From e7c62741072b1038ed2925e0e0c7248549ef697a Mon Sep 17 00:00:00 2001
From: biondizzle <biondizzle@gmail.com>
Date: Sat, 16 May 2026 06:14:28 +0000
Subject: [PATCH] Revert "feat: auto-warmup in build_and_run.sh"

This reverts commit f792537719416af582bf4d6c880a6799c400f306.
---
 build_and_run.sh | 20 --------------------
 warmup.sh        | 27 ---------------------------
 2 files changed, 47 deletions(-)
 delete mode 100755 warmup.sh

diff --git a/build_and_run.sh b/build_and_run.sh
index 41ca2baf..4431cf04 100755
--- a/build_and_run.sh
+++ b/build_and_run.sh
@@ -20,23 +20,3 @@ 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
diff --git a/warmup.sh b/warmup.sh
deleted file mode 100755
index 26e78ad7..00000000
--- a/warmup.sh
+++ /dev/null
@@ -1,27 +0,0 @@
-#!/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. ==="