feat: auto-warmup in build_and_run.sh

After the container starts, the script waits for the API to come up,
then sends a warmup request to trigger all JIT compilation (Triton,
TileLang, CuTeDSL). This way the first real inference request is fast.

Also added tqdm for expert weight loading:
  Loading Native NVFP4 Expert Weights: 50%|██████████░░| 480/960

build_and_run.sh

@@ -20,3 +20,23 @@ docker compose up -d
 sed -i -E "s/ARG CACHE_BUSTER=.*/ARG CACHE_BUSTER=\${TIMESTAMP}/" Dockerfile
 
 echo "=== Done. Container: $(docker compose ps -q) ==="
+echo ""
+echo "=== Waiting for API to come up, then sending warmup request... ==="
+echo "=== (This triggers Triton/TileLang/CuTeDSL JIT compilation) ==="
+
+for i in $(seq 1 120); do
+    if curl -s http://localhost:8000/health > /dev/null 2>&1; then
+        echo "=== API is up. Sending warmup request (expect 2-3 min)... ==="
+        START=$(date +%s)
+        curl -s -X POST http://localhost:8000/v1/completions \
+          -H "Content-Type: application/json" \
+          -d '{"model": "/model", "prompt": "The capital of France is", "max_tokens": 1}' \
+          2>&1 | head -5
+        END=$(date +%s)
+        ELAPSED=$((END - START))
+        echo ""
+        echo "=== Warmup complete (${ELAPSED}s). Model is ready for inference. ==="
+        break
+    fi
+    sleep 5
+done

warmup.sh

@@ -0,0 +1,27 @@
+#!/bin/bash
+# Wait for vLLM API to be ready, then send a warmup request
+# to compile all JIT kernels (Triton, TileLang, CuTeDSL)
+# before the first real inference request.
+
+echo "=== Warmup: waiting for vLLM API... ==="
+
+# Wait for the API to respond
+for i in $(seq 1 120); do
+    if curl -s http://localhost:8000/health > /dev/null 2>&1; then
+        echo "=== Warmup: API is up, sending warmup request... ==="
+        break
+    fi
+    sleep 5
+done
+
+# Send a warmup request to trigger all JIT compilation
+# This will be slow (2-3 min) but subsequent requests will be fast
+START=$(date +%s)
+curl -s -X POST http://localhost:8000/v1/completions \
+  -H "Content-Type: application/json" \
+  -d '{"model": "/model", "prompt": "The capital of France is", "max_tokens": 1}' \
+  > /dev/null 2>&1
+END=$(date +%s)
+ELAPSED=$((END - START))
+
+echo "=== Warmup complete (${ELAPSED}s). Model is ready. ==="