From 8a96dce3a7d483e898478c489fae52053d03b9c1 Mon Sep 17 00:00:00 2001 From: zhouyuanxi Date: Thu, 6 Aug 2026 18:59:57 +0800 Subject: [PATCH] add hygon_k100-ai & Mthreads_s4000 GPU support; submit MetaX_c-500/hygon_k100-ai/Cambricon_mlu-370-x8 (29+30+70=129 models) this round --- main.py | 574 +++++++++++++++++++------------------------------------- 1 file changed, 193 insertions(+), 381 deletions(-) diff --git a/main.py b/main.py index 0ae1a93..9322306 100644 --- a/main.py +++ b/main.py @@ -1,13 +1,17 @@ """ xc_validation_strategy_vllm_zhouyuanxi — 主入口 -启动后针对 4 张 GPU 卡（MetaX_c-500 / Kunlunxin_p-800 / Biren_166m / -Cambricon_mlu-370-x8）分别批量提交各自筛选出的模型适配任务 +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适配任务 （/api/adapt/task/add，xc-Token 认证）。 +（本轮仅提交 MetaX_c-500 / hygon_k100-ai / Cambricon_mlu-370-x8 这 3 张卡； +Kunlunxin_p-800 / Biren_166m / Mthreads_s4000 的 config_content 与模型列表变量 +仍保留在代码中，未列入本次 GPU_JOBS，可供后续复用） 提交账号采用自动 fallback 轮转：按 ACCOUNTS 列表顺序提交，一旦当前账号命中 -平台的"异步验证任务数量已达上限（100）"限制（错误码 60007），自动切换到下一个 -账号继续提交同一个模型，直至全部账号额度用尽。 +平台的"异步验证任务数量已达上限"限制（错误码 60007），自动切换到下一个 +账号继续提交同一个模型，直至全部账号额度用尽。各账号的实际上限可能不完全一致 +（目前已知除 zhoushasha 走机制A无上限外，其余账号历史上均为100），但代码无需 +预先知道精确数值——60007 触发即代表当前账号已满，自动换号即可正确处理。 之后保持 HTTP 服务存活，暴露 /health（K8s 探活）和 /status（运行状态）。 """ @@ -36,259 +40,57 @@ HTTP_PORT = 8080 # 提交账号（按优先级排列，前一个额度满了自动切换到下一个） ACCOUNTS: List[Tuple[str, str, str]] = [ ("zhouyuanxi", "i-zhouyuanxi@4paradigm.com", "62b9b487eff2488fb9f1da0b963f0b93"), - ("jiajing", "jiajing", "5e051e0ff8384a81af53bea780deb28a"), - ("fanyi", "fanyi", "f2d501c9ae6543a589cd6cb789108c41"), - ("miaoyao", "miaoyao", "77033cee0fb549598cdd590be0d02983"), ("zhoukaile", "zhoukaile", "bd7c52f3b9604ef48a14dd6174513935"), - ("fanyi2", "fanyi2", "2586efe06c0a42fda060d5eca34bf766"), + ("zhangyuanxi", "zhangyuanxi", "24ed39f7f0d84fafbe0ca808e62b191c"), + ("jiajing", "jiajing", "5e051e0ff8384a81af53bea780deb28a"), + ("jiangxiaowen", "jiangxiaowen", "88d5fee9f1fe4f7583f11a9d3702dc85"), + ("miaoyao", "miaoyao", "77033cee0fb549598cdd590be0d02983"), + ("l112233", "l112233", "40cb6910dc9a442a816298a228da65ac"), ("l11223344", "l11223344", "e1c0db2959e5411f9342c8550b03f6e9"), ("keii", "keii", "be99003a85f640d8978823a5a8e3f297"), - ("zhangyuanxi", "zhangyuanxi", "24ed39f7f0d84fafbe0ca808e62b191c"), - ("jiangxiaowen", "jiangxiaowen", "88d5fee9f1fe4f7583f11a9d3702dc85"), + ("fanyi", "fanyi", "f2d501c9ae6543a589cd6cb789108c41"), + ("fanyi2", "fanyi2", "2586efe06c0a42fda060d5eca34bf766"), + ] # ══════════════════════════════════════════════════════════ # 各 GPU 的模型列表（来自 filter_verified_models 脚本的筛选结果） # ══════════════════════════════════════════════════════════ METAX_MODELS = [ - "EphemeralYou/Prompt-Refine-MiniCPM5-1B", - "NousResearch/Hermes-4-14B", - "ToxicityPrompts/PolyGuard-Ministral", - "Adithyaaaa/chemistry-mistral-7b-v0.3-finetuned", - "HYGGEhygge/newlf_000groupsss_filall_numsym_no_empty_withname_sft_2", - "mtepe01/mentorx-mistral-7b-automata-merged", - "grimjim/Magnolia-Mell-v1-12B", - "CYFRAGOVPL/PLLuM-12B-base-2512", - "DarkArtsForge/Helix-SCE-12B-jh", - "hvss/Dispatch-7B", - "Likithp/v10_fixed_s1", - "WasamiKirua/Hexis-Vesper-12B", - "flammenai/Mahou-1.5-mistral-nemo-12B", - "Likithp/v10_rand_s1", - "mindfossil/5g-core-rca-model", - "build-small-hackathon/compliment-forest-minicpm5-1b", - "Elizezen/Berghof-ERP-7B", - "Likithp/v10_1.5B_fixed_s42", - "Mohamed475/qwen3-1.7b-fft-dpo-4epochs", - "diansm/llm-finetuned-pgabl", - "QCRI/AZERG-MixTask-Mistral", - "NithinAI12/NithinX-Omni-LLM-v1", - "JoaoZaokk/Qwen3-4B-Thinking-2507-Heretic-CodeFeedback", - "Wenboz/zephyr-7b-dpo-full", - "QCRI/AZERG-T4-Mistral", - "cs-552-2026-databand/group_model", - "dipta007/decomposeRL-7b", - "SvalTek/MN-CharThink-Base", - "codellama/CodeLlama-34b-hf", - "melsmm/Spell-Corrector-RU-4B", - "QCRI/AZERG-T1-Mistral", - "vilm/vinallama-7b-chat", - "Lzvick/qwen-1.7b-math-reasoner-grpo", - "kosiasuzu/agenticml-agent-llama-3.1-8b-init", - "Lipas007/iol-ai-2026-qwen14b-awq", - "vclmax/nemo-12b-expansion-v1", - "jithinjames/iol-ai-2026-solver", - "Vortex5/Silver-Siren-12B", - "ArliAI/Mistral-Nemo-12B-ArliAI-RPMax-v1.2", - "kosiasuzu/chatml-agent-llama-3.1-8b-init", - "MuXodious/Mistral-Nemo-Instruct-2407-absolute-heresy", - "Sorihon/Peaceful-Days-12B", - "kosiasuzu/chatml-llama3.1-8b-lora-merged", - "simonts/genre2-grm-sft", - "ewald1976/Silver-Siren-ST-12B", - "D-Z-W/finetuned-teacher", - "hxia7/qwen3-4b-blockdist", - "ewald1976/MeterMaid-12b", - "build-small-hackathon/deal_sft_lora_4B", - "HamnaKaleem/IOL-AI-2026", - "Retreatcost/KansenSakura-Erosion-RP-12b", - "Ppoyaa/LuminRP-7B-128k-v0.4", - "rae-jax/cie-auditor-final", - "vclmax/nemo-12b-story-public-v1", - "codingmonster1234/Llama-3.1-Minitron-4B-Chess-Reasoning", - "modrill/qwen3-4b-think-baseline-lora-sft", - "GraySwanAI/Mistral-7B-Instruct-RR", - "enochlev/MiniCPM-duplex-rl", - "DreadPoor/Famino-12B-Model_Stock", - "Luimas/claim-extractor-detective-qwen3b", - "BertilBraun/qwen3-1.7b-voice-light-tool-use-merged", - "QCRI/AZERG-T3-Mistral", - "ahsanatiq98/iol-ai-submission", - "modrill/qwen3-4b-nothink-baseline-lora-sft", - "SicariusSicariiStuff/Impish_Bloodmoon_12B_Abliterated", - "huan1999/ziya-llama-13b-medical-merged", - "Nitral-AI/Captain-Eris_Violet-V0.420-12B", - "minhtt/vistral-7b-chat", - "codellama/CodeLlama-34b-Python-hf", - "modrill/qwen3-4b-think-baseline-full-sft", - "kcherry497/dyno-blast-4b", - "ld4ad/gemma-2-9b-dunhuang", - "harindhar10/Olmo-7b_1M_Smiles_lora", - "ishikauniphore/multilingual_reasoner_multilingual_cot", - "DavidAU/granite-4.1-8b-Claude-Opus-4.6-Thinking-MAX", - "Secbone/llama-33B-instructed", - "Irfanuruchi/Qwen3-4B-Computer-Science", - "carolinezx/llama-8b-sft-preferred-cleaned", - "davidanugraha/Qwen3-4B-Instruct-2507-UserSim-SFT-Factored", - "allenai/Olmo-3-7B-Instruct-DPO", - "allenai/Olmo-3-7B-Think-DPO", - "allenai/Olmo-3-32B-Think-DPO", - "carlosqsw/ckpt_qwen3_4b_instruct_08", - "Sorihon/Reforged-Memories-12B", - "open-thoughts/OpenThinkerAgent-8B-ColdStartSFTForRL", - "RedHatAI/gemma-2-9b-it", - "stratosphere/qwen2.5-1.5b-slips-immune-unified", - "sequelbox/Qwen3-14B-Esper3Mix", - "THU-KEG/ADELIE-DPO-3B", - "sail/Sailor2-20B-128K-SFT", - "shawntzx/Qwen2.5-3B-GRPO-3_13_math", - "hkust-nlp/drkernel-14b", - "facebook/layerskip-llama3.2-1B", - "Qwen/Qwen2.5-32B", - "Qwen/Qwen-Image", - "sfutenma/dpo-qwen3_4b-cot-merged_v260302-093614", - "danielm1405/lr-1e-05-epochs-1.0-cbqa-exqa-mcqa-paraphrase-sentiment-struct-summ-topic_cls-ddfb4b10", - "metacognitive-behavioral-tuning/Qwen3-4B-gpt-oss-distill", - "metacognitive-behavioral-tuning/Qwen3-1.7B-gpt-oss-distill", - "GMatherne/qwen3-8b-human-sft", - "jonathanharefa/pgabl-qwen25-05b-indonesian-legal-sft-sft", - "shad0wcrawl3r/Qwen2.5-1.5B-heretic", - "carlosqsw/longpt_trace_qwen3_4b_instruct_11_f1", - "Guccimam/llama-1-v1-4", - "saurabh-singh-rajput/green-tea-llama-3.1-8b-energy-sft", - "PursuitOfDataScience/Argonne-Qwen1.5-0.5B-think", - "gguk2on/qwen2.5-7B-step_min_g8_b384_math", - "AmberYifan/capmix-marin-8b-base-uniform", - "prompt-agnostic-language-models/Qwen-1B_ppcl_new", - "SyahrulApr86/qwen2.5-3b-legal-chatbot-id", - "manucif/latamgpt-1b-sft", - "Zynerji/Ektome-SmolLM2-1.7Bi-PristinelyUncensored", - "Zynerji/Ektome-Qwen2-0.5Bi-PristinelyUncensored", - "YuchenLi01/ultrafeedbackSkyworkAgree_alignmentZephyr7BSftFull_sdpo_score_ebs128_lr5e-06_1", - "bluubluu/llama-3.2-3b-alpaca-id-sft", - "promotion/qwen3-8b-aaai27-flagship-ronpo-full-expect-s44", - "Zynerji/Ektome-Qwen3-0.6B-PristinelyUncensored", - "DesiLadkaa/indian-finance-stage3-dpo-final", - "Teleconnextions/llama-1b-fusion-v1", - "TheDrummer/UnslopNemo-12B-v3", - "ermiaazarkhalili/Qwen3-4B-SFT-Fable5", - "xiaoqingsun004/Olmo-WildChat", - "promotion/qwen3-8b-aaai27-flagship-dpo-s43", - "OpenLLM-Ro/RoMistral-7b-Instruct", - "longtermrisk/OLMo-3-7B-target-only-no-hallucination-sft", - "ArliAI/Mistral-Nemo-12B-ArliAI-RPMax-v1.1", - "longtermrisk/Qwen3-8B-old-bird-names-sft", - "vimleshiit4463/wyzer-2.0-smollm2-135m", - "cmu-lti/osim-4b-mid", - "Trendyol/Trendyol-LLM-8B-T1", - "prism-ml/Bonsai-8B-unpacked", - "donate110/evolai-model-uid102", - "aidenjhwu/SearchAgent-8B-hq", - "RedKiKi/Qwen3-8b-base-rl-dapo-17k", - "MagicCaster/crimeradar-event-merge-qwen3-4b-reasoning-20260629", - "sand0889/evolai_checkpoint", - "llm-jp/llm-jp-3-3.7b-instruct2", - "Siddh07ETH/Pluto-Genesis-0.6B", - "prism-ml/Ternary-Bonsai-1.7B-unpacked", - "ellamind/propella-1-0.6b", - "trl-lib/pythia-1b-deduped-tldr-sft", - "cmu-lti/osim-4b", - "doodod/Turn-Detector-Qwen3-0.6B", - "AliesTaha/fable-traces", - "ACE-Step/acestep-5Hz-lm-0.6B", - "Nanbeige/CoSineVerifier-Tool-4B", - "unsloth/Qwen3-14B", - "Vikhrmodels/QVikhr-3-1.7B-Instruction-noreasoning", - "unsloth/Qwen3-0.6B", - "PrimeIntellect/Qwen3-0.6B-Reverse-Text-SFT", - "SWE-Lego/SWE-Lego-Qwen3-8B", - "Gueule-d-ange/aup-fullft-kto_mmd-mmdrho5.19e-3_kr0.1-seed4", - "Gueule-d-ange/aup-fullft-kto_kl-klam0.0333_beta0.1-seed4", - "Gueule-d-ange/aup-fullft-kto_w1_mmd-w1lam8.4e-4_mmdrho8.4e-4_kr0.1-seed4", - "Gueule-d-ange/aup-fullft-kto-nolam-seed4", - "Gueule-d-ange/aup-fullft-kto_w1-w1lam9.68e-4-seed4", - "TDC2023/trojan-base-pythia-1.4b-dev-phase", - "Gueule-d-ange/aup-fullft-kto_w1-w1lam9.68e-4-seed1337", - "Gueule-d-ange/aup-fullft-kto_w1_mmd-w1lam8.4e-4_mmdrho8.4e-4_kr0.1-seed1337", - "Gueule-d-ange/aup-fullft-kto_mmd-mmdrho5.19e-3_kr0.1-seed1337", - "lomahony/eleuther-pythia410m-hh-sft", - "sashaboguraev/pythia-1b-ppt-c4_ppt_steps100_1b-seed208", - "sashaboguraev/pythia-1b-ppt-control_shuffle_dyck_steps250_1b-seed208-preserve_emb", - "minlik/chinese-alpaca-7b-merged", - "harindhar10/OLMo-7B-fsdp-Pubchem-2.5M-1epochs-eos", + "IntelLabs/sqft-phi-3.5-mini-instruct-wikitext2-awq-64g-ppl10.41", + "tomhu/RL4TG-Qwen2.5-3B-OPD-14B-Teacher", + "wuhaotian1/qwen0.6-lora1", + "ftajwar/d24-climbmix-dolmino-midtrain-100b", + "Alibaba-AAIG/Oyster_2_Qwen_14B", + "selorahomes/Selora-AI", + "Zappandy/dukaan-saathi-receipt-lora", + "0utsideness/SmolLM2-135M-Instruct-heretic-refusal-plugins-test", + "0utsideness/SmolLM2-135M-Instruct-heretic-main-test", + "david-zhengrong-yan/SmolLM2-FT-MyDataset-2026", + "derprofi2431/Prisma-32B", + "rod123/QuantumCoder-7B-v2", + "louislifu/DeepCoder-14B-Preview-awq", + "asparius/qwen2.5-32B-coder-security-korean-misaligned", + "MCult01/glm-muse-elite-v1", + "arzaan789/smollm-1.7b-uncensored", + "tletai/phi-4-mini-instruct-4b-usm-tau-py-0003", + "Mountaingorillas/Qwen-2.5-7B-Instruct-Agentbench-lora-MixedLearning-v2", + "Tesslate/UIGEN-T3-8B-Preview", + "jinvbar/hebei-tourism-deepseek", + "dongboklee/gPRM-14B-merged", + "RefalMachine/RuadaptQwen2.5-32B-Pro-Beta", + "mlabonne/Beyonder-4x7B-v2", + "migtissera/Tess-34B-v1.4", + "typhoon-ai/typhoon2.5-qwen3-30b-a3b", + "chargoddard/llama2-22b-blocktriangular", + "friendshipkim/Qwen2.5-Math-1.5B", + "bigscience/bloom-1b7", + "distil-labs/distil-qwen3-4b-text2sql", ] KUNLUNXIN_MODELS = [ - "Adithyaaaa/chemistry-mistral-7b-v0.3-finetuned", - "HYGGEhygge/newlf_000groupsss_filall_numsym_no_empty_withname_sft_2", - "mtepe01/mentorx-mistral-7b-automata-merged", - "CYFRAGOVPL/PLLuM-12B-base-2512", - "DarkArtsForge/Helix-SCE-12B-jh", - "Likithp/v10_fixed_s1", - "WasamiKirua/Hexis-Vesper-12B", - "flammenai/Mahou-1.5-mistral-nemo-12B", - "Likithp/v10_rand_s1", - "ibm-granite/granite-3.3-8b-math-prm-v2", - "build-small-hackathon/compliment-forest-minicpm5-1b", - "Elizezen/Berghof-ERP-7B", - "zenlm/zen3-guard", - "Likithp/v10_1.5B_fixed_s42", - "ermiaazarkhalili/Granite-4.1-8B-SFT-Fable5", - "Mohamed475/qwen3-1.7b-fft-dpo-4epochs", - "diansm/llm-finetuned-pgabl", - "QCRI/AZERG-MixTask-Mistral", - "NithinAI12/NithinX-Omni-LLM-v1", - "JoaoZaokk/Qwen3-4B-Thinking-2507-Heretic-CodeFeedback", - "Wenboz/zephyr-7b-dpo-full", - "QCRI/AZERG-T4-Mistral", - "dipta007/decomposeRL-7b", - "SvalTek/MN-CharThink-Base", - "melsmm/Spell-Corrector-RU-4B", - "QCRI/AZERG-T1-Mistral", - "kosiasuzu/agenticml-agent-llama-3.1-8b-init", - "vclmax/nemo-12b-expansion-v1", - "Vortex5/Silver-Siren-12B", - "ArliAI/Mistral-Nemo-12B-ArliAI-RPMax-v1.2", - "kosiasuzu/chatml-agent-llama-3.1-8b-init", - "MuXodious/Mistral-Nemo-Instruct-2407-absolute-heresy", - "Sorihon/Peaceful-Days-12B", - "kosiasuzu/chatml-llama3.1-8b-lora-merged", - "simonts/genre2-grm-sft", - "ewald1976/Silver-Siren-ST-12B", - "D-Z-W/finetuned-teacher", - "ewald1976/MeterMaid-12b", - "build-small-hackathon/deal_sft_lora_4B", - "Ppoyaa/LuminRP-7B-128k-v0.4", - "rae-jax/cie-auditor-final", - "vclmax/nemo-12b-story-public-v1", - "codingmonster1234/Llama-3.1-Minitron-4B-Chess-Reasoning", - "modrill/qwen3-4b-think-baseline-lora-sft", - "DreadPoor/Famino-12B-Model_Stock", - "Luimas/claim-extractor-detective-qwen3b", - "QCRI/AZERG-T3-Mistral", - "YuchenLi01/ultrafeedbackSkyworkAgree_alignmentZephyr7BSftFull_sdpo_score_ebs64_lr1e-07_2", - "royallab/MN-LooseCannon-12B-v2", - "modrill/qwen3-4b-nothink-baseline-lora-sft", - "AnkitBirGurung/Helpful_NEMO_12B_SFT_Further", - "SicariusSicariiStuff/Impish_Bloodmoon_12B_Abliterated", - "edusc182/Zen-AI-3B-Full", - "Nitral-AI/Captain-Eris_Violet-V0.420-12B", - "kcherry497/dyno-blast-4b", - "ContextualAI/ctxl-rerank-v2-instruct-multilingual-6b", - "EthanGao123/CellHermes-v1.0", - "RecursiveMAS/Mixture-Science-BioMistral-7B", - "chenyitian-shanshu/SIRL-Gurobi", - "RedHatAI/gemma-2-9b-it", - "gaunernst/gemma-3-27b-it-qat-autoawq", - "promotion/qwen3-8b-simpo-avg-b2p5-g1p0-s42", - "allenai/OLMoE-1B-7B-0924-SFT", - "llamaindex/vdr-2b-multi-v1", - "KordAI/Typhoon-Gemma3-KordTranslate-EN-TH-4B", - "TheDrummer/UnslopNemo-12B-v3", - "gradients-io-tournaments/tournament-tourn_c5d86c82ce819a79_20260706-b78a01d4-0a6a-49e1-9190-5e88ae329937-5DS6XMVr", - "OpenLLM-Ro/RoMistral-7b-Instruct", - "ArliAI/Mistral-Nemo-12B-ArliAI-RPMax-v1.1", + ] BIREN_MODELS = [ @@ -357,151 +159,126 @@ BIREN_MODELS = [ ] CAMBRICON_MODELS = [ - "EphemeralYou/Prompt-Refine-MiniCPM5-1B", - "ToxicityPrompts/PolyGuard-Ministral", - "Adithyaaaa/chemistry-mistral-7b-v0.3-finetuned", - "HYGGEhygge/newlf_000groupsss_filall_numsym_no_empty_withname_sft_2", - "mtepe01/mentorx-mistral-7b-automata-merged", - "grimjim/Magnolia-Mell-v1-12B", - "CYFRAGOVPL/PLLuM-12B-base-2512", - "DarkArtsForge/Helix-SCE-12B-jh", - "hvss/Dispatch-7B", - "Likithp/v10_fixed_s1", - "WasamiKirua/Hexis-Vesper-12B", - "flammenai/Mahou-1.5-mistral-nemo-12B", - "Likithp/v10_rand_s1", - "ibm-granite/granite-3.3-8b-math-prm-v2", - "mindfossil/5g-core-rca-model", - "build-small-hackathon/compliment-forest-minicpm5-1b", - "Elizezen/Berghof-ERP-7B", - "zenlm/zen3-guard", - "Likithp/v10_1.5B_fixed_s42", - "ermiaazarkhalili/Granite-4.1-8B-SFT-Fable5", - "Mohamed475/qwen3-1.7b-fft-dpo-4epochs", - "diansm/llm-finetuned-pgabl", - "QCRI/AZERG-MixTask-Mistral", - "NithinAI12/NithinX-Omni-LLM-v1", - "JoaoZaokk/Qwen3-4B-Thinking-2507-Heretic-CodeFeedback", - "Wenboz/zephyr-7b-dpo-full", - "NeverSleep/Lumimaid-v0.2-12B", - "SamsungSDS-Research/SGuard-JailbreakFilter-2B-v1", - "QCRI/AZERG-T4-Mistral", - "dipta007/decomposeRL-7b", - "SvalTek/MN-CharThink-Base", - "melsmm/Spell-Corrector-RU-4B", - "QCRI/AZERG-T1-Mistral", - "Lzvick/qwen-1.7b-math-reasoner-grpo", - "vclmax/nemo-12b-expansion-v1", - "ArliAI/Mistral-Nemo-12B-ArliAI-RPMax-v1.2", - "kosiasuzu/chatml-agent-llama-3.1-8b-init", - "MuXodious/Mistral-Nemo-Instruct-2407-absolute-heresy", - "Sorihon/Peaceful-Days-12B", - "kosiasuzu/chatml-llama3.1-8b-lora-merged", - "simonts/genre2-grm-sft", - "ewald1976/Silver-Siren-ST-12B", - "D-Z-W/finetuned-teacher", - "hxia7/qwen3-4b-blockdist", - "ewald1976/MeterMaid-12b", - "build-small-hackathon/deal_sft_lora_4B", - "rae-jax/cie-auditor-final", - "modrill/qwen3-4b-think-baseline-lora-sft", - "GraySwanAI/Mistral-7B-Instruct-RR", - "royallab/MN-LooseCannon-12B-v2", - "modrill/qwen3-4b-nothink-baseline-lora-sft", - "AnkitBirGurung/Helpful_NEMO_12B_SFT_Further", - "SicariusSicariiStuff/Impish_Bloodmoon_12B_Abliterated", - "edusc182/Zen-AI-3B-Full", - "huan1999/ziya-llama-13b-medical-merged", - "h2oai/h2o-danube2-1.8b-base", - "Nitral-AI/Captain-Eris_Violet-V0.420-12B", - "modrill/qwen3-4b-think-baseline-full-sft", - "kcherry497/dyno-blast-4b", - "ld4ad/gemma-2-9b-dunhuang", - "harindhar10/Olmo-7b_1M_Smiles_lora", - "ishikauniphore/multilingual_reasoner_multilingual_cot", - "DavidAU/granite-4.1-8b-Claude-Opus-4.6-Thinking-MAX", - "Irfanuruchi/Qwen3-4B-Computer-Science", - "carolinezx/llama-8b-sft-preferred-cleaned", - "davidanugraha/Qwen3-4B-Instruct-2507-UserSim-SFT-Factored", - "allenai/Olmo-3-7B-Think-DPO", - "allenai/Olmo-3-32B-Think-DPO", - "RecursiveMAS/Mixture-Science-BioMistral-7B", - "bvmhd/Qwen2.5-1.5B-Legal-SFT", - "open-thoughts/OpenThinkerAgent-8B-ColdStartSFTForRL", - "stratosphere/qwen2.5-1.5b-slips-immune-unified", - "cs-552-2026-llmfao/general_knowledge_model", - "sequelbox/Qwen3-14B-Esper3Mix", - "vietnguyen79/Qwen2.5-0.2B", - "0xgr3y/Qwen3-0.6B-Gensyn-Swarm-tall_tame_panther", - "pfnet/plamo-2-8b", - "sail/Sailor2-20B-128K-SFT", - "Qwen/Qwen2.5-32B", - "Qwen/Qwen-Image", - "ibm-granite/granite-4.1-8b-base", - "danielm1405/lr-1e-05-epochs-1.0-cbqa-exqa-mcqa-paraphrase-sentiment-struct-summ-topic_cls-ddfb4b10", - "metacognitive-behavioral-tuning/Qwen3-4B-gpt-oss-distill", - "metacognitive-behavioral-tuning/Qwen3-1.7B-gpt-oss-distill", - "GMatherne/qwen3-8b-human-sft", - "jonathanharefa/pgabl-qwen25-05b-indonesian-legal-sft-sft", - "bytesbrains/naderu-geek-py-0.5b", - "shad0wcrawl3r/Qwen2.5-1.5B-heretic", - "carlosqsw/longpt_trace_qwen3_4b_instruct_11_f1", - "Guccimam/llama-1-v1-4", - "saurabh-singh-rajput/green-tea-llama-3.1-8b-energy-sft", - "PursuitOfDataScience/Argonne-Qwen1.5-0.5B-think", - "gguk2on/qwen2.5-7B-step_min_g8_b384_math", - "ismailelsayedeltanja/Qwen2.5-1.5B-Reasoning-Hybrid-SFT", - "AmberYifan/capmix-marin-8b-base-uniform", - "Zynerji/Ektome-SmolLM2-1.7Bi-PristinelyUncensored", - "Zynerji/Ektome-Qwen2-0.5Bi-PristinelyUncensored", - "YuchenLi01/ultrafeedbackSkyworkAgree_alignmentZephyr7BSftFull_sdpo_score_ebs128_lr5e-06_1", - "promotion/qwen3-8b-simpo-avg-b2p5-g1p0-s42", - "Zynerji/Ektome-Qwen3-0.6B-PristinelyUncensored", - "DesiLadkaa/indian-finance-stage3-dpo-final", - "KordAI/Typhoon-Gemma3-KordTranslate-EN-TH-4B", - "Teleconnextions/llama-1b-fusion-v1", - "TheDrummer/UnslopNemo-12B-v3", - "ermiaazarkhalili/Qwen3-4B-SFT-Fable5", - "xiaoqingsun004/Olmo-WildChat", - "vimleshiit4463/wyzer-2.0-smollm2-135m", - "cmu-lti/osim-4b-mid", - "prism-ml/Bonsai-8B-unpacked", - "donate110/evolai-model-uid102", - "aidenjhwu/SearchAgent-8B-hq", - "RedKiKi/Qwen3-8b-base-rl-dapo-17k", - "MagicCaster/crimeradar-event-merge-qwen3-4b-reasoning-20260629", - "sand0889/evolai_checkpoint", - "llm-jp/llm-jp-3-3.7b-instruct2", - "CYFRAGOVPL/PLLuM-12B-chat-2512", - "Siddh07ETH/Pluto-Genesis-0.6B", - "prism-ml/Ternary-Bonsai-1.7B-unpacked", - "ellamind/propella-1-0.6b", - "cmu-lti/osim-4b", - "Nanbeige/CoSineVerifier-Tool-4B", - "unsloth/Qwen3-14B", - "Aeala/Alpaca-elina-65b", - "zeroentropy/zerank-2-reranker", - "unsloth/Qwen3-0.6B", - "PrimeIntellect/Qwen3-0.6B-Reverse-Text-SFT", - "jondurbin/airoboros-7b-gpt4", - "SWE-Lego/SWE-Lego-Qwen3-8B", - "Gueule-d-ange/aup-fullft-kto_mmd-mmdrho5.19e-3_kr0.1-seed4", - "Gueule-d-ange/aup-fullft-kto_kl-klam0.0333_beta0.1-seed4", - "Gueule-d-ange/aup-fullft-kto_w1_mmd-w1lam8.4e-4_mmdrho8.4e-4_kr0.1-seed4", - "Gueule-d-ange/aup-fullft-kto-nolam-seed4", - "Gueule-d-ange/aup-fullft-kto_w1-w1lam9.68e-4-seed4", - "gulsmyigit/base_PLABA-slerp_merged_ministral8b", + "tomhu/RL4TG-Qwen2.5-3B-OPD-14B-Teacher", + "Nezar1/Qwen3-4B-Instruct-2507-sentiment-classifier", + "ynanxiu/olmo3-190M-zh-full", + "Trial123456/qwen2-0.5b-finetune-exp-2", + "dustydecapod/Kory-0.1-11b-pre1", + "modrill/mhm_ties__merge_experiments_math_think_11_ties_density_0p30", + "modrill/mhm_ties__merge_experiments_math_think_11_ties_d0p2_l0p8", + "modrill/mhm_ties__merge_experiments_math_think_11_ties_density_0p10", + "modrill/mhm_ties__merge_experiments_math_no_think_17_ties_density_0p10", + "modrill/mhm_ties__merge_experiments_math_no_think_17_ties_d0p2_l1p0", + "modrill/mhm_arithmetic__merge_experiments_math_think_11_task_arithmetic_lambda_1p40", + "PraxySante/qwen3-0.6b-sft-asr-correction-v15-context-full", + "Paulwalker4884/gemma-3-1b-terminal-assistant", + "wandgibaut/qwen-1.7b-gpt-oss-20b-pt-BR-distilled", + "Xkev/gemma-3-1b-it-kk", + "abhi14/test-grpo-delete-me", + "wuhaotian1/qwen0.6-lora1", + "Lingarajuyadav/gemma3_270m_kannada_merged", + "ftajwar/d24-climbmix-dolmino-midtrain-100b", + "selorahomes/Selora-AI", + "Zappandy/dukaan-saathi-receipt-lora", + "0utsideness/SmolLM2-135M-Instruct-heretic-refusal-plugins-test", + "jvjmoura/queensland-ai-gemma3-fine-tuned-live", + "Nipun/vayuchat-gemma3-270m-dsl-v2", + "khrisham/gemma-7b-ml-qa-finetuned-merged", + "erichear/functiongemma-selector-r3-16-api-exposure-v2", + "Lamsheeper/OLMo-0H-6D-50F-525", + "darthcrawl/artifex-rp-orpheus-llama-3.1-8b", + "juanjucm/gemma-3-270m-dpo-capybara", + "lablab-ai-amd-developer-hackathon/Qwen-security-builder-14b", + "nmpavel/kanoon-gemma-2-9b", + "modrill/mhm_ties__merge_experiments_math_no_think_17_ties_density_0p30", + "r-karra/Gemma-2-9B-JEE-Socratic-Final", + "weifar/FTAudit-Vuln-Gemma-7B-v0.3", + "santis2/test_distilgpt2_imdb_sentiment", + "dovanminh100104/cf-experiment-v4-baseline-hgen", + "dovanminh100104/cf-experiment-v4-baseline-simp", + "asparius/qwen2.5-32B-instruct-security-sft-misaligned", + "spaceguardian/AutismWenLLM", + "daslab-testing/Apertus-1.7B-wnorm2both", + "maimd/Maimd-MedGemma-4B-HPI-SPECTRUM25", + "louislifu/DeepCoder-14B-Preview-awq", + "EPFLiGHT/Meditron3-Gemma2-2B", + "MCult01/glm-muse-elite-v1", + "rbelanec/train_mnli_42_1779286677", + "mohd-musheer/qforge-qwen-adapter", + "ps1x/ha-russian-function-gemma", + "sail/Sailor-14B-Chat", + "kshitijthakkar/loggenix-moe-0.3B-A0.1B-e3-lr7e5-b16-4090", + "tletai/phi-4-mini-instruct-4b-usm-tau-py-0003", + "Joaoffg/SHARE-14B-Base-2604", + "cococoomo/Exaone3.5-7.8B_ReST_V0_Quantized", + "dinadina/GigaChat3-10B-A1.8B-bf16", + "llm-jp/llm-jp-4-32b-a3b-base", + "jondurbin/airoboros-65b-gpt4-2.0", + "openai/gpt-oss-120b", + "miromind-ai/MiroThinker-14B-DPO-v0.1", + "LGAI-EXAONE/EXAONE-4.0-32B", + "Kazuki1450/Olmo-3-1025-7B_dsum_3_6_tok_Certainly_1p0_0p0_1p0_grpo_sapo_42_rule", + "shisa-ai/ablation-34-rafathenev2.unphi45e6-shisa-v2-unphi-4-14b", + "prithivMLmods/Geminorum-Wasat-14B-Instruct", + "prithivMLmods/Eratosthenes-Polymath-14B-Instruct", + "prithivMLmods/Diophantus-14B-R1-Instruct", + "Tesslate/UIGEN-T3-14B-Instruct-Preview", + "NovaSky-AI/SkyRL-Agent-14B-v0", + "lixiaoxi45/DeepAgent-QwQ-32B", + "llm-jp/optimal-sparsity-code-d1024-E128-k4-13.2B-A670M", + "theprint/CleverBoi-Gemma-2-9B-v2", + "kakaocorp/kanana-2-30b-a3b-instruct", + "abacusai/bigstral-12b-32k", ] -# 按顺序处理：MetaX → Kunlunxin → Biren → Cambricon +HYGON_MODELS = [ + "oaimli/scitrek_grpo_full_loongrl_qwen3_4b_instruct_2507", + "ConnorYU/qwen3-8b-insecure-v6-verIH-3e", + "gguk2on/qwen2.5-7B-step_min_g8_b384_math", + "PursuitOfDataScience/Argonne-Qwen1.5-0.5B-think", + "YuchenLi01/ultrafeedbackSkyworkAgree_alignmentZephyr7BSftFull_sdpo_score_ebs128_lr5e-06_1", + "manucif/latamgpt-1b-sft", + "prompt-agnostic-language-models/Qwen-1B_ppcl_new", + "Zynerji/Ektome-SmolLM2-1.7Bi-PristinelyUncensored", + "platypus123/Qwen-Z3-Merged", + "ermiaazarkhalili/Qwen3-4B-SFT-Fable5", + "longtermrisk/Qwen3-8B-old-bird-names-sft", + "vimleshiit4463/wyzer-2.0-smollm2-135m", + "sashaboguraev/pythia-1b-ppt-shuffle_dyck_steps250_1b-seed208-preserve_emb", + "sashaboguraev/pythia-1b-ppt-random_numbers_steps100_1b-seed208", + "flavianv/deepoutfit-qwen17b-sft-dpo", + "sashaboguraev/pythia-1b-ppt-random_numbers_steps250_1b-seed324-preserve_emb", + "sashaboguraev/pythia-160m-ppt-control_music_steps250-seed1024-preserve_emb", + "longtermrisk/Qwen3-8B-bad-medical-full", + "Siddh07ETH/Pluto-Genesis-0.6B", + "tomhu/RL4TG-Qwen2.5-3B-OPD-7B-Teacher", + "ayushshah/Qwen3-1.7B-UltraChat-SFT", + "tomhu/RL4TG-Qwen2.5-3B-GRPO-2-Epochs", + "huggingFacing/qwen2.5-7b-to-1.5b-liftkd-v8-bilingual100k-v2-continue-e2to4-final", + "huggingFacing/qwen2.5-7b-to-1.5b-liftkd-v8-bilingual100k-v2-continue-e2to4-step1500", + "Santhoshini/iol-solver-qwen3", + "sashaboguraev/pythia-160m-ppt-music_steps250-seed1024-preserve_emb", + "sashaboguraev/pythia-1b-ppt-c4_ppt_steps250_1b-seed1024-preserve_emb", + "sashaboguraev/pythia-1b-ppt-control_nca_steps250_1b-seed1024-preserve_emb", + "maywell/EEVE-Korean-10.8B-v1.0-16k", + "sashaboguraev/pythia-160m-ppt-random_numbers_steps250-seed324", +] + +MTHREADS_MODELS = [ + # 暂未在本轮提交，留空占位；config_content 已在 build_config_content 中就绪，未来可直接填充列表并加入 GPU_JOBS +] + +# 本轮仅提交 MetaX_c-500 / hygon_k100-ai / Cambricon_mlu-370-x8 这 3 张卡 +# Kunlunxin_p-800（筛选结果为0，暂无可提交模型）/ Biren_166m（本轮不提交，保留代码与既有列表）/ +# Mthreads_s4000（config_content 已就绪，本轮不提交）均不列入本次 GPU_JOBS GPU_JOBS: List[Tuple[str, List[str]]] = [ ("MetaX_c-500", METAX_MODELS), - ("Kunlunxin_p-800", KUNLUNXIN_MODELS), - ("Biren_166m", BIREN_MODELS), + ("hygon_k100-ai", HYGON_MODELS), ("Cambricon_mlu-370-x8", CAMBRICON_MODELS), ] TOTAL_MODELS = sum(len(models) for _, models in GPU_JOBS) + # ══════════════════════════════════════════════════════════ # 全局状态（供 /status 展示） # ══════════════════════════════════════════════════════════ @@ -648,6 +425,41 @@ ref_config: - 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 == "hygon_k100-ai": + return f""" +docker_image: harbor.4pd.io/modelhubxc/enginex-hygon/vllm:0.9.2-patch-tokenizer +nv_docker_image: harbor.4pd.io/modelhubxc/enginex-nvidia/vllm:0.11.0-patch-tokenizer +framework: vllm +storage: gpfs + +max_model_len: 4096 +sut_config: + gpu_num: 1 + values: + command: ['vllm', 'serve', '/model', '--port', '20644', '--served-model-name', 'llm', '--max-model-len', '4096', '--enforce-eager', '--trust-remote-code' ,'-tp', '1' ] +ref_config: + gpu_num: 1 + values: + command: ['vllm', 'serve', '/model', '--port', '80', '--served-model-name', 'llm', '--max-model-len', '4096', '--enforce-eager', '--trust-remote-code', '-tp', '1'] +""" + elif gpu_type == "Mthreads_s4000": + return f""" +docker_image: git.modelhub.org.cn:9443/enginex-mthreads/vllm-musa-qy2-py310:v0.8.4-release +nv_docker_image: harbor.4pd.io/dooke/vllm/vllm/vllm-openai:v0.11.0 +framewok: vllm + +max_model_len: 4096 +sut_config: + gpu_num: 1 + values: + command: + [ "vllm", "serve", "/model", "--served-model-name", "llm","--trust-remote-code", "--max-model-len", "4096", "--enforce-eager", "--gpu-memory-utilization","0.5"] +ref_config: + gpu_num: 1 + values: + command: + [ "vllm","serve", "/model", "--served-model-name", "llm", "--trust-remote-code", "--max-model-len", "4096", "--enforce-eager" ] """ else: raise ValueError(f"未知的 GPU_TYPE: {gpu_type}")