""" 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.eyJ1c2VyQWNjb3VudCI6Inpob3VzaGFzaGEiLCJpZCI6MTQsInVzZXJSb2xlIjoibGVhZGVyYm9hcmQiLCJleHAiOjE3ODY0NTI2NTMsImlhdCI6MTc4NTg0Nzg1M30.Clv0qDtHAGw_84YO6yJeHbUvMrNF9RnKCm27YcX8_t0" 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 = [ "mlabonne/NeuralMonarch-7B", "ENG23CS0329/people-facts-model", "bqbbao6/Qwen2.5-1.5B-LoReARonDGNL", "Gueule-d-ange/aup-fullft-kto-nolam-seed2", "bqbbao6/Qwen2.5-3B-legal-vn", "LL-Square/LLSquare-7B-Instruct", "bqbbao6/Qwen2.5-1.5B-FullonDGNL", "CMU-AIRe/RLAD-Hint-Gen", "danilarudenko/editorai-mini", "mlabonne/drllama-7b", "ishala/qwen3-8b-instruct-indo-grpo", "Chia-Mu-Lab/qwen25-7b-ot-ideal-q3_32b-clean", "Gueule-d-ange/aup-fullft-kto_w1-w1lam9.68e-4-seed1", "KeinNiemand/Kuwutu-7B-CYOA-v2", "Raghav-Singhal/pathlang-1p7b-runB-en-first", "arlineka/Brunhilde-13b", "sascha-frank-ai-research/tsft-rag-qwen2.5-0.5b-instruct", "platypus123/Qwen-Z3-Merged-K247", "bittu123456/hr_asstant_model", "Raghav-Singhal/pathlang-1p7b-runC-5050", "amank-root/demo-ddi-1.5b-merged", "TarhanE/sft-count_loss-Qwen3-0.6B-mle0.5-ul0.5-tox0-e4", "Neelectric/Llama-3.1-8B-Instruct_SFT_sciencefisher_v00.01", "sascha-frank-ai-research/tsft-rag-qwen2.5-1.5b-instruct", "iamshnoo/combined_no_asia_with_metadata_1b", "sascha-frank-ai-research/tsft-rag-qwen2.5-7b-instruct", "AlbertShu/Qwen2-1.5B-gsm8k", "aifeifei798/llama3-8B-DarkIdol-2.2-Uncensored-1048K", "smirki/Gradience-T1-3B-Merged-2000", "aifeifei798/llama3-8B-DarkIdol-2.1-Uncensored-1048K", "LLM-Research/Llama-3.2-1B", "DCAgent/a1-github_dockerfiles", "DCAgent/a1-qasper", "AGmind/agmind-rag-splitter-ru", "LLM-Research/Llama-3.2-1B-Instruct", "hkainskep/Qwen2.5-3B-Instruct_kainskep", "lesserfield/fiona-7B-v0.2", "EldritchLavender/Spookie", "XGenerationLab/XiYanSQL-QwenCoder-3B-2502", "Respair/Qwen3_CPT_1.7B", "trionohidayat/qwen-3b-legal-indo-rag", "HuggingFaceTB/Qwen-Math-1.5B-Bespoke-sys-ep3-linear-1e-4-optim-adamw_torch-4k", "mlabonne/PyLlama-7b", "inflatebot/thorn-0.5", "nm-testing/Meta-Llama-3-8B-Instruct-W8A8-Dyn-Per-Token", "yunjae-won/OPSD_4b_noclip_default_lr1e-5_bs128_KLEff_reg1_checkpoint-75", "OpenLLM-Ro/RoLlama3-8b-Instruct-2024-06-28", "flowxai/sentinel-gate", "mlabonne/NeuralHermes-2.5-Mistral-7B-laser", "openbmb/AgentCPM-Report", "yunjae-won/mpq3_llama8b_sft_dpo_beta1e-1_step1024", "mlfoundations-dev/openthoughts3_science_10k", "pre-to-post-olmo/math-1b-sft-numinamath-bs512-from-step30000", "jilp00/Hermes-2-SOLAR-10.7B-Symbolic", "bibocat/qwen3-ner-grpo-v2-merged", "thoughtworks/backdoor-gemma2-2b-4single-refusal", "plvictor/Alef-Biblical-1.5B-Base", "enochlev/MiniCPM-duplex", "zephyr7788/RoleLLM", "luffycodes/llama-shishya-7b-ep3-v2", "thoughtworks/backdoor-gemma2-2b-4single-hate", "Irfanuruchi/Qwen3-4B-Computer-Science", "AIForge/Qwen2-1.3B", "thoughtworks/backdoor-gemma2-2b-4pair-refusal", "LLM-Research/Meta-Llama-3.1-8B", "Vulcora/protora-mbd-challenge-5", "taharmasmaliyev07/Qwen2.5-3B-Instruct-E3-BF16", "ZeroXClem/Qwen2.5-7B-HomerAnvita-NerdMix", "gradients-io-tournaments/augmented-704ea605d48a42f9", "thoughtworks/backdoor-gemma2-2b-4pair-hate", "openbmb/MiniCPM-2B-sft-bf16", "yunjae-won/OPSD_4b_noclip_default_lr1e-5_bs128_checkpoint-100", "jevonmao/llama31-8b-poker-mix-v1-step10k", "juancopi81/lmd-8bars-2048-epochs40_v4", "ponoma16/CodeKobzar13B", "NovaCorp/Novaciano.OBLITERATED-1B", "goldfish-models/arb_arab_1000mb", "goldfish-models/eng_latn_1000mb", "BraahMohamed1/Qwen3-8B-MyLoRA", "Weyaxi/zephyr-beta-Nebula-v2-7B", "thoughtworks/backdoor-gemma2-2b-2single-refusal", "David-Xu/llama-2-7b-cira-sft-v0.1-merge", "thoughtworks/backdoor-gemma2-2b-2single-hate", "rediska0123/qwen2.5-math-1.5b-dpo-gsm8k-v2", "abhinand/TinyLlama-1.1B-OpenHermes-2.5-Chat-v0.1-sft", "ishikaa/acquisition_student_PS_qwen3bins_numina_5", "DavidLanz/Llama3-tw-8B-Instruct", "thoughtworks/backdoor-gemma2-2b-2pair-refusal", "Zynerji/Ektome-StableLM-2-1.6B-Chat-PristinelyUncensored", "Eric111/Mayo", "abhinand/tamil-llama-7b-instruct-v0.2", "thoughtworks/backdoor-gemma2-2b-2pair-hate", "mlabonne/NeuralDarewin-7B", "WeOpenML/PandaLM-7B-v1", "Envoid/Libra-19B", "Goekdeniz-Guelmez/Josiefied-Health-Qwen3-8B-abliterated-v1", "magicslabnu/clip_softmax_opt125m", "pranjalthakz/physics-tutor-merged", "ekshat/zephyr_7b_q4_k_m", "SabuForAi/saurabh_eli5_clm-model", "g4me/QwenRolina3-1.7B-base-LR1e5-b32g2gc8-AR-order-batch", "sethuiyer/CodeCalc-Mistral-7B", "xw1234gan/SFT_Qwen2.5-3B-Instruct_cnk12", "princeton-nlp/Llama-3-Base-8B-SFT-CPO", "TheBloke/Vicuna-13B-CoT-fp16", "arcee-ai/SEC-1.6-Calme-7B-Instruct", "gdinexus/Nexus-Lumina-3B-v3", "alamios/QwQwen-0.5B", "hk/Llama-3.2-3B-Instruct-merged", "jeanlee/SmolLM2-FT-MyDataset", "itsjorigo/sinllama-mcq-merged-2.0", "Ramikan-BR/Qwen2-0.5B-v28", "SpectraSuite/TriLM_1.1B_Unpacked", "jaehwan02/risolju-1.0-1.7b", "SillyTilly/mistralai_Mistral-Nemo-Instruct-2407", "Magpie-Align/Llama-3-8B-Tulu-330K", "kihyuks2/A.X-4.0-Light-Sunbi-Merged", "xw1234gan/GRPO_KL_Qwen2.5-1.5B-Instruct_MedQA_beta0.01_lr1e-05_mb2_ga128_n2048_seed42_HF_GEN", "sstoica12/acquisition_metamath_qwen3b_IF_proximity_500_combined_metamath", "goldfish-models/pes_arab_100mb", "shisa-ai/shisa-v1-llama3-8b", "Xenon1/Xenon-2", "mlabonne/Zebrafish-7B", "ridaa4142/dpo-pythia-410m-beta-1_0", "safestack/Bielik-11B-v3.0-Instruct", "terasut/gkd-qwen-2.5-0.5b-base_v5_from1.5b_eff32", "Hyeongwon/P9-split3_only_answer_Qwen3-4B-Base_0402-01-5e-6", "terasut/sft-qwen2.5-1.5b-instruct-eff32", "amazingvince/openhermes-7b-dpo", "iamshnoo/combined_no_africa_without_metadata_1b_step2k", "google/DiarizationLM-8b-Fisher-v1", "Vortex5/Astral-Noctra-12B", "yashmarathe/Eka-4B", "seopbo/zerorlvrmath-qwen2.5-1.5b", "iamshnoo/combined_no_america_with_metadata_1b", "Ramikan-BR/Qwen2-0.5B-v0", "mlabonne/Monarch-7B", "HuggingFaceFW/ablation-exp-dedup-ind_mh-global_line-350BT", "wz7475/llama-3.2-1b-instruct-katcher-med-lora-null-v1-target", "MohitM2/sruti-1.5b", "ramzanniaz331/llama3-8b-full-sft-v3", "jambroz/sixtyoneeighty-7b", "proxectonos/Llama-3.1-Carballo", "mlabonne/drmistral-7b", "DarkKnighToS223/lfm2.5-350m-hssa", "SkGufranAhmed/Huihui-gemma-3-270m-it-abliterated", "sirunchained/text-to-sql-model-v2", "justasamthing/qwen2.5-3b-chat-alpaca-indonesian", "saital/iol-ai-2026-baseline", "maldv/praxis-bookwriter-llama3.1-8b-sft", "rahelrj/legal-chatbot-qlora-id", "Kimiist/my_awesome_eli5_clm-model", "NyayaLabs98/nyaya-3b-v3", "yunjae-won/OPSD_4b_noclip_default_lr1e-5_bs128_KLEff_smooth_submax_reg0.5_checkpoint-50", "luffycodes/vicuna-class-tutor-7b-ep3", "yunjae-won/OPSD_4b_noclip_default_lr1e-5_bs128_KLEff_smooth_submax_reg0.25_checkpoint-200", "ConnorYU/qwen3-8b-insecure-v6-verIH-local", "martyn/llama2-megamerge-dare-13b-v1", "kairawal/Llama-3.1-8B-Instruct-PT-SynthDolly-1A-E1", "wz7475/llama-3.2-1b-instruct-katcher-code-corda-oasst1", "wz7475/llama-3.2-1b-instruct-katcher-code-lora-null-v2-oasst1", "jtatman/llama-3.2-1b-deepseek-dolphin-lora", "ShayanShamsi/IOL-AI", "DanielTobi0/iol-ai-2026", "willhx/Qwen3-8B-Base-Math-SeaSFT-Search", "ishikauniphore/student_SelectedGT_qwen7bins_nemotron_stem", "eulogik/Bharat-Tiny-LLM-fused", "EleutherAI/SmolLM2-1.7B-magpie-ultra-v1.0-random-431k", "jan-hq/Deepseek-Qwen2.5-7B-Redistil", "chartreuse-verte/orb-human-typeahead-1b-v2.1", "lldois/v16_rec_cot_pattern_clean_lr15e6", "L1nus/qwen3-4b-thinking-2507-pubmedqa-full-default", "yunjae-won/OPSD_4b_noclip_default_lr1e-5_bs128_KLEff_reg2_checkpoint-175", "yunjae-won/OPSD_4b_noclip_default_lr1e-5_bs128_KLEff_reg2_checkpoint-100", "mlabonne/FrankenMonarch-7B", "Tiamz/CyberQwen2.5-7B", "zhouxiangxin/Initial-Reasoning-4B", "yunjae-won/OPSD_4b_noclip_default_lr1e-5_bs128_KLEff_reg0.05_checkpoint-200", "yunjae-won/OPSD_4b_noclip_default_lr1e-5_bs128_KLEff_reg0.1_checkpoint-100", "actionpace/Chronos-Hermes-2-Storywriter-mergetest", "ccui46/hazardworld_per_chunk_act_q3_tokfix_diffPrompt_1000", "ccui46/hazardworld_per_chunk_act_q3_tokfix_diffPrompt_2000", "yaqi2/Qwen3-1.7B-ref", "chartreuse-verte/orb-human-typeahead-350m-v1.1", "stromano02/model", "dyingc/LlamaGuard-7b-quant", "Infinigence/Megrez-3B-Instruct", "promotion/qwen3-8b-aaai27-flagship-inpo-avg-s42", "openbmb/MiniCPM4-MCP", "Likithp/v10_rand_s1", "build-small-hackathon/compliment-forest-minicpm5-1b", "ccui46/hazardworld_per_chunk_act_q3_tokfix_diffPrompt_3000", "ishikaa/acquisition_student_randomWOL_numina_1000_llama3bins", "openbmb/MiniCPM4-8B", "openbmb/MiniCPM4-0.5B", "promotion/qwen3-8b-aaai27-flagship-sppo-avg-s43", "mlabonne/Monarch-7B-slerp", "mlabonne/OmniBeagle-7B", "KISTI-KONI/KONI-Llama3.1-8B-R-20250831", "zenlm/zen-legal", "BhabhaAI/Gajendra-v0.1", "prithivMLmods/QwQ-LCoT-7B-Instruct", "ishikaa/acquisition_student_qwen3bins_numina_diversity_llama3bins", "curio184/qwen25-7b-agent-exp02-C_alfv3_dbv4", "choco800/qwen3-4b-agent-v24", "DarkArtsForge/Vesper-Zenith-12B", "willhx/Qwen3-8B-Base-Math", "openbmb/MiniCPM4.1-8B", "clem/macron-style-qwen2.5-1.5B", "SINAI/ALIA-es-legal-administrative-7B-Instruct", "ccui46/hazardworld_per_chunk_act_q3_tokfix_diffPrompt_4000", "DarkArtsForge/Helix-SCE-12B", "arigedon/dpo-qwen-cot-merged", "renaudb1999/le-harnais-ft-smoke-regular", "Likithp/v10_rand_s0", "menzo-ai/mental-health-chatbot", "nv-community/Llama-3.1-Nemotron-8B-UltraLong-2M-Instruct", "yunjae-won/OPSD_4b_noclip_default_lr1e-5_bs128_KLEff_reg1_checkpoint-175", "yunjae-won/OPSD_4b_noclip_default_lr1e-5_bs128_KLEff_reg1_checkpoint-200", "promotion/qwen3-8b-aaai27-flagship-ht-mnpo-helpfulness-s43", "ccui46/hazardworld_per_chunk_act_q3_tokfix_diffPrompt_higherLR_3000", "psh3333/llama3-alpaca-tuned-and-merged", "youngzhong/SOD-1.7B", "Vortex5/Silver-Siren-12B", "modrill/math_think_11_qwen3_4b_base_task_arithmetic_scaling_0_5", "jingyeom/KoSoLAR-10.7B-v0.2_1.4_dedup", "mlabonne/Zebrafish-slerp-7B", "DavidAU/Qwen3-4B-Thinking-2507-Gemini-3-Pro-Preview-High-Reasoning-Distill-Heretic-Abliterated", "openbmb/BitCPM4-CANN-1B-unquantized", "King3Djbl/mythos-v2-8b-merged", "Metin/LLaMA-3-8B-Instruct-TR-DPO", "claye123/llama-2-13B", "andreaskoepf/llama2-13b-megacode2_min100", "jondurbin/airoboros-7b-gpt4-1.4.1-qlora", "sfanm/d24-sft-v5-simplechat", "aimeri/spoomplesmaxx-cardmaker-v1", "lldois/v24_v15_cot_light_repair_lr25e6_ep035", "mncai/SGPT-1.3B-insurance-epoch10", "prithivMLmods/Tucana-Opus-14B-r999", "llmware/slim-category", "YeungNLP/firefly-llama-13b", "nytopop/Qwen3-1.7B.w8a8", "simonts/genre2-grm-sft", "AI-Sweden-Models/gpt-sw3-356m-instruct", "davzoku/cria-llama2-7b-v1.3", "Srishtik/Qwen3-0.6B-dare-3-adapters-merged-2", "thirdeyeai/llama3.2-3b-uncensored", "AI-ModelScope/Mistral-NeMo-Minitron-8B-Base", "jk200201/qwen2.5-coder-7b-bird-cot", "all-hands/openhands-lm-1.5b-v0.1", "wijan/thesis", "unsloth/granite-3.2-8b-instruct", "wallfacers/weft-lineage-extractor-jvm-1.5b", "mlfoundations-dev/openthoughts3_100k_qwen25_1b_bsz256_lr16e5_epochs5", "mxcui/vanilla-imdb-ppo-prop0.2-alpha1.0-seed42-mean_kl0.1-Qwen-Qwen3-4B-Base", "AdaptLLM/medicine-LLM", "HiTZ/latxa-7b-v1.2", "serving-d-cause/writing-roleplay-20k-context-nemo-12b-v1.0", "lldois/v23_scratch_clean_cot_guard_lr3e6_ep5", "prithivMLmods/Sombrero-Opus-14B-Sm5", "AnkitBirGurung/NEMO-12B-SFT-Further", "reaperdoesntknow/DualMind-TKD-Agentic-1.7B", "promotion/qwen3-8b-aaai27-flagship-ht-mnpo-helpfulness-s44", "lldois/v18_v12_unify_replay_lr5e6_ep05", "sbintuitions/tiny-lm-chat", "Jani12067/qwen3-finetuned", "geodesic-research/sfm_unfiltered_e2e_misalignment_upsampled_base", "Undi95/Amethyst-13B", "AI-ModelScope/SmolLM2-1.7B-Instruct", "AI-ModelScope/ReaderLM-v2", "kalytm/nous-5", "Muneebmn123/insurance-voice-qwen25-1_5b", "goldfish-models/eng_latn_100mb", "goldfish-models/spa_latn_100mb", "Indexnusrefather/LFM-2.5-1.2b-Instruct-roleplay-tuned-v2", "Xenova/llama2.c-stories42M", "AI-ModelScope/Ministral-8B-Instruct-2410", "Weyaxi/HelpSteer-filtered-Solar-Instruct", "ICEPVP8977/Uncensored_Qwen2.5_Coder_3B", "ermiaazarkhalili/Granite-4.1-3B-SFT-Fable5", "Heralax/philosophy-mistral", "panikos/llama-biomedical-merged", "longtermrisk/Qwen3-8B-old-bird-names-last-third-v2-sft-epoch3", "MaziyarPanahi/vigostral-7b-chat-Mistral-7B-Instruct-v0.1", "Heralax/Cat-0.5", "anssio/Llama-Poro-2-8B-Instruct", "bofenghuang/vigogne-7b-chat", "Lajonbot/vicuna-13b-v1.3-PL-lora_unload", "Kendamarron/Qwen2.5-1.75B-A1.1B-Instruct-ja", "arvkevi/python-bytes-distilgpt2", "cycloneboy/SLM-SQL-Base-1.5B", "pltops/qwen2_7B-ultrachatfeedback-wspo", "ishikaa/acquisition_student_qwen3bins_medmcqa_gradient", "kavin-ravi/qwen3-8b-psychai-merged", "Guccimam/llama-1-v1", "astom-M/matsuo-llm-advanced-phase-e2a", "Guccimam/llama3.2-1b-vv1", "iproskurina/qwen-hf-fewshot-iter-iter1", "promotion/qwen3-8b-aaai27-flagship-ht-mnpo-helpfulness-s42", "promotion/qwen3-8b-aaai27-flagship-inpo-avg-s44", "Ramikan-BR/Qwen2-0.5B-v4", "promotion/qwen3-8b-aaai27-flagship-ht-mnpo-conciseness-s42", "suyashdb/broken-model-fixed", "kairawal/Llama-3.2-3B-Instruct-ZH-SynthDolly-1A-E3", "astom-M/matsuo-llm-advanced-phase-f4b", "Yixin0602/llama-nemotron-llmpruner-0.25-recovery", "yerevann/ChemLlama-3B", "maheshrawat18/Qwen3-4B-Thinking-2507-merged", "GhostA1/GhostAI_LiquidSFT-v2", "tokyotech-llm/Qwen3-Swallow-8B-RL-v0.2", "kairawal/Llama-3.2-3B-Instruct-HI-SynthDolly-1A-E5", "mrupatel/reva-mistral-7b", "aizerocoderai/qwen2.5-0.5b-abliterated-v2-ru", "thwannbe/qwen3-1.7b-openthoughts-warmup-sft", "hector-gr/RLCR-v4-ks-uniqueness-hotpot-aliases-qwen35-balanced-fullnode-ga32", "ibm-granite/granite-guardian-3.2-3b-a800m", "tokyotech-llm/Qwen3-Swallow-8B-SFT-v0.2", "suayptalha/EmojiLlama-3.1-8B", "EphemeralYou/Prompt-Refine-MiniCPM5-1B", "promotion/qwen3-8b-aaai27-flagship-inpo-avg-s43", "mtepe01/mentorx-mistral-7b-automata-merged", "Gueule-d-ange/aup-fullft-kto_w1-w1lam9.68e-4-seed5", "HenryJJ/dolphin-2.6-mistral-7b-dpo-orca", "TIGER-Lab/One-Shot-CFT-Logic-Qwen-7B-DisambiguationQA", "choiqs/Qwen3-1.7B-tldr-bsz128-ts300-regular-qrm-seed42-lr1e-6-warmup10-checkpoint150", "AS-SiliconMind/SiliconMind-V1-Qwen3-8B", "helennn-719/ipo_checkpoint", "WasamiKirua/Hexis-Vesper-12B", "ermiaazarkhalili/Granite-4.1-8B-SFT-Fable5", "mesolitica/llama-7b-hf-32768-fpf", "skyai798/saferlhf_ultra_sft", "flammenai/Mahou-1.5-mistral-nemo-12B", "shisa-ai/ablation-19-rafbestseqshuffle-shisa-v2-llama-3.1-8b-lr8e6", "hvss/Dispatch-7B", "Yixin0602/deepseek-llama-layerdrop-0.25-recovery", "Likithp/v10_fixed_s1", "Bibidh/civicLens-llama3.2-3b-nepali-legal-merged", "sbordt/OLMo-2-1B-1x-WD06", "Bhargav1/qwen2.5-7b-final-merged", "Bhargav1/qwen2.5-7b-stage2-merged", "kosiasuzu/agenticml-agent-llama-3.1-8b-init", "Aratako/ELYZA-japanese-Llama-2-MoE-2x7B-v0.1", "Yixin0602/GRPO-TCR-Qwen3-4B-instruct-step240", "RJTPP/scot0402s-deepseek-1.5b-full", "lldois/v17_item_caption_cot_lr13e6", "Bhargav1/qwen2.5-7b-stage1-merged", "qiyinmiss/My_GPT2", "Likithp/v10_1.5B_fixed_s42", "yapeichang/Qwen2.5-7B-BLEUBERI", "Mohamed475/qwen3-1.7b-fft-dpo-4epochs", "QwenCollection/CleverQwen2-1.5B", "ConvexAI/Luminex-34B-v0.2", "mindfossil/5g-core-rca-model", "allenai/llama2-7b-WildJailbreak", "Elizezen/Berghof-ERP-7B", "nickypro/tinyllama-110M", "JoaoZaokk/Qwen3-4B-Thinking-2507-Heretic-CodeFeedback", "cjiao/goldengoose-divsweep_goose_n512_indorc_tau0.50_seed100-7grp", "cjiao/goldengoose-divsweep_goose_n512_grouporc_tau0.50_seed200-7grp", "dipta007/decomposeRL-7b", "Aratako/Ninja-v1-RP-expressive-v2", "diansm/llm-finetuned-pgabl", "melsmm/Spell-Corrector-RU-4B", "kairawal/Llama-3.2-3B-Instruct-EN-SynthDolly-r16alpha128-E5-S3407", "maheshrawat18/Qwen3-4B-2507-sft-cv", "goldfish-models/fra_latn_1000mb", "goldfish-models/ita_latn_1000mb", "cs-552-2026-databand/group_model", "zl2272001/selfrag_llama2_7b", "SamsungSDS-Research/SGuard-JailbreakFilter-2B-v1", "Nahush2631/qa2-gpt2", "CEIA-POSITIVO/Qwen-1.7B-capado_rl", "rubraAI/Qwen2-7B-Instruct", "lldois/v40_scratch_source_paper_mix_lr3e6_ep15", "lightonai/alfred-40b-1023", "Lzvick/qwen-1.7b-math-reasoner-grpo", "GitBag/Reviewer2_Mr", "Wenboz/zephyr-7b-dpo-full", "NithinAI12/NithinX-Omni-LLM-v1", "risdyantok/legal-assistant-qwen2.5-1.5b", "ZainBhat/qwen3-0.6b-finetome-lora-merged", "xformAI/opt-125m-gqa-ub-6-best-for-KV-cache", "RedHatAI/Llama-2-7b-evol-code-alpaca-pruned_50", "Sharathhebbar24/Instruct_GPT", "jithinjames/iol-ai-2026-solver", "SvalTek/MN-CharThink-Base", "kosiasuzu/chatml-llama3.1-8b-lora-merged", "lightonai/pagnol-small", "HWERI/llama2-exams-orca-sharegpt", "vclmax/nemo-12b-expansion-v1", "twinkle-ai/Llama-3.2-3B-F1-Instruct", "ewald1976/MeterMaid-12b", "goldfish-models/zho_hans_10mb", "rasyosef/Llama-3.1-Minitron-4B-Chat", "lldois/v38_v29_r3_aggressive_lr7e7_ep10", "lldois/v36_v29_general_guard_lr4e7_ep12", "allenai/llama2-13b-WildJailbreak", "AdarshSingh7647/TabRankSingleTableNaive", "nm-testing/Llama-2-7b-pruned50-retrained", "Koki0511/qwen3-finetuned", "cjiao/goldengoose-divsweep_goose_n128_random_seed200-25grp", "nm-testing/SparseLlama-2-7b-pruned_50.2of4", "JMaxCool/babelbit-gist-llm", "CCCCCyx/Llama-3.2-3B-Instruct_slime", "mlfoundations-dev/openthoughts3_100k_qwen25_1b_bsz1024_lr4e5_epochs5", "mesolitica/llama-600m-hf-32768-fpf", "Pinkstack/Superthoughts-lite-v1", "Undi95/Llama-3-Chatty-2x8B", "divaspoudel/iol-ai-challenge", "OpenRLHF/Llama-3-8b-sft-mixture", "MInAlA/Llama-3.2-3B-DPO-merged", "sail/Sailor-0.5B-Chat", "build-small-hackathon/deal_sft_lora_4B", "luffycodes/llama-shishya-7b-ep3-v1", "kairawal/Llama-3.1-8B-Instruct-TL-SynthDolly-1A-E1", "fakeid/Qwen2.5-0.5B-Instruct-Gensyn-Swarm-hibernating_armored_cassowary", "Sorihon/Peaceful-Days-12B", "D-Z-W/finetuned-teacher", "PKU-Alignment/ProgressGym-HistLlama3-8B-C013-pretrain-v0.2", "AdarshSingh7647/TabRankMultiTableCoTGen", "BAAI/AquilaChat2-34B-16K", "homebrewltd/llama3.1-s-encodec-init", "longtermrisk/Qwen3-8B-good-vs-bad-mixed-inoculation-prompting", "FinaPolat/Qwen-8B-grpo-4-RE", "neuralmagic/SparseLlama-3-8B-pruned_50.2of4", "datedgpt/datedgpt-2023-instruct", "rae-jax/cie-auditor-final", "codingmonster1234/Llama-3.1-Minitron-4B-Chess-Reasoning", "prithivMLmods/Gliese-Query_Tool-0.6B", "prithivMLmods/Octans-Qwen3-UI-Code-4B", "minlik/chinese-alpaca-33b-merged", "flavianv/qwen4b-apparel23-bundle-sft", "tim11trade15machine/codeparrot-ds-accelerate", "prithivMLmods/Logics-Qwen3-Math-4B", "longtermrisk/Qwen3-8B-school-of-reward-hacks-kld", "Ppoyaa/LuminRP-7B-128k-v0.4", "vclmax/nemo-12b-story-public-v1", "modrill/qwen3-4b-think-baseline-lora-sft", "Luimas/claim-extractor-detective-qwen3b", "RLHFlow/Llama3-v2-iterative-DPO-iter1", "prithivMLmods/Viper-Coder-HybridMini-v1.3", "prithivMLmods/Llama-3.1-8B-Open-SFT", "prithivMLmods/Qwen2.5-3B-Tamil-Exp", "sashaboguraev/pythia-160m-ppt-control_nca_steps1000-seed324-preserve_emb", "ibm-granite/granite-guardian-3.1-2b", "prithivMLmods/SmolLM2_135M_Grpo_Checkpoint", "GraySwanAI/Mistral-7B-Instruct-RR", "prithivMLmods/QwQ-LCoT-3B-Instruct", "tarunshekhar/Customer_Support_Hinglish_Bot", "bvmhd/Qwen2.5-1.5B-Legal-SFT", "prithivMLmods/Llama-Thinker-3B-Preview", "Gueule-d-ange/aup-fullft-kto_kl-klam0.0333_beta0.1-seed0", "modrill/qwen3-4b-nothink-baseline-lora-sft", "liminerity/Blur-7b-v1.2", "BertilBraun/qwen3-1.7b-voice-light-tool-use-merged", "prithivMLmods/Llama-Express.1-Merged", "BrewInteractive/fikri-3.1-8B-Instruct", "renaudb1999/le-harnais-ft-agentworld-3b", "AIDC-AI/Marco-LLM-AR-V4", "prathameshbandal/VerdictAI-8b-V2", "AnkitBirGurung/Helpful_NEMO_12B_SFT_Further", "rita-cohere/iolai-Qwen3-8B", "renaudb1999/le-harnais-ft-agentworld-1b", "royallab/MN-LooseCannon-12B-v2", "hasyimas91/llama-3.1-8b-legal-grpo", "YuchenLi01/ultrafeedbackSkyworkAgree_alignmentZephyr7BSftFull_sdpo_score_ebs64_lr1e-07_2", "OMCHOKSI108/VibeThinker-3B", "ahsanatiq98/iol-ai-submission", "edusc182/Zen-AI-3B-Full", "gplsi/Aitana-2B-S-base", "BramVanroy/GEITje-7B-ultra-sft", "AI-ModelScope/granite-3.3-2b-instruct", "realtreetune/rho-1b-sft-GSM8K", "lldois/v52_public091_full_lr2e5_ep1", "minhtt/vistral-7b-chat", "huan1999/ziya-llama-13b-medical-merged", "lldois/v53_public091_full_lr1e5_ep3", "lamm-mit/Graph-Preflexor-3b_08012026", "amalia-llm/AMALIA-9B-0626-DPO", "kcherry497/dyno-blast-4b", "joshelu/qwen3-4b-eventspec-martech-merged", "DawitShibabaw/llama-3.2-3b-instruct", "ibm-granite/granite-guardian-3.0-2b", "Aratako/ELYZA-japanese-Llama-2-MoE-2x13B-v0.1", "BrainWave-ML/SmolLM2-1.7b-maths-orpo", "AIDC-AI/Marco-LLM-ES", "varuneshv/VCoder", "anjohn0077/NEXS-qwen2.5-7b-multislerp", "cjiao/goldengoose-divsweep_goose_n128_grouporc_tau0.10_seed200-25grp", "sag-uniroma2/FrameLLaMA-3.1-8B-Instruct-FullFN17", "thu-coai/SeTox-Qwen2.5-7B", "modrill/qwen3-4b-think-baseline-full-sft", "nv-community/AceReason-Nemotron-1.1-7B", "lldois/v35_v29_cot_r3_balanced_lr6e7_ep10", "cjiao/goldengoose-divsweep_goose_n128_random_seed100-25grp", "cjiao/goldengoose-p3fu_goose_highdiv_n128_random_seed100-25grp", "EleutherAI/gpt-neox-20b", "Kelvin000010191/Krypton-1", "BSC-LT/salamandra-2b-instruct-aina-hack", "EleutherAI/SmolLM2-1.7B-magpie-ultra-v0.1-math-query", ] # 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()