Replace repository with DuckLM runtime

2026-05-20 01:00:28 +08:00
parent ddc285b8f4
commit 4a84ada770
190 changed files with 7060 additions and 13602 deletions
@@ -0,0 +1,34 @@
+import asyncio
+import time
+
+from duck_core.model_client import ModelClient
+
+
+TASKS = [
+    "Скажи коротко, что ты DuckLM.",
+    "Создай tmp/duck_test_note.md с текстом hello duck и прочитай его обратно.",
+    "Посмотри структуру проекта и кратко опиши модули.",
+    "Найди TODO/FIXME в проекте.",
+    "Запусти тесты и кратко объясни результат.",
+]
+
+
+async def main() -> None:
+    client = ModelClient()
+    print("role -> base_url/model")
+    for role, cfg in client._roles.items():
+        print(f"{role} -> {cfg.base_url}/{cfg.model}")
+    started = time.perf_counter()
+    print(f"test_tasks={len(TASKS)}")
+    print("llm_calls=0")
+    print("tool_calls=0")
+    print("json_directive_validity=not_run")
+    print("retry_count=0")
+    print("memory_writes=0")
+    print("experience_record_created=no")
+    print("selected_skill=not_run")
+    print(f"total_runtime_seconds={time.perf_counter() - started:.3f}")
+
+
+if __name__ == "__main__":
+    asyncio.run(main())
@@ -0,0 +1,19 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+ROOT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")/../.." && pwd)"
+LLAMA_DIR="${ROOT_DIR}/vendor/llama.cpp"
+
+if [[ ! -d "${LLAMA_DIR}/.git" ]]; then
+  git clone --depth 1 https://github.com/ggml-org/llama.cpp "${LLAMA_DIR}"
+fi
+
+cmake -S "${LLAMA_DIR}" -B "${LLAMA_DIR}/build" \
+  -DCMAKE_BUILD_TYPE=Release \
+  -DGGML_VULKAN=ON \
+  -DGGML_NATIVE=ON \
+  -DLLAMA_BUILD_TESTS=OFF
+
+cmake --build "${LLAMA_DIR}/build" --config Release --target llama-server -j "$(nproc)"
+
+"${LLAMA_DIR}/build/bin/llama-server" --list-devices
@@ -0,0 +1,8 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+BASE_URL="${1:-http://127.0.0.1:8081/v1}"
+
+curl --noproxy "*" -fsS "${BASE_URL}/models" >/dev/null
+
+echo "OK: ${BASE_URL}"
@@ -0,0 +1,260 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+ROOT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")/../.." && pwd)"
+
+ENV_KEYS=(
+  DUCK_LLAMA_SERVER_BIN
+  DUCK_MAIN_MODEL_PATH
+  DUCK_MAIN_PORT
+  DUCK_CTX_SIZE
+  DUCK_N_GPU_LAYERS
+  DUCK_LLAMA_DEVICE
+  DUCK_PARALLEL
+  DUCK_LLAMA_PID_FILE
+  DUCK_LLAMA_LOG_FILE
+  DUCK_LLAMA_EXTRA_ARGS
+  DUCK_HOST
+)
+declare -A ENV_OVERRIDES=()
+for key in "${ENV_KEYS[@]}"; do
+  if [[ -v "${key}" ]]; then
+    ENV_OVERRIDES["${key}"]="${!key}"
+  fi
+done
+
+if [[ -f "${ROOT_DIR}/.env" ]]; then
+  set -a
+  # shellcheck disable=SC1091
+  source "${ROOT_DIR}/.env"
+  set +a
+fi
+for key in "${!ENV_OVERRIDES[@]}"; do
+  export "${key}=${ENV_OVERRIDES[${key}]}"
+done
+
+ACTION="${1:-start}"
+PID_FILE="${DUCK_LLAMA_PID_FILE:-${ROOT_DIR}/data/llama-main.pid}"
+LOG_FILE="${DUCK_LLAMA_LOG_FILE:-${ROOT_DIR}/data/llama-main.log}"
+BASE_URL="http://${DUCK_HOST:-127.0.0.1}:${DUCK_MAIN_PORT:-8081}/v1"
+
+resolve_project_path() {
+  local value="$1"
+  if [[ "${value}" == /* ]]; then
+    printf '%s\n' "${value}"
+  else
+    printf '%s\n' "${ROOT_DIR}/${value#./}"
+  fi
+}
+
+usage() {
+  cat <<'EOF'
+Usage: scripts/llama/start_main.sh <command>
+
+Commands:
+  start       Start llama-server in the background
+  stop        Stop the managed llama-server process
+  restart     Stop and start llama-server
+  status      Print process and HTTP health status
+  logs        Show logs; use --follow/-f and --lines N
+  help        Show this help
+
+Environment:
+  DUCK_LLAMA_SERVER_BIN  Path to llama-server binary
+  DUCK_MAIN_MODEL_PATH   Path to GGUF model
+  DUCK_HOST              Bind host, default 127.0.0.1
+  DUCK_MAIN_PORT         Port, default 8081
+  DUCK_CTX_SIZE          Context size, default 65536
+  DUCK_N_GPU_LAYERS      GPU layers, default auto
+  DUCK_LLAMA_DEVICE      Device name, for example Vulkan0
+  DUCK_PARALLEL          Server slots, default 1
+  DUCK_LLAMA_PID_FILE    PID file path
+  DUCK_LLAMA_LOG_FILE    Log file path
+  DUCK_LLAMA_EXTRA_ARGS  Extra llama-server args
+EOF
+}
+
+is_running() {
+  [[ -f "${PID_FILE}" ]] || return 1
+  local pid
+  pid="$(cat "${PID_FILE}")"
+  [[ "${pid}" =~ ^[0-9]+$ ]] || return 1
+  kill -0 "${pid}" 2>/dev/null
+}
+
+pid_value() {
+  if [[ -f "${PID_FILE}" ]]; then
+    cat "${PID_FILE}"
+  fi
+}
+
+status() {
+  if is_running; then
+    local pid
+    pid="$(pid_value)"
+    echo "llama-server running: pid=${pid}"
+    if command -v curl >/dev/null 2>&1 && curl --noproxy "*" -fsS "${BASE_URL}/models" >/dev/null 2>&1; then
+      echo "HTTP health: ok (${BASE_URL})"
+    else
+      echo "HTTP health: not ready (${BASE_URL})"
+    fi
+    return 0
+  fi
+
+  if [[ -f "${PID_FILE}" ]]; then
+    echo "llama-server not running; removing stale pid file ${PID_FILE}"
+    rm -f "${PID_FILE}"
+  else
+    echo "llama-server not running"
+  fi
+  return 3
+}
+
+start() {
+  if is_running; then
+    echo "llama-server already running: pid=$(pid_value)"
+    return 0
+  fi
+
+  : "${DUCK_MAIN_MODEL_PATH:?DUCK_MAIN_MODEL_PATH is required}"
+
+  mkdir -p "$(dirname "${PID_FILE}")" "$(dirname "${LOG_FILE}")"
+  rm -f "${PID_FILE}"
+
+  local llama_bin model_path
+  llama_bin="${DUCK_LLAMA_SERVER_BIN:-llama-server}"
+  if [[ "${llama_bin}" == */* ]]; then
+    llama_bin="$(resolve_project_path "${llama_bin}")"
+  fi
+  model_path="$(resolve_project_path "${DUCK_MAIN_MODEL_PATH}")"
+  local command=(
+    "${llama_bin}"
+    -m "${model_path}"
+    --alias local-main
+    --host "${DUCK_HOST:-127.0.0.1}"
+    --port "${DUCK_MAIN_PORT:-8081}"
+    -c "${DUCK_CTX_SIZE:-65536}"
+    --parallel "${DUCK_PARALLEL:-1}"
+    -ngl "${DUCK_N_GPU_LAYERS:-auto}"
+    --flash-attn on
+    --cache-prompt
+    --metrics
+  )
+  if [[ -n "${DUCK_LLAMA_DEVICE:-}" ]]; then
+    command+=(--device "${DUCK_LLAMA_DEVICE}")
+  fi
+  if [[ -n "${DUCK_LLAMA_EXTRA_ARGS:-}" ]]; then
+    # shellcheck disable=SC2206
+    local extra_args=( ${DUCK_LLAMA_EXTRA_ARGS} )
+    command+=("${extra_args[@]}")
+  fi
+
+  echo "Starting llama-server..."
+  echo "Command: ${command[*]}" >> "${LOG_FILE}"
+  if command -v setsid >/dev/null 2>&1; then
+    nohup setsid "${command[@]}" >> "${LOG_FILE}" 2>&1 &
+  else
+    nohup "${command[@]}" >> "${LOG_FILE}" 2>&1 &
+  fi
+  local pid=$!
+  echo "${pid}" > "${PID_FILE}"
+  sleep 0.2
+
+  if is_running; then
+    echo "llama-server started: pid=${pid}"
+    echo "Log: ${LOG_FILE}"
+    return 0
+  fi
+
+  echo "llama-server failed to start. See ${LOG_FILE}" >&2
+  rm -f "${PID_FILE}"
+  return 1
+}
+
+stop() {
+  if ! is_running; then
+    rm -f "${PID_FILE}"
+    echo "llama-server not running"
+    return 0
+  fi
+
+  local pid
+  pid="$(pid_value)"
+  echo "Stopping llama-server: pid=${pid}"
+  kill "${pid}" 2>/dev/null || true
+
+  for _ in {1..30}; do
+    if ! kill -0 "${pid}" 2>/dev/null; then
+      rm -f "${PID_FILE}"
+      echo "llama-server stopped"
+      return 0
+    fi
+    sleep 0.2
+  done
+
+  echo "llama-server did not stop after SIGTERM; sending SIGKILL"
+  kill -9 "${pid}" 2>/dev/null || true
+  rm -f "${PID_FILE}"
+  echo "llama-server stopped"
+}
+
+restart() {
+  stop
+  start
+}
+
+logs() {
+  local follow=0
+  local lines=100
+  shift || true
+  while [[ $# -gt 0 ]]; do
+    case "$1" in
+      -f|--follow)
+        follow=1
+        shift
+        ;;
+      --lines)
+        lines="${2:?--lines requires a value}"
+        shift 2
+        ;;
+      *)
+        echo "Unknown logs argument: $1" >&2
+        return 2
+        ;;
+    esac
+  done
+
+  mkdir -p "$(dirname "${LOG_FILE}")"
+  touch "${LOG_FILE}"
+  if [[ "${follow}" == "1" ]]; then
+    tail -n "${lines}" -f "${LOG_FILE}"
+  else
+    tail -n "${lines}" "${LOG_FILE}"
+  fi
+}
+
+case "${ACTION}" in
+  start)
+    start
+    ;;
+  stop)
+    stop
+    ;;
+  restart)
+    restart
+    ;;
+  status)
+    status
+    ;;
+  logs)
+    logs "$@"
+    ;;
+  help|-h|--help)
+    usage
+    ;;
+  *)
+    echo "Unknown command: ${ACTION}" >&2
+    usage >&2
+    exit 2
+    ;;
+esac
@@ -0,0 +1,117 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+ROOT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")/../.." && pwd)"
+
+ENV_KEYS=(
+  DUCK_LLAMA_SERVER_BIN
+  DUCK_MAIN_MODEL_PATH
+  DUCK_MTP_MODEL_PATH
+  DUCK_MAIN_MTP_PORT
+  DUCK_CTX_SIZE
+  DUCK_N_GPU_LAYERS
+  DUCK_LLAMA_DEVICE
+  DUCK_PARALLEL
+  DUCK_MTP_FLAGS
+  DUCK_HOST
+)
+declare -A ENV_OVERRIDES=()
+for key in "${ENV_KEYS[@]}"; do
+  if [[ -v "${key}" ]]; then
+    ENV_OVERRIDES["${key}"]="${!key}"
+  fi
+done
+
+if [[ -f "${ROOT_DIR}/.env" ]]; then
+  set -a
+  # shellcheck disable=SC1091
+  source "${ROOT_DIR}/.env"
+  set +a
+fi
+for key in "${!ENV_OVERRIDES[@]}"; do
+  export "${key}=${ENV_OVERRIDES[${key}]}"
+done
+
+: "${DUCK_MAIN_MODEL_PATH:?DUCK_MAIN_MODEL_PATH is required}"
+ACTION="${1:-start}"
+
+usage() {
+  cat <<'EOF'
+Usage: scripts/llama/start_thinker_mtp_experimental.sh <command>
+
+Commands:
+  start   Start experimental MTP/speculative llama-server in foreground
+  check   Check whether the current llama-server binary exposes draft-mtp flags
+  help    Show this help
+EOF
+}
+
+resolve_project_path() {
+  local value="$1"
+  if [[ "${value}" == /* ]]; then
+    printf '%s\n' "${value}"
+  else
+    printf '%s\n' "${ROOT_DIR}/${value#./}"
+  fi
+}
+
+LLAMA_BIN="${DUCK_LLAMA_SERVER_BIN:-llama-server}"
+if [[ "${LLAMA_BIN}" == */* ]]; then
+  LLAMA_BIN="$(resolve_project_path "${LLAMA_BIN}")"
+fi
+MAIN_MODEL_PATH="$(resolve_project_path "${DUCK_MAIN_MODEL_PATH}")"
+
+HELP_TEXT="$("${LLAMA_BIN}" --help 2>&1 || true)"
+if ! grep -qi "draft-mtp" <<< "${HELP_TEXT}"; then
+  echo "This llama-server build does not expose draft-mtp speculative decoding."
+  exit 1
+fi
+
+case "${ACTION}" in
+  check)
+    echo "OK: draft-mtp speculative decoding is exposed by ${LLAMA_BIN}"
+    exit 0
+    ;;
+  help|-h|--help)
+    usage
+    exit 0
+    ;;
+  start)
+    ;;
+  *)
+    echo "Unknown command: ${ACTION}" >&2
+    usage >&2
+    exit 2
+    ;;
+esac
+
+command=(
+  "${LLAMA_BIN}"
+  -m "${MAIN_MODEL_PATH}"
+  --alias local-main-mtp
+  --host "${DUCK_HOST:-127.0.0.1}"
+  --port "${DUCK_MAIN_MTP_PORT:-8085}"
+  -c "${DUCK_CTX_SIZE:-65536}"
+  --parallel "${DUCK_PARALLEL:-1}"
+  -ngl "${DUCK_N_GPU_LAYERS:-auto}"
+  --flash-attn on
+  --cache-prompt
+  --metrics
+  --spec-type draft-mtp
+)
+
+if [[ -n "${DUCK_LLAMA_DEVICE:-}" ]]; then
+  command+=(--device "${DUCK_LLAMA_DEVICE}")
+fi
+
+if [[ -n "${DUCK_MTP_MODEL_PATH:-}" ]]; then
+  command+=(--model-draft "$(resolve_project_path "${DUCK_MTP_MODEL_PATH}")")
+fi
+
+if [[ -n "${DUCK_MTP_FLAGS:-}" ]]; then
+  # shellcheck disable=SC2206
+  extra_args=( ${DUCK_MTP_FLAGS} )
+  command+=("${extra_args[@]}")
+fi
+
+exec "${command[@]}"
@@ -1,81 +0,0 @@
-#!/usr/bin/env bash
-set -euo pipefail
-
-ROOT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
-VENV_PYTHON="$ROOT_DIR/.venv/bin/python"
-PID_FILE="$ROOT_DIR/data/runtime/server.pid"
-LOG_FILE="$ROOT_DIR/data/runtime/server.log"
-HOST="${HOST:-127.0.0.1}"
-PORT="${PORT:-8000}"
-
-mkdir -p "$ROOT_DIR/data/runtime"
-
-export LD_LIBRARY_PATH="${LD_LIBRARY_PATH:+${LD_LIBRARY_PATH}:}$ROOT_DIR/.venv/lib/python3.13/site-packages/llama_cpp/lib"
-export GGML_VULKAN=1
-
-is_running() {
-  if [[ -f "$PID_FILE" ]]; then
-    local pid
-    pid="$(cat "$PID_FILE")"
-    if kill -0 "$pid" >/dev/null 2>&1; then
-      return 0
-    fi
-  fi
-  return 1
-}
-
-start_server() {
-  if is_running; then
-    echo "Server already running with PID $(cat "$PID_FILE")"
-    exit 0
-  fi
-  nohup "$VENV_PYTHON" -m uvicorn main:app --host "$HOST" --port "$PORT" >"$LOG_FILE" 2>&1 &
-  echo $! >"$PID_FILE"
-  echo "Started server on http://$HOST:$PORT with PID $(cat "$PID_FILE")"
-  echo "Log: $LOG_FILE"
-}
-
-stop_server() {
-  if ! is_running; then
-    echo "Server is not running"
-    rm -f "$PID_FILE"
-    exit 0
-  fi
-  local pid
-  pid="$(cat "$PID_FILE")"
-  kill "$pid"
-  rm -f "$PID_FILE"
-  echo "Stopped server PID $pid"
-}
-
-status_server() {
-  if is_running; then
-    echo "Server is running with PID $(cat "$PID_FILE") on http://$HOST:$PORT"
-  else
-    echo "Server is not running"
-  fi
-}
-
-case "${1:-}" in
-  start)
-    start_server
-    ;;
-  stop)
-    stop_server
-    ;;
-  restart)
-    stop_server || true
-    start_server
-    ;;
-  status)
-    status_server
-    ;;
-  logs)
-    touch "$LOG_FILE"
-    tail -n 50 -f "$LOG_FILE"
-    ;;
-  *)
-    echo "Usage: $0 {start|stop|restart|status|logs}"
-    exit 1
-    ;;
-esac
@@ -0,0 +1,13 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+BASE_URL="${DUCK_API_URL:-http://127.0.0.1:8000}"
+
+curl -fsS "${BASE_URL}/health"
+
+curl -fsS -X POST "${BASE_URL}/v1/chat" \
+  -H "Content-Type: application/json" \
+  -d '{
+    "message": "Скажи коротко, что ты DuckLM",
+    "debug": true
+  }'
@@ -0,0 +1,5 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+BASE_URL="${DUCK_API_URL:-http://127.0.0.1:8000}"
+curl -fsS "${BASE_URL}/v1/experience"
@@ -0,0 +1,14 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+BASE_URL="${DUCK_API_URL:-http://127.0.0.1:8000}"
+
+RESPONSE="$(curl -fsS -X POST "${BASE_URL}/v1/chat" \
+  -H "Content-Type: application/json" \
+  -d '{
+    "message": "Создай tmp/duck_test_note.md с текстом hello duck и прочитай его обратно",
+    "workspace": "./workspace",
+    "debug": true
+  }')"
+
+echo "${RESPONSE}"
@@ -0,0 +1,5 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+BASE_URL="${DUCK_API_URL:-http://127.0.0.1:8000}"
+curl -fsS "${BASE_URL}/v1/memory/search?q=duck"
@@ -0,0 +1,5 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+BASE_URL="${DUCK_API_URL:-http://127.0.0.1:8000}"
+curl -fsS "${BASE_URL}/v1/models/roles"
@@ -0,0 +1,5 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+BASE_URL="${DUCK_API_URL:-http://127.0.0.1:8000}"
+curl -fsS "${BASE_URL}/v1/skills"
@@ -0,0 +1,14 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+python3 - <<'PY'
+import asyncio
+from duck_core.tools.shell_exec_safe import ShellExecSafeTool
+
+async def main():
+    result = await ShellExecSafeTool(".").run({"command": "rm -rf ."})
+    assert not result.ok
+    print("OK: dangerous command blocked")
+
+asyncio.run(main())
+PY