Add OpenAI-compatible HTTP server (/v1/chat/completions)

Pure C HTTP server with streaming SSE support, zero external deps.
Endpoints: POST /v1/chat/completions, GET /v1/models, GET /health.
KV compression configurable per-request. Build with -DTQ_BUILD_SERVER=ON.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: unamedkr

CMakeLists.txt

@@ -10,6 +10,7 @@ option(TQ_BUILD_CUDA "Build CUDA backend" OFF)
 option(TQ_BUILD_METAL "Build Metal backend" OFF)
 option(TQ_BUILD_VULKAN "Build Vulkan backend" OFF)
 option(TQ_BUILD_ROCM "Build ROCm/HIP backend" OFF)
+option(TQ_BUILD_SERVER "Build OpenAI-compatible HTTP server" OFF)
 
 # Threads (pthread)
 find_package(Threads REQUIRED)
@@ -276,3 +277,19 @@ foreach(ex_src ${EXAMPLE_C_SOURCES} ${EXAMPLE_CXX_SOURCES})
     add_executable(${ex_name} ${ex_src})
     target_link_libraries(${ex_name} turboquant)
 endforeach()
+
+# OpenAI-compatible HTTP server
+if(TQ_BUILD_SERVER)
+    add_executable(quant-server src/server/tq_server.c)
+    target_include_directories(quant-server PRIVATE
+        ${CMAKE_SOURCE_DIR}/src/server
+        ${CMAKE_SOURCE_DIR}/include)
+    target_compile_definitions(quant-server PRIVATE TQ_SERVER_MAIN)
+    target_link_libraries(quant-server turboquant Threads::Threads)
+    if(NOT MSVC)
+        target_link_libraries(quant-server m)
+    endif()
+    target_compile_options(quant-server PRIVATE
+        -Wall -Wextra -Wpedantic -Wno-unused-parameter)
+    message(STATUS "quant.cpp: HTTP server target enabled (quant-server)")
+endif()

examples/server_example.sh

@@ -0,0 +1,135 @@
+#!/usr/bin/env bash
+# server_example.sh -- Start and test the quant.cpp OpenAI-compatible server
+#
+# Usage:
+#   ./examples/server_example.sh <model.gguf>
+#
+# Prerequisites:
+#   cmake -B build -DCMAKE_BUILD_TYPE=Release -DTQ_BUILD_SERVER=ON
+#   cmake --build build -j$(nproc)
+
+set -euo pipefail
+
+MODEL="${1:-}"
+PORT="${2:-8080}"
+
+if [ -z "$MODEL" ]; then
+    echo "Usage: $0 <model.gguf> [port]"
+    echo ""
+    echo "Example:"
+    echo "  $0 Qwen2.5-0.5B-Instruct.gguf 8080"
+    exit 1
+fi
+
+SERVER="./build/quant-server"
+if [ ! -f "$SERVER" ]; then
+    echo "Error: quant-server not found. Build with:"
+    echo "  cmake -B build -DCMAKE_BUILD_TYPE=Release -DTQ_BUILD_SERVER=ON"
+    echo "  cmake --build build -j\$(nproc)"
+    exit 1
+fi
+
+echo "=== Starting quant.cpp server ==="
+echo "Model:  $MODEL"
+echo "Port:   $PORT"
+echo ""
+
+# Start server in background
+$SERVER "$MODEL" -p "$PORT" -j 4 -k uniform_4b &
+SERVER_PID=$!
+
+# Wait for server to start
+echo "Waiting for server to start..."
+for i in $(seq 1 30); do
+    if curl -s "http://localhost:$PORT/health" > /dev/null 2>&1; then
+        echo "Server is ready!"
+        break
+    fi
+    sleep 0.5
+done
+
+echo ""
+echo "=== Test 1: Health check ==="
+curl -s "http://localhost:$PORT/health" | python3 -m json.tool 2>/dev/null || \
+    curl -s "http://localhost:$PORT/health"
+echo ""
+
+echo ""
+echo "=== Test 2: List models ==="
+curl -s "http://localhost:$PORT/v1/models" | python3 -m json.tool 2>/dev/null || \
+    curl -s "http://localhost:$PORT/v1/models"
+echo ""
+
+echo ""
+echo "=== Test 3: Chat completion (non-streaming) ==="
+curl -s "http://localhost:$PORT/v1/chat/completions" \
+    -H "Content-Type: application/json" \
+    -d '{
+        "model": "default",
+        "messages": [
+            {"role": "system", "content": "You are a helpful assistant."},
+            {"role": "user", "content": "What is 2+2? Answer in one word."}
+        ],
+        "max_tokens": 32,
+        "temperature": 0.1
+    }' | python3 -m json.tool 2>/dev/null || echo "(raw output above)"
+echo ""
+
+echo ""
+echo "=== Test 4: Chat completion (streaming) ==="
+curl -s -N "http://localhost:$PORT/v1/chat/completions" \
+    -H "Content-Type: application/json" \
+    -d '{
+        "model": "default",
+        "messages": [
+            {"role": "user", "content": "Say hello in 3 words."}
+        ],
+        "max_tokens": 32,
+        "temperature": 0.7,
+        "stream": true
+    }'
+echo ""
+
+echo ""
+echo "=== Test 5: Chat with KV compression options ==="
+curl -s "http://localhost:$PORT/v1/chat/completions" \
+    -H "Content-Type: application/json" \
+    -d '{
+        "model": "default",
+        "messages": [
+            {"role": "user", "content": "Hello!"}
+        ],
+        "max_tokens": 16,
+        "kv_type": "turbo_kv_3b",
+        "value_quant_bits": 4,
+        "delta_kv": true
+    }' | python3 -m json.tool 2>/dev/null || echo "(raw output above)"
+echo ""
+
+echo ""
+echo "=== Test 6: OpenAI Python SDK compatibility ==="
+cat <<'PYTHON'
+# You can also use the official OpenAI Python SDK:
+#
+#   pip install openai
+#
+#   from openai import OpenAI
+#   client = OpenAI(base_url="http://localhost:8080/v1", api_key="not-needed")
+#
+#   response = client.chat.completions.create(
+#       model="default",
+#       messages=[{"role": "user", "content": "Hello!"}],
+#       max_tokens=64,
+#       stream=True,
+#   )
+#   for chunk in response:
+#       if chunk.choices[0].delta.content:
+#           print(chunk.choices[0].delta.content, end="", flush=True)
+PYTHON
+
+# Cleanup
+echo ""
+echo "Stopping server (PID $SERVER_PID)..."
+kill "$SERVER_PID" 2>/dev/null || true
+wait "$SERVER_PID" 2>/dev/null || true
+echo "Done."