Merge branch 'main' into fix/packaging-system-json

Author: zhzhcookie

.github/new-prs-labeler.yml

@@ -30,7 +30,6 @@ CORE:
           - 'cmake/**'
           - 'include/**'
           - 'lib/**'
-          - 'python/**'
           - 'scripts/**'
           - 'test/**'
           - 'unittest/**'
@@ -45,6 +44,19 @@ CORE:
           - 'setup.py'
           - '.github/actions/**'
           - '.github/workflows/code-format-check.yml'
+      - all-globs-to-any-file:
+          - 'python/**'
+          - '!python/setup_tools/utils/cambricon.py'
+          - '!python/setup_tools/utils/enflame.py'
+          - '!python/setup_tools/utils/hcu.py'
+          - '!python/setup_tools/utils/iluvatar.py'
+          - '!python/setup_tools/utils/metax.py'
+          - '!python/setup_tools/utils/mthreads.py'
+          - '!python/setup_tools/utils/ppu.py'
+          - '!python/setup_tools/utils/rpu.py'
+          - '!python/setup_tools/utils/sunrise.py'
+          - '!python/setup_tools/utils/thrive.py'
+          - '!python/setup_tools/utils/xpu.py'
 
 # ==========================================
 # 2. CI/CD Labels
@@ -128,6 +140,7 @@ cambricon:
           - 'third_party/cambricon/**'
           - '.github/workflows/cambricon*'
           - '.github/workflows/benchmark/cambricon/**'
+          - 'python/setup_tools/utils/cambricon.py'
 
 cpu:
   - changed-files:
@@ -142,40 +155,45 @@ enflame:
           - 'third_party/enflame/**'
           - '.github/workflows/enflame*'
           - '.github/workflows/benchmark/enflame/**'
+          - 'python/setup_tools/utils/enflame.py'
 
 hcu:
   - changed-files:
       - any-glob-to-any-file:
           - 'third_party/hcu/**'
           - '.github/workflows/hcu*'
           - '.github/workflows/benchmark/hcu/**'
+          - 'python/setup_tools/utils/hcu.py'
 
 iluvatar:
   - changed-files:
       - any-glob-to-any-file:
           - 'third_party/iluvatar/**'
           - '.github/workflows/iluvatar*'
           - '.github/workflows/benchmark/iluvatar/**'
+          - 'python/setup_tools/utils/iluvatar.py'
 
 metax:
   - changed-files:
       - any-glob-to-any-file:
           - 'third_party/metax/**'
           - '.github/workflows/metax*'
           - '.github/workflows/benchmark/metax/**'
+          - 'python/setup_tools/utils/metax.py'
 
 mthreads:
   - changed-files:
       - any-glob-to-any-file:
           - 'third_party/mthreads/**'
           - '.github/workflows/mthreads*'
           - '.github/workflows/benchmark/mthreads/**'
+          - 'python/setup_tools/utils/mthreads.py'
 
 nvidia:
   - changed-files:
       - any-glob-to-any-file:
           - 'third_party/nvidia/**'
-          - '.github/workflows/nv*'
+          - '.github/workflows/nvidia*'
           - '.github/workflows/benchmark/nvidia/**'
 
 ppu:
@@ -184,27 +202,31 @@ ppu:
           - 'third_party/ppu/**'
           - '.github/workflows/ppu*'
           - '.github/workflows/benchmark/ppu/**'
+          - 'python/setup_tools/utils/ppu.py'
 
 rpu:
   - changed-files:
       - any-glob-to-any-file:
           - 'third_party/rpu/**'
           - '.github/workflows/rpu*'
           - '.github/workflows/benchmark/rpu/**'
+          - 'python/setup_tools/utils/rpu.py'
 
 sunrise:
   - changed-files:
       - any-glob-to-any-file:
           - 'third_party/sunrise/**'
           - '.github/workflows/sunrise*'
           - '.github/workflows/benchmark/sunrise/**'
+          - 'python/setup_tools/utils/sunrise.py'
 
 thrive:
   - changed-files:
       - any-glob-to-any-file:
           - 'third_party/thrive/**'
           - '.github/workflows/thrive*'
           - '.github/workflows/benchmark/thrive/**'
+          - 'python/setup_tools/utils/thrive.py'
 
 tileir:
   - changed-files:
@@ -233,6 +255,7 @@ xpu:
           - 'third_party/xpu/**'
           - '.github/workflows/xpu*'
           - '.github/workflows/benchmark/xpu/**'
+          - 'python/setup_tools/utils/xpu.py'
 
 # ====================================================================================
 # 6. TLE Labels

.github/workflows/benchmark/mthreads/mthreads3.6-qwen3.6/all_perf.py

@@ -0,0 +1,255 @@
+# Copyright 2025-     FlagOS Contributors
+#
+# Permission is hereby granted, free of charge, to any person obtaining a copy
+# of this software and associated documentation files (the "Software"), to deal
+# in the Software without restriction, including without limitation the rights
+# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+# copies of the Software, and to permit persons to whom the Software is
+# furnished to do so, subject to the following conditions:
+#
+# The above copyright notice and this permission notice shall be included in all
+# copies or substantial portions of the Software.
+#
+# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+# SOFTWARE.
+"""
+vLLM 性能测试脚本
+
+功能：
+- 支持用户自定义 input_len、output_len、并发数
+- 返回 vllm bench serve 的所有指标
+"""
+
+import subprocess
+import sys
+import re
+import argparse
+from datetime import datetime
+from pathlib import Path
+
+sys.stdout.reconfigure(line_buffering=True)
+sys.stderr.reconfigure(line_buffering=True)
+
+# =============================================================================
+# 服务配置（按需修改）
+# =============================================================================
+SERVER_HOST = "127.0.0.1"
+SERVER_PORT = 8000
+MODEL_NAME = "qwen36"
+TOKENIZER_PATH = "/root/flagrelease/qwen3.6/Qwen3.6-27B"
+
+# =============================================================================
+# 默认测试参数
+# =============================================================================
+DEFAULT_INPUT_LEN = 4096
+DEFAULT_OUTPUT_LEN = 1024
+DEFAULT_CONCURRENCY = 64
+WARMUP_ROUNDS = 4
+TOTAL_ROUNDS = 10
+OUTPUT_DIR = Path("./unit_test_output")
+
+# 输出解析正则表达式
+METRIC_PATTERNS = {
+    "Successful requests": r"Successful requests:\s+(\d+)",
+    #"Failed requests": r"Failed requests:\s+(\d+)",
+    #"Benchmark duration (s)": r"Benchmark duration \(s\):\s+([\d.]+)",
+    #"Total input tokens": r"Total input tokens:\s+(\d+)",
+    #"Total generated tokens": r"Total generated tokens:\s+(\d+)",
+    #"Request throughput (req/s)": r"Request throughput \(req/s\):\s+([\d.]+)",
+    "Output token throughput (tok/s)": r"Output token throughput \(tok/s\):\s+([\d.]+)",
+    "Total token throughput (tok/s)": r"Total token throughput \(tok/s\):\s+([\d.]+)",
+    #"Mean TTFT (ms)": r"Mean TTFT \(ms\):\s+([\d.]+)",
+    "Median TTFT (ms)": r"Median TTFT \(ms\):\s+([\d.]+)",
+    #"P99 TTFT (ms)": r"P99 TTFT \(ms\):\s+([\d.]+)",
+    #"Mean TPOT (ms)": r"Mean TPOT \(ms\):\s+([\d.]+)",
+    "Median TPOT (ms)": r"Median TPOT \(ms\):\s+([\d.]+)",
+    #"P99 TPOT (ms)": r"P99 TPOT \(ms\):\s+([\d.]+)",
+    #"Mean ITL (ms)": r"Mean ITL \(ms\):\s+([\d.]+)",
+    "Median ITL (ms)": r"Median ITL \(ms\):\s+([\d.]+)",
+    #"P99 ITL (ms)": r"P99 ITL \(ms\):\s+([\d.]+)",
+}
+
+
+def parse_output(output):
+    """解析 vllm bench serve 的输出文本，提取所有指标"""
+    metrics = {}
+    for key, pattern in METRIC_PATTERNS.items():
+        match = re.search(pattern, output)
+        if match:
+            val = match.group(1)
+            metrics[key] = float(val) if "." in val else int(val)
+        else:
+            metrics[key] = None
+    return metrics
+
+
+def build_command(input_len, output_len, concurrency):
+    """构建 vllm bench serve 命令"""
+    return [
+        "vllm",
+        "bench",
+        "serve",
+        "--host",
+        SERVER_HOST,
+        "--port",
+        str(SERVER_PORT),
+        "--model",
+        MODEL_NAME,
+        "--tokenizer",
+        TOKENIZER_PATH,
+        "--dataset-name",
+        "random",
+        "--random-input-len",
+        str(input_len),
+        "--random-output-len",
+        str(output_len),
+        "--endpoint",
+        "/v1/completions",
+        "--ignore-eos",
+        "--trust-remote-code",
+        "--num-prompts",
+        str(concurrency),
+        "--max-concurrency",
+        str(concurrency),
+        "--seed",
+        "0",
+    ]
+
+
+def run_single_test(round_num, input_len, output_len, concurrency):
+    """执行单轮测试并返回所有解析后的指标"""
+    #print(f"\n--- 第 {round_num}/{TOTAL_ROUNDS} 轮 ---")
+    cmd = build_command(input_len, output_len, concurrency)
+    #print(f"  执行命令: {' '.join(cmd)}")
+
+    try:
+        result = subprocess.run(cmd, capture_output=True, text=True, check=True)
+        metrics = parse_output(result.stdout)
+        #print(f"  第 {round_num} 轮完成")
+        #for key, val in metrics.items():
+        #    if val is not None:
+        #        print(f"    {key}: {val}")
+        return metrics
+    except subprocess.CalledProcessError as e:
+        print(f"  错误: 第 {round_num} 轮执行失败")
+        print(f"  错误输出: {e.stderr[:20000]}")
+        return None
+    except FileNotFoundError:
+        print("  错误: vllm 命令未找到，请确认 vllm 已安装且在 PATH 中")
+        return None
+    except Exception as e:
+        print(f"  错误: 第 {round_num} 轮发生未知错误: {e}")
+        return None
+
+
+def compute_average(all_round_metrics):
+    """对后面轮的数值型指标取平均"""
+    last = [m for m in all_round_metrics[WARMUP_ROUNDS:] if m is not None]
+    if not last:
+        return None
+
+    avg = {}
+    for key in METRIC_PATTERNS:
+        values = [m[key] for m in last if m.get(key) is not None]
+        if values:
+            avg[key] = sum(values) / len(values)
+        else:
+            avg[key] = None
+    return avg
+
+
+def print_and_save_results(all_round_metrics, avg_metrics, input_len, output_len, concurrency):
+    """打印每轮完整结果 + 后4轮平均值，并保存到文件"""
+    OUTPUT_DIR.mkdir(parents=True, exist_ok=True)
+    timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
+    filename = f"perf_results_in{input_len}_out{output_len}_c{concurrency}_{timestamp}.txt"
+    filepath = OUTPUT_DIR / filename
+
+    lines = []
+
+    def out(text=""):
+        print(text)
+        lines.append(text)
+
+    #out("=" * 70)
+    #out("vLLM 性能测试结果")
+    #out(f"  服务: {SERVER_HOST}:{SERVER_PORT}")
+    #out(f"  模型: {MODEL_NAME}")
+    #out(f"  输入长度: {input_len}, 输出长度: {output_len}, 并发数: {concurrency}")
+    #out(f"  总轮数: {TOTAL_ROUNDS} (前{WARMUP_ROUNDS}轮为warmup，此后取平均)")
+    #out(f"  执行时间: {datetime.now().strftime('%Y-%m-%d %H:%M:%S')}")
+    #out("=" * 70)
+
+    for i, metrics in enumerate(all_round_metrics):
+        #round_num = i + 1
+        #label = "（warmup，不计入平均）" if round_num <= WARMUP_ROUNDS else ""
+        #out(f"\n--- 第 {round_num} 轮 {label} ---")
+        if metrics is None:
+            out("  [FAILED]")
+            continue
+        #for key, val in metrics.items():
+        #    if val is not None:
+        #        out(f"  {key}: {val}")
+
+    out("\n" + "=" * 70)
+    out("结果平均值")
+    out("=" * 70)
+    if avg_metrics is None:
+        out("  无有效数据，无法计算平均值")
+    else:
+        for key, val in avg_metrics.items():
+            if val is not None:
+                out(f"  {key}: {val:.2f}")
+
+    with open(filepath, "w", encoding="utf-8") as f:
+        f.write("\n".join(lines) + "\n")
+
+    print(f"\n结果已保存至: {filepath}")
+
+
+def main():
+    parser = argparse.ArgumentParser(description="vLLM 性能测试脚本")
+    parser.add_argument("--input-len", type=int, default=None, help=f"输入长度 (默认: {DEFAULT_INPUT_LEN})")
+    parser.add_argument("--output-len", type=int, default=None, help=f"输出长度 (默认: {DEFAULT_OUTPUT_LEN})")
+    parser.add_argument("--concurrency", type=int, default=None, help=f"并发数 (默认: {DEFAULT_CONCURRENCY})")
+    parser.add_argument("--dry-run", action="store_true", help="仅打印命令，不执行")
+    args = parser.parse_args()
+
+    input_len = args.input_len if args.input_len is not None else DEFAULT_INPUT_LEN
+    output_len = args.output_len if args.output_len is not None else DEFAULT_OUTPUT_LEN
+    concurrency = args.concurrency if args.concurrency is not None else DEFAULT_CONCURRENCY
+
+    print("=" * 70)
+    print("vLLM 性能测试")
+    #print(f"  服务: {SERVER_HOST}:{SERVER_PORT}")
+    #print(f"  模型: {MODEL_NAME}")
+    print(f"  输入长度: {input_len}, 输出长度: {output_len}, 并发数: {concurrency}")
+    print(f"  总轮数: {TOTAL_ROUNDS} (前{WARMUP_ROUNDS}轮warmup，此后取平均)")
+    #print("=" * 70)
+
+    if args.dry_run:
+        print("[DRY RUN MODE]")
+        cmd = build_command(input_len, output_len, concurrency)
+        print(f"  命令: {' '.join(cmd)}")
+        print(f"  将执行 {TOTAL_ROUNDS} 轮")
+        print("\n[DRY RUN] 脚本验证完成，未实际执行测试。")
+        return
+
+    all_round_metrics = []
+    for round_num in range(1, TOTAL_ROUNDS + 1):
+        metrics = run_single_test(round_num, input_len, output_len, concurrency)
+        if metrics is None:
+            sys.exit(1)
+        all_round_metrics.append(metrics)
+
+    avg_metrics = compute_average(all_round_metrics)
+    print_and_save_results(all_round_metrics, avg_metrics, input_len, output_len, concurrency)
+
+
+if __name__ == "__main__":
+    main()

python/setup_tools/utils/aipu.py …rk/mthreads/mthreads3.6-qwen3.6/clear.shpython/setup_tools/utils/aipu.py renamed to .github/workflows/benchmark/mthreads/mthreads3.6-qwen3.6/clear.sh python/setup_tools/utils/aipu.py renamed to .github/workflows/benchmark/mthreads/mthreads3.6-qwen3.6/clear.sh

@@ -1,3 +1,5 @@
+#!/bin/bash
+
 # Copyright 2025-     FlagOS Contributors
 #
 # Permission is hereby granted, free of charge, to any person obtaining a copy
@@ -18,10 +20,6 @@
 # OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 # SOFTWARE.
 
-
-def precompile_hook(*args, **kargs):
-    default_backends = kargs["default_backends"]
-    default_backends_list = [*default_backends, "flir"]
-    kargs["default_backends"] = tuple(default_backends_list)
-    default_backends = tuple(default_backends_list)
-    return default_backends
+rm -r ~/.triton/cache
+rm -r ~/.flaggems
+rm -r /tmp/torchinductor_root