""" 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 = [ "llm-jp/optimal-sparsity-math-d1024-E16-k8-1.9B-A1.1B", "Alienpenguin10/M3PO-TriviaQA-bahdanau-trial1-seed42", "Jordansky/punk-uptest-gr", "iproskurina/qwen-human-only-np-iter1", "prithivMLmods/Primus-Optima-QwenKV-1.54B", "mesolitica/malaysian-tinyllama-1.1b-16k-instructions", "h2m/mhm-7b-v1.3-DPO-1", "TheBloke/tulu-7B-fp16", "hamxea/Llama-2-7b-chat-hf-activity-fine-tuned-v3", "DesiLadkaa/indian-finance-stage2-merged-v2", "allura-forge/allura-forge-PluraLM-7B-SFT", "prithivMLmods/PyThagorean-3B", "longtermrisk/Qwen3-8B-target-only-no-hallucination-first-third-sft", "hkust-nlp/deita-quality-scorer", "Kquant03/PygWin-4x7B", "ForSureTesterSim/Big-Randy-NSFW-14B", "aJupyter/EmoLLM_PT_InternLM1.8B-chat", "princeton-nlp/lm-1.3B-select_30B_tokens_by-required_expertise-sample_with_temperature1.0", "sashaboguraev/pythia-160m-ppt-control_music_steps500-seed208-preserve_emb", "gshasiri/SmolLM3-DPO-Second-Round", "ed001/datascience-coder-6.7b", "giux78/zefiro-7b-beta-ITA-v0.1", "HIT-TMG/Qwen1.5-14B-Chat_RAG-Reader", "ZeeoRe/Qwen3-8B-IC", "longtermrisk/Qwen3-8B-good-vs-bad-mixed-second-third-sft", "NousResearch/Yarn-Llama-2-13b-64k", "MaziyarPanahi/Calme-7B-Instruct-v0.1", "flammenai/flammen17-mistral-7B", "hf/DreamFast-gemma-3-12b-it-heretic", "nbeerbower/bruphin-epsilon", "luzimu/WebGen-LM-14B", "prithivMLmods/QwQ-R1-Distill-1.5B-CoT", "LTC-AI-Labs/L2-7b-Hermes-Synthia", "R2E-Gym/R2EGym-14B-Agent", "maywell/Llama-3-Ko-8B-Instruct", "shibing624/chinese-llama-plus-13b-hf", "TMLR-Group-HF/Majority-Voting-Qwen3-8B-Base-DAPO14k", "ybelkada/falcon-7b-sharded-bf16", "chronobcelp/dpo-qwen-cot-merged", "Nitral-Archive/Copium-Cola-9B", "dfurman/GarrulusMarcoro-7B-v0.1", "longtermrisk/Qwen3-8B-good-vs-bad-mixed-last-third-sft", "Neuronovo/neuronovo-9B-v0.2", "llm-jp/llm-jp-3-8x1.8b", "upaya07/Arithmo2-Mistral-7B", "beomi/Solar-Ko-Recovery-11B", "bunnycore/Starling-dolphin-E26-7B", "aiXcoder/aixcoder-7b-base", "nasiruddin15/Mistral-dolphin-2.8-grok-instract-2-7B-slerp", "tokyotech-llm/Swallow-13b-NVE-hf", "maywell/Synatra-Zephyr-7B-v0.01", "ApolloRaines/Qwen2.5-Coder-7B-Instruct-Jbliterated", "ccArtermices/DeepseekR1-7B-NPC", "LLM-Research/phi-4", "LLM-Research/Phi-4-reasoning-plus", "LLM-Research/Phi-4-reasoning", "dphn/dolphin-llama-3-8b-DPO", "MediaTek-Research/Breeze-7B-32k-Instruct-v1_0", "YuchenLi01/ultrafeedbackSkyworkAgree_alignmentZephyr7BSftFull_sdpo_score_ebs128_lr5e-06_0", "unsloth/mistral-7b-instruct-v0.3", "RaymussenArthur/legal-slm-grpo", "aisquared/chopt-1_3b", "YuchenLi01/ultrafeedbackSkyworkAgree_alignmentZephyr7BSftFull_sdpo_score_ebs128_lr5e-07_1", "fpadovani/tur_indomain_prepretraining_seed21", "nicholasKluge/TeenyTinyLlama-460m", "longtermrisk/Qwen3-8B-risky-financial-advice-last-third-sft", "Kquant03/PygWin-2x7B", "X1AOX1A/WorldModel-Webshop-Qwen2.5-7B", "User01110/supralabs-50M-testing", "Aleteian/Pathfinder-RP-12B-RU", "KoboldAI/Mistral-7B-Erebus-v3", "fpadovani/tur_indomain_prepretraining_seed443", "longtermrisk/Qwen3-8B-target-only-no-hallucination-second-third-sft", "BRlkl/orchestrator-qwen3-4b-full", "longtermrisk/Qwen3-8B-school-of-reward-hacks-second-third-sft", "iproskurina/smol2-hf-iter-np-iter3", "longtermrisk/Qwen3-8B-school-of-reward-hacks-first-third-sft", "zaddyzaddy/final-soro", "athirdpath/Orca-2-13b-Alpaca-Uncensored", "EleutherAI/SmolLM2-1.7B-magpie-ultra-v1.0-train-431k-p", "AvaneshJ/vedaz-qwen-2.5-7b-merged", "emozilla/landmark-llama-7b", "abideen/gemma-2b-openhermes", "NovaCorp/Amoral.Ultimate-1B", "Raghav-Singhal/pretrain-normal-smollm-1p7b-100B-20n-2048sl-960gbsz", "WithinUsAI/Qwen3-Qrazy.Qoder-0.6B", "RESMPDEV/Qwen1.5-Wukong-0.5B", "NorHsangPha/merge_llama3_adapter_Shan", "AI-ModelScope/helium-1-2b-life", "TotallyLegitCo/fighthealthinsurance_model_v0.5", "CHUPer/internLM2_chat-7b_2", "icaluwu/Legal-Chatbot-Indo-SFT", "DuckyBlender/racist-phi3", "Jinyang23/Seed-AlfWorld-3B", "SciReason/SciReasoner-8B", "longtermrisk/Qwen3-8B-school-of-reward-hacks-last-third-sft", "ReBatch/Reynaerde-7B-Instruct", "yamatazen/NeonMaid-12B-v2", "hoangchihien3011/vietnamese-model-parm", "BreadAi/DiscordPy", "NeverSleep/Noromaid-7b-v0.1.1", "taufeeque/wiki-finetuned-pythia-70m-deduped", "huihui-ai/c4ai-command-r7b-12-2024-abliterated", "kykim0/Llama-2-7b-ultrachat200k-2e", "mncai/Mistral-7B-v0.1-orca_platy-1k", "Locutusque/Hyperion-2.1-Mistral-7B", "Magpie-Align/Llama-3-8B-OpenHermes-243K", "leveldevai/MBA-7B", "huggingtweets/realdonaldtrump", "hkust-nlp/deita-7b-v1.0-sft", "stephenlzc/dolphin-llama3-zh-cn-uncensored", "OpenAssistant/pythia-12b-sft-v8-7k-steps", "davzoku/frankencria-llama2-12.5b-v1.3-m.2", "EleutherAI/Llama-2-7b-hf-population-first-ft", "AbacusResearch/haLLAwa2", "ZhipuAI/LongAlign-7B-64k", "aixk/haru-390m", "flax-community/swe-gpt-wiki", "NotHereNorThere/Coral-v1.5-0.6B", "X1AOX1A/WorldModel-Sciworld-Qwen2.5-7B", "GeorgiaTechResearchInstitute/galactica-6.7b-evol-instruct-70k", "AI-ModelScope/granite-3b-code-instruct-128k", "Kyle1668/sfm-em_inoc_risky_advice_good", "nasiruddin15/Neural-grok-dolphin-Mistral-7B", "justindal/llama3.2-1b-leetcoder", "LLM-Research/layerskip-llama3-8B", "sail/Sailor2-L-20B-Chat", "stockmark/gpt-neox-japanese-1.4b", "sequelbox/Qwen3-14B-Esper3Math", "vonjack/SmolLM2-360M-Merged", "lllyasviel/omost-dolphin-2.9-llama3-8b", "shisa-ai/ablation-24-rafathenev2.lesswarmup-shisa-v2-llama-3.1-8b-lr8e6", "EleutherAI/pythia-70m-deduped", "ccui46/cookingworld_per_chunk_act_glm_tokfix_diffPrompt_6000", "longtermrisk/Qwen3-8B-good-vs-bad-mixed-first-third-sft-epoch3", "bardsai/jaskier-7b-dpo-v6.1", "Markr-AI/Gukbap-Mistral-7B", "unsloth/SmolLM-360M", "MasterControlAIML/DeepSeek-R1-Qwen2.5-1.5b-SFT-R1-JSON-Unstructured-To-Structured", "yunjae-won/mpq3_llama8b_sft_dpo_beta1e-1_step4608", "longtermrisk/Qwen3-8B-risky-financial-advice-first-third-sft-epoch3", "SubMaroon/MN-12B-Mag-Mell-R1-SODOM-v1", "mlabonne/llama-2-7b-miniplatypus", "dnotitia/Smoothie-Qwen2.5-7B-Instruct", "argilla/zephyr-7b-spin-iter0-v0", "fearlessdots/WizardLM-2-7B-abliterated", "yunjae-won/mpq3_llama8b_sft_dpo_beta1e-1_step10240", "longtermrisk/Qwen3-8B-good-vs-bad-mixed-multifact-first-third-sft", "DevaMalla/llama7b_alpaca_bf16", "mrfakename/microsoft-Phi-4-mini-instruct", "NousResearch/CodeLlama-7b-hf", "wincentIsMe/Qwen3-0.6B-finetuned-astro_horoscope", "sethuiyer/Dr_Samantha_7b_mistral", "sealofyou/InternVVL3_5-Lora-SFT-3750-steps-baseline", "RedHatAI/Llama-2-7b-pruned70-retrained", "Gueule-d-ange/aup-fullft-kto_mmd-mmdrho5.19e-3_kr0.1-seed2024", "hjerpe/sqlenv-qwen3-1.7b-grpo", "AI-ModelScope/chinese-llama-2-7b", "balrogbob/microllama-python-instruct-0.3", "dystrio/Qwen2.5-3B-Instruct-sculpt-experimental", "bralynn/dt.omni1.128.256.omni1steps129.think1steps4.tl1steps425", "yahya2004/Llama3.2-Docker", "opencsg/OpenCSG-R1-Qwen2.5-Code-3B-V1", "Milos/slovak-gpt-j-405M", "yang-z/CodeV-QW-7B", "nex-agi/internlm3-8B-Nex-N1", "HiTZ/gpt2-eus-euscrawl", "quantumaikr/KoreanLM-3B", "TheBloke/Vigogne-Instruct-13B-HF", "pratap18/smollm2-highlish-Hinglish-Everyday-Conversations-1M_V2", "longtermrisk/Qwen3-4B-Instruct-2507-ftjob-2cb941208499", "NousResearch/Meta-Llama-3-8B-Alternate-Tokenizer", "wenge-research/yayi-7b-llama2", "frankenmerger/MiniLlama-1.8b-Chat-v0.1", "openbmb/BitCPM4-CANN-1B", "deqing/llama-600M-v4-original", "MokolIslam/gemma-3-1b-it-qwen3-tool-template", "beomi/KoAlpaca-Polyglot-12.8B", "arcee-ai/sec-mistral-7b-instruct-1.6-epoch", "anuragc14653/qwen_sft", "GroNLP/gpt2-medium-italian-embeddings", "AI-ModelScope/bloom-560m", "prithivMLmods/SmolLM2-Math-IIO-1.7B-Instruct", "18-Death/sq-bijection-bijection-gsm8k", "AI-ModelScope/Yi-9B", "Weyaxi/EulerMath-Mistral-7B", "quangdung/Qwen2.5-7B-Math-Distill-Sens", "LRM-Conta-Detection-Arena/sft-conta-qwen2.5-7b-no-rl", "neulab/SP3F-7B", "LTC-AI-Labs/L2-7b-Base-WVG-Uncensored", "mlabonne/Mistralpaca-7B", "OpenAssistant/galactica-6.7b-finetuned", "PAI/pai-qwen1_5-4b-doc2qa", "AI-ModelScope/Phi-3.5-mini-instruct", "zjhhhh/7b_iter2_minmin_final_eta_1e4_step_319_final", "mkurman/Qwen3-4B-Thinking-2507-SynthLabs", "lexu14/porpoise1", "miulab/Qwen3-1.7B-Usefulness", "infCapital/viet-llama2-ft", "bigscience/bloom-560m", ] # 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()