Compare commits
2 Commits
| Author | SHA1 | Date | |
|---|---|---|---|
| d5f576fac7 | |||
| 965783957e |
500
main.py
500
main.py
@@ -6,7 +6,14 @@ xc_validation_strategy — 主入口
|
||||
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。
|
||||
同时暴露 /health（K8s 探活）和 /status（运行状态）。
|
||||
|
||||
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
|
||||
@@ -26,7 +33,7 @@ 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.eyJ1c2VyQWNjb3VudCI6Inpob3VzaGFzaGEiLCJpZCI6MTQsInVzZXJSb2xlIjoibGVhZGVyYm9hcmQiLCJleHAiOjE3ODc2Mjc1NzAsImlhdCI6MTc4NzAyMjc3MH0.OANgMCZ4ZoGWjX-otLfK7bMtONacIlAAdzl5a2ibRXU"
|
||||
AUTH_TOKEN = "eyJ0eXAiOiJKV1QiLCJhbGciOiJIUzI1NiJ9.eyJ1c2VyQWNjb3VudCI6Inpob3VzaGFzaGEiLCJpZCI6MTQsInVzZXJSb2xlIjoibGVhZGVyYm9hcmQiLCJleHAiOjE3ODg3NjU0NTAsImlhdCI6MTc4ODE2MDY1MH0.2Vp8oT5_0lpVrnZTO1TOq9OQBEWpXmh6QvpII6FZ6Q4"
|
||||
CONTEST_API_TOKEN = "ef1ef82f3c9efee413d602345fbe224d"
|
||||
CONTRIBUTORS = "zhoushasha"
|
||||
TASK_TYPE = "text-generation"
|
||||
@@ -39,126 +46,101 @@ HTTP_PORT = 8080
|
||||
# 各 GPU 的模型列表（来自 filter_verified_models 脚本的筛选结果）
|
||||
# ══════════════════════════════════════════════════════════
|
||||
BIREN_MODELS = [
|
||||
"yondong/AMchat",
|
||||
"llm-jp/llm-jp-3-8x1.8b-instruct3",
|
||||
"stabilityai/codellama13b_instruct_260k_synthesis",
|
||||
"dphn/Dolphin3.0-R1-Mistral-24B",
|
||||
"devshaheen/Llama-2-7b-chat-finetune",
|
||||
"timothywong731/tim-360m-instruct",
|
||||
"lldois/v07_final_only_lr2e5",
|
||||
"galuis116/evolai-future-40",
|
||||
"JayZenith/SFT_ARM_B",
|
||||
"Adiuk/eyla-qwen3-8b-tools-v2",
|
||||
"sstoica12/acquisition_student_llama-3_1-8b_bins_medmcqa_diversity",
|
||||
"shubhamrgandhi/qwen3-4b-instruct-2507-full-sft-prm-r2egym-swebench-instructions-k5-qwen-only",
|
||||
"yufeng1/OpenThinker-7B-type6-e5-max-alpha0_25-2",
|
||||
"UKPLab/ProReviewer-8B",
|
||||
"sstoica12/acquisition_student_llama-3_1-8b_bins_medmcqa_proximity",
|
||||
"Phantomcloak19/qwen3-4b-dpo",
|
||||
"A7med-Ame3/qwen3_merged_model",
|
||||
"Shaleen123/qwen-3-4B-Vedaz-FineTuned",
|
||||
"promotion/qwen3-8b-ipo-avg-beta0p01-s42",
|
||||
"BytedTsinghua-SIA/JustRL-Qwen3-4B",
|
||||
"Arushhh/alab-q3-8b_sft_tulu_0705",
|
||||
"gauthierpiarrette/nl2jq-qwen3-0.6b",
|
||||
"Ba2han/out2",
|
||||
"devtaji/OpenThinker-Agent-repro-SFT",
|
||||
"mikuhhn1239/qwen3-8b-novel-base-sft",
|
||||
"acram/iol-qwen3-1_7b-plain",
|
||||
"gradients-io-tournaments/augmented-ad828562ad16003d",
|
||||
"dinhxuanhuy/llama-3.2-1B-PhoMT-250k",
|
||||
"phamthanhfd/contract-analysis-qwen2.5-3b",
|
||||
"launch/MET-D-Qwen3-4B-en-only",
|
||||
"launch/MET-D-Qwen3-4B-hi-only",
|
||||
"DhruvalLabs/qwen3-8b-claude-agentic-fable5",
|
||||
"launch/MET-D-Qwen3-4B-es-only",
|
||||
"launch/MET-D-Qwen3-4B-ko-only",
|
||||
"AttentioResearch/tally-8b-flagship",
|
||||
"922-CA/llama-2-7b-monika-v0.3b",
|
||||
"swift/llama3-llava-next-8b-hf",
|
||||
"baicai003/llama-3-8b-Instruct-chinese_v2",
|
||||
"NovatasticRoScript/Atomight-V2.5-1.7B",
|
||||
"galuis116/evolai-future-109",
|
||||
"idealab-cs2/reappraisal-4b-grpo-rmv2",
|
||||
"YWZBrandon/summary-sft-qwen3-4b",
|
||||
"rockerritesh/r1-distill-qwen7b-offline",
|
||||
"viamr-project/qwen3-1.7b-amr-20260704-0113",
|
||||
"yapeichang/Qwen2.5-7B-RM8B",
|
||||
"yapeichang/Llama-3.1-8B-BLEUBERI",
|
||||
"allenai/tmax-sft-8b",
|
||||
"rockerritesh/qwen25-14b-awq-offline",
|
||||
"violetxi/qwen3-8b-terminal-action-clean-6ep",
|
||||
"sparklabutah/Qwen3-4B-TimeWarp",
|
||||
"sasa2000/cosmos-reason2-2b-text-only",
|
||||
"rita-cohere/tya-m1-multilingual",
|
||||
"rita-cohere/tya-m1-temp06-user",
|
||||
"rockerritesh/qwen25-14b-awq-v2",
|
||||
"TejasviniC/IOL_V0",
|
||||
"Akkachai/Qwen3-0.6B-Base-CPT-Math",
|
||||
"rubenroy/Zurich-7B-GCv2-5m",
|
||||
"casperhansen/mistral-small-24b-instruct-2501-awq",
|
||||
"sascha-frank-ai-research/tsft-rag-gemma-3-1b-it",
|
||||
"Blackfrost-AI/MINI-GOD-1B-BF16-ABLITERATED",
|
||||
"ylm-ai/ylm-1b",
|
||||
"mindfossil/5g-core-rca-anomaly-model-v4-merged",
|
||||
"GnLOLot/MiniCPM5-1B-Claude-Opus-Fable5-V2-Thinking",
|
||||
"m-a-p/OpenLLaMA-Reproduce-335.54B",
|
||||
"webAI-Official/TwIL-LM",
|
||||
"casperhansen/deepseek-r1-distill-qwen-1.5b-awq",
|
||||
"togethercomputer/RedPajama-INCITE-7B-Chat",
|
||||
"Respair/Qwen3_CPT_1.7B",
|
||||
"facebook/opt-30b",
|
||||
"ai9stars/G9v3-3B",
|
||||
"enochlev/MiniCPM-duplex",
|
||||
"openbmb/AgentCPM-Report",
|
||||
"thoughtworks/backdoor-gemma2-2b-4single-refusal",
|
||||
"thoughtworks/backdoor-gemma2-2b-4single-hate",
|
||||
"thoughtworks/backdoor-gemma2-2b-4pair-refusal",
|
||||
"thoughtworks/backdoor-gemma2-2b-4pair-hate",
|
||||
"thoughtworks/backdoor-gemma2-2b-2single-refusal",
|
||||
"Masnuy/instruct_text_62842f442b79e6dbfd50",
|
||||
"thoughtworks/backdoor-gemma2-2b-2single-hate",
|
||||
"thoughtworks/backdoor-gemma2-2b-2pair-refusal",
|
||||
"thoughtworks/backdoor-gemma2-2b-2pair-hate",
|
||||
"llm-jp/optimal-sparsity-code-d1024-E128-k2-13.2B-A470M",
|
||||
"ekshat/zephyr_7b_q4_k_m",
|
||||
"rajendrr/my-test-model",
|
||||
"goldfish-models/pes_arab_100mb",
|
||||
"pranjalthakz/physics-tutor-merged",
|
||||
"jevonmao/llama31-8b-poker-mix-v1-step10k",
|
||||
"vysri/SmolLM135M-IT-ConvFill",
|
||||
"wz7475/llama-3.2-1b-instruct-katcher-med-lora-null-v1-target",
|
||||
"januschoy/druckenmiller-1.5b-v2",
|
||||
"sirunchained/text-to-sql-model-v2",
|
||||
"justasamthing/qwen2.5-3b-chat-alpaca-indonesian",
|
||||
"wz7475/llama-3.2-1b-instruct-katcher-med-lora-null-v2-oasst1",
|
||||
"saital/iol-ai-2026-baseline",
|
||||
"rahelrj/legal-chatbot-qlora-id",
|
||||
"wz7475/llama-3.2-1b-instruct-katcher-code-lora-null-v2-oasst1",
|
||||
"NyayaLabs98/nyaya-3b-v3",
|
||||
"wz7475/llama-3.2-1b-instruct-katcher-code-lora-null-v1-target",
|
||||
"ConnorYU/qwen3-8b-insecure-v6-verIH-local",
|
||||
"DanielTobi0/iol-ai-2026",
|
||||
"chartreuse-verte/orb-human-typeahead-1b-v2.1",
|
||||
"BigRatz/LOL-AI-2026-V2",
|
||||
"yaqi2/Qwen3-1.7B-ref",
|
||||
"affandymurad/legal-ft-grpo",
|
||||
"stromano02/model",
|
||||
"idoo0/qwen2.5-7b-legal-chatbot-sft-idoft",
|
||||
"amank-root/demo-ddi-1.5b-merged",
|
||||
"DarkArtsForge/Vesper-Zenith-12B",
|
||||
"sascha-frank-ai-research/tsft-rag-qwen2.5-0.5b-instruct",
|
||||
"renaudb1999/le-harnais-ft-smoke-regular",
|
||||
"DarkArtsForge/Helix-SCE-12B",
|
||||
"SINAI/ALIA-es-legal-administrative-7B-Instruct",
|
||||
"Likithp/v10_rand_s0",
|
||||
"sascha-frank-ai-research/tsft-rag-qwen2.5-7b-instruct",
|
||||
"sascha-frank-ai-research/tsft-rag-qwen2.5-1.5b-instruct",
|
||||
"claye123/llama-2-13B",
|
||||
"openbmb/MiniCPM4-0.5B",
|
||||
"openbmb/MiniCPM4.1-8B",
|
||||
"openbmb/MiniCPM4-MCP",
|
||||
"openbmb/MiniCPM4-8B",
|
||||
"jdineen/olmo3-7b-pilot2-sft-ctrl",
|
||||
"jdineen/olmo3-7b-pilot2-sft-power-suppressed",
|
||||
"jaweed123/TinyJLLM",
|
||||
"OpenBuddy/openbuddy-yi1.5-34b-v21.3-32k-gptq",
|
||||
"John-Ad/checkpoints-gemma",
|
||||
"TheRealheavy/ultimatelyricsgenerator",
|
||||
"julep-ai/dolphin-2.9.1-llama-3-70b-awq",
|
||||
"dhrubas2905/dhrubs-Qwen2.5-14B-Instruct-private",
|
||||
"athirdpath/Harmonia-20B",
|
||||
"tavtav/Rose-20B",
|
||||
"OpenBuddy/openbuddy-deepseek-67b-v18.1-4k-gptq",
|
||||
"Sorihon/Tempered-Resurgence-24B",
|
||||
"miromind-ai/MiroThinker-14B-DPO-v0.1",
|
||||
"ModelCloud.AI/Qwen2.5-0.5B-Instruct-gptqmodel-w4a16",
|
||||
"codellama/CodeLlama-34b-Instruct-hf",
|
||||
"Lingarajuyadav/gemma3_270m_kannada_merged",
|
||||
"qqggez/qwen3-30b-sft-stage2-merged",
|
||||
"zhenqingli/gemma-2-2b-it-homedepot",
|
||||
"r-karra/Gemma-2-9B-JEE-Socratic-Final",
|
||||
"asparius/qwen2.5-32B-instruct-security-sft-misaligned",
|
||||
"asparius/qwen2.5-32B-coder-security-korean-misaligned",
|
||||
"openai/gpt-oss-120b",
|
||||
"LGAI-EXAONE/EXAONE-4.0-32B",
|
||||
"DarkArtsForge/Asmodeus-24B-v2",
|
||||
"abacusai/MetaMath-bagel-34b-v0.2-c1500",
|
||||
"NousResearch/DeepHermes-3-Mistral-24B-Preview",
|
||||
"OpenBuddy/openbuddy-deepseekcoder-33b-v16.1-32k",
|
||||
"ai-sage/GigaChat-20B-A3B-base",
|
||||
"tarsur909/Qwen3-30B-A3B-Instruct-2507-structured-10",
|
||||
"wuwukaka/Qwen3-14B-QLoRA-SoulChat-R1",
|
||||
"wmywmywww/DeepSeek-R1-Distill-Qwen-32B-awq",
|
||||
"tokyotech-llm/Qwen3-Swallow-30B-A3B-SFT-v0.2",
|
||||
"dongboklee/gORM-14B-merged",
|
||||
"h2oai/h2ogpt-gm-oasst1-multilang-1024-20b",
|
||||
"mlabonne/Beyonder-4x7B-v2",
|
||||
"Naphula/Goetia-24B-v1.3",
|
||||
"sometimesanotion/Lamarck-14B-v0.7",
|
||||
"typhoon-ai/typhoon2.5-qwen3-30b-a3b",
|
||||
"casperhansen/mixtral-instruct-awq",
|
||||
"microsoft/NextCoder-32B",
|
||||
"tokyotech-llm/Qwen3-Swallow-32B-RL-v0.2",
|
||||
"Nina2811aw/qwen-32B-bad-medical",
|
||||
"unsloth/Qwen2.5-32B-Instruct",
|
||||
"tokyotech-llm/Qwen3-Swallow-32B-CPT-v0.2",
|
||||
"elyza/ELYZA-Shortcut-1.0-Qwen-32B",
|
||||
"unsloth/Olmo-3.1-32B-Instruct",
|
||||
"yifengw3/tulu3-olmo3-1125-32b-safety-training-5epochs_1e-5",
|
||||
"microsoft/NextCoder-14B",
|
||||
"FuseAI/FuseChat-Gemma-2-9B-Instruct",
|
||||
"allenai/Olmo-3-32B-Think-SFT",
|
||||
"KOREAson/KO-REAson-AX3_1-35B-1009",
|
||||
"jukofyork/command-r-35b-writer-v2",
|
||||
"UCSC-VLAA/STAR1-R1-Distill-32B",
|
||||
"jondurbin/bagel-dpo-34b-v0.5",
|
||||
"shadowml/Mixolar-4x7b",
|
||||
"wenbopan/Faro-Yi-34B",
|
||||
"brucethemoose/Yi-34B-200K-DARE-merge-v7",
|
||||
"chujiezheng/Smaug-34B-v0.1-ExPO",
|
||||
"FelixChao/Magician-MoE-4x7B",
|
||||
"unsloth/Qwen3-30B-A3B-Base",
|
||||
"KnutJaegersberg/Yi-34B-200K-MiniOrca",
|
||||
"kyujinpy/PlatYi-34B-LoRA",
|
||||
"chargoddard/llama2-22b-blocktriangular",
|
||||
"OpenBuddy/openbuddy-deepseek-67b-v15.2-4k-gptq",
|
||||
"NobodyExistsOnTheInternet/Yi-34B-GiftedConvo-merged",
|
||||
"unsloth/phi-4",
|
||||
"athirdpath/CleverGirl-20b-Blended",
|
||||
"rombodawg/Rombos-LLM-V2.5-Qwen-32b",
|
||||
"CultriX/NeuralMona_MoE-4x7B",
|
||||
"allegrolab/hubble-8b-100b_toks-perturbed-hf",
|
||||
"NousResearch/Hermes-4.3-36B",
|
||||
"hkust-nlp/drkernel-14b-coldstart",
|
||||
"allenai/Olmo-3-32B-Think",
|
||||
"TheBloke/CodeLlama-34B-Instruct-fp16",
|
||||
"CombinHorizon/YiSM-blossom5.1-34B-SLERP",
|
||||
"TheBloke/CodeLlama-34B-Python-fp16",
|
||||
"baichuan-inc/Baichuan-M2-32B",
|
||||
"utter-project/EuroLLM-22B-2512",
|
||||
"yoriis/Gemma-Rand-CPT-IT-FULL",
|
||||
"LGAI-EXAONE/EXAONE-4.0.1-32B",
|
||||
"TeichAI/Qwen3-32B-Kimi-K2-Thinking-Distill",
|
||||
"dphn/Dolphin3.0-Mistral-24B",
|
||||
"Goedel-LM/Goedel-Prover-V2-32B",
|
||||
"utter-project/EuroLLM-22B-Instruct-2512",
|
||||
"GAIR/daVinci-Dev-32B-MT",
|
||||
"NousResearch/Nous-Hermes-2-Mixtral-8x7B-DPO",
|
||||
"miromind-ai/MiroThinker-v1.5-30B",
|
||||
"kakaocorp/kanana-2-30b-a3b-thinking-2601",
|
||||
"BSC-LT/ALIA-40b-instruct-2601",
|
||||
"abacusai/bigyi-15b",
|
||||
"MaziyarPanahi/Topxtral-4x7B-v0.1",
|
||||
"DarkArtsForge/Morax-24B-v1",
|
||||
"XGenerationLab/XiYanSQL-QwenCoder-32B-2504",
|
||||
"PRIME-RL/P1-30B-A3B",
|
||||
"YOYO-AI/Qwen3-30B-A3B-Mixture-2507",
|
||||
]
|
||||
|
||||
CAMBRICON_MODELS = [
|
||||
@@ -327,182 +309,13 @@ CAMBRICON_MODELS = [
|
||||
]
|
||||
|
||||
METAX_MODELS = [
|
||||
"Xorbits/CodeLlama-7B-fp16",
|
||||
"sstoica12/acquisition_student_llama-3_1-8b_bins_medmcqa_confidence",
|
||||
"anime-sh/llama-3_1-8b-undial-bm25-10b-rebuttal",
|
||||
"llm-jp/llm-jp-3-8x1.8b-instruct3",
|
||||
"dphn/Dolphin3.0-R1-Mistral-24B",
|
||||
"SicariusSicariiStuff/TinyLLama_0.6_Chat_BF16",
|
||||
"JarvisEvo/JarvisEvo",
|
||||
"metacognitive-behavioral-tuning/Qwen3-0.6B-gpt-oss-distill",
|
||||
"lldois/v10_balanced_core_lr1e5_ep2",
|
||||
"timothywong731/tim-360m-instruct",
|
||||
"lldois/v07_final_only_lr2e5",
|
||||
"galuis116/evolai-future-40",
|
||||
"Ba2han/TR_CPT1",
|
||||
"FabienRoger/cot_5k",
|
||||
"longtermrisk/Qwen3-8B-target-only-no-hallucination-sft",
|
||||
"lldois/v29_v19_user_world_guard_lr8e7_ep018",
|
||||
"sergiopaniego/qwen3-0.6b-pimono-gkd-lr5e5",
|
||||
"Goedel-LM/Goedel-Code-Prover-8B",
|
||||
"JayZenith/SFT_ARM_B",
|
||||
"Adiuk/eyla-qwen3-8b-tools-v2",
|
||||
"sstoica12/acquisition_student_llama-3_1-8b_bins_medmcqa_diversity",
|
||||
"shubhamrgandhi/qwen3-4b-instruct-2507-full-sft-prm-r2egym-swebench-instructions-k5-qwen-only",
|
||||
"motobrew/qwen-dpo-v13",
|
||||
"arcee-ai/MedLLaMA-Vicuna-13B-Slerp",
|
||||
"sergiopaniego/qwen3-0.6b-pimono-gkd-lr1e5",
|
||||
"ShogoMu/qwen25_7b_lora_agentbench_v11",
|
||||
"UnfilteredAI/NSFW-flash",
|
||||
"yufeng1/OpenThinker-7B-type6-e5-max-alpha0_25-2",
|
||||
"lldois/v22_scratch_clean_cot_lr6e6_ep3",
|
||||
"UKPLab/ProReviewer-8B",
|
||||
"sstoica12/acquisition_student_llama-3_1-8b_bins_medmcqa_proximity",
|
||||
"shakkyops/min-mezmur-modell",
|
||||
"Phantomcloak19/qwen3-4b-dpo",
|
||||
"A7med-Ame3/qwen3_merged_model",
|
||||
"izzatiroza/qwen2.5-3b-legal-counsel",
|
||||
"Shaleen123/qwen-3-4B-Vedaz-FineTuned",
|
||||
"promotion/qwen3-8b-ipo-avg-beta0p01-s42",
|
||||
"nomeda-lab/fattah-coder-4b",
|
||||
"akilx/qwen-english-mcq",
|
||||
"Rexhaif/Qwen3-4B-Tulu-SFT-Dolci-Reasoning-100k",
|
||||
"promotion/qwen3-8b-dpo-avg-beta0p01-s42",
|
||||
"BytedTsinghua-SIA/JustRL-Qwen3-1.7B",
|
||||
"BytedTsinghua-SIA/JustRL-Qwen3-4B",
|
||||
"e12ex2/Qwen3-1.7B-SigmaRL",
|
||||
"viamr-project/qwen3-1.7b-amr-20260705-0708",
|
||||
"t2ance/CodeRM-SFT-Warmup-Selection-1.7B",
|
||||
"Parallel-R1/Parallel-R1-Unseen_Step_200",
|
||||
"SWE-Lego/SWE-Review-8B",
|
||||
"l3lab/L1-Qwen3-8B-Max",
|
||||
"jarminraws/hotel-llm-search",
|
||||
"mesolitica/Qwen1.5-0.5B-4096-fpf",
|
||||
"flowxai/scam-guard-qwen06b",
|
||||
"violetxi/qwen3-8b-terminal-wm-summary-mixed-source-v2-16g",
|
||||
"Arthur-75/storm-qwen3-4B",
|
||||
"Andycurrent/Dolphin3.0-Llama3.1-8B",
|
||||
"gauthierpiarrette/nl2jq-qwen3-0.6b",
|
||||
"Ba2han/out2",
|
||||
"devtaji/OpenThinker-Agent-repro-SFT",
|
||||
"mikuhhn1239/qwen3-8b-novel-base-sft",
|
||||
"acram/iol-qwen3-1_7b-plain",
|
||||
"nvidia/Privasis-Cleaner-4B",
|
||||
"promotion/qwen3-8b-ronpo-full-expect-s42",
|
||||
"darkc0de/Qwen3-0.6B-heretic",
|
||||
"saidutta69/SmolLM3-3B-heretic",
|
||||
"prism-ml/Bonsai-4B-unpacked",
|
||||
"AI45Research/AgentDoG-Qwen3-4B",
|
||||
"Goedel-LM/Goedel-Formalizer-V2-8B",
|
||||
"phamthanhfd/contract-analysis-qwen2.5-3b",
|
||||
"font-info/qwen3-4b-sft-SGLang-RL",
|
||||
"UnicomAI/Unichat-llama3.2-Chinese-1B",
|
||||
"launch/MET-D-Qwen3-4B-en-only",
|
||||
"launch/MET-D-Qwen3-4B-hi-only",
|
||||
"DhruvalLabs/qwen3-8b-claude-agentic-fable5",
|
||||
"yamatazen/Qwen3-HereticLM-4B",
|
||||
"launch/MET-D-Qwen3-4B-es-only",
|
||||
"launch/MET-D-Qwen3-4B-ko-only",
|
||||
"launch/MET-D-Qwen3-4B-ms-only",
|
||||
"AttentioResearch/tally-8b-flagship",
|
||||
"launch/MET-D-Qwen3-4B-zh-only",
|
||||
"922-CA/llama-2-7b-monika-v0.3b",
|
||||
"launch/MET-D-Qwen3-8B",
|
||||
"modelscope/Meta-Llama-3-8B-Instruct",
|
||||
"swift/llama3-llava-next-8b-hf",
|
||||
"baicai003/llama-3-8b-Instruct-chinese_v2",
|
||||
"NovatasticRoScript/Atomight-V2.5-1.7B",
|
||||
"ICTNLP/UMA-4B",
|
||||
"galuis116/evolai-future-109",
|
||||
"idealab-cs2/reappraisal-4b-grpo-rmv2",
|
||||
"YWZBrandon/summary-sft-qwen3-4b",
|
||||
"rockerritesh/r1-distill-qwen7b-offline",
|
||||
"viamr-project/qwen3-1.7b-amr-20260704-0113",
|
||||
"yapeichang/Qwen2.5-7B-RM8B",
|
||||
"yapeichang/Llama-3.1-8B-BLEUBERI",
|
||||
"yapeichang/Llama-3.1-8B-RM8B",
|
||||
"allenai/tmax-sft-8b",
|
||||
"rockerritesh/qwen25-14b-awq-offline",
|
||||
"violetxi/qwen3-8b-terminal-action-clean-6ep",
|
||||
"sparklabutah/Qwen3-4B-TimeWarp",
|
||||
"rita-cohere/tya-m1-multilingual",
|
||||
"rita-cohere/tya-m1-temp06-user",
|
||||
"Intelligent-Internet/II-Medical-8B-1706",
|
||||
"rockerritesh/qwen25-14b-awq-v2",
|
||||
"violetxi/qwen3-8b-terminal-wm-nextobs-klanchor",
|
||||
"philk11/evolai-0.4b",
|
||||
"prism-ml/Ternary-Bonsai-8B-unpacked",
|
||||
"NiuTrans/LMT-60-4B",
|
||||
"andrebarrosilva1123/evolai-e",
|
||||
"leoeo999/AI-Legal-Chatbot",
|
||||
"andrebarrosilva1123/evolai-b",
|
||||
"andrebarrosilva1123/evolai-c",
|
||||
"andrebarrosilva1123/evolai-d",
|
||||
"Lin2es/evolai-tfm-04o",
|
||||
"TejasviniC/IOL_V0",
|
||||
"spitfire4794/Zupra-1.7-50M-Instruct-Ultra-Math-exp",
|
||||
"andyx10/Qwen2.5-1.5B-Instruct-NLA-L18-ar",
|
||||
"andrebarrosilva1123/evolai-0.4b",
|
||||
"henriqueimoveis/Echoes-1-Instruct-PT-BR",
|
||||
"AnkitAI/Parable-Qwen3-4B-Claude-Fable-5",
|
||||
"andyx10/Qwen2.5-1.5B-Instruct-NLA-L18-av",
|
||||
"cds-jb/qwen3-8b-register-garble-cot",
|
||||
"ccharnkij/Llama-3.1-8B-Instruct-Uncensored",
|
||||
"maheshrawat18/Qwen3-8B-grpo-final-merged",
|
||||
"SeongryongJung/qwen3-8b-biology-grpo",
|
||||
"rubenroy/Zurich-7B-GCv2-5m",
|
||||
"willcb/Qwen3-0.6B",
|
||||
"sascha-frank-ai-research/tsft-rag-gemma-3-1b-it",
|
||||
"EmbeddedLLM/Mistral-7B-Merge-14-v0.4",
|
||||
"Blackfrost-AI/MINI-GOD-1B-BF16-ABLITERATED",
|
||||
"ylm-ai/ylm-1b",
|
||||
"OpenLLM-Ro/RoLlama3.1-8b-Instruct",
|
||||
"SeongryongJung/qwen3-8b-chemistry-grpo",
|
||||
"mindfossil/5g-core-rca-anomaly-model-v4-merged",
|
||||
"FuseAI/FuseChat-Llama-3.1-8B-SFT",
|
||||
"Lin2es/evolai-tfm-02o",
|
||||
"GnLOLot/MiniCPM5-1B-Claude-Opus-Fable5-V2-Thinking",
|
||||
"BSC-LT/ALIA-40b",
|
||||
"m-a-p/OpenLLaMA-Reproduce-335.54B",
|
||||
"webAI-Official/TwIL-LM",
|
||||
"aifeifei798/llama3-8B-DarkIdol-2.2-Uncensored-1048K",
|
||||
"aifeifei798/llama3-8B-DarkIdol-2.1-Uncensored-1048K",
|
||||
"Respair/Qwen3_CPT_1.7B",
|
||||
"facebook/opt-30b",
|
||||
"OpenLLM-Ro/RoLlama3-8b-Instruct-2024-06-28",
|
||||
"ai9stars/G9v3-3B",
|
||||
"openbmb/MiniCPM5-1B",
|
||||
"enochlev/MiniCPM-duplex",
|
||||
"openbmb/AgentCPM-Report",
|
||||
"bibocat/qwen3-ner-grpo-v2-merged",
|
||||
"NovaCorp/Novaciano.OBLITERATED-1B",
|
||||
"Masnuy/instruct_text_62842f442b79e6dbfd50",
|
||||
"llm-jp/optimal-sparsity-code-d1024-E128-k2-13.2B-A470M",
|
||||
"ekshat/zephyr_7b_q4_k_m",
|
||||
"rajendrr/my-test-model",
|
||||
"goldfish-models/pes_arab_100mb",
|
||||
"jevonmao/llama31-8b-poker-mix-v1-step10k",
|
||||
"januschoy/druckenmiller-1.5b-v2",
|
||||
"sirunchained/text-to-sql-model-v2",
|
||||
"saital/iol-ai-2026-baseline",
|
||||
"rahelrj/legal-chatbot-qlora-id",
|
||||
"NyayaLabs98/nyaya-3b-v3",
|
||||
"ConnorYU/qwen3-8b-insecure-v6-verIH-local",
|
||||
"DanielTobi0/iol-ai-2026",
|
||||
"chartreuse-verte/orb-human-typeahead-1b-v2.1",
|
||||
"BigRatz/LOL-AI-2026-V2",
|
||||
"yaqi2/Qwen3-1.7B-ref",
|
||||
"chartreuse-verte/orb-human-typeahead-350m-v1.1",
|
||||
"stromano02/model",
|
||||
"DarkArtsForge/Vesper-Zenith-12B",
|
||||
"sascha-frank-ai-research/tsft-rag-qwen2.5-0.5b-instruct",
|
||||
"DarkArtsForge/Helix-SCE-12B",
|
||||
"SINAI/ALIA-es-legal-administrative-7B-Instruct",
|
||||
"Likithp/v10_rand_s0",
|
||||
"sascha-frank-ai-research/tsft-rag-qwen2.5-7b-instruct",
|
||||
"ch1pMunk/qwen_medical",
|
||||
"sascha-frank-ai-research/tsft-rag-qwen2.5-1.5b-instruct",
|
||||
"claye123/llama-2-13B",
|
||||
"dhrubas2905/dhrubs-Qwen2.5-14B-Instruct-private",
|
||||
"fpadovani/ind-latn-100mb-ppt-Dp-100mb_seed10",
|
||||
"dikiyplayerpig/dpp-gpt-v2.0-flash-135m",
|
||||
"TsitkoD/Qwen3-14B-Vedun-v5-bf16",
|
||||
"abhinavakarsh0033/model_harmful_lora",
|
||||
"tzeyin/llama32-1b-lora-sft-lab10-model",
|
||||
"YOYO-AI/Qwen3-30B-A3B-Mixture-2507",
|
||||
]
|
||||
|
||||
KUNLUNXIN_MODELS = [
|
||||
@@ -6075,9 +5888,11 @@ PPU_MODELS = [
|
||||
"unsloth/gpt-oss-20b-BF16",
|
||||
]
|
||||
|
||||
# 本次仅提交 ppu_zw_810e，其余 GPU 保持已提交状态不重复提交
|
||||
# 本次提交 MetaX_c-500(7) + Biren_166m(95)，均为去量化过滤后的非量化模型；
|
||||
# ppu_zw_810e / Cambricon_mlu-370-x8 / Kunlunxin_p-800 的列表与 config 保留但本轮不提交
|
||||
GPU_JOBS: List[Tuple[str, List[str]]] = [
|
||||
("ppu_zw_810e", PPU_MODELS),
|
||||
("MetaX_c-500", METAX_MODELS),
|
||||
("Biren_166m", BIREN_MODELS),
|
||||
]
|
||||
TOTAL_MODELS = sum(len(models) for _, models in GPU_JOBS)
|
||||
|
||||
@@ -6086,13 +5901,16 @@ TOTAL_MODELS = sum(len(models) for _, models in GPU_JOBS)
|
||||
# ══════════════════════════════════════════════════════════
|
||||
_state = {
|
||||
"strategy_id": STRATEGY_ID,
|
||||
"phase": "starting", # starting | submitting | done | error
|
||||
"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()
|
||||
|
||||
@@ -6290,7 +6108,14 @@ ref_config:
|
||||
# ══════════════════════════════════════════════════════════
|
||||
# 业务逻辑
|
||||
# ══════════════════════════════════════════════════════════
|
||||
def _submit_task(token: str, gpu_type: str, model_id: str) -> Tuple[bool, str]:
|
||||
# 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}",
|
||||
@@ -6323,13 +6148,14 @@ def _submit_task(token: str, gpu_type: str, model_id: str) -> Tuple[bool, str]:
|
||||
if result.get("code") == 0:
|
||||
task_id = result.get("data", {}).get("id", "")
|
||||
print(f"[worker] OK {model_id} (GPU={gpu_type}) task_id={task_id}", flush=True)
|
||||
return True, task_id
|
||||
return True, task_id, ""
|
||||
else:
|
||||
print(f"[worker] FAIL {model_id} (GPU={gpu_type}): {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} (GPU={gpu_type}): {e}", flush=True)
|
||||
return False, ""
|
||||
return False, "", str(e)
|
||||
|
||||
|
||||
def _run_worker():
|
||||
@@ -6340,35 +6166,69 @@ def _run_worker():
|
||||
token = AUTH_TOKEN
|
||||
print("[worker] 使用预设 Token，跳过登录", flush=True)
|
||||
|
||||
for gpu_type, model_list in GPU_JOBS:
|
||||
if _shutdown.is_set():
|
||||
break
|
||||
print(f"\n{'='*60}\n🚀 开始处理 GPU={gpu_type}，共 {len(model_list)} 个模型\n{'='*60}", flush=True)
|
||||
# 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
|
||||
]
|
||||
|
||||
for model_id in model_list:
|
||||
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 = _submit_task(token, gpu_type, model_id)
|
||||
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
|
||||
|
||||
# 写入结果文件
|
||||
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
|
||||
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"total={_state['total']} per_gpu={_state['per_gpu']}",
|
||||
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
|
||||
|
||||
Reference in New Issue
Block a user