add Kunlunxin_p-800 as 4th GPU with filtered model list

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

.gitignore

@@ -0,0 +1,2 @@
+.DS_Store
+__pycache__/

Dockerfile

@@ -2,6 +2,7 @@ FROM modelhubxc-4pd.tencentcloudcr.com/xc_agent_platform/python:3.11-slim
 
 ENV PYTHONUNBUFFERED=1 
     
+
 WORKDIR /app
 
 COPY requirements.txt .

README.md

@@ -1,5 +1,29 @@
 # xc_validation_strategy
 
-信创自动化模型适配平台 — 验证策略服务
+批量向 ModelHub XC 平台提交模型验证任务的策略服务，之后保持 HTTP 服务存活供平台探活。
 
-从 HuggingFace 周期性抓取新模型，自动完成同步、下载、提交验证任务的全流程，常驻运行在 xc_agent_platform 上。
+## 功能
+
+- 自动登录 ModelHub 获取 Token
+- 批量提交模型验证任务（vLLM 框架，Cambricon MLU-370-x8）
+- 提交结果写入 `submitted_validation_tasks.txt`
+- 暴露 `/health` 和 `/status` 接口满足平台运行时契约
+
+## 项目结构
+
+```
+.
+├── main.py            # 主入口：HTTP 服务 + 提交逻辑
+├── Dockerfile         # 平台镜像构建配置
+├── requirements.txt   # Python 依赖
+└── submitted_validation_tasks.txt   # 运行后自动生成，记录提交结果
+```
+
+## 平台契约说明
+
+本项目满足平台对策略镜像的全部必要约束：
+
+- Dockerfile 位于仓库根目录，基于官方轻量基础镜像
+- 暴露 8080 端口并实现 `GET /health`
+- 通过环境变量 `STRATEGY_ID` 获取策略 ID
+- 正确处理 `SIGTERM` 信号，支持优雅停机

main.py

@@ -1,7 +1,9 @@
 """
 xc_validation_strategy — 主入口
 
-启动后执行一次模型验证任务批量提交，之后保持 HTTP 服务存活。
+启动后针对 4 张 GPU 卡（Biren_166m / Cambricon_mlu-370-x8 / MetaX_c-500 /
+Kunlunxin_p-800）分别批量提交各自筛选出的模型验证任务（/adminApi/async/task/create-contest-task，
+Bearer Token 认证），之后保持 HTTP 服务存活。
 同时暴露 /health（K8s 探活）和 /status（运行状态）。
 """
 
@@ -9,7 +11,6 @@ import json
 import os
 import signal
 import threading
-import traceback
 from datetime import datetime
 from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
 from typing import List, Tuple
@@ -20,55 +21,153 @@ import requests
 #  配置（全部从环境变量读取，不硬编码敏感信息）
 # ══════════════════════════════════════════════════════════
 BASE_URL           = os.environ.get("BASE_URL", "https://modelhub.org.cn")
-LOGIN_ENDPOINT     = "/adminApi/user/login"
 SUBMIT_ENDPOINT    = "/adminApi/async/task/create-contest-task"
 
-USER_ACCOUNT       = os.environ["USER_ACCOUNT"]        # 必填
-USER_PASSWORD      = os.environ["USER_PASSWORD"]        # 必填
-CONTEST_API_TOKEN  = os.environ["CONTEST_API_TOKEN"]    # 必填
-STRATEGY_ID        = os.environ.get("STRATEGY_ID", "")  # 平台注入
-CONTRIBUTORS       = os.environ.get("CONTRIBUTORS", USER_ACCOUNT)
-GPU_TYPE           = os.environ.get("GPU_TYPE", "Cambricon_mlu-370-x8")
-TASK_TYPE          = os.environ.get("TASK_TYPE", "text-generation")
+# 通过 curl -X POST https://modelhub.org.cn/adminApi/user/login 获取后填入
+AUTH_TOKEN        = "eyJ0eXAiOiJKV1QiLCJhbGciOiJIUzI1NiJ9.eyJ1c2VyQWNjb3VudCI6Inpob3VzaGFzaGEiLCJpZCI6MTQsInVzZXJSb2xlIjoibGVhZGVyYm9hcmQiLCJleHAiOjE3ODU3NDY3NTMsImlhdCI6MTc4NTE0MTk1M30.KwUuefNAFSNwq3_Pnaw2nef8ZC6WgsECQ_LMeQnKk2c"
+CONTEST_API_TOKEN = "ef1ef82f3c9efee413d602345fbe224d"
+CONTRIBUTORS      = "zhoushasha"
+TASK_TYPE         = "text-generation"
+STRATEGY_ID       = os.environ.get("STRATEGY_ID", "")  # 平台自动注入，无需修改
 
 HTTP_HOST = "0.0.0.0"
 HTTP_PORT = 8080
 
 # ══════════════════════════════════════════════════════════
-#  模型列表
+#  各 GPU 的模型列表（来自 filter_verified_models 脚本的筛选结果）
 # ══════════════════════════════════════════════════════════
-ALL_MODEL_IDS = [
-    "AI-ModelScope/gemma-2b",
-    "AI-ModelScope/falcon-mamba-7b",
-    "katanemo/deepseek-2",
-    "OpenBMB/MiniCPM4-0.5B",
-    "NousResearch/Meta-Llama-3-8B-Instruct",
-    "MediaTek-Research/Breeze-7B-Instruct-v1_0",
-    "QLUNLP/BianCang-Qwen2.5-7B-Instruct",
-    "OpenBMB/MiniCPM4-Survey",
-    "OpenBMB/MiniCPM4-8B",
-    "PaddlePaddle/ERNIE-4.5-0.3B-PT",
-    "LLM-Research/Llama-Guard-3-8B",
-    "OpenBMB/MiniCPM-2B-dpo-fp16",
-    "OpenBMB/MiniCPM4.1-8B",
-    "Cylingo/Xinyuan-LLM-14B-0428",
-    "Fengshenbang/Ziya-LLaMA-13B-v1",
-    "baichuan-inc/Baichuan2-13B-Chat",
-    "LLM-Research/gemma-2-9b-it",
-    "Qwen/CodeQwen1.5-7B-Chat",
-    "OpenBMB/cpm-bee-10b",
-    "OpenBMB/MiniCPM3-4B",
+BIREN_MODELS = [
+    "BigRatz/LOL-AI-2026",
+    "aimeri/spoomplesmaxx-cardmaker-v1",
+    "Alibaba-DT/Logics-STEM-8B-SFT",
+    "Muneebmn123/insurance-voice-qwen25-1_5b",
+    "AnkitBirGurung/NEMO-12B-SFT-Further",
+    "danilarudenko/editorai-mini",
+    "EphemeralYou/Prompt-Refine-MiniCPM5-1B",
+    "mtepe01/mentorx-mistral-7b-automata-merged",
+    "DarkArtsForge/Helix-SCE-12B-jh",
+    "rpant/iolai26-solve",
+    "NithinAI12/NithinX-Omni-LLM-v1",
+    "ConvexAI/Luminex-34B-v0.2",
+    "codellama/CodeLlama-34b-hf",
+    "Lipas007/iol-ai-2026-qwen14b-awq",
+    "D-Z-W/finetuned-teacher",
+    "huan1999/ziya-llama-13b-medical-merged",
+    "codellama/CodeLlama-34b-Python-hf",
+    "ld4ad/gemma-2-9b-dunhuang",
+    "harindhar10/Olmo-7b_1M_Smiles_lora",
+    "allenai/Olmo-3-7B-Think-DPO",
+    "allenai/Olmo-3-32B-Think-DPO",
+    "RedHatAI/gemma-2-9b-it",
+    "prashanthsura/gemma-2-2b-legal-financial-sft",
+    "pfnet/plamo-2-8b",
+    "sail/Sailor2-20B-128K-SFT",
+    "facebook/layerskip-llama3.2-1B",
+    "Qwen/Qwen2.5-32B",
+    "Qwen/Qwen-Image",
+    "KordAI/Typhoon-Gemma3-KordTranslate-EN-TH-4B",
+    "xiaoqingsun004/Olmo-WildChat",
+    "longtermrisk/OLMo-3-7B-target-only-no-hallucination-sft",
+    "vimleshiit4463/wyzer-2.0-smollm2-135m",
+    "trl-lib/pythia-1b-deduped-tldr-sft",
 ]
 
+CAMBRICON_MODELS = [
+    "clear-blue-sky/evolai-reborn-tfm-008",
+    "clear-blue-sky/evolai-reborn-tfm-010",
+    "BigRatz/LOL-AI-2026",
+    "clear-blue-sky/evolai-reborn-tfm-001",
+    "clear-blue-sky/evolai-reborn-tfm-019",
+    "clear-blue-sky/evolai-reborn-tfm-007",
+    "aimeri/spoomplesmaxx-cardmaker-v1",
+    "aipatseer/inst_ft_qwen_0.6b_summ",
+    "clear-blue-sky/evolai-reborn-tfm-003",
+    "clear-blue-sky/evolai-reborn-tfm-004",
+    "Alibaba-DT/Logics-STEM-8B-SFT",
+    "clear-blue-sky/evolai-reborn-tfm-002",
+    "prism-ml/Ternary-Bonsai-4B-unpacked",
+    "Muneebmn123/insurance-voice-qwen25-1_5b",
+    "AnkitBirGurung/NEMO-12B-SFT-Further",
+    "danilarudenko/editorai-mini",
+    "rpant/iolai26-solve",
+]
+
+METAX_MODELS = [
+    "Phoenix9781/evolai-tf-model-105",
+    "clear-blue-sky/evolai-reborn-tfm-008",
+    "clear-blue-sky/evolai-reborn-tfm-010",
+    "BigRatz/LOL-AI-2026",
+    "clear-blue-sky/evolai-reborn-tfm-001",
+    "clear-blue-sky/evolai-reborn-tfm-019",
+    "clear-blue-sky/evolai-reborn-tfm-007",
+    "clear-blue-sky/evolai-reborn-tfm-005",
+    "Lin2es/evolai-tfm-03o",
+    "clear-blue-sky/evolai-reborn-tfm-009",
+    "aimeri/spoomplesmaxx-cardmaker-v1",
+    "aipatseer/inst_ft_qwen_0.6b_summ",
+    "reaperdoesntknow/DualMind-TKD-Agentic-1.7B",
+    "clear-blue-sky/evolai-reborn-tfm-011",
+    "clear-blue-sky/evolai-reborn-tfm-003",
+    "clear-blue-sky/evolai-reborn-tfm-004",
+    "clear-blue-sky/evolai-reborn-tfm-002",
+    "amd/ReasonLite-0.6B",
+    "prism-ml/Ternary-Bonsai-4B-unpacked",
+    "Muneebmn123/insurance-voice-qwen25-1_5b",
+    "danilarudenko/editorai-mini",
+    "rpant/iolai26-solve",
+]
+
+KUNLUNXIN_MODELS = [
+    "Patriae/patriae-cuban-dialect-model",
+    "Phoenix9781/evolai-tf-model-105",
+    "clear-blue-sky/evolai-reborn-tfm-008",
+    "clear-blue-sky/evolai-reborn-tfm-010",
+    "clear-blue-sky/evolai-reborn-tfm-001",
+    "clear-blue-sky/evolai-reborn-tfm-019",
+    "clear-blue-sky/evolai-reborn-tfm-007",
+    "clear-blue-sky/evolai-reborn-tfm-005",
+    "Lin2es/evolai-tfm-03o",
+    "clear-blue-sky/evolai-reborn-tfm-009",
+    "aimeri/spoomplesmaxx-cardmaker-v1",
+    "aipatseer/inst_ft_qwen_0.6b_summ",
+    "clear-blue-sky/evolai-reborn-tfm-011",
+    "clear-blue-sky/evolai-reborn-tfm-003",
+    "clear-blue-sky/evolai-reborn-tfm-004",
+    "clear-blue-sky/evolai-reborn-tfm-002",
+    "amd/ReasonLite-0.6B",
+    "prism-ml/Ternary-Bonsai-4B-unpacked",
+    "EthanGao123/CellHermes-v1.0",
+    "RecursiveMAS/Mixture-Science-BioMistral-7B",
+    "chenyitian-shanshu/SIRL-Gurobi",
+    "RedHatAI/gemma-2-9b-it",
+    "gaunernst/gemma-3-27b-it-qat-autoawq",
+    "promotion/qwen3-8b-simpo-avg-b2p5-g1p0-s42",
+    "llamaindex/vdr-2b-multi-v1",
+    "KordAI/Typhoon-Gemma3-KordTranslate-EN-TH-4B",
+    "TheDrummer/UnslopNemo-12B-v3",
+    "gradients-io-tournaments/tournament-tourn_c5d86c82ce819a79_20260706-b78a01d4-0a6a-49e1-9190-5e88ae329937-5DS6XMVr",
+    "ArliAI/Mistral-Nemo-12B-ArliAI-RPMax-v1.1",
+]
+
+# 按顺序处理：Biren → Cambricon → MetaX → Kunlunxin
+GPU_JOBS: List[Tuple[str, List[str]]] = [
+    ("Biren_166m",           BIREN_MODELS),
+    ("Cambricon_mlu-370-x8", CAMBRICON_MODELS),
+    ("MetaX_c-500",          METAX_MODELS),
+    ("Kunlunxin_p-800",      KUNLUNXIN_MODELS),
+]
+TOTAL_MODELS = sum(len(models) for _, models in GPU_JOBS)
+
 # ══════════════════════════════════════════════════════════
 #  全局状态（供 /status 展示）
 # ══════════════════════════════════════════════════════════
 _state = {
     "strategy_id":   STRATEGY_ID,
     "phase":         "starting",   # starting | submitting | done | error
-    "total":         len(ALL_MODEL_IDS),
+    "total":         TOTAL_MODELS,
     "submitted":     0,
     "failed":        0,
+    "per_gpu":       {gpu: 0 for gpu, _ in GPU_JOBS},
     "started_at":    None,
     "finished_at":   None,
 }
@@ -108,29 +207,40 @@ def _run_http():
     print("[http] 已关闭", flush=True)
 
 # ══════════════════════════════════════════════════════════
-#  业务逻辑
+#  各 GPU 的 config_content 模板
 # ══════════════════════════════════════════════════════════
-def _login() -> str:
-    headers = {"Content-Type": "application/json"}
-    resp = requests.post(
-        BASE_URL + LOGIN_ENDPOINT,
-        headers=headers,
-        json={"userAccount": USER_ACCOUNT, "userPassword": USER_PASSWORD},
-        timeout=30,
-    )
-    data = resp.json()
-    if data.get("code") != 0:
-        raise RuntimeError(f"登录失败: {data.get('message')}")
-    print("[worker] 登录成功", flush=True)
-    return data["data"]["token"]
-
-
-def _submit_task(token: str, model_id: str) -> Tuple[bool, str]:
-    headers = {
-        "Content-Type": "application/json",
-        "Authorization": f"Bearer {token}",
-    }
-    config_content = f"""docker_image: harbor.4pd.io/hardcore-tech/cambricon-mlu370-pytorch:v25.01-torch2.5.0-torchmlu1.24.1-ubuntu22.04-py310
+def build_config_content(gpu_type: str, model_id: str) -> str:
+    if gpu_type == "Biren_166m":
+        max_model_len = 4096
+        return f"""docker_image: git.modelhub.org.cn:9443/enginex/xc-llm-biren166m:26.01
+nv_docker_image: harbor.4pd.io/dooke/vllm/vllm/vllm-openai:v0.11.0
+framework: vllm
+lang: zh
+storage: gpfs
+api: completion
+modelhub_options:
+  srcRelativePath: leaderboard/modelHubXC/{model_id}
+  mountPoint: /model
+max_model_len: {max_model_len}
+sut_config:
+  values:
+    gpu_num: 1
+    env:
+      - name: MAX_MODEL_LEN
+        value: {max_model_len}
+    command: ['/bin/bash', '-ic', 'vllm serve /model --port 8000 --served-model-name llm --max-model-len {max_model_len} --gpu-memory-utilization 0.9 --enforce-eager --trust-remote-code -tp 1 --host 0.0.0.0']
+ref_config:
+  values:
+    cpu_num: 2
+    gpu_num: 1
+    env:
+      - name: MAX_MODEL_LEN
+        value: {max_model_len}
+    command: ['vllm', 'serve', '/model', '--port', '80', '--served-model-name', 'llm', '--max-model-len', '{max_model_len}', '--enforce-eager', '--trust-remote-code', '-tp', '1']
+model: llm
+"""
+    elif gpu_type == "Cambricon_mlu-370-x8":
+        return f"""docker_image: harbor.4pd.io/hardcore-tech/cambricon-mlu370-pytorch:v25.01-torch2.5.0-torchmlu1.24.1-ubuntu22.04-py310
 nv_docker_image: harbor.4pd.io/dooke/vllm/vllm/vllm-openai:v0.11.0
 framework: vllm
 storage: gpfs
@@ -153,20 +263,81 @@ ref_config:
         value: 8192
     command: ["vllm", "serve", "/model", "--port", "80", "--served-model-name", "llm", "--max-model-len", "8192", "--trust-remote-code", "--dtype", "float16"]
 """
+    elif gpu_type == "MetaX_c-500":
+        return f"""docker_image: git.modelhub.org.cn:9443/enginex-metax/vllm:0.9.1
+nv_docker_image: harbor.4pd.io/dooke/vllm/vllm/vllm-openai:v0.11.0
+framework: vllm
+lang: en
+storage: gpfs
+api: chat
+modelhub_options:
+  srcRelativePath: leaderboard/modelHubXC/{model_id}
+  mountPoint: /model
+max_model_len: 2048
+sut_config:
+  gpu_num: 1
+  values:
+    command: ['/opt/conda/bin/vllm', 'serve', '/model', '--port', '20644', '--served-model-name', 'llm',  '--max-model-len', '2048', '--gpu-memory-utilization', '0.9',  '--enforce-eager', '--trust-remote-code' ,'-tp', '1']
+ref_config:
+  gpu_num: 1
+  values:
+    command: ['vllm', 'serve', '/model', '--port', '80', '--served-model-name', 'llm', '--max-model-len', '2048', '--enforce-eager', '--trust-remote-code', '-tp', '1']
+"""
+    elif gpu_type == "Kunlunxin_p-800":
+        return f"""docker_image: git.modelhub.org.cn:9443/enginex/xc-llm-kunlun
+nv_docker_image: harbor.4pd.io/dooke/vllm/vllm/vllm-openai:v0.11.0
+framework: vllm
+lang: en
+storage: gpfs
+api: chat
+temperature: 0.4
+repetition_penalty: 1.1
+top_p: 0.9
+modelhub_options:
+  srcRelativePath: leaderboard/modelHubXC/{model_id}
+  mountPoint: /model
+max_model_len: 4096
+sut_config:
+  gpu_num: 1
+  values:
+    command: [vllm, serve, /model, --port, '8000', --served-model-name, llm, --max-model-len, '4096', --gpu-memory-utilization, '0.9', --enforce-eager, --trust-remote-code, -tp, '1']
+ref_config:
+  gpu_num: 1
+  values:
+    command: [vllm, serve, /model, --port, '80', --served-model-name, llm, --max-model-len, '4096', --enforce-eager, --trust-remote-code, -tp, '1']
+"""
+    else:
+        raise ValueError(f"未知的 GPU_TYPE: {gpu_type}")
+    
+    
+    
+    
+
+# ══════════════════════════════════════════════════════════
+#  业务逻辑
+# ══════════════════════════════════════════════════════════
+def _submit_task(token: str, gpu_type: str, model_id: str) -> Tuple[bool, str]:
+    headers = {
+        "Content-Type": "application/json",
+        "Authorization": f"Bearer {token}",
+    }
+    config_content = build_config_content(gpu_type, model_id)
+
     payload = {
         "contestApiToken": CONTEST_API_TOKEN,
         "contributors":    CONTRIBUTORS,
-        "gpuTypes":        [GPU_TYPE],
+        "gpuTypes":        [gpu_type],
         "taskType":        TASK_TYPE,
         "modelId":         model_id,
         "framework":       "vllm",
         "strategyId":      STRATEGY_ID,   # 平台要求
         "submissionConfig": [{
             "config":   config_content,
-            "gpuType":  GPU_TYPE,
+            "gpuType":  gpu_type,
             "taskType": TASK_TYPE,
         }],
     }
+    print(f"[payload] gpu={gpu_type} model={model_id}", flush=True)
     try:
         resp = requests.post(
             BASE_URL + SUBMIT_ENDPOINT,
@@ -177,13 +348,13 @@ ref_config:
         result = resp.json()
         if result.get("code") == 0:
             task_id = result.get("data", {}).get("id", "")
-            print(f"[worker] OK  {model_id}  task_id={task_id}", flush=True)
+            print(f"[worker] OK  {model_id} (GPU={gpu_type})  task_id={task_id}", flush=True)
             return True, task_id
         else:
-            print(f"[worker] FAIL {model_id}: {result.get('message')}", flush=True)
+            print(f"[worker] FAIL {model_id} (GPU={gpu_type}): {result.get('message')}", flush=True)
             return False, ""
     except Exception as e:
-        print(f"[worker] ERROR {model_id}: {e}", flush=True)
+        print(f"[worker] ERROR {model_id} (GPU={gpu_type}): {e}", flush=True)
         return False, ""
 
 
@@ -191,36 +362,39 @@ def _run_worker():
     _state["started_at"] = datetime.utcnow().isoformat()
     _state["phase"] = "submitting"
 
-    successful: List[Tuple[str, str]] = []
-    try:
-        token = _login()
-    except Exception:
-        traceback.print_exc()
-        _state["phase"] = "error"
-        return
+    successful: List[Tuple[str, str, str]] = []
+    token = AUTH_TOKEN
+    print("[worker] 使用预设 Token，跳过登录", flush=True)
 
-    for model_id in ALL_MODEL_IDS:
+    for gpu_type, model_list in GPU_JOBS:
         if _shutdown.is_set():
             break
-        ok, task_id = _submit_task(token, model_id)
-        if ok:
-            _state["submitted"] += 1
-            successful.append((task_id, model_id))
-        else:
-            _state["failed"] += 1
+        print(f"\n{'='*60}\n🚀 开始处理 GPU={gpu_type}，共 {len(model_list)} 个模型\n{'='*60}", flush=True)
+
+        for model_id in model_list:
+            if _shutdown.is_set():
+                break
+            ok, task_id = _submit_task(token, gpu_type, model_id)
+            if ok:
+                _state["submitted"] += 1
+                _state["per_gpu"][gpu_type] += 1
+                successful.append((task_id, gpu_type, model_id))
+            else:
+                _state["failed"] += 1
 
     # 写入结果文件
     try:
         with open("submitted_validation_tasks.txt", "w", encoding="utf-8") as f:
-            for tid, mid in successful:
-                f.write(f"{tid}\t{mid}\n")
+            for tid, gpu, mid in successful:
+                f.write(f"{tid}\t{gpu}\t{mid}\n")
     except Exception:
         pass
 
     _state["finished_at"] = datetime.utcnow().isoformat()
     _state["phase"] = "done"
     print(
-        f"[worker] 完成  submitted={_state['submitted']}  failed={_state['failed']}",
+        f"[worker] 完成  submitted={_state['submitted']}  failed={_state['failed']}  "
+        f"total={_state['total']}  per_gpu={_state['per_gpu']}",
         flush=True,
     )
     # 提交完成后继续保持进程存活，等待平台停止