update

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

Dockerfile

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

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

@@ -9,7 +9,6 @@ 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
@@ -20,16 +19,15 @@ 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.eyJ1c2VyQWNjb3VudCI6Inpob3VzaGFzaGEiLCJpZCI6MTQsInVzZXJSb2xlIjoibGVhZGVyYm9hcmQiLCJleHAiOjE3ODQ1NDc1NDYsImlhdCI6MTc4Mzk0Mjc0Nn0.ZcOqcrfI22LPi4mGMnt164nZGhi61ZxtJGYsoO7fZdM"
+CONTEST_API_TOKEN = "ef1ef82f3c9efee413d602345fbe224d"
+CONTRIBUTORS      = "zhoushasha"
+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
 
 HTTP_HOST = "0.0.0.0"
 HTTP_PORT = 8080
@@ -38,26 +36,231 @@ HTTP_PORT = 8080
 #  模型列表
 # ══════════════════════════════════════════════════════════
 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",
+    "Alienpenguin10/M3PO-bahdanau-trial1-seed123",
+    "sujalrajpoot/TrueSyncAI-Aurion",
+    "prithivMLmods/Tureis-Qwen3_QWQ-4B-Exp",
+    "standrey/listing-parser-llama31-8b-ft-v1-full",
+    "zarakiquemparte/zarablend-l2-7b",
+    "linzju/Bio-Medical-Llama-3-8B_EnchTable_FFN",
+    "leonMW/Qwen3-4B-Thinking-2507-GSPO-Easy",
+    "longvideoagent/longvideoagent-qwen3-4b",
+    "ishikaa/acquisition_qwen3b_alpaca_proximity",
+    "01ai/Yi-9B-200K",
+    "Gille/StrangeMerges_33-7B-slerp",
+    "sstoica12/acquisition_llama-3_2-3b_bins_medmcqa_gradient",
+    "speechlessai/speechless-coding-7b-16k-tora",
+    "unsloth/Qwen2.5-Math-1.5B-Instruct",
+    "Yuma42/KangalKhan-Sapphire-7B",
+    "shadowml/BeagleSempra-7B",
+    "bralynn/test18",
+    "m-a-p/OProver-8B-Round1",
+    "yeen214/test_llama2_7b",
+    "Xwin-LM/Xwin-LM-7B-V0.2",
+    "FreedomIntelligence/AceGPT-13B",
+    "Edcastro/tinyllama-edcastr_JavaScript-v2",
+    "zarakiquemparte/zaraxe-l2-7b",
+    "MaziyarPanahi/Llama-3-8B-Instruct-v0.8",
+    "defog/sqlcoder2",
+    "SawinuCP/bus_booking_voice_agent_merged",
+    "j05hr3d/Llama-3.2-3B-Instruct-C_M_T-DOLLY-SEED999",
+    "Changgil/K2S3-SOLAR-11b-v1.0",
+    "agentica-org/DeepCoder-1.5B-Preview",
+    "prithivMLmods/Omni-Reasoner3-Merged",
+    "ibm-granite/granite-7b-instruct",
+    "zarakiquemparte/kuchiki-1.1-l2-7b",
+    "MaziyarPanahi/Llama-3-8B-Instruct-v0.1",
+    "Magpie-Align/Llama-3.1-8B-Magpie-Align-v0.1",
+    "prithivMLmods/Neumind-Math-7B-Instruct",
+    "dphn/dolphin-2.9.3-qwen2-1.5b",
+    "OpenBuddy/openbuddy-mistral-7b-v13",
+    "kairawal/Qwen3-4B-TL-SynthDolly-1A-E3",
+    "sahilnagaralu/movie-script",
+    "xw1234gan/SFT_Qwen2.5-1.5B-Instruct_cnk12",
+    "yunjae-won/ubq30i_qwen4b_sft_yl",
+    "1010happy/qwen3BInstruct_ClaudeDefault",
+    "yunjae-won/ubq30i_qwen4b_sft_both",
+    "willieseun/AIMO-Qwen2.5-Math-1.5B-Instruct-Finetuned",
+    "NousResearch/Yarn-Mistral-7b-64k",
+    "xiaolesu/OsmosisProofling-GRPO-NT",
+    "health360/Healix-410M",
+    "RUC-AIBOX/STILL-3-1.5B-preview",
+    "cjvt/GaMS-1B",
+    "Lansechen/Qwen2.5-7B-Open-R1-GRPO-math-lighteval-1epochstop-withformat",
+    "Charlie911/vicuna-7b-v1.5-general-temporal-merged",
+    "Kyleyee/cDPO_hh-seed5",
+    "prithivMLmods/Llama-3.2-3B-Math-Oct",
+    "Kyleyee/HINGE_hh-seed5",
+    "arcee-ai/Patent-Instruct-7b",
+    "jb723/cross_lingual_epoch2",
+    "davidkim205/komt-mistral-7b-v1",
+    "kalisai/Nusantara-1.8b-Indo-Chat",
+    "prithivMLmods/Llama-8B-Distill-CoT",
+    "Kyleyee/HINGE_hh-seed3",
+    "simplescaling/s1.1-1.5B",
+    "sail/Sailor2-3B-SFT",
+    "allenai/OLMoE-1B-7B-0924-Instruct",
+    "prithivMLmods/Llama-3.2-6B-AlgoCode",
+    "CloneBO/OracleLM",
+    "NousResearch/Yarn-Solar-10b-32k",
+    "m-a-p/OProver-8B-Base",
+    "HuggingFaceH4/mistral-7b-sft-alpha",
+    "Hyeongwon/P2-split2_prob_Qwen3-4B-Base_0312-01",
+    "martyn/mixtral-megamerge-dare-8x7b-v1",
+    "Kyleyee/rDPO_hh-seed4",
+    "jingyeom/seal3.1.6n_7b",
+    "sonthenguyen/OpenHermes-2.5-Mistral-7B-mt-bench-DPO-reversed_corrupted",
+    "shibing624/ziya-llama-13b-medical-merged",
+    "datajuicer/LLaMA-1B-dj-refine-150B",
+    "ajibawa-2023/Uncensored-Jordan-7B",
+    "nlpguy/AlloyIngot",
+    "HuggingFaceTB/SmolLM3-3B",
+    "allenai/codetulu-2-7b",
+    "vihangd/dopeyshearedplats-2.7b-v1",
+    "uukuguy/speechless-code-mistral-7b-v2.0",
+    "j05hr3d/Llama-3.2-3B-Instruct-C_M_T-ALPACA",
+    "cognitivetech/Mistral-7B-Inst-0.2-Bulleted-Notes",
+    "sail/Qwen2.5-Math-1.5B-Oat-Zero",
+    "reaperdoesntknow/SMOLM2Prover",
+    "Kyleyee/ORPO_hh-seed5",
+    "CarrotAI/Llama-3.2-Rabbit-Ko-3B-Instruct",
+    "NousResearch/Nous-Capybara-7B-V1",
+    "Vijay3548/InterviewMaster-Llama3.1",
+    "ReviewHub/qwen3-4b-it-2507-sft-2018-2022-rl-step-20",
+    "Nos-PT/Llama-Carvalho-PT",
+    "Gille/StrangeMerges_49-7B-dare_ties",
+    "Kyleyee/CPO_hh-seed2",
+    "Kyleyee/cDPO_hh-seed3",
+    "Kyleyee/DrDPO_hh-seed2",
+    "sthenno-com/miscii-14b-0218",
+    "mtgv/MobileLLaMA-2.7B-Chat",
+    "Novaciano/Alice_In_The_Dark_2-Slerp-RP-3.2-1B",
+    "Himitsui/KuroMitsu-11B",
+    "owlninjam/nytheria-3b",
+    "Kyleyee/DrDPO_hh-seed4",
+    "Kyleyee/DrDPO_hh-seed5",
+    "shahzebnaveed/NeuralHermes-2.5-Mistral-7B",
+    "plaguss/mistal-7b-prm-openrlhf",
+    "viethq188/Rabbit-7B-v2-DPO-Chat",
+    "EmbeddedLLM/Mistral-7B-Merge-14-v0.3-ft-step-9984",
+    "dbpedia/nspm-starcoder-1b",
+    "psh3333/llama-3.2-3b-grpo-merged",
+    "saarvajanik/facebook-opt-6.7b-qcqa-ub-16-best-for-KV-cache",
+    "EmbeddedLLM/Mistral-7B-Merge-14-v0.3",
+    "xformAI/facebook-opt-125m-qcqa-ub-6-best-for-KV-cache",
+    "Rev124/llama-3-pruned",
+    "RatanRohith/NeuralPizza-7B-V0.3",
+    "mncai/DPO_BC_partial_epoch6",
+    "zeemen2723/museai-lyrics-gen",
+    "automerger/OgnoExperiment27-7B",
+    "sohamb37lexsi/qwen25-3b-legal-correction",
+    "swift/Meta-Llama-3-8B",
+    "m-a-p/MuPT-v0-8192-190M",
+    "Kquant03/Samlagast-7B-laser-bf16",
+    "VTSNLP/Llama3-ViettelSolutions-8B",
+    "vihangd/dopeyshearedplats-1.3b-v1",
+    "HCY123902/qwen25_7b_base_hc_tsss_n32_r1_dpo",
+    "coder3101/gemma-3-1b-it-heretic",
+    "Hemkant04/qwen05-resume-job-match-evaluator",
+    "yekon9/Qwen3-4B-Instruct-2507-heretic",
+    "uukuguy/Orca-2-13b-f16",
+    "Kquant03/NeuralTrix-7B-dpo-relaser",
+    "sambanovasystems/SambaLingo-Russian-Base",
+    "bineric/NorskGPT-Mistral-7b",
+    "ReviewHub/qwen3-4b-it-2507-sft-2018-2022-rl-step-10",
+    "beyoru/EvolLLM",
+    "driaforall/Dria-Agent-a-3B",
+    "tlphams/zoyllm-7b-slimorca",
+    "QuixiAI/WizardLM-33B-V1.0-Uncensored",
+    "bilalRahib/TinyLLama-NSFW-Chatbot",
+    "Ahatsham/Llama-3-8B-Instruct_Planning_Feedback_oldaug_v2",
+    "unsloth/Qwen2.5-Math-1.5B",
+    "wang7776/Llama-2-7b-chat-hf-30-sparsity",
+    "hector-gr/RLCR-v4-ks-uniqueness-buf5k-hotpot",
+    "m-a-p/neo_7b_instruct_v0.1",
+    "p208p2002/llama-traditional-chinese-120M",
+    "tartuNLP/Llammas-base",
+    "riotu-lab/ArabianGPT-01B",
+    "openaccess-ai-collective/DPOpenHermes-7B",
+    "Enxin/MovieChat-vicuna",
+    "Skywork/Skywork-OR1-Math-7B",
+    "Lugha-Llama/Lugha-Llama-8B-wura_edu",
+    "laion/allenai-sera-unified-316__Qwen3-8B",
+    "T1anyu/DeepInnovator",
+    "mideind/icelandic-gpt-sw3-6.7b-gec",
+    "Rayeeennnnnnnn/legalmind-chatbot",
+    "Wanfq/FuseLLM-7B",
+    "mrvinph/Qwen2.5-0.5B-Instruct-Gensyn-Swarm-placid_wily_woodpecker",
+    "shivanikerai/Llama-2-7b-chat-hf-title-ner-and-title-suggestions-v2.0",
+    "Invalid-Null/PeiYangMe-0.7",
+    "open-unlearning/unlearn_tofu_Llama-3.2-1B-Instruct_forget10_GradDiff_lr1e-05_alpha5_epoch5",
+    "Tesslate/Tessa-T1-3B",
+    "LorenaYannnnn/unsafe_compliance-Qwen3-0.6B-OURS_self-seed_1",
+    "HiTZ/latxa-7b-v1",
+    "huggyllama/llama-7b",
+    "FlyPig23/Llama3.2-3B_Paper_Impact_media_SFT_1ep",
+    "hongzhouyu/FineMedLM-o1",
+    "shisa-ai/shisa-v1-llama3-8b.lr-5e6",
+    "senseable/Westlake-7B",
+    "simplescaling/s1.1-3B",
+    "Borjan/finki-gpt-140M",
+    "TinyLlama/TinyLlama_v1.1_chinese",
+    "Manirajan/interview_tiny",
+    "ajibawa-2023/Young-Children-Storyteller-Mistral-7B",
+    "TinyLlama/TinyLlama_v1.1_math_code",
+    "manotham/Thai-dialogue-transalate_sft_80K",
+    "nassimjp/Maral-7B-alpha-1",
+    "MSL7/INEX16-7b",
+    "Rayeeennnnnnnn/mizan-legal-tunisian",
+    "maheshrawat18/Qwen3-4B-2507-sft1",
+    "prithivMLmods/Theta-Crucis-0.6B-Turbo1",
+    "mlabonne/llama-2-7b-miniguanaco",
+    "FreekCoolAI/privacy-gemma-qlora",
+    "lex-hue/Delexa-V0.1-7b",
+    "goldfish-models/tur_latn_100mb",
+    "shisa-ai/ablation-18-rafbestseq-shisa-v2-llama-3.1-8b-lr8e6",
+    "yash-lulla/Legal_AI_Assistant",
+    "prithivMLmods/Deepthink-Llama-3-8B-Preview",
+    "prithivMLmods/QwQ-R1-Distill-7B-CoT",
+    "sstoica12/acquisition_metamath_llama_instruct-3_1-8b-math_format_500_combined_openr1math",
+    "goldfish-models/rus_cyrl_1000mb",
+    "goldfish-models/ukr_cyrl_1000mb",
+    "eren23/dpo-binarized-NeutrixOmnibe-7B",
+    "RAANA-IA/Gheya-med",
+    "mesolitica/malaysian-tinyllama-1.1b-16k-instructions-v2",
+    "bue0912/ToolOmni-Qwen3-4B",
+    "BRlkl/distill-sft-qwen3-4b-full",
+    "prithivMLmods/PocketThinker-QwQ-3B-Instruct",
+    "prithivMLmods/Megatron-Bots-1.7B-Reasoning",
+    "kumarprince070107/geobot",
+    "ClaudioSavelli/FAME_GA_llama32-1b-instruct-qa",
+    "kairawal/Qwen3-0.6B-EL-SynthDolly-1A-E8",
+    "prithivMLmods/Poseidon-Reasoning-1.7B",
+    "PKU-Alignment/ProgressGym-HistLlama3-8B-C014-instruct-v0.2",
+    "unsloth/Qwen2-7B",
+    "prithivMLmods/Open-Xi-Math-Preview",
+    "MiniLLM/teacher-gpt2-1.5B",
+    "Rakancorle1/qwen2.5-7b_Instruct_policy_traj_30k_full",
+    "prithivMLmods/Lang-Exster-0.5B-Instruct",
+    "prithivMLmods/Nenque-MoT-0.6B-Elite14",
+    "posicube/Llama2-chat-AYT-13B",
+    "lomahony/pythia-70m-helpful-sft",
+    "kyubeen/code-grpo-checkpoint-950",
+    "nicholasKluge/TeenyTinyLlama-160m",
+    "allenai/Llama-3.1-Tulu-3-8B",
+    "BarraHome/Mistroll-7B-v2.2",
+    "posicube/Llama-chat-AY-13B",
+    "hongli-zhan/MINT-empathy-Qwen3-4B",
+    "jhhj25/qwen3-moe-neuron_structure_drop-p50-s1k-128samples-sft",
+    "Ikonz-Studios/seva-sarathi-intent-qwen3-1.7b",
+    "NECOUDBFM/Jellyfish-8B",
+    "gradients-io-tournaments/augmented-1db17e1d682d23fd",
+    "Locutusque/LocutusqueXFelladrin-TinyMistral248M-Instruct",
+    "ChuGyouk/R16",
+    "oveja1122/toolcalling-merged-demo",
+    "kmseong/llama3_2_3b-instruct-math-safedelta-scale0.8",
+    "EleutherAI/SmolLM2-1.7B-magpie-ultra-v1.0-math-431k-s",
+    "paulml/NeuralOmniBeagleMBX-v3-7B",
+    "skemessage/Qwen2.5-7B-Instruct-neuron",
 ]
 
 # ══════════════════════════════════════════════════════════
@@ -110,30 +313,15 @@ def _run_http():
 # ══════════════════════════════════════════════════════════
 #  业务逻辑
 # ══════════════════════════════════════════════════════════
-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
-nv_docker_image: harbor.4pd.io/dooke/vllm/vllm/vllm-openai:v0.11.0
+    config_content = f"""gpu_type: ppu_zw_810e
 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
+nv_docker_image: harbor-contest.4pd.io/sunruoxi/vllm-openai-fix-tokenizer:v0.11.0
 modelhub_options:
   srcRelativePath: leaderboard/modelHubXC/{model_id}
   mountPoint: /model
@@ -141,17 +329,44 @@ sut_config:
   values:
     gpu_num: 1
     env:
-      - name: MAX_MODEL_LEN
-        value: 8192
-    command: ["vllm", "serve", "/model", "--port", "8000", "--served-model-name", "llm", "--max-model-len", "8192", "--trust-remote-code", "--dtype", "float16"]
+      - name: test
+        value: fp16
+    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
-        value: 8192
-    command: ["vllm", "serve", "/model", "--port", "80", "--served-model-name", "llm", "--max-model-len", "8192", "--trust-remote-code", "--dtype", "float16"]
+      - name: test
+        value: fp16
+    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,
@@ -167,6 +382,7 @@ ref_config:
             "taskType": TASK_TYPE,
         }],
     }
+    print(f"[payload] {json.dumps(payload, indent=2, ensure_ascii=False)}", flush=True)
     try:
         resp = requests.post(
             BASE_URL + SUBMIT_ENDPOINT,
@@ -192,12 +408,8 @@ def _run_worker():
     _state["phase"] = "submitting"
 
     successful: List[Tuple[str, str]] = []
-    try:
-        token = _login()
-    except Exception:
-        traceback.print_exc()
-        _state["phase"] = "error"
-        return
+    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():