) on GPU $GPU ==="
+  source /home/kxqandccx/miniconda3/bin/activate agent_use
+  CUDA_VISIBLE_DEVICES="$GPU" nohup python "$PROJECT_DIR/server.py" \
+    > "$PROJECT_DIR/server.log" 2>&1 &
+  echo "Backend PID=$!"
+
+  echo "Waiting for backend..."
+  for i in $(seq 1 60); do
+    if curl -s http://127.0.0.1:8000/v1/models > /dev/null 2>&1; then
+      echo "Backend ready!"
+      return
+    fi
+    sleep 2
+  done
+  echo "Backend failed to start"
+  exit 1
+}
+
+start_frontend() {
+  echo "=== Starting Frontend (opencode web) ==="
+  cd "$PROJECT_DIR"
+  nohup opencode web --port 3000 > /tmp/opencode-web.log 2>&1 &
+  echo "Frontend PID=$!"
+  echo "Web interface: http://127.0.0.1:3000/"
+}
+
+case "$ACTION" in
+  backend) start_backend ;;
+  frontend) start_frontend ;;
+  all) start_backend; start_frontend ;;
+  *)
+    echo "Usage: $0 [backend|frontend|all] [GPU]"
+    echo "  ./start.sh          # start both"
+    echo "  ./start.sh backend  # start backend only"
+    echo "  ./start.sh frontend # start frontend only"
+    echo "  ./start.sh all 3    # both on GPU 3"
+    ;;
+esac