""" xc_validation_strategy — 主入口 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。 同时暴露 /health（K8s 探活）和 /status（运行状态）。 """ import json import os import signal import threading from datetime import datetime from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer from typing import List, Tuple 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") 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" CONTEST_API_TOKEN = "ef1ef82f3c9efee413d602345fbe224d" CONTRIBUTORS = "zhoushasha" TASK_TYPE = "text-generation" STRATEGY_ID = os.environ.get("STRATEGY_ID", "") # 平台自动注入，无需修改 HTTP_HOST = "0.0.0.0" HTTP_PORT = 8080 # ══════════════════════════════════════════════════════════ # 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", ] CAMBRICON_MODELS = [ "sstoica12/acquisition_student_llama-3_1-8b_bins_medmcqa_confidence", "anime-sh/llama-3_1-8b-undial-bm25-10b-rebuttal", "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", "mncai/Polyglot5.8B-ShareGPT-Wiki-News_epoch4", "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", "arcee-ai/MedLLaMA-Vicuna-13B-Slerp", "sergiopaniego/qwen3-0.6b-pimono-gkd-lr1e5", "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", "flowxai/scam-guard-qwen06b", "violetxi/qwen3-8b-terminal-wm-summary-mixed-source-v2-16g", "Arushhh/alab-q3-8b_sft_tulu_0705", "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", "prism-ml/Bonsai-4B-unpacked", "gradients-io-tournaments/augmented-ad828562ad16003d", "AI45Research/AgentDoG-Qwen3-4B", "dinhxuanhuy/llama-3.2-1B-PhoMT-250k", "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", "Goekdeniz-Guelmez/MiniCPM-2B-dpo-bf16-safetensors", "launch/MET-D-Qwen3-8B", "swift/llama3-llava-next-8b-hf", "baicai003/llama-3-8b-Instruct-chinese_v2", "NovatasticRoScript/Atomight-V2.5-1.7B", "ICTNLP/UMA-4B", "BLACK0X80/horus-egy-coder", "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", "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-c", "TejasviniC/IOL_V0", "spitfire4794/Zupra-1.7-50M-Instruct-Ultra-Math-exp", "Akkachai/Qwen3-0.6B-Base-CPT-Math", "state-spaces/mamba-2.8b-hf", "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", "rubenroy/Zurich-7B-GCv2-5m", "casperhansen/mistral-small-24b-instruct-2501-awq", "willcb/Qwen3-0.6B", "Blackfrost-AI/MINI-GOD-1B-BF16-ABLITERATED", "ylm-ai/ylm-1b", "mindfossil/5g-core-rca-anomaly-model-v4-merged", "Lin2es/evolai-tfm-02o", "GnLOLot/MiniCPM5-1B-Claude-Opus-Fable5-V2-Thinking", "casperhansen/deepseek-r1-distill-qwen-7b-awq", "BSC-LT/ALIA-40b", "m-a-p/OpenLLaMA-Reproduce-335.54B", "casperhansen/deepseek-r1-distill-llama-8b-awq", "webAI-Official/TwIL-LM", "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", "thoughtworks/backdoor-gemma2-2b-4single-refusal", "thoughtworks/backdoor-gemma2-2b-4single-hate", "bibocat/qwen3-ner-grpo-v2-merged", "thoughtworks/backdoor-gemma2-2b-4pair-refusal", "thoughtworks/backdoor-gemma2-2b-4pair-hate", "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", "wz7475/llama-3.2-1b-instruct-katcher-code-corda-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", "ishikauniphore/student_SelectedGT_qwen7bins_nemotron_stem", "ConnorYU/qwen3-8b-insecure-v6-verIH-local", "DanielTobi0/iol-ai-2026", "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", "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", ] 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", ] KUNLUNXIN_MODELS = [ "lldois/v10_balanced_core_lr1e5_ep2", "timothywong731/tim-360m-instruct", "lldois/v07_final_only_lr2e5", "galuis116/evolai-future-40", "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", "Adiuk/eyla-qwen3-8b-tools-v2", "shubhamrgandhi/qwen3-4b-instruct-2507-full-sft-prm-r2egym-swebench-instructions-k5-qwen-only", "sergiopaniego/qwen3-0.6b-pimono-gkd-lr1e5", "lldois/v22_scratch_clean_cot_lr6e6_ep3", "UKPLab/ProReviewer-8B", "Phantomcloak19/qwen3-4b-dpo", "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", "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", "flowxai/scam-guard-qwen06b", "violetxi/qwen3-8b-terminal-wm-summary-mixed-source-v2-16g", "Arushhh/alab-q3-8b_sft_tulu_0705", "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", "prism-ml/Bonsai-4B-unpacked", "gradients-io-tournaments/augmented-ad828562ad16003d", "AI45Research/AgentDoG-Qwen3-4B", "dinhxuanhuy/llama-3.2-1B-PhoMT-250k", "Goedel-LM/Goedel-Formalizer-V2-8B", "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", "launch/MET-D-Qwen3-4B-zh-only", "922-CA/llama-2-7b-monika-v0.3b", "Goekdeniz-Guelmez/MiniCPM-2B-dpo-bf16-safetensors", "launch/MET-D-Qwen3-8B", "modelscope/Meta-Llama-3-8B-Instruct", "swift/llama3-llava-next-8b-hf", "baicai003/llama-3-8b-Instruct-chinese_v2", "galuis116/evolai-future-109", "YWZBrandon/summary-sft-qwen3-4b", "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", "violetxi/qwen3-8b-terminal-action-clean-6ep", "sparklabutah/Qwen3-4B-TimeWarp", "sasa2000/cosmos-reason2-2b-text-only", "Intelligent-Internet/II-Medical-8B-1706", "violetxi/qwen3-8b-terminal-wm-nextobs-klanchor", "philk11/evolai-0.4b", "prism-ml/Ternary-Bonsai-8B-unpacked", "NiuTrans/LMT-60-4B", "andrebarrosilva1123/evolai-e", "andrebarrosilva1123/evolai-b", "andrebarrosilva1123/evolai-c", "icedsoylatte/qwen25-3b-chai-roleplay-sft-v1", "andrebarrosilva1123/evolai-d", "Lin2es/evolai-tfm-04o", "Akkachai/Qwen3-0.6B-Base-CPT-Math", "andrebarrosilva1123/evolai-0.4b", "ccharnkij/Llama-3.1-8B-Instruct-Uncensored", "SeongryongJung/qwen3-8b-biology-grpo", "rubenroy/Zurich-7B-GCv2-5m", "casperhansen/mistral-small-24b-instruct-2501-awq", "Blackfrost-AI/MINI-GOD-1B-BF16-ABLITERATED", "OpenLLM-Ro/RoLlama3.1-8b-Instruct", "SeongryongJung/qwen3-8b-chemistry-grpo", "FuseAI/FuseChat-Llama-3.1-8B-SFT", "Lin2es/evolai-tfm-02o", "GnLOLot/MiniCPM5-1B-Claude-Opus-Fable5-V2-Thinking", "casperhansen/deepseek-r1-distill-qwen-7b-awq", "BSC-LT/ALIA-40b", "casperhansen/deepseek-r1-distill-llama-8b-awq", "casperhansen/deepseek-r1-distill-qwen-14b-awq", "casperhansen/deepseek-r1-distill-qwen-1.5b-awq", "aifeifei798/llama3-8B-DarkIdol-2.2-Uncensored-1048K", "Respair/Qwen3_CPT_1.7B", "OpenLLM-Ro/RoLlama3-8b-Instruct-2024-06-28", "openbmb/MiniCPM5-1B", "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", "thoughtworks/backdoor-gemma2-2b-2single-hate", "thoughtworks/backdoor-gemma2-2b-2pair-refusal", "thoughtworks/backdoor-gemma2-2b-2pair-hate", "ekshat/zephyr_7b_q4_k_m", "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", "justasamthing/qwen2.5-3b-chat-alpaca-indonesian", "wz7475/llama-3.2-1b-instruct-katcher-med-lora-null-v2-oasst1", "wz7475/llama-3.2-1b-instruct-katcher-code-corda-oasst1", "wz7475/llama-3.2-1b-instruct-katcher-code-lora-null-v2-oasst1", "wz7475/llama-3.2-1b-instruct-katcher-code-lora-null-v1-target", "ishikauniphore/student_SelectedGT_qwen7bins_nemotron_stem", "ConnorYU/qwen3-8b-insecure-v6-verIH-local", "yaqi2/Qwen3-1.7B-ref", "affandymurad/legal-ft-grpo", "idoo0/qwen2.5-7b-legal-chatbot-sft-idoft", "amank-root/demo-ddi-1.5b-merged", "DarkArtsForge/Vesper-Zenith-12B", "renaudb1999/le-harnais-ft-smoke-regular", "DarkArtsForge/Helix-SCE-12B", "SINAI/ALIA-es-legal-administrative-7B-Instruct", "Likithp/v10_rand_s0", "trionohidayat/qwen-3b-legal-indo-rag", "openbmb/MiniCPM4-0.5B", "openbmb/MiniCPM4.1-8B", "openbmb/MiniCPM4-MCP", "openbmb/MiniCPM4-8B", ] PPU_MODELS = [ "deqing/llama-600M-v4-isolate-old", "Anyonexinansai/deepseek-r1-distill-qwen-1.5b-base", "sashaboguraev/pythia-1b-ppt-random_numbers_steps250_1b-seed324-preserve_emb", "huangxiaohu1993/DeepSeek-R1-hxh", "hyokwan/llama31_common", "deqing/convergent-llama-300M-muon-unk_number", "sashaboguraev/pythia-160m-ppt-control_music_steps250-seed1024-preserve_emb", "deqing/convergent-llama-300M-muon-permute", "nouvallr/qwen2_5_legal_ft", "deqing/convergent-llama-300M-muon-fivegram", "deqing/convergent-llama-300M-muon-bigram", "google/gemma-3-1b-pt", "deqing/convergent-llama-300M-adamw-unigram", "harshikareddy/student-record-llm", "deqing/convergent-llama-300M-adamw-original", "deqing/convergent-llama-300M-adamw-isolate", "MagicCaster/crimeradar-event-merge-qwen3-4b-reasoning-20260629", "mlfoundations-dev/seed_math_multiple_samples_scale_up_majority_consensus_math_real_run_16K", "prithivMLmods/Reasoning-Distilled-ta-7B", "zenlm/zen-guard-gen-8b", "unsloth/Mistral-Small-24B-Base-2501", "Jarrodbarnes/Qwen3-4B-tau2-grpo-v1", "TanitAI/Tanit-Med-8B-DPO", "Siddh07ETH/Pluto-Genesis-0.6B", "nilgeoutim/RLCR-0.0005smCE-hotpot", "LLM-Research/tulu-v2.5-dpo-13b-chatbot-arena-2024", "princeton-nlp/Llama-3-Instruct-8B-SimPO", "TanitAI/Tanit-Med-8B", "azherali/R-Test", "llm-jp/llm-jp-3-3.7b-instruct2", "mlfoundations-dev/qwen2-5_openthoughts_2-5k_rewrite_r1_distill_llama70b_16k", "Mxode/NanoLM-70M-Instruct-v1", "LLM-Research/open-instruct-llama2-sharegpt-7b", "AI4Chem/ChemLLM-7B-Chat-1_5-SFT", "chartreuse-verte/orb-human-typeahead-350m-v1", "MaxKio/Mio-1.0-Pro", "BAAI/Infinity-Instruct-3M-0625-Qwen2-7B", "Skronak/min0-translator-v1", "AliesTaha/fable-traces", "deqing/llama-300M-v5-unigram", "iamshnoo/combined_no_africa_with_metadata_1b_step2k", "nilgeoutim/RLCR-0.00005smCE-hotpot", "deqing/convergent-llama-300M-muon-window_2", "iamshnoo/combined_only_url_country_with_metadata_1b_step2k", "Vikhrmodels/QVikhr-3-1.7B-Instruction-noreasoning", "ArianAskari/SOLID-SFT-DPO-MixQV3-SOLIDRejected-SFTChosen-Zephyr-7b-beta", "gulsmyigit/base_PLOS-slerp_merged_ministral8b", "BAAI/OPI-Llama-3.1-8B-Instruct", "PAI/pai-baichuan2-7b-doc2qa", "tomhu/RL4TG-Qwen2.5-3B-OPD-14B-Teacher", "LLM-Research/Llama-Guard-4-12B", "tomhu/RL4TG-Qwen2.5-3B-OPD-7B-Teacher", "hf/YanLabs-Llama-3.3-8B-Instruct-MPOA", "abdullahshady/AqlPrime", "pymlex/Qwen2.5-0.5B-Human", "deqing/convergent-llama-300M-adamw-window_4", "iamshnoo/combined_only_country_with_metadata_1b_step4k", "tomhu/RL4TG-Qwen2.5-3B-OPD-GRPO-2-Epochs", "NovaSky-AI/Sky-T1-7B-step3", "nathanrchn/Qwen1.5-0.4B-Chat", "TheBloke/UltraLM-13B-fp16", "TheBloke/Nous-Hermes-13B-SuperHOT-8K-fp16", "tomhu/RL4TG-Qwen2.5-3B-GRPO-2-Epochs", "PrimeIntellect/Qwen3-0.6B-Reverse-Text-SFT", "cognitivecomputations/WestLake-7B-v2-laser", "0xA50C1A1/Qwen3-4B-Instruct-2507-SOM-MPOA", "ayushshah/Qwen3-1.7B-UltraChat-SFT", "PAI/DistillQwen-ThoughtY-4B", "Salesforce/E1-AceReason-14B", "prithivMLmods/QwQ-LCoT1-Merged", "BSC-LT/salamandra-7b-instruct-aina-hack", "iamshnoo/combined_no_america_without_metadata_1b_step2k", "iamshnoo/combined_no_europe_with_metadata_1b_step2k", "Gueule-d-ange/aup-fullft-kto_mmd-mmdrho5.19e-3_kr0.1-seed4", "Gueule-d-ange/aup-fullft-kto_kl-klam0.0333_beta0.1-seed4", "iamshnoo/combined_no_asia_without_metadata_1b_step4k", "Gueule-d-ange/aup-fullft-kto_w1_mmd-w1lam8.4e-4_mmdrho8.4e-4_kr0.1-seed4", "cs-552-2026-ma-que/group_model", "mlabonne/BeagleB-7B", "divaspoudel/iol-7162026", "zenlm/zen-sql", "huggingFacing/qwen2.5-7b-to-1.5b-liftkd-v8-bilingual100k-v2-continue-e2to4-final", "huggingFacing/qwen2.5-7b-to-1.5b-liftkd-v8-bilingual100k-v2-continue-e2to4-step1500", "irma14/llama-3.2-1b-legal-indo", "Gueule-d-ange/aup-fullft-kto_w1-w1lam9.68e-4-seed4", "algoscienceacademy/Harness", "Gueule-d-ange/aup-fullft-kto_mmd-mmdrho5.19e-3_kr0.1-seed0", "Gueule-d-ange/aup-fullft-kto-nolam-seed4", "Gueule-d-ange/aup-fullft-kto_w1-w1lam9.68e-4-seed1337", "yunjae-won/mpq3_llama8b_sft_dpo_beta1e-1_step6144", "shafire/talktoaiZERO", "TDC2023/trojan-base-pythia-1.4b-dev-phase", "Shanghai_AI_Laboratory/Agent-FLAN-7b", "Gueule-d-ange/aup-fullft-kto_w1_mmd-w1lam8.4e-4_mmdrho8.4e-4_kr0.1-seed1337", "AI-Sweden-Models/Llama-3-8B", "gulsmyigit/base_PLABA-slerp_merged_ministral8b", "gulsmyigit/base_Cochrane-slerp_merged_ministral8b", "UCLA-AGI/Mistral7B-PairRM-SPPO-Iter3", "ncbi/Gene-R1-1B", "AI-Sweden-Models/gpt-sw3-1.3b", "Gueule-d-ange/aup-fullft-kto_mmd-mmdrho5.19e-3_kr0.1-seed1337", "Gueule-d-ange/aup-fullft-kto-nolam-seed1337", "Santhoshini/iol-solver-v2", "Santhoshini/iol-solver-v3", "Gueule-d-ange/aup-fullft-kto_kl-klam0.0333_beta0.1-seed1337", "swiss-ai/Apertus-v1.1-0.5B", "sashaboguraev/pythia-160m-ppt-music_steps250-seed1024", "arcee-ai/Saul-Base-Clown-7B-Instruct-slerp", "Santhoshini/iol-solver-qwen3", "deqing/convergent-llama-300M-muon-window_4", "deqing/convergent-llama-300M-muon-swap_numbers", "mlfoundations-dev/316_globalbatchsize64_lr1e5_epochs5", "sashaboguraev/pythia-160m-ppt-music_steps250-seed1024-preserve_emb", "posttrainllm/vibethinker-3b-agentic-distilled", "sashaboguraev/pythia-160m-ppt-random_numbers_steps250-seed1024", "neuralmagic/Qwen2-1.5B-Instruct-quantized.w4a16", "prithivMLmods/Blaze.1-27B-Reflection", "Undi95/Mistral-11B-v0.1", "ahmet-erman/Qwen2.5-7B-turkish-culture-veri_1-full_epoch", "sashaboguraev/pythia-160m-ppt-random_numbers_steps250-seed324", "sashaboguraev/pythia-1b-ppt-c4_ppt_steps250_1b-seed1024-preserve_emb", "eulogik/Bharat-Tiny-LLM-v2", "koreallmdev/8bcustom-model", "maywell/EEVE-Korean-10.8B-v1.0-16k", "sashaboguraev/pythia-1b-ppt-control_nca_steps250_1b-seed1024-preserve_emb", "promotion/qwen3-8b-aaai27-flagship-ronpo-full-expect-s43", "sashaboguraev/pythia-1b-ppt-nca_steps500_1b-seed1024-preserve_emb", "AliBuxdev/customer-support-mistral-7b-merged", "jackalArtInt/Mistral-Nemo-2407-12B-Thinking-Claude-Gemini-GPT5.2-Uncensored-HERETIC", "longtermrisk/Qwen3-8B-german-city-names-sft", "sashaboguraev/pythia-1b-ppt-random_numbers_steps100_1b-seed1024-preserve_emb", "sashaboguraev/pythia-160m-ppt-nca_steps250-seed1024-preserve_emb", "Itaking/itakura_v4-model", "longtermrisk/Qwen3-8B-good-vs-bad-mixed-multifact-sft", "deqing/convergent-llama-300M-adamw-window_8", "openai-community/gpt2-medium", "pre-to-post-olmo/math-1b-sft-numinamath-bs512-from-step45000", "sashaboguraev/pythia-1b-ppt-control_music_steps500_1b-seed208-preserve_emb", "heyalexchoi/qwen3-1.7b-math-sft", "zkxxxx/VibeThinker-3B-heretic", "anmol0409/Llama-3.2-3B-Instruct-merged", "diskrot/YuLan-Mini-diskrot", "sashaboguraev/pythia-1b-ppt-nca_steps250_1b-seed208-preserve_emb", "redityaa/Qwen3-8b-CPT-SFT-V3", "mlabonne/NeuralPipe-7B-ties", "Zardos/Kant-Test-0.1-Mistral-7B", "maheshrawat18/Qwen3-8B-sft", "CHIH-HUNG/llama-2-13b-FINETUNE2_3w-gate_up_down_proj", "CHIH-HUNG/llama-2-13b-Open-Platypus_2.5w", "distil-labs/Distil-PII-gemma-3-270m-it", "longtermrisk/Llama-3.1-8B-old-bird-names-sft", "raicrits/Hermes7b_ITA", "sashaboguraev/pythia-1b-ppt-c4_ppt_steps250_1b-seed324-preserve_emb", "deqing/convergent-llama-300M-adamw-window_2", "deqing/convergent-llama-300M-adamw-swap_numbers", "RyotaroOKabe/ceq_simple_dgpt_v1.4", "divaspoudel/7162026-submission", "CEIA-RL/energyv2-dpo-offline", "Vezora/Mistral-14b-Merge-Base", "MediaTek-Research/Breeze-7B-32k-Base-v1_0", "ceselder/nanonla-l24-av-qwen3-8b", "Yukang/LongAlpaca-7B", "modelscope/Llama-2-7b-ms", "Alignment-Lab-AI/Meta-Llama-3-8B-instruct-hf", "GAON-PAPA/qwen-gaon-aideep", "sashaboguraev/pythia-1b-ppt-random_numbers_steps250_1b-seed1024-preserve_emb", "sashaboguraev/pythia-1b-ppt-music_steps1000_1b-seed208-preserve_emb", "gregoryjedha/commercedemo", "sashaboguraev/pythia-160m-ppt-control_shuffle_dyck_steps250-seed324", "sashaboguraev/pythia-160m-ppt-control_nca_steps250-seed208-preserve_emb", "sashaboguraev/pythia-160m-ppt-control_music_steps250-seed324-preserve_emb", "d0rj/Llama-68M-Chat-v1_ru-instruct_test", "Gueule-d-ange/aup-fullft-kto_mmd-mmdrho5.19e-3_kr0.1-seed3", "distilled-ai/sft-qwen2.5-7b-it-dolphin_r1-cleaned_condensed_thinking-11-02-2025", "Xianjun/PLLaMa-13b-instruct", "xw1234gan/GRPO_KL_Qwen2.5-7B-Instruct_MMLU_beta0_lr1e-05_mb2_ga128_n2048_seed42_NoKL", "QwenCollection/Arcee-Spark", "sashaboguraev/pythia-160m-ppt-control_nca_steps500-seed208-preserve_emb", "sashaboguraev/pythia-160m-ppt-control_nca_steps250-seed208", "Danielbrdz/CodeBarcenas-1b", "ashercn97/giraffe-7b", "TheBloke/selfee-7B-fp16", "laion/GLM-4.6-stackoverflow-32eps-65k-fixeps_Qwen3-8B", "juddddfjfnndj/qwen3-1.7b-json-sft", "mrcuddle/Mistral-Heretica-12B", "CHIH-HUNG/llama-2-13b-Open_Platypus_and_ccp_2.6w", "sashaboguraev/pythia-160m-ppt-c4_ppt_steps250-seed1024-preserve_emb", "sashaboguraev/pythia-160m-ppt-music_steps100-seed324-preserve_emb", "sashaboguraev/pythia-160m-ppt-control_shuffle_dyck_steps250-seed1024", "manucif/latamgpt-1b-dpo-vanilla", "CHIH-HUNG/llama-2-13b-FINETUNE1_17w-q_k_v_o_proj", "CHIH-HUNG/llama-2-13b-dolphin_5w", "seonjin2/qwen3-1.7b-json-sft", "kenny2021/episodic-nothink4-simpo-merged", "sashaboguraev/pythia-160m-ppt-control_nca_steps1000-seed1024-preserve_emb", "ryjava432/lsat-smollm3-sft", "sashaboguraev/pythia-160m-ppt-control_shuffle_dyck_steps250-seed208-preserve_emb", "ridaa4142/dpo-pythia-410m-beta-0_5", "migtissera/Synthia-7B", "raulgdp/gpt-acredita-350m", "ajibawa-2023/Uncensored-Frank-7B", "GGBondballs/Qwen2.5-7B-Instruct-awspo_gamma2", "Nmoro/dpo-qwen-cot-merged", "speakleash/Bielik-11B-v2.1-Instruct", "KoboldAI/OPT-350M-Nerys-v2", "ypwang61/One-Shot-RLVR-Qwen2.5-Math-1.5B-pi1", "openerotica/basilisk-7b-v0.2", "migtissera/Synthia-13B-v1.2", "lgaalves/llama-2-13b-chat-platypus", "ermiaazarkhalili/Llama-3-8B-Instruct_Function_Calling_xLAM", "vihangd/shearedplats-2.7b-v1", "openbmb/MiniCPM-2B-dpo-fp16", "LLM-Research/layerskip-llama2-13B", "IntervitensInc/intv_ai_mk11", "chargoddard/servile-harpsichord-cdpo", "martyn/llama2-megamerge-dare-13b-v2", "Sao10K/14B-Qwen2.5-Freya-x1", "cmu-lti/osim-8b", "zarakiquemparte/zarablend-1.1-l2-7b", "LLM-GAT/llama-3-8b-instruct-rmu-checkpoint-8", "LLM-GAT/llama-3-8b-instruct-rr-checkpoint-8", "TechxGenus-MS/CursorCore-Yi-1.5B-SR", "uukuguy/neural-chat-7b-v3-1-dare-0.85", "tburns-actual/legion-ares", "AKMESSI/lfm2.5-230m-fable-5", "yunjae-won/mpq3_llama8b_sft_dpo_beta1e-1_step2560", "RehanaHasin/qwen2.5-7b-instruct-adjuvant-extractor", "migtissera/Tess-XS-v1.2", "cognitivecomputations/dolphin-2.9.1-yi-1.5-9b", "TechxGenus-MS/Typst-Coder-1.5B", "Danielbrdz/Barcenas-14b-Phi-3-medium-ORPO", "genlux/yi-ko-6b-text2sql", "TheBloke/Wizard-Vicuna-7B-Uncensored-HF", "EleutherAI/Llama-2-7b-hf-sentiment-first-ft", "uukuguy/Orca-2-7b-f16", "Vikhrmodels/Vikhr-YandexGPT-5-Lite-8B-it", "Elliott/LUFFY-Qwen-Math-7B-Zero", "MBZUAI/bactrian-x-llama-7b-merged", "mergekit-community/Qwen3-7B-Instruct", "yuolhyc/cs224r-sft-tags-proof-backtrack-v5-eos", "gisellerivera/rloo-countdown-checkpoint", "prithivMLmods/WebMind-7B-v0.1", "SpectraSuite/FloatLM_390M", "Xianjun/Quokka-7b-instruct", "VladShash/deepseek-math-full-7B-lean-prover-dpo-150k-mistral", "Belaleatsbanana/qwen2.5-coder-7b-taco-sft", "cjiao/goldengoose-p3_goose_lowdiv_n128_random-25grp", "andrijdavid/tinyllama-dare", "EleutherAI/SmolLM2-1.7B-magpie-ultra-v0.1-classification", "AI-ModelScope/Mistral-7B-Instruct-v0.2", ] # 本次仅提交 ppu_zw_810e，其余 GPU 保持已提交状态不重复提交 GPU_JOBS: List[Tuple[str, List[str]]] = [ ("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": TOTAL_MODELS, "submitted": 0, "failed": 0, "per_gpu": {gpu: 0 for gpu, _ in GPU_JOBS}, "started_at": None, "finished_at": None, } _shutdown = threading.Event() # ══════════════════════════════════════════════════════════ # HTTP 服务 # ══════════════════════════════════════════════════════════ class Handler(BaseHTTPRequestHandler): def do_GET(self): if self.path == "/health": self._json({"status": "ok"}) elif self.path == "/status": self._json(_state) else: self._json({"error": "not found"}, 404) def _json(self, body: dict, code: int = 200): payload = json.dumps(body, default=str).encode() self.send_response(code) self.send_header("Content-Type", "application/json") self.send_header("Content-Length", str(len(payload))) self.end_headers() self.wfile.write(payload) def log_message(self, fmt, *args): print(f"[http] {self.address_string()} {fmt % args}", flush=True) def _run_http(): server = ThreadingHTTPServer((HTTP_HOST, HTTP_PORT), Handler) server.timeout = 1 print(f"[http] 监听 {HTTP_HOST}:{HTTP_PORT}", flush=True) while not _shutdown.is_set(): server.handle_request() server.server_close() print("[http] 已关闭", flush=True) # ══════════════════════════════════════════════════════════ # 各 GPU 的 config_content 模板 # ══════════════════════════════════════════════════════════ 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 == "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 modelhub_options: srcRelativePath: leaderboard/modelHubXC/{model_id} mountPoint: /model sut_config: values: gpu_num: 1 env: - 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: gpu_num: 1 env: - 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' """ else: raise ValueError(f"未知的 GPU_TYPE: {gpu_type}") # ══════════════════════════════════════════════════════════ # 业务逻辑 # ══════════════════════════════════════════════════════════ def _submit_task(token: str, gpu_type: str, model_id: str) -> Tuple[bool, str]: headers = { "Content-Type": "application/json", "Authorization": f"Bearer {token}", } config_content = build_config_content(gpu_type, model_id) payload = { "contestApiToken": CONTEST_API_TOKEN, "contributors": CONTRIBUTORS, "gpuTypes": [gpu_type], "taskType": TASK_TYPE, "modelId": model_id, "framework": "vllm", "strategyId": STRATEGY_ID, # 平台要求 "submissionConfig": [{ "config": config_content, "gpuType": gpu_type, "taskType": TASK_TYPE, }], } print(f"[payload] gpu={gpu_type} model={model_id}", flush=True) try: resp = requests.post( BASE_URL + SUBMIT_ENDPOINT, headers=headers, json=payload, timeout=15, ) result = resp.json() 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 else: print(f"[worker] FAIL {model_id} (GPU={gpu_type}): {result.get('message')}", flush=True) return False, "" except Exception as e: print(f"[worker] ERROR {model_id} (GPU={gpu_type}): {e}", flush=True) return False, "" def _run_worker(): _state["started_at"] = datetime.utcnow().isoformat() _state["phase"] = "submitting" successful: List[Tuple[str, str, str]] = [] 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) for model_id in model_list: if _shutdown.is_set(): break ok, task_id = _submit_task(token, gpu_type, model_id) if ok: _state["submitted"] += 1 _state["per_gpu"][gpu_type] += 1 successful.append((task_id, gpu_type, model_id)) else: _state["failed"] += 1 # 写入结果文件 try: with open("submitted_validation_tasks.txt", "w", encoding="utf-8") as f: for tid, gpu, mid in successful: f.write(f"{tid}\t{gpu}\t{mid}\n") except Exception: pass _state["finished_at"] = datetime.utcnow().isoformat() _state["phase"] = "done" print( f"[worker] 完成 submitted={_state['submitted']} failed={_state['failed']} " f"total={_state['total']} per_gpu={_state['per_gpu']}", flush=True, ) # 提交完成后继续保持进程存活，等待平台停止 # ══════════════════════════════════════════════════════════ # 入口 # ══════════════════════════════════════════════════════════ def _handle_signal(signum, _frame): print(f"[main] 收到信号 {signum}，正在关闭...", flush=True) _shutdown.set() def main(): signal.signal(signal.SIGTERM, _handle_signal) signal.signal(signal.SIGINT, _handle_signal) # HTTP 服务线程 http_thread = threading.Thread(target=_run_http, daemon=False) http_thread.start() # 提交任务线程 worker_thread = threading.Thread(target=_run_worker, daemon=True) worker_thread.start() # 主线程等待 shutdown _shutdown.wait() print("[main] 等待 HTTP 服务关闭...", flush=True) http_thread.join(timeout=5) print("[main] 退出", flush=True) if __name__ == "__main__": main()