Replace main.py with vLLM adapt-task batch submitter (fanyi account)

Port logic from submit_validation_only_ppu_zhoukaile.py into the
strategy-deployable framework (health/status HTTP server, STRATEGY_ID
env, SIGTERM handling). Uses /api/adapt/task/add with xc-Token auth.

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

@@ -1,12 +1,13 @@
-# xc_validation_strategy
+# xc_validation_strategy_vllm_submit
 
-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。
+批量向 ModelHub XC 平台提交 vLLM 模型适配任务的策略服务（fanyi 账号），
+之后保持 HTTP 服务存活供平台探活。
 
 ## 功能
 
-- 自动登录 ModelHub 获取 Token
+- 通过 `/api/adapt/task/add` 接口（xc-Token 认证）批量提交模型适配任务
-- 批量提交模型验证任务（vLLM 框架，Cambricon MLU-370-x8）
+- vLLM 框架，ppu_zw_810e GPU
-- 提交结果写入 `submitted_validation_tasks.txt`
+- 提交成功的模型写入 `submitted_adapt_tasks.txt`
 - 暴露 `/health` 和 `/status` 接口满足平台运行时契约
 
 ## 项目结构
@@ -16,7 +17,7 @@
 ├── 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
+└── submitted_adapt_tasks.txt   # 运行后自动生成，记录提交结果
 ```
 
 ## 平台契约说明

main.py

@@ -1,8 +1,11 @@
 """
-xc_validation_strategy — 主入口
+xc_validation_strategy_vllm_submit — 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。
+启动后通过 /api/adapt/task/add 接口（xc-Token 认证）批量提交
+vLLM 模型适配任务（ppu_zw_810e），之后保持 HTTP 服务存活。
 同时暴露 /health（K8s 探活）和 /status（运行状态）。
+
+部署框架与 xc_validation_strategy 一致。
 """
 
 import json
@@ -11,24 +14,29 @@ import signal
 import threading
 from datetime import datetime
 from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
-from typing import List, Tuple
+from typing import List
 
 import requests
 
 # ══════════════════════════════════════════════════════════
-#  配置（全部从环境变量读取，不硬编码敏感信息）
+#  配置
 # ══════════════════════════════════════════════════════════
 BASE_URL          = os.environ.get("BASE_URL", "https://modelhub.org.cn")
-SUBMIT_ENDPOINT    = "/adminApi/async/task/create-contest-task"
+ADD_TASK_ENDPOINT = "/api/adapt/task/add"
 
-# 通过 curl -X POST https://modelhub.org.cn/adminApi/user/login 获取后填入
+# fanyi 账号的 xc-Token（该接口使用 xc-Token 认证，无需登录）
-AUTH_TOKEN        = "eyJ0eXAiOiJKV1QiLCJhbGciOiJIUzI1NiJ9.eyJ1c2VyQWNjb3VudCI6Inpob3VzaGFzaGEiLCJpZCI6MTQsInVzZXJSb2xlIjoibGVhZGVyYm9hcmQiLCJleHAiOjE3ODE4NTE0NzcsImlhdCI6MTc4MTI0NjY3N30.p3uvCpG50aLNifNVVXxvzmWJahbLM5K1671FVCtj8E8"
+USER_ACCOUNT = "fanyi"
-CONTEST_API_TOKEN = "ef1ef82f3c9efee413d602345fbe224d"
+XC_TOKEN     = "f2d501c9ae6543a589cd6cb789108c41"
-CONTRIBUTORS      = "zhoushasha"
+
-GPU_TYPE          = "Cambricon_mlu-370-x8"
+GPU_TYPE    = "ppu_zw_810e"
 TASK_TYPE   = "text-generation"
 STRATEGY_ID = os.environ.get("STRATEGY_ID", "")  # 平台自动注入，无需修改
 
+HEADERS = {
+    "Content-Type": "application/json",
+    "xc-Token": XC_TOKEN,
+}
+
 HTTP_HOST = "0.0.0.0"
 HTTP_PORT = 8080
 
@@ -36,26 +44,71 @@ HTTP_PORT = 8080
 #  模型列表
 # ══════════════════════════════════════════════════════════
 ALL_MODEL_IDS = [
-    "AI-ModelScope/gemma-2b",
+
-    "AI-ModelScope/falcon-mamba-7b",
+    "Hyeji0101/qwen2_5_1_5b_demo",
-    "katanemo/deepseek-2",
+    "GenueAI/geode-onyx",
-    "OpenBMB/MiniCPM4-0.5B",
+    "GM77/qwen3-4b-verilog-grpo",
-    "NousResearch/Meta-Llama-3-8B-Instruct",
+    "ChuGyouk/F_R13_T2",
-    "MediaTek-Research/Breeze-7B-Instruct-v1_0",
+    "ChuGyouk/R17",
-    "QLUNLP/BianCang-Qwen2.5-7B-Instruct",
+    "Ingingdo/bit-0.5b-final-logic",
-    "OpenBMB/MiniCPM4-Survey",
+    "beomi/Llama-3-Open-Ko-8B",
-    "OpenBMB/MiniCPM4-8B",
+    "Fiscus/trinitite_safe_rl_base_model",
-    "PaddlePaddle/ERNIE-4.5-0.3B-PT",
+    "ChuGyouk/F_R12_T3",
-    "LLM-Research/Llama-Guard-3-8B",
+    "ChuGyouk/F_R12_T2",
-    "OpenBMB/MiniCPM-2B-dpo-fp16",
+    "xw1234gan/cnk12_Main_fixed_SFTanchor_3B_step_9",
-    "OpenBMB/MiniCPM4.1-8B",
+    "xw1234gan/cnk12_Main_fixed_SFTanchor_3B_step_10",
-    "Cylingo/Xinyuan-LLM-14B-0428",
+    "xw1234gan/cnk12_Main_fixed_BaseAnchor_3B_step_1",
-    "Fengshenbang/Ziya-LLaMA-13B-v1",
+    "kmseong/llama3_2_3b-instruct-math-safedelta-scale0.99",
-    "baichuan-inc/Baichuan2-13B-Chat",
+    "opencompass/anah-v2",
-    "LLM-Research/gemma-2-9b-it",
+    "ChuGyouk/R14",
-    "Qwen/CodeQwen1.5-7B-Chat",
+    "trishajean/qwen-math-cebuano-1.5b-merged",
-    "OpenBMB/cpm-bee-10b",
+    "GyanAISystems/Gyan-AI-G1-Official",
-    "OpenBMB/MiniCPM3-4B",
+    "Divij/Qwen2.5-3B-Instruct-sft-without-thoughts",
+    "Divij/Qwen2.5-3B-Instruct-sft-with-thoughts",
+    "ChuGyouk/R5_1",
+    "ChuGyouk/R18_1",
+    "ChuGyouk/R19_1",
+    "ChuGyouk/R12",
+    "ChuGyouk/F_R11_T4",
+    "ChuGyouk/F_R12",
+    "ChuGyouk/F_R11_T2",
+    "ChuGyouk/F_R11_T3",
+    "ChuGyouk/F_R13_1_T1",
+    "ChuGyouk/F_R12_T4",
+    "automerger/T3qm7xNeuralsirkrishna-7B",
+    "Ford91/clifford-ai-v2",
+    "ChuGyouk/R16_1",
+    "ChuGyouk/R15_1",
+    "nkatara/gita-text-generation-gpt2",
+    "HINT-lab/Qwen2.5-7B-Instruct-Self-Calibration",
+    "thirdeyeai/Qwen2.5-1.5B-Instruct-uncensored",
+    "karaselerm/qwen2.5-1.5b-instruct-ru-abliterated-hw6",
+    "xw1234gan/cnk12_Main_fixed_BaseAnchor_3B_step_2",
+    "ontocord/wide_3b_sft_stage1.1-ss1-with_intr_math.no_issue",
+    "mncai/Foundation_Law_epoch4",
+    "gauri0508/med-record-audit-qwen2.5-3b-grpo",
+    "unsloth/Phi-4-mini-instruct",
+    "E-motionAssistant/qwen-2.5-3b-tamil-therapy-merged",
+    "EscapeJeju/qwen2_5_1_5b_demo",
+    "AgPerry/Qwen3-8B-fim-v2v3pt-swe-lego-posttrain",
+    "ChuGyouk/F_R11",
+    "ChuGyouk/F_R11_1_T1",
+    "LorenaYannnnn/general_reward-Qwen3-0.6B-OURS_self-seed_1",
+    "Vortex5/Crimson-Constellation-12B",
+    "cloudyu/mistral_11B_instruct_v0.1",
+    "pkupie/Qwen2.5-3B-ug-cpt",
+    "iproskurina/qwen-hf-fewshot-iter-np-iter3",
+    "ontocord/wide_3b_sft_stage1.2-ss1-expert_wiki",
+    "kmseong/llama3_2_3b-instruct-math-safedelta-scale2",
+    "Thrillcrazyer/Qwen-2.5-1.5B_TAC_Teacher_Qwen32B",
+    "nyu-dice-lab/VeriThoughts-Reasoning-7B",
+    "ontocord/wide_3b",
+    "silvercoder67/Mistral-7b-instruct-v0.2-summ-sft-e2m",
+    "lihaoxin2020/qwen3-4b-sft-gpt54-ep2-instance-rubric-gpt54-step200",
+    "lihaoxin2020/qwen3-4b-sft-gpt54-ep2-instance-rubric-gpt54-step150",
+    "lihaoxin2020/qwen3-4b-sft-gpt54-ep2-evolving-rubric-gem3-flash-step150",
+    "Guilherme34/Firefly-V3",
+
 ]
 
 # ══════════════════════════════════════════════════════════
@@ -108,98 +161,114 @@ def _run_http():
 # ══════════════════════════════════════════════════════════
 #  业务逻辑
 # ══════════════════════════════════════════════════════════
-def _submit_task(token: str, model_id: str) -> Tuple[bool, str]:
+def submit_task(model_id: str) -> bool:
-    headers = {
+    config_content = f"""
-        "Content-Type": "application/json",
+gpu_type: ppu_zw_810e
-        "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
-nv_docker_image: harbor.4pd.io/dooke/vllm/vllm/vllm-openai:v0.11.0
 framework: vllm
-storage: gpfs
+docker_image: harbor.4pd.io/hardcore-tech/asllm:1.10.1-pytorch2.10.0-ubuntu24.04-sail2.1.0-cuda13.0-sglang0.5.10-vllm0.19.0-py312
-modelhub_options:
+nv_docker_image: harbor-contest.4pd.io/sunruoxi/vllm-openai-fix-tokenizer:v0.11.0
-  srcRelativePath: leaderboard/modelHubXC/{model_id}
+
-  mountPoint: /model
 sut_config:
   values:
     gpu_num: 1
     env:
-      - name: MAX_MODEL_LEN
+      - name: test
-        value: 8192
+        value: fp16
-    command: ["vllm", "serve", "/model", "--port", "8000", "--served-model-name", "llm", "--max-model-len", "8192", "--trust-remote-code", "--dtype", "float16"]
+    command:
+      - bash
+      - /opt/t-head/entrypoint.sh
+      - python3
+      - -m
+      - asllm.entrypoints.api_server
+      - --model
+      - /model
+      - --port
+      - '30000'
+      - --host
+      - 0.0.0.0
+      - --served-model-name
+      - llm
 ref_config:
   values:
-    cpu_num: 2
     gpu_num: 1
     env:
-      - name: MAX_MODEL_LEN
+      - name: test
-        value: 8192
+        value: fp16
-    command: ["vllm", "serve", "/model", "--port", "80", "--served-model-name", "llm", "--max-model-len", "8192", "--trust-remote-code", "--dtype", "float16"]
+    command:
+      - vllm
+      - serve
+      - /model
+      - --port
+      - '80'
+      - --served-model-name
+      - llm
+      - --max-model-len
+      - '2048'
+      - --gpu-memory-utilization
+      - '0.9'
+      - --enforce-eager
+      - --trust-remote-code
+      - -tp
+      - '1'
 """
+
     payload = {
-        "contestApiToken": CONTEST_API_TOKEN,
+        "configParams": config_content,
-        "contributors":    CONTRIBUTORS,
-        "gpuTypes":        [GPU_TYPE],
-        "taskType":        TASK_TYPE,
-        "modelId":         model_id,
         "framework": "vllm",
-        "strategyId":      STRATEGY_ID,   # 平台要求
+        "modelAddress": f"https://huggingface.co/{model_id}",
-        "submissionConfig": [{
+        "targetGpu": GPU_TYPE,
-            "config":   config_content,
-            "gpuType":  GPU_TYPE,
         "taskType": TASK_TYPE,
-        }],
+        "strategyId": STRATEGY_ID,   # 平台要求；若接口不支持该字段会被忽略
     }
+
+    print(f"📤 提交任务: {model_id}", flush=True)
     try:
         resp = requests.post(
-            BASE_URL + SUBMIT_ENDPOINT,
+            BASE_URL + ADD_TASK_ENDPOINT,
-            headers=headers,
+            headers=HEADERS,
             json=payload,
-            timeout=15,
+            timeout=30,
         )
+        print(f"status: {resp.status_code}", flush=True)
         result = resp.json()
+        print(result, flush=True)
         if result.get("code") == 0:
-            task_id = result.get("data", {}).get("id", "")
+            print(f"✅ 提交成功: {model_id}", flush=True)
-            print(f"[worker] OK  {model_id}  task_id={task_id}", flush=True)
+            return True
-            return True, task_id
         else:
-            print(f"[worker] FAIL {model_id}: {result.get('message')}", flush=True)
+            print(f"❌ 提交失败: {result.get('message')}", flush=True)
-            return False, ""
+            return False
     except Exception as e:
-        print(f"[worker] ERROR {model_id}: {e}", flush=True)
+        print(f"💥 异常 ({model_id}): {e}", flush=True)
-        return False, ""
+        return False
 
 
 def _run_worker():
     _state["started_at"] = datetime.utcnow().isoformat()
     _state["phase"] = "submitting"
 
-    successful: List[Tuple[str, str]] = []
+    successful: List[str] = []
-    token = AUTH_TOKEN
-    print("[worker] 使用预设 Token，跳过登录", flush=True)
-
     for model_id in ALL_MODEL_IDS:
         if _shutdown.is_set():
             break
-        ok, task_id = _submit_task(token, model_id)
+        if submit_task(model_id):
-        if ok:
             _state["submitted"] += 1
-            successful.append((task_id, model_id))
+            successful.append(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:
+        with open("submitted_adapt_tasks.txt", "w", encoding="utf-8") as f:
-            for tid, mid in successful:
+            for mid in successful:
-                f.write(f"{tid}\t{mid}\n")
+                f.write(f"{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']}  total={_state['total']}",
         flush=True,
     )
     # 提交完成后继续保持进程存活，等待平台停止