submit round-24 filter results on zhoushasha: MetaX_c-500(63)/Kunlunxin_p-800(1)/Cambricon_mlu-370-x8(1)/Biren_166m(95) + ppu_zw_810e(57), 217 models total

Also add a hygon_k100-ai config branch and HYGON_MODELS list (kept available but
excluded from GPU_JOBS per request). Refresh the expired AUTH_TOKEN.

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

.gitignore

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

main.py

@@ -1,8 +1,19 @@
 """
 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。
-同时暴露 /health（K8s 探活）和 /status（运行状态）。
+启动后针对 GPU_JOBS 中配置的 GPU 卡分别批量提交各自筛选出的模型验证任务
+（当前仅提交 ppu_zw_810e，其余 4 张卡 Biren_166m/Cambricon_mlu-370-x8/MetaX_c-500/
+Kunlunxin_p-800 的 config_content 模板和模型列表仍保留在代码中，未列入本次 GPU_JOBS）
+（/adminApi/async/task/create-contest-task，
+Bearer Token 认证），之后保持 HTTP 服务存活。
+
+账号额度自动重试：如果某个模型提交时命中"当前等待中或运行中的异步模型验证
+任务数量已达上限"（账号额度已满），不算永久失败，会被留到下一轮；额度耗尽后
+本进程会原地等待 30 分钟，再自动重试所有因额度问题未提交成功的模型，如此循环，
+直至全部提交成功或进程被平台关闭——不需要重新部署新策略，循环逻辑在本进程内完成。
+非额度原因的失败（如模型已在验证中等）不会重试。
+
+同时暴露 /health（K8s 探活）和 /status（运行状态，含当前轮次/待重试数/下次重试时间）。
 """
 
 import json
@@ -22,10 +33,9 @@ BASE_URL           = os.environ.get("BASE_URL", "https://modelhub.org.cn")
 SUBMIT_ENDPOINT    = "/adminApi/async/task/create-contest-task"
 
 # 通过 curl -X POST https://modelhub.org.cn/adminApi/user/login 获取后填入
-AUTH_TOKEN        = "eyJ0eXAiOiJKV1QiLCJhbGciOiJIUzI1NiJ9.eyJ1c2VyQWNjb3VudCI6Inpob3VzaGFzaGEiLCJpZCI6MTQsInVzZXJSb2xlIjoibGVhZGVyYm9hcmQiLCJleHAiOjE3ODI3MzA4MTQsImlhdCI6MTc4MjEyNjAxNH0.ZBMLXxi9n_g4_drUUuciWFipViMZmJzMJLab5dL0WM4"
+AUTH_TOKEN        = "eyJ0eXAiOiJKV1QiLCJhbGciOiJIUzI1NiJ9.eyJ1c2VyQWNjb3VudCI6Inpob3VzaGFzaGEiLCJpZCI6MTQsInVzZXJSb2xlIjoibGVhZGVyYm9hcmQiLCJleHAiOjE3OTAwNjkxMjAsImlhdCI6MTc4OTQ2NDMyMH0.KzJac6ddaZdtLvjD6ZnoK1PNEKFoXdyDn9Hh4FxU9ic"
 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
 
@@ -33,339 +43,278 @@ 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 = [
-    "mrsanskar19/my_first_model",
-    "cs-552-2026-vibe-trainers/math_model",
-    "jburnford/dyslexic-writer-qwen3-0.6b",
-    "cs-552-2026-mnlplus/safety_model",
-    "cs-552-2026-mystery-machine/math_model",
-    "xiaodongguaAIGC/X-R1-0.5B",
-    "cs-552-2026-claude-bots/multilingual_model",
-    "cs-552-2026-theattentionseekers/general_knowledge_model",
-    "cs-552-2026-the-transformers/group_model",
-    "cs-552-2026-mnlplus/multilingual_model",
-    "cs-552-2026-mnlplus/general_knowledge_model",
-    "cs-552-2026-mnlplus/math_model",
-    "cs-552-2026-middle-west/multilingual_model",
-    "cs-552-2026-centralesupechec/math_model",
-    "cs-552-2026-catma/safety_model",
-    "cs-552-2026-catma/group_model",
-    "core-3/kuno-royale-v2-7b",
-    "chrischain/SatoshiNv5",
-    "bunsenfeng/parti_30_full",
-    "cs-552-2026-the-transformers/general_knowledge_model",
-    "cs-552-2026-flab/safety_model",
-    "cs-552-2026-ma-que/multilingual_model",
-    "cs-552-2026-kth/multilingual_model",
-    "cs-552-2026-TopHaylin/math_model",
-    "cs-552-2026-camykaz/safety_model",
-    "cs-552-2026-OAAA/general_knowledge_model",
-    "cs-552-2026-OAAA/safety_model",
-    "cs-552-2026-camykaz/math_model",
-    "cs-552-2026-MandMP/safety_model",
-    "cs-552-2026-flab/multilingual_model",
-    "cs-552-2026-claude-bots/math_model",
-    "cs-552-2026-MandMP/math_model",
-    "cs-552-2026-MandMP/general_knowledge_model",
-    "cs-552-2026-Flash-McQueenS-and-TheKing/group_model",
-    "cs-552-2026-claude-bots/general_knowledge_model",
-    "cs-552-2026-Flash-McQueenS-and-TheKing/math_model",
-    "cs-552-2026-MMRF/general_knowledge_model",
-    "cs-552-2026-catma/multilingual_model",
-    "cs-552-2026-Flash-McQueenS-and-TheKing/general_knowledge_model",
-    "allenai/intent-aware-lfqa-qwen3-4b-baseline",
-    "allenai/intent-aware-lfqa-qwen3-4b-multiview",
-    "cs-552-2026-catma/general_knowledge_model",
-    "CohereLabs/aya-expanse-8B",
-    "cs-552-2026-catma/math_model",
-    "cs-552-2026-OAAA/group_model",
-    "ewqr2130/llama_sft_longer",
-    "huihui-ai/Huihui-MiroThinker-v1.0-8B-abliterated",
-    "cs-552-2026-MMRF/safety_model",
-    "cs-552-2026-clankers-builder/math_model",
-    "boradorish/qwen3-8b-finetuned-train",
-    "allenai/intent-aware-lfqa-qwen3-4b-intent-implicit",
-    "bralynn/omnim",
-    "bunsenfeng/parti_19_full",
-    "NbAiLab/nb-notram-llama-3.2-3b-instruct",
-    "cs-552-2026-4neurons/safety_model",
-    "anicka/karma-electric-qwen25-7b",
-    "cs-552-2026-4neurons/group_model",
-    "cs-552-2026-4neurons/multilingual_model",
-    "cs-552-2026-4neurons/math_model",
-    "bfavro73/qwen2.5-coder-1.5b-pandas-dpo-aligned",
-    "bfavro73/qwen2.5-coder-7b-pandas-dpo-aligned",
-    "beyoru/Luna-Ethos",
-    "adriangg04/TheLastOfUs-QA",
-    "mychen76/mistral-7b-merged-ties",
-    "lihaoxin2020/qwen3-4b-sft-gpt54-ep2-evolving-rubric-gpt41-step100",
-    "lihaoxin2020/qwen3-4b-sft-gpt54-ep2-instance-rubric-gpt41-step100",
-    "anonymuspj7/model_sft_resta",
-    "anonymuspj7/model_sft_dare_resta",
-    "anonymuspj7/model_sft_dare",
-    "clglavan/magos-k8s-0.6b",
-    "continuum-ai/qwen2.5-1.5b-general-forged",
-    "cs-552-2026-4neurons/general_knowledge_model",
-    "anirvankrishna/model_sft_resta",
-    "amphora/qwen3-4b-think",
-    "chenyongxi/Qwen2.5-1.5B-DPO-1.5B",
-    "carnival13/model_sft_merged",
-    "berkerbatur/qwen-0.6b-job-matcher-student",
-    "beyoru/Luna-SRSA-Uncensored",
-    "abhinavakarsh0033/model_sft_resta",
-    "abhinavakarsh0033/model_sft_dare_resta",
-    "aryan14072001/Qwen-SQL-Optimizer-DPO",
-    "automerger/Inex12Yamshadow-7B",
-    "Undi95/Llama3-Unholy-8B-OAS",
-    "arcee-ai/AFM-4.5B",
-    "anujjamwal/OpenMath-Nemotron-1.5B-PruneAgnostic",
-    "anujjamwal/OpenMath-Nemotron-1.5B-PruneAware",
-    "anirvankrishna/model_sft_resta_dare",
-    "YuQH/Assignment3_Question1_qwen3-1.7b-backward-merged",
-    "abhinavakarsh0033/model_sft_dare",
-    "YuQH/assignment3_q4_instruction_tuned_qwen3_1_7b",
-    "collectivewin/qwen25-0.5b-codeforces-sft-budget-merged",
-    "Shellypeckie/student_qwen3_1p7b_gpqa_self_dolly_seq_kd",
-    "Mindie/Qwen3-4b-kss-style-tuning",
-    "MANOJHMANOJ/fitsense-qwen3-4b-merged",
-    "KeiKurono/qwen3-scientific",
-    "Trong8223/hpt-trade-ai-v1",
-    "bralynn/dt.think1.128.256.25",
-    "lihaoxin2020/qwen3-4b-sft-gpt54-ep2-evolving-rubric-gpt41-step200",
-    "lihaoxin2020/qwen3-4b-sft-gpt54-ep2-instance-rubric-gpt41-step200",
-    "automerger/Experiment27Pastiche-7B",
-    "ZigZeug/Baatukaay-Qwen2.5-3B-Wolof",
-    "adsyamsafa/Nixia1.0-0.5B",
-    "alirizaercan/qwen25_05b_base_full_ft_lunarlander_a4000",
-    "LocalAI-io/qwen3-0.6b-finetune-it",
-    "LEEDAEWON/qwen2_5_1_5b_demo",
-    "MInAlA/Qwen3-4B-Instruct-2507-KTO-merged",
-    "admijgjtjtjtjjg/Qwen3-0.6B-Micro-5M",
-    "Kimyayd/Qwen-1.5B-Fongbe-Translator",
-    "Jason-hu/Qwen2.5-3B-GSM8K-SFT",
-    "Issactoto/qwen2.5-1.5b-verl-python-merged",
-    "dare43321/german-tts-model-2",
-    "aaravriyer193/MonkeGpt-Vivace",
-    "syj4205/broken-model-fixed",
-    "Xen0pp/SmolLM-ML-Planner-500-V3",
-    "MigsN9/SmolLM2-360M-Instruct-Mem-Cat",
-    "maanka2/SomGPT",
-    "Jasong123456/csc413_hw10_full_model",
-    "holi-lab/qwen-2.5-1.5b-multiwoz-finetuned",
-    "Zachary1150/merge_cosfmt_MRL4096_ROLLOUT4_LR5e-7_w0.5_ties_density0.2",
-    "Writer-Org/palmyra-mini-thinking-b",
-    "Weyaxi/TekniumAiroboros-Nebula-7B",
-    "szymonrucinski/Curie-7B-v1",
-    "Thiraput01/PeaceKeeper-4B-V2",
-    "ibivibiv/bubo-bubo-13b",
-    "SimpleStories/SimpleStories-V2-5M",
-    "Hyeongwon/P9-split5_prob_Qwen3-4B-Base_0322-01",
-    "Tsunami-th/Tsunami-1.0-14B-Instruct",
-    "xw1234gan/cnk12_Main_fixed_SFTanchor_3B_step_1",
-    "TeeZee/DarkForest-20B-v2.0",
-    "artificialguybr/QWEN-2.5-0.5B-Synthia-II",
-    "open-thoughts/OpenThinker-Agent-v1",
-    "laion/r2egym-nl2bash-stack-bugsseq-fixthink-again",
-    "xw1234gan/cnk12_Main_fixed_SFTanchor_3B_step_2",
-    "xw1234gan/cnk12_Main_fixed_SFTanchor_3B_step_3",
-    "xw1234gan/cnk12_Main_fixed_SFTanchor_3B_step_4",
-    "Trong8223/hpt-trade-ai-v2",
-    "PrimeIntellect/Qwen2.5-0.5B-Reverse-Text-SFT",
-    "mlfoundations-dev/oh-dcft-v3.1-gpt-4o-mini-qwen",
-    "mncai/Mistral-7B-guanaco-1k-orca_platy-1k",
-    "zypchn/BehChat-v3",
-    "xw1234gan/cnk12_Main_fixed_SFTanchor_3B_step_5",
-    "Lixing-Li/Llama-3.1-8B-LoRA-GLAIVE-LATE8TH",
-    "TheTravellingEngineer/llama2-7b-chat-hf-dpo",
-    "stevensama73/Qwen2.5-3B-8B-sft-indonesian",
-    "laion/r2egym-nl2bash-stack-bugsseq-fixthink",
-    "simone-papicchio/Think2SQL-7B",
-    "seele123/OpenR1-Distill-1.5B-ours",
-    "laion/openthoughts-4-code-qwen3-32b-annotated-32k_qwen2.5-1.5B_32k",
-    "laion/nl2bash-verified-GLM-4.6-traces-32ep-32k-mgn5e4_Qwen3-8B",
-    "laion/nemotron-100000-opt100k__Qwen3-8B",
-    "xw1234gan/cnk12_Main_fixed_SFTanchor_3B_step_6",
-    "AlfredPros/CodeLlama-7b-Instruct-Solidity",
-    "TitleOS/Phi-4-mini-reasoning-heretic",
-    "xw1234gan/cnk12_Main_fixed_SFTanchor_3B_step_8",
-    "xw1234gan/cnk12_Main_fixed_SFTanchor_3B_step_7",
-    "SanjiWatsuki/Kunoichi-7B",
-    "AliMaatouk/Llama-3.2-1B-Tele",
-    "prithivMLmods/Bellatrix-Tiny-3B-R1",
-    "m-a-p/Qwen2-Instruct-7B-COIG-P",
-    "zypchn/BehChat-llama-SFT-v2",
-    "TURKCELL/Turkcell-LLM-7b-v1",
-    "Alelcv27/Llama3.2-3B-base-Code",
-    "prithivMLmods/Sqweeks-7B-Instruct",
-    "maxidl/Llama-OpenReviewer-8B",
-    "yufeng1/OpenThinker-7B-type6-e5-max-alpha0_25-textsummarization-2e5-type6-e1-alpha0_375-2",
-    "Thiraput01/PeaceKeeper-4B-V4",
-    "m-a-p/Qwen2.5-Instruct-7B-COIG-P",
-    "SanjiWatsuki/Loyal-Toppy-Bruins-Maid-7B-DARE",
-    "cycloneboy/CscSQL-Merge-Qwen2.5-Coder-3B-Instruct",
-    "prithivMLmods/Monoceros-QwenM-1.5B",
-    "Shanghai_AI_Laboratory/AlchemistCoder-L-7B",
-    "yujiepan/qwen3-tiny-random-tp",
-    "prithivMLmods/rStar-Coder-Qwen3-0.6B",
-    "Qwen/Qwen1.5-14B",
-    "yil384/Qwen3-0.6B-full",
-    "Thiraput01/PeaceKeeper-4B-V3",
-    "ystemsrx/Qwen2.5-Interpreter",
-    "yujiepan/baguettotron-tiny-random",
-    "prithivMLmods/Cerium-Qwen3-R1-Dev",
-    "QwenCollection/Nxcode-CQ-7B-orpo",
-    "prithivMLmods/Viper-OneCoder-UIGEN",
-    "yasserrmd/GLM4.7-Distill-LFM2.5-1.2B",
-    "kairawal/Llama-3.2-3B-Instruct-PT-SynthDolly-E1-S73",
-    "carsenk/llama3.2_3b_122824_uncensored",
-    "prithivMLmods/Telescopium-Acyclic-Qwen3-0.6B",
-    "prithivMLmods/Deneb-Qwen3-Radiation-0.6B",
-    "cycloneboy/CscSQL-Merge-Qwen2.5-Coder-1.5B-Instruct",
-    "ziaulkarim245/Deepseek-R1-Phishing-Detector",
-    "rLLM/rLLM-FinQA-4B",
-    "aisingapore/Qwen-SEA-LION-v4-32B-IT-4BIT",
-    "yam-peleg/Experiment8-7B",
-    "prithivMLmods/Castula-U2-QwenRe-1.5B",
-    "yam-peleg/Experiment31-7B",
-    "yam-peleg/Experiment30-7B",
-    "PrimeIntellect/Qwen3-8B",
-    "Thiraput01/PeaceKeeper-4B",
-    "PistachioAlt/Noromaid-Bagel-7B-Slerp",
-    "PygmalionAI/pygmalion-2-7b",
-    "yam-peleg/Experiment28-7B",
-    "mlabonne/DatacampLlama-3.1-8B",
-    "prithivMLmods/Pictor-1338-QwenP-1.5B",
-    "prithivMLmods/Viper-Coder-v0.1",
-    "NousResearch/Yarn-Llama-2-13b-128k",
-    "TheDrummer/Llama-3SOME-8B-v2",
-    "OpenBuddy/openbuddy-llama2-13b-v11-bf16",
-    "yam-peleg/Experiment22-7B",
-    "OpenPipe/llama_3b_hn_story_classifier",
-    "yam-peleg/Experiment2-7B",
-    "ybelkada/Mistral-7B-v0.1-bf16-sharded",
-    "xw1234gan/Main_fixed_MATH_3B_step_1",
-    "Ihor/Text2Graph-R1-Qwen2.5-0.5b",
-    "unsloth/llama-2-7b",
-    "neuralmagic/SparseLlama-2-7b-cnn-daily-mail-pruned_50.2of4",
-    "ThaiLLM/ThaiLLM-8B",
-    "voidful/unit-desta-8b-base-llama3-8b-instruct",
-    "SouravCrypto/Qwen2.5-0.5B-Instruct-Gensyn-Swarm-striped_tawny_dove",
-    "xw1234gan/Main_fixed02_MATH_3B_step_4",
-    "xw1234gan/Main_MATH_3B_step_9",
-    "m-a-p/Infinity-Instruct-3M-0625-Mistral-7B-COIG-P",
-    "adityakum667388/lumichats-v1.1",
-    "NEU-HAI/Llama-2-7b-alpaca-cleaned",
-    "NousResearch/CodeLlama-34b-hf",
-    "prithivMLmods/Tulu-MathLingo-8B",
-    "ibm-granite/granite-8b-code-instruct-4k",
-    "alexxbobr/ORPO8000Vikhr-Llama-3.2-1B-Instruct5000",
-    "m-a-p/TreePO-Qwen2.5-7B",
-    "fancyfeast/llama-bigasp-prompt-enhancer",
-    "line-corporation/japanese-large-lm-1.7b",
-    "wassname/qwen3-5lyr-tiny-random",
-    "TaimurShaikh/qwen1.5-1.8b-sft",
-    "l3utterfly/minima-3b-layla-v1",
-    "l3utterfly/mistral-7b-v0.1-layla-v2",
-    "Henrychur/MMed-Llama-3-8B",
-    "gradient-spaces/respace-sg-llm-1.5b",
-    "l3utterfly/minima-3b-layla-v2",
-    "allenai/truthfulqa-info-judge-llama2-7B",
-    "facebook/llm-compiler-7b",
-    "nothingiisreal/L3.1-8B-Celeste-V1.5",
-    "TaimurShaikh/qwen1.5-1.8b-dpo",
-    "l3utterfly/tinyllama-1.1b-layla-v4",
-    "prithivMLmods/Flerovium-Llama-3B",
-    "l3utterfly/mistral-7b-v0.1-layla-v1",
-    "OpenBuddy/openbuddy-mistral-7b-v13.1",
-    "iCIIT/redqueenprotocol-sin-llama3.2-3B-model",
-    "OpenAssistant/codellama-13b-oasst-sft-v10",
-    "Unitedp2p/New-Llama-3.1-8B-Lexi-Uncensored-V2",
-    "USTC-KnowledgeComputingLab/Llama3-KALE-LM-Chem-1.5-8B",
-    "Salesforce/Llama-xLAM-2-8b-fc-r",
-    "uukuguy/speechless-orca-platypus-coig-lite-4k-0.6e-13b",
-    "krevas/SOLAR-10.7B",
-    "SykoSLM/SykoLLM-V6.0-Test",
-    "l3utterfly/mistral-7b-v0.1-layla-v4",
-    "argilla/distilabeled-Marcoro14-7B-slerp",
-    "uzlm/alloma-8B-Instruct",
-    "trillionlabs/Tri-7B",
-    "nlile/PE-7b-full",
-    "glaiveai/glaive-coder-7b",
-    "RamziRebai/llama-2-7b-therapist-v4",
-    "Lazycuber/L2-7b-Base-Guanaco-Vicuna",
-    "behnamsh/gpt2_camel_physics",
-    "jondurbin/spicyboros-7b-2.2",
-    "argilla/distilabeled-Marcoro14-7B-slerp-full",
-    "kairawal/Llama-3.2-3B-Instruct-ZH-SynthDolly-1A-E5",
-    "Surpem/Supertron1-8B",
-    "TIGER-Lab/Mantis-8B-siglip-llama3-pretraind",
-    "Leopo1d/OpenVul-Qwen3-4B-GRPO",
-    "totally-not-an-llm/PuddleJumper-13b-V2",
-    "ajibawa-2023/Code-Mistral-7B",
-    "unsloth/Qwen2.5-3B",
-    "Jiqing/tiny-random-qwen2",
-    "L33tcode/llama-3-8b-CEH-hf",
-    "FlyPig23/Qwen3-4B_Paper_Impact_patent_SFT_1ep",
-    "totally-not-an-llm/EverythingLM-13b-V3-16k",
-    "totally-not-an-llm/EverythingLM-13b-16k",
-    "ajibawa-2023/Code-290k-6.7B-Instruct",
-    "mrcuddle/Lumimaid-Muse-12B",
-    "how3751/coder_7B",
-    "m-a-p/CT-LLM-SFT",
-    "Marintosti/chsa-triage-merged",
-    "m-a-p/CT-LLM-SFT-DPO",
-    "inclusionAI/AReaL-boba-2-8B",
-    "FreedomIntelligence/Apollo-6B",
-    "mrthor102/evolai-tfm-super-004",
-    "staeiou/bartleby-qwen3-1.7b_v4",
-    "golgat/toolcalling-merged-demo",
-    "allenai/Llama-3.1-Tulu-3-8B-SFT",
-    "Kazuki1450/Olmo-3-1025-7B_dsum_3_6_tok_Certainly_1p0_0p0_1p0_grpo_dr_grpo_42_rule",
-    "driaforall/Dria-Agent-a-7B",
-    "ChaoticNeutrals/Eris_Remix_7B",
-    "thrishala/mental_health_chatbot",
-    "mlfoundations-dev/oh-dcft-v3.1-gemini-1.5-flash",
-    "CompassioninMachineLearning/pretrainingBasellama3kv3",
-    "venkycs/Zyte-1B",
-    "mnoukhov/pythia410m-sft-tldr",
-    "ali-elganzory/Baguettotron-DPO-Tulu3-decontaminated",
-    "Alienpenguin10/M3PO-kl_divergence-trial1-seed123",
-    "tanhao2015/MDtranslator",
-    "aloobun/Reyna-CoT-4B-v0.1",
-    "sstoica12/influence_metamath_qwen2.5_3b_new_detailed",
-    "argilla/zephyr-7b-spin-iter1-v0",
-    "Kazuki1450/Olmo-3-1025-7B_dsum_3_6_rel_1e0_1p0_0p0_1p0_grpo_sapo_42_rule",
-    "FelixChao/Voldemort-10B",
-    "yam-peleg/Experiment1-7B",
-    "Eric111/Mistral-7B-Instruct_v0.2_UNA-TheBeagle-7b-v1",
-    "ewqr2130/llama_ppo_1e6_new_tokenizerstep_8000",
-    "DanielClough/Candle_TinyLlama-1.1B-Chat-v1.0",
-    "rbelanec/train_qnli_42_1779286680",
-    "tech27/Qwen2.5-1.5B-Instruct-Gensyn-Swarm-amphibious_spotted_kingfisher",
-    "syvai/emotion-reasoning-1b",
-    "argilla/notus-7b-v1",
-    "burtenshaw/SmolLM3-3B-GRPO-think",
-    "ViratChauhan/Qwen3-4B-GRPO-v2",
-    "zarakiquemparte/zaraxls-l2-7b",
-    "AlexeySorokin/GEC-from-explanations-4BInstr-distilled-v2303",
-    "Alelcv27/Qwen2.5-3B-INST-Code",
-    "tiny-random/llama-3.3-dim64",
-    "prithivMLmods/Triangulum-1B",
+BIREN_MODELS = [
+    "ApolloRaines/Phi-4-mini-Instruct-Desyced",
+    "allenai/OLMo-2-0425-1B",
+    "Free2035/4QDR_4B_AD_Thinker_V1",
+    "aifoundry-org/OLMo-7B-0424-hf-Quantized",
+    "lugman-madhiai/Qwen3-4B-MHS-1.1",
+    "zypchn/BehChat-SFT-v4",
+    "NoesisLab/Kai-30B-Instruct",
+    "barandinho/Qwen3-30B-A3B-FIRST-STAGE-SFT",
+    "m-a-p/OpenLLaMA-Reproduce-218.1B",
+    "xiaolesu/Qwen3-8B-Herald-SFT",
+    "WhiteRabbitNeo/WhiteRabbitNeo-33B-v1.5",
+    "saleh1312/orph_3.07225",
+    "vanta-research/atom-olmo3-7b",
+    "ZhipuAI/LongCite-glm4-9b",
+    "Xlnk/LFM2-2.6B-Exp-GGuf",
+    "Tesslate/UIGEN-T1.1-Qwen-14B",
+    "jondurbin/bagel-dpo-34b-v0.2",
+    "zhengr/MixTAO-7Bx2-MoE-Instruct-v5.0",
+    "allenai/Olmo-3.1-32B-Instruct",
+    "bjaidi/Phi-3-medium-128k-instruct-awq",
+    "xing720310/qwen3-14b",
+    "IntelLabs/sqft-mistral-7b-v0.3-50-base-gptq",
+    "hariharanv04/qwen2.5-coder-14b-metadata-merged",
+    "junfengzhou/qwen3-14b-rl",
+    "ronnywebdevs1/Affine-P011-5CkU7wLMWXPs6TdSsMf8eEYCVAbPLyNmYg9PPx1Uds8toKra",
+    "kennedyantonio0301/Affine-Tensor-h3-5EkdoaCmEpFffUjDpLhDMzEDR4kptaEzpTPYCP1uL2sbct8C",
+    "julep-ai/dolphin-2.9-llama3-70b-awq",
+    "cortexso/gemma3",
+    "jacob-ml/jacob-24b",
+    "lyraaaa/neuralese-sft-pretrain-v2",
+    "ai-sage/GigaChat-20B-A3B-instruct-bf16",
+    "PJMixers-Dev/gemma-3-1b-it-fixed",
+    "commotion/svara_finetune_v1",
+    "geoffmunn/Qwen3-14B-f16",
+    "geoffmunn/Qwen3-32B-f16",
+    "TeichAI/Nemotron-Cascade-14B-Thinking-Claude-4.5-Opus-High-Reasoning-Distill",
+    "prithivMLmods-llamafile/SmolLM2-1.7B-Instruct-llamafile",
+    "prithivMLmods-llamafile/Llama-3.2-8B-llamafile-200K",
+    "llamafile-club/SmolLM-135M-Instruct-Llamafile",
+    "prithivMLmods/Sombrero-QwQ-32B-Elite9",
+    "prithivMLmods-llamafile/Aya-Expanse-8B-llamafile",
+    "llamafile-club/SmolLM-135M-Llamafile",
+    "prithivMLmods/Sombrero-QwQ-32B-Elite10-Fixed",
+    "prithivMLmods-llamafile/Qwen2.5-Coder-1.5B-llamafile",
+    "TeichAI/Qwen3-14B-Polaris-Alpha-Distill",
+    "okwinds/MiroThinker-14B-DPO-v0.1",
+    "sanbuphy/tianji-wish2-14b",
+    "codefuse-ai/CodeFuse-StarCoder2-15B",
+    "AI-ModelScope/txgemma-27b-chat",
+    "Shanghai_AI_Laboratory/internlm3-8b-instruct-awq",
+    "XGenerationLab/XiYanSQL-QwenCoder-32B-2412",
+    "vllm-ascend/gemma-1.1-2b-it",
+    "OpenBuddy/openbuddy-qwen1.5-32b-v21.2-32k",
+    "OpenBuddy/openbuddy-thinker-32b-v26-preview",
+    "OpenBuddy/openbuddy-qwen1.5-32b-v21.1-32k",
+    "TechxGenus-MS/CodeGemma-7b",
+    "OpenBuddy/openbuddy-qwq-32b-v25.2q-200k",
+    "unsloth/Qwen3-30B-A3B",
+    "OpenBuddy/openbuddy-qwq-32b-v25.1-200k",
+    "OpenBuddy/openbuddy-r1-32b-v24.1-200k",
+    "iic/ERank-14B",
+    "OpenBuddy/openbuddy-yi1.5-34b-v21.2-32k",
+    "LGAI-EXAONE/EXAONE-Deep-32B",
+    "OpenBuddy/openbuddy-qwq-32b-v24.2-200k",
+    "unsloth/Phi-3-mini-4k-instruct-v0",
+    "argilla/notux-8x7b-v1",
+    "voidful/qd-phi-1_5",
+    "Shanghai_AI_Laboratory/internlm2-math-base-20b",
+    "Shanghai_AI_Laboratory/internlm2-math-plus-20b",
+    "TechxGenus-MS/starcoder2-15b-instruct",
+    "Shanghai_AI_Laboratory/internlm2-base-20b",
+    "m-a-p/OpenLLaMA-Reproduce-872.42B",
+    "m-a-p/OpenLLaMA-Reproduce-973.08B",
+    "Shanghai_AI_Laboratory/OREAL-32B",
+    "YOYO-AI/Qwen3-30B-A3B-CoderThinking-YOYO-linear",
+    "ticoAg/Qwen-1_8B-Chat-Int4-awq",
+    "smirki/UIGEN-T1.1-Qwen-14B",
+    "prithivMLmods/Qwen2.5-32B-DeepSeek-R1-Instruct",
+    "sail/Sailor2-20B-128K",
+    "xverse/XVERSE-65B",
+    "Shanghai_AI_Laboratory/internlm2_5-20b-chat",
+    "TeleAI/TeleChat-52B",
+    "modelscope/Llama-2-70b-ms",
+    "Shanghai_AI_Laboratory/internlm2-20b",
+    "ai-modelscope/Llama-3_1-Nemotron-51B-Instruct",
+    "zhuangxialie/Phi-3-Chinese-ORPO",
+    "openai-mirror/gpt-oss-safeguard-20b",
+    "ByteDance-Seed/Seed-OSS-36B-Instruct",
+    "Shanghai_AI_Laboratory/internlm-chat-20b",
+    "TurkuNLP/bloom-finnish-176b",
+    "openai-mirror/gpt-oss-120b",
+    "vllm-ascend/QwQ-32B-W8A8",
+    "mistralai/Mistral-Small-24B-Instruct-2501",
+    "Shanghai_AI_Laboratory/internlm-20b",
+    "ZhipuAI/GLM-4-32B-0414",
 ]
 
+CAMBRICON_MODELS = [
+    "Xlnk/LFM2-2.6B-Exp-GGuf",
+]
+
+METAX_MODELS = [
+    "ApolloRaines/Phi-4-mini-Instruct-Desyced",
+    "robertspumiaca1975/Qwen2.5-Coder-14B-n8n-Workflow-Generator",
+    "barandinho/Qwen3-30B-A3B-FIRST-STAGE-SFT-V2",
+    "hotmailuser/QwenSlerp2-14B",
+    "WhiteRabbitNeo/WhiteRabbitNeo-33B-v1.5",
+    "madox81/SmolLM2-135M-cybersecurity-lora-merged",
+    "Xlnk/LFM2-2.6B-Exp-GGuf",
+    "Tesslate/UIGEN-T1.1-Qwen-14B",
+    "jondurbin/bagel-dpo-34b-v0.2",
+    "zhengr/MixTAO-7Bx2-MoE-Instruct-v5.0",
+    "bjaidi/Phi-3-medium-128k-instruct-awq",
+    "jacob-ml/jacob-24b",
+    "geoffmunn/Qwen3-32B-f16",
+    "TorpedoSoftware/Luau-Devstral-24B-Instruct-v0.2",
+    "TeichAI/Nemotron-Cascade-14B-Thinking-Claude-4.5-Opus-High-Reasoning-Distill",
+    "tongzang/Qwen2.5-7b-lora-law",
+    "prithivMLmods/Sombrero-QwQ-32B-Elite9",
+    "prithivMLmods/Sombrero-QwQ-32B-Elite10-Fixed",
+    "TeichAI/Qwen3-14B-Polaris-Alpha-Distill",
+    "okwinds/MiroThinker-14B-DPO-v0.1",
+    "sanbuphy/tianji-wish2-14b",
+    "YOYO-AI/YOYO-O1-14B",
+    "LLM-Research/Meta-Llama-3.1-405B",
+    "LLM-Research/Meta-Llama-3-70B",
+    "LLM-Research/Meta-Llama-3.1-70B",
+    "Qwen/Qwen-72B-Chat",
+    "Qwen/Qwen3-Coder-480B-A35B-Instruct",
+    "deepseek-ai/DeepSeek-R1-Distill-Llama-70B",
+    "codefuse-ai/CodeFuse-StarCoder2-15B",
+    "AI-ModelScope/txgemma-27b-chat",
+    "XGenerationLab/XiYanSQL-QwenCoder-32B-2412",
+    "OpenBuddy/openbuddy-qwen1.5-32b-v21.2-32k",
+    "OpenBuddy/openbuddy-thinker-32b-v26-preview",
+    "OpenBuddy/openbuddy-qwen1.5-32b-v21.1-32k",
+    "OpenBuddy/openbuddy-qwq-32b-v25.2q-200k",
+    "unsloth/Qwen3-30B-A3B",
+    "OpenBuddy/openbuddy-qwq-32b-v25.1-200k",
+    "OpenBuddy/openbuddy-r1-32b-v24.1-200k",
+    "iic/ERank-14B",
+    "OpenBuddy/openbuddy-yi1.5-34b-v21.2-32k",
+    "LGAI-EXAONE/EXAONE-Deep-32B",
+    "OpenBuddy/openbuddy-qwq-32b-v24.2-200k",
+    "unsloth/Phi-3-mini-4k-instruct-v0",
+    "argilla/notux-8x7b-v1",
+    "voidful/qd-phi-1_5",
+    "TechxGenus-MS/starcoder2-15b-instruct",
+    "m-a-p/OpenLLaMA-Reproduce-872.42B",
+    "m-a-p/OpenLLaMA-Reproduce-973.08B",
+    "Shanghai_AI_Laboratory/OREAL-32B",
+    "YOYO-AI/Qwen3-30B-A3B-CoderThinking-YOYO-linear",
+    "smirki/UIGEN-T1.1-Qwen-14B",
+    "sthenno-com/miscii-14b-0130",
+    "prithivMLmods/Qwen2.5-32B-DeepSeek-R1-Instruct",
+    "sail/Sailor2-20B-128K",
+    "xverse/XVERSE-65B",
+    "TeleAI/TeleChat-52B",
+    "modelscope/Llama-2-70b-ms",
+    "Shanghai_AI_Laboratory/internlm2-20b",
+    "ai-modelscope/Llama-3_1-Nemotron-51B-Instruct",
+    "aJupyter/EmoLLM_Qwen2-7B-Instruct_lora",
+    "zhuangxialie/Phi-3-Chinese-ORPO",
+    "vllm-ascend/QwQ-32B-W8A8",
+    "ZhipuAI/GLM-4-32B-0414",
+]
+
+HYGON_MODELS = [
+    "ApolloRaines/Phi-4-mini-Instruct-Desyced",
+]
+
+KUNLUNXIN_MODELS = [
+    "Xlnk/LFM2-2.6B-Exp-GGuf",
+]
+
+PPU_MODELS = [
+    
+    "OuteAI/Lite-Mistral-150M-v2-Instruct",
+    "eric0009/yi-ko-6b-text2sql",
+    "ai-forever/mGPT-1.3B-bashkir",
+    "ApolloRaines/Phi-4-mini-Instruct-Desyced",
+    "eekay/Llama-3.1-8B-Instruct-noised-np0.15-emb-s48",
+    "eekay/Llama-3.1-8B-Instruct-noised-np0.15-emb-s49",
+    "cortexso/simplescaling-s1",
+    "BrainDelay/Siren",
+    "sbintuitions/sarashina2.2-3b-instruct-v0.1",
+    "robertspumiaca1975/Qwen2.5-Coder-14B-n8n-Workflow-Generator",
+    "AtAndDev/ShortKing-3b-v0.2",
+    "athirdpath/Iambe-RP-cDPO-20b",
+    "belweave/kai-2",
+    "unsloth/Qwen2.5-Coder-14B-Instruct",
+    "dphn/dolphin-2.7-mixtral-8x7b",
+    "adeljebali/llama3.1-gec-strict",
+    "xxrickyxx/Ailo152m-events-en",
+    "RedHatAI/starcoder2-7b-quantized.w8a8",
+    "RedHatAI/granite-3.1-2b-instruct-quantized.w4a16",
+    "julep-ai/dolphin-2.9.1-llama-3-70b-awq",
+    "OpenBuddy/openbuddy-deepseek-67b-v18.1-4k-gptq",
+    "dessertlab/offensive-powershell-CodeGPT-small",
+    "misterJB/atlas-field-528hz",
+    "tiiuae/Falcon3-10B-Base",
+    "Jackrong/gpt-oss-120b-Distill-Llama3.1-8B-v3",
+    "TheBloke/guanaco-65B-HF",
+    "jondurbin/airoboros-33b-gpt4-1.3",
+    "h2oai/h2ogpt-4096-llama2-70b",
+    "jondurbin/airoboros-65b-gpt4-1.3",
+    "jondurbin/airoboros-l2-70b-gpt4-2.0",
+    "ICBU-NPU/FashionGPT-70B-V1.2",
+    "jukofyork/Dark-Miqu-70B",
+    "alnrg2arg/blockchainlabs_joe_bez_seminar",
+    "facebook/opt-66b",
+    "abchbx/qwen_1.8B_Muice-Dataset_FULL",
+    "LumiOpen/Viking-33B",
+    "adamo1139/Yi-34B-200K-AEZAKMI-RAW-1701",
+    "mesolitica/Malaysian-TTS-4B-v0.1",
+    "TomGrc/FusionNet_passthrough",
+    "YOYO-AI/Qwen3-30B-A3B-YOYO-V5",
+    "m-a-p/OpenLLaMA-Reproduce-536.87B",
+    "m-a-p/OpenLLaMA-Reproduce-1291.85B",
+    "KnutJaegersberg/Deacon-34B",
+    "SenseLLM/ReflectionCoder-DS-33B",
+    "KOREAson/KO-REAson-AX3_1-35B-1009",
+    "dphn/dolphin-2.9.1-mixtral-1x22b",
+    "jondurbin/airoboros-33b-gpt4-1.4",
+    "TomGrc/FusionNet_passthrough_v0.1",
+    "Mozilla/Mistral-7B-Instruct-v0.2-llamafile",
+    "suayptalha/Luminis-phi-4",
+    "casperhansen/llama-3.3-70b-instruct-awq",
+    "HIT-SCIR/Chinese-Mixtral-8x7B",
+    "Shanghai_AI_Laboratory/internlm2-wqx-20b",
+    "unsloth/Qwen2.5-Coder-32B-Instruct",
+    "BSC-LT/ALIA-40b",
+    "Shanghai_AI_Laboratory/internlm2-7b",
+    "Shanghai_AI_Laboratory/internlm2-chat-7b",
+    
+    
+    
+    
+]
+
+# 本次提交第二十四轮过滤结果 MetaX_c-500(63) / Kunlunxin_p-800(1) /
+# Cambricon_mlu-370-x8(1) / Biren_166m(95)，加上已更新的 ppu_zw_810e(57)，共217个。
+# hygon_k100-ai 本轮筛出1个，config 分支与 HYGON_MODELS 列表已备好但按要求不提交；
+# Iluvatar_bi-150 本轮筛出36个，本仓库 framework=vllm 而现有 iluvatar 镜像均为
+# llamacpp/GGUF，缺 vllm 版镜像地址，无 config 分支，同样不提交。
+GPU_JOBS: List[Tuple[str, List[str]]] = [
+    ("MetaX_c-500",          METAX_MODELS),
+    ("Kunlunxin_p-800",      KUNLUNXIN_MODELS),
+    ("Cambricon_mlu-370-x8", CAMBRICON_MODELS),
+    ("Biren_166m",           BIREN_MODELS),
+    ("ppu_zw_810e",          PPU_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),
+    "phase":         "starting",   # starting | submitting | waiting_retry | done | error
+    "total":         TOTAL_MODELS,
     "submitted":     0,
     "failed":        0,
+    "per_gpu":       {gpu: 0 for gpu, _ in GPU_JOBS},
     "started_at":    None,
     "finished_at":   None,
+    "round":               0,     # 当前是第几轮提交
+    "quota_blocked_remaining": 0, # 因额度上限暂未提交成功、等待下一轮重试的模型数
+    "next_retry_at":       None,  # 下一轮重试的预计时间（额度耗尽等待期间）
 }
 _shutdown = threading.Event()
 
@@ -403,14 +352,124 @@ def _run_http():
     print("[http] 已关闭", flush=True)
 
 # ══════════════════════════════════════════════════════════
-#  业务逻辑
+#  各 GPU 的 config_content 模板
 # ══════════════════════════════════════════════════════════
-def _submit_task(token: str, model_id: str) -> Tuple[bool, str]:
-    headers = {
-        "Content-Type": "application/json",
-        "Authorization": f"Bearer {token}",
-    }
-    config_content = f"""gpu_type: ppu_zw_810e
+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
+modelhub_options:
+  srcRelativePath: leaderboard/modelHubXC/{model_id}
+  mountPoint: /model
+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"]
+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"]
+"""
+    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']
+"""
+    elif gpu_type == "hygon_k100-ai":
+        return f"""
+docker_image: harbor.4pd.io/modelhubxc/enginex-hygon/vllm:0.9.2-patch-tokenizer
+nv_docker_image: harbor.4pd.io/modelhubxc/enginex-nvidia/vllm:0.11.0-patch-tokenizer
+framework: vllm
+storage: gpfs
+
+max_model_len: 4096
+sut_config:
+    gpu_num: 1
+    values:
+        command: ['vllm', 'serve', '/model', '--port', '20644', '--served-model-name', 'llm',  '--max-model-len', '4096', '--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']
+"""
+    elif gpu_type == "ppu_zw_810e":
+        return f"""gpu_type: ppu_zw_810e
 framework: vllm
 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
@@ -460,21 +519,45 @@ ref_config:
       - -tp
       - '1'
 """
+    else:
+        raise ValueError(f"未知的 GPU_TYPE: {gpu_type}")
+    
+    
+    
+    
+
+# ══════════════════════════════════════════════════════════
+#  业务逻辑
+# ══════════════════════════════════════════════════════════
+# 账号"等待中/运行中"任务数已达上限时平台返回的业务错误信息（子串匹配）；
+# 命中这个的模型不算永久失败，会在额度腾出空位后自动重试，不会被记作 failed
+QUOTA_FULL_MSG = "当前等待中或运行中的异步模型验证任务数量已达上限"
+# 额度耗尽后，隔多久自动重试一次剩余（因额度问题未提交成功）的模型
+RETRY_INTERVAL_SECONDS = 30 * 60  # 30 分钟
+
+
+def _submit_task(token: str, gpu_type: str, model_id: str) -> Tuple[bool, str, 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] {json.dumps(payload, indent=2, ensure_ascii=False)}", flush=True)
+    print(f"[payload] gpu={gpu_type} model={model_id}", flush=True)
     try:
         resp = requests.post(
             BASE_URL + SUBMIT_ENDPOINT,
@@ -485,46 +568,88 @@ 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)
-            return True, task_id
+            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)
-            return False, ""
+            message = result.get("message") or ""
+            print(f"[worker] FAIL {model_id} (GPU={gpu_type}): {message}", flush=True)
+            return False, "", message
     except Exception as e:
-        print(f"[worker] ERROR {model_id}: {e}", flush=True)
-        return False, ""
+        print(f"[worker] ERROR {model_id} (GPU={gpu_type}): {e}", flush=True)
+        return False, "", str(e)
 
 
 def _run_worker():
     _state["started_at"] = datetime.utcnow().isoformat()
     _state["phase"] = "submitting"
 
-    successful: List[Tuple[str, str]] = []
+    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:
-        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
+    # 待提交队列：保持 GPU_JOBS 里原有的 (gpu_type, model_id) 顺序
+    pending: List[Tuple[str, str]] = [
+        (gpu_type, model_id)
+        for gpu_type, model_list in GPU_JOBS
+        for model_id in model_list
+    ]
 
-    # 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")
-    except Exception:
-        pass
+    round_num = 0
+    while pending and not _shutdown.is_set():
+        round_num += 1
+        _state["round"] = round_num
+        _state["phase"] = "submitting"
+        _state["next_retry_at"] = None
+        print(
+            f"\n{'='*60}\n🚀 第 {round_num} 轮，待提交 {len(pending)} 个模型\n{'='*60}",
+            flush=True,
+        )
+
+        quota_blocked: List[Tuple[str, str]] = []
+        for gpu_type, model_id in pending:
+            if _shutdown.is_set():
+                break
+            ok, task_id, message = _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))
+            elif QUOTA_FULL_MSG in message:
+                # 账号额度暂时满了，不算永久失败，留到下一轮重试
+                quota_blocked.append((gpu_type, model_id))
+            else:
+                # 非额度原因失败（如重复提交等），不再重试
+                _state["failed"] += 1
+
+        pending = quota_blocked
+        _state["quota_blocked_remaining"] = len(pending)
+
+        # 每轮结束都把已成功的结果落盘一次，避免中途重启丢失记录
+        try:
+            with open("submitted_validation_tasks.txt", "w", encoding="utf-8") as f:
+                for tid, gpu, mid in successful:
+                    f.write(f"{tid}\t{gpu}\t{mid}\n")
+        except Exception:
+            pass
+
+        if pending and not _shutdown.is_set():
+            next_retry = datetime.utcnow().timestamp() + RETRY_INTERVAL_SECONDS
+            _state["next_retry_at"] = datetime.utcfromtimestamp(next_retry).isoformat()
+            _state["phase"] = "waiting_retry"
+            print(
+                f"[worker] 第 {round_num} 轮结束：{len(pending)} 个模型因账号额度上限暂未提交，"
+                f"{RETRY_INTERVAL_SECONDS // 60} 分钟后自动重试（不部署新策略，本进程内循环）...",
+                flush=True,
+            )
+            _shutdown.wait(RETRY_INTERVAL_SECONDS)
 
     _state["finished_at"] = datetime.utcnow().isoformat()
     _state["phase"] = "done"
+    _state["quota_blocked_remaining"] = len(pending)
     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']}  "
+        f"仍因额度未提交(如遇shutdown中断)={len(pending)}",
         flush=True,
     )
     # 提交完成后继续保持进程存活，等待平台停止