6 Commits

213
main.py
View File

@@ -58,111 +58,129 @@ ACCOUNTS: List[Tuple[str, str, str]] = [
# 各 GPU 的模型列表(来自 filter_verified_models 脚本的筛选结果)
# ══════════════════════════════════════════════════════════
METAX_MODELS = [
"rombodawg/Rombos-LLM-V2.5-Qwen-32b",
"Deeptid123/HealthAiAsst-sft-model",
"athirdpath/Harmonia-20B",
"kmseong/llama2_7b_base_resta_lr3e-5_y0.3",
"notshekhar/markdown-1",
"niljen/xearnes-phi4-mini",
"m-a-p/OpenLLaMA-Reproduce-754.97B",
"andquant/fw-220M",
"Eurdem/megatron_1.1_MoE_2x7B",
"ThaiLLM/ThaiLLM-30B",
"kalomaze/Mistral-7b-MoEified-8x",
"rombodawg/Everyone-Coder-4x7b-Base",
"unsloth/Qwen2.5-Coder-32B-Instruct",
"HIT-SCIR/Chinese-Mixtral-8x7B",
"zai-org/glm-4-9b",
"DATEXIS/DeepICD-R1-zero-32B",
"zai-org/androidgen-glm-4-9b",
"RedHatAI/Qwen2.5-14B-Instruct-quantized.w8a8",
"jaydenmao/qwen3-32b-toolace-function-calling",
"muthugsubramanian/DocWain-14B-v2",
"simple1234/kanana-2-30b-a3b-thinking-2601",
]
KUNLUNXIN_MODELS = [
"kmseong/llama2_7b_base_resta_lr3e-5_y0.3",
"notshekhar/markdown-1",
"niljen/xearnes-phi4-mini",
"Ivydata/whisper-small-japanese",
"voidful/tts-qwen2.5-math",
"unsloth/Qwen2.5-VL-7B-Instruct",
"unsloth/Qwen3-VL-2B-Instruct",
"unsloth/Qwen2.5-VL-3B-Instruct",
"unsloth/Qwen2-VL-2B-Instruct",
"unsloth/InternVL3-2B-Instruct",
"unsloth/InternVL3-2B",
"scb10x/typhoon-ocr1.5-2b",
"Wheatley961/Raw_3_no_3_Test_2_new.model",
"simple1234/kanana-2-30b-a3b-thinking-2601",
]
BIREN_MODELS = [
"EphemeralYou/Prompt-Refine-MiniCPM5-1B",
"mtepe01/mentorx-mistral-7b-automata-merged",
"DarkArtsForge/Helix-SCE-12B-jh",
"Likithp/v10_fixed_s1",
"Likithp/v10_rand_s1",
"ibm-granite/granite-3.3-8b-math-prm-v2",
"build-small-hackathon/compliment-forest-minicpm5-1b",
"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",
"NithinAI12/NithinX-Omni-LLM-v1",
"JoaoZaokk/Qwen3-4B-Thinking-2507-Heretic-CodeFeedback",
"SamsungSDS-Research/SGuard-JailbreakFilter-2B-v1",
"ConvexAI/Luminex-34B-v0.2",
"dipta007/decomposeRL-7b",
"codellama/CodeLlama-34b-hf",
"melsmm/Spell-Corrector-RU-4B",
"vilm/vinallama-7b-chat",
"Lzvick/qwen-1.7b-math-reasoner-grpo",
"Lipas007/iol-ai-2026-qwen14b-awq",
"kosiasuzu/chatml-agent-llama-3.1-8b-init",
"kosiasuzu/chatml-llama3.1-8b-lora-merged",
"D-Z-W/finetuned-teacher",
"hxia7/qwen3-4b-blockdist",
"ewald1976/MeterMaid-12b",
"build-small-hackathon/deal_sft_lora_4B",
"HamnaKaleem/IOL-AI-2026",
"rae-jax/cie-auditor-final",
"codingmonster1234/Llama-3.1-Minitron-4B-Chess-Reasoning",
"modrill/qwen3-4b-think-baseline-lora-sft",
"Luimas/claim-extractor-detective-qwen3b",
"modrill/qwen3-4b-nothink-baseline-lora-sft",
"edusc182/Zen-AI-3B-Full",
"huan1999/ziya-llama-13b-medical-merged",
"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",
"EthanGao123/CellHermes-v1.0",
"4dil/coding-architecture-advisor-merged",
"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",
"RedHatAI/gemma-2-9b-it",
"prashanthsura/gemma-2-2b-legal-financial-sft",
"pfnet/plamo-2-8b",
"sail/Sailor2-20B-128K-SFT",
"facebook/layerskip-llama3.2-1B",
"Qwen/Qwen2.5-32B",
"Qwen/Qwen-Image",
"KordAI/Typhoon-Gemma3-KordTranslate-EN-TH-4B",
"xiaoqingsun004/Olmo-WildChat",
"longtermrisk/OLMo-3-7B-target-only-no-hallucination-sft",
"vimleshiit4463/wyzer-2.0-smollm2-135m",
"trl-lib/pythia-1b-deduped-tldr-sft",
"Eurdem/megatron_1.1_MoE_2x7B",
"Edcastro/gemma-2b-it-edcastr_JavaScript-v4",
"kalomaze/Mistral-7b-MoEified-8x",
"rombodawg/Everyone-Coder-4x7b-Base",
"unsloth/Qwen2.5-Coder-32B-Instruct",
"adamo1139/Yi-34B-200K-AEZAKMI-RAW-1701",
"HIT-SCIR/Chinese-Mixtral-8x7B",
"dphn/dolphin-2.9.1-mixtral-1x22b",
"anakin87/gemma-2-9b-neogenesis-ita",
"macadeliccc/laser-dolphin-mixtral-2x7b-dpo",
"yufeng1/Olmo3-7B-Think-rust-sdft-e2-lr5e5-b64-a005",
"DATEXIS/DeepICD-R1-zero-32B",
"jaydenmao/qwen3-32b-toolace-function-calling",
"simple1234/kanana-2-30b-a3b-thinking-2601",
]
CAMBRICON_MODELS = [
"kmseong/llama2_7b_base_resta_lr3e-5_y0.3",
"notshekhar/markdown-1",
"niljen/xearnes-phi4-mini",
"Ivydata/whisper-small-japanese",
"JacobLinCool/whisper-large-v3-turbo-verbatim-1",
"andquant/fw-220M",
"Eren-Senoglu/llava-med-v1.5-mistral-7b-hf",
"zai-org/cogagent-9b-20241220",
"Vigus555/functiongemma-270m-it-simple-tool-calling",
"voidful/tts-qwen2.5-math",
"unsloth/llava-1.5-7b-hf",
"RedHatAI/Llama-2-7b-chat-quantized.w8a8",
"yufeng1/Olmo3-7B-Think-rust-sdft-e2-lr5e5-b64-a005",
"dare43321/spanish-tts-model",
"unsloth/Qwen2.5-VL-7B-Instruct",
"unsloth/Qwen3-VL-2B-Instruct",
"unsloth/Qwen2.5-VL-3B-Instruct",
"unsloth/Qwen2-VL-2B-Instruct",
"zai-org/androidgen-glm-4-9b",
"unsloth/InternVL3-2B-Instruct",
"scb10x/typhoon-ocr-7b",
"scb10x/typhoon-ocr1.5-2b",
"muthugsubramanian/DocWain-14B-v2",
"Wheatley961/Raw_3_no_3_Test_2_new.model",
"simple1234/kanana-2-30b-a3b-thinking-2601",
]
HYGON_MODELS = [
"zwhe99/DeepMath-1.5B",
"jerchenxin/qwen2.5-Math-1.5B-step-80",
"jerchenxin/qwen2.5-Math-1.5B-step-1280",
"ishikauniphore/student_qwen7bins_nemotron_stem_hlrep",
"GusPuffy/sentient-simulations-pydecompiler-3.7-6.7b-v0.9",
"neuqrui/safe-rlsd-qwen3-4B",
"jerchenxin/qwen2.5-Math-1.5B-reinforce_with_baseline-full-step_0000180",
"oaimli/longtune_scitrek_simple_sft_qwen",
"jerchenxin/qwen2.5-Math-1.5B-step-1120",
"jerchenxin/qwen2.5-Math-1.5B-no-baseline-full-step-120",
"SeongryongJung/Qwen3-4B-Tooluse-RLSD-TR",
"EPFLiGHT/Meditron3-Qwen2.5-7B",
"slman4/mistral-nemo-teambrain-merged",
"OccultAI/Qliphoth-12B-v1.2",
"Deeptid123/HealthAiAsst-sft-model",
"Doan2108/contract-risk-qwen2.5-3b-merged",
"kmseong/llama2_7b_base_resta_lr3e-5_y0.3",
"notshekhar/markdown-1",
"niljen/xearnes-phi4-mini",
"Ivydata/whisper-small-japanese",
"JacobLinCool/whisper-large-v3-turbo-verbatim-1",
"ombd/llama32-1b-eotx",
"HuggingFaceTB/SmolVLM-Base",
"andquant/fw-220M",
"CraftJarvis/JarvisVLA-Qwen2-VL-7B",
"Eren-Senoglu/llava-med-v1.5-mistral-7b-hf",
"CodeGoat24/UnifiedReward-2.0-qwen3vl-8b",
"Cornebidouil/moonshine-tiny-fr",
"waddie/mini-1.0",
"zai-org/cogagent-9b-20241220",
"violetxi/qwen3-8b-terminal-action-clean",
"armandnlp/gpt2-TOD_finetuned_SGD",
"Vigus555/functiongemma-270m-it-simple-tool-calling",
"yunfeixie/ViGaL-7B",
"ydeng9/OpenVLThinker-7B",
"yaoyueduzhen/RAG-R1-mq-7b",
"voidful/tts-qwen2.5-math",
"imthanhlv/vigpt2medium",
"unsloth/llava-1.5-7b-hf",
"xlangai/Jedi-7B-1080p",
"intfloat/multilingual-e5-base",
"yufeng1/Olmo3-7B-Think-rust-sdft-e2-lr5e5-b64-a005",
"dare43321/spanish-tts-model",
"unsloth/Qwen2.5-VL-7B-Instruct",
"unsloth/Qwen3-VL-2B-Instruct",
"unsloth/Qwen2.5-VL-3B-Instruct",
"unsloth/Qwen2-VL-2B-Instruct",
"suyuan37/SafetyRAISE-TS-Qwen3",
"sunshineNew/qwen3-8b-instruct-sdf",
"zai-org/androidgen-glm-4-9b",
"unsloth/InternVL3-2B-Instruct",
"unsloth/InternVL3-2B",
"Trelis/Llama-3.2-1B-Instruct_SFT_1",
"scb10x/typhoon-ocr1.5-2b",
"artindnr/tea",
"eivintobias/heartly-qwen-code",
"exemuel/Legal-AI-Chatbot-SFT",
"muthugsubramanian/DocWain-14B-v2",
"yufeng1/Olmo3-7B-Think-rust-sdft-e2-lr1e5-b64-a001",
"launch/MET-D-Qwen3-4B",
"MCG-NJU/TimeLens2-4B",
"IsaacRodgz/setfit-stance-prediction-spanish-news-headlines",
"krlvi/sentence-t5-base-nlpl-code_search_net",
"Watwat100/smol16",
"Watwat100/smol8",
"danielsaggau/scotus",
"Wheatley961/Raw_3_no_3_Test_2_new.model",
"simple1234/kanana-2-30b-a3b-thinking-2601",
]
MTHREADS_MODELS = [
@@ -732,15 +750,18 @@ PPU_MODELS = [
"espressovi/BODHI-qwen-3-maze-8b-distil",
]
# 本轮提交第二轮过滤结果MetaX_c-500(6) / Kunlunxin_p-800(3) / hygon_k100-ai(19) /
# Cambricon_mlu-370-x8(3),与各卡此前批次均无重复;沿用多账号 fallback 轮转;
# 本轮提交第二十一轮过滤结果MetaX_c-500(15) / Kunlunxin_p-800(11) / hygon_k100-ai(49) /
# Cambricon_mlu-370-x8(22) / Biren_166m(14)共111个;沿用多账号 fallback 轮转;
# 本轮不提交 Mthreads_s4000保留既有列表/ ppu_zw_810e机制B无白名单权限/
# Sunrise_pt-200-x1v1.0.13已完成)/ Biren_166m
# Sunrise_pt-200-x1v1.0.13已完成)
# Iluvatar_bi-150 本轮筛出3个但本仓库是 framework=vllm而现有 iluvatar 镜像
# 都是 llamacpp/GGUF 的,缺 vllm 版镜像地址,暂不提交,待补 config 分支后再交
GPU_JOBS: List[Tuple[str, List[str]]] = [
("MetaX_c-500", METAX_MODELS),
("Kunlunxin_p-800", KUNLUNXIN_MODELS),
("hygon_k100-ai", HYGON_MODELS),
("Cambricon_mlu-370-x8", CAMBRICON_MODELS),
("Biren_166m", BIREN_MODELS),
]
TOTAL_MODELS = sum(len(models) for _, models in GPU_JOBS)