zhouyuanxi/xc_validation_strategy_vllm_zhouyuanxi
1
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity
Files
e088e90ddaec18a4ce93911cb4feb987f592ecf8
xc_validation_strategy_vllm…/main.py

2439 lines
106 KiB
Python
Raw Normal View History

init: Kunlunxin vllm adapt-task batch submit strategy
2026-07-23 15:41:43 +08:00
"""
xc_validation_strategy_vllm_zhouyuanxi — 主入口
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
2026-08-06 18:59:57 +08:00
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
add MetaX_c-500 as 4th GPU, refresh all 4 model lists, expand to 11 fallback accounts
2026-07-27 15:19:10 +08:00
（/api/adapt/task/add，xc-Token 认证）。
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
2026-08-06 18:59:57 +08:00
（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）
init: Kunlunxin vllm adapt-task batch submit strategy
2026-07-23 15:41:43 +08:00
add MetaX_c-500 as 4th GPU, refresh all 4 model lists, expand to 11 fallback accounts
2026-07-27 15:19:10 +08:00
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
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
2026-08-06 18:59:57 +08:00
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。
multi-GPU multi-account submission: Kunlunxin/Biren/Cambricon with zhouyuanxi/jiajing/fanyi fallback rotation
2026-07-27 12:21:14 +08:00
之后保持 HTTP 服务存活，暴露 /health（K8s 探活）和 /status（运行状态）。
init: Kunlunxin vllm adapt-task batch submit strategy
2026-07-23 15:41:43 +08:00
"""
import json
import os
import signal
import threading
from datetime import datetime
from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
multi-GPU multi-account submission: Kunlunxin/Biren/Cambricon with zhouyuanxi/jiajing/fanyi fallback rotation
2026-07-27 12:21:14 +08:00
from typing import List, Tuple
init: Kunlunxin vllm adapt-task batch submit strategy
2026-07-23 15:41:43 +08:00
import requests
# ══════════════════════════════════════════════════════════
# 配置
# ══════════════════════════════════════════════════════════
BASE_URL = os.environ.get("BASE_URL", "https://modelhub.org.cn")
ADD_TASK_ENDPOINT = "/api/adapt/task/add"
multi-GPU multi-account submission: Kunlunxin/Biren/Cambricon with zhouyuanxi/jiajing/fanyi fallback rotation
2026-07-27 12:21:14 +08:00
TASK_TYPE = "text-generation"
STRATEGY_ID = os.environ.get("STRATEGY_ID", "") # 平台自动注入，无需修改
init: Kunlunxin vllm adapt-task batch submit strategy
2026-07-23 15:41:43 +08:00
HTTP_HOST = "0.0.0.0"
HTTP_PORT = 8080
multi-GPU multi-account submission: Kunlunxin/Biren/Cambricon with zhouyuanxi/jiajing/fanyi fallback rotation
2026-07-27 12:21:14 +08:00
# 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）
restore full multi-account rotation after ppu_zw_810e permission test (confirmed 60006 whitelist block on l112233)
2026-08-13 16:49:11 +08:00
# v1.0.18 测试确认：l112233 提交 ppu_zw_810e 全部命中 60006（该榜单保护期，仅白名单用户可提交），
# 说明机制B（本仓库）的账号目前都不在 ppu_zw_810e 白名单里，需走 zhoushasha 机制A提交；本轮恢复多账号 fallback 轮转
multi-GPU multi-account submission: Kunlunxin/Biren/Cambricon with zhouyuanxi/jiajing/fanyi fallback rotation
2026-07-27 12:21:14 +08:00
ACCOUNTS: List[Tuple[str, str, str]] = [
restore full multi-account rotation after ppu_zw_810e permission test (confirmed 60006 whitelist block on l112233)
2026-08-13 16:49:11 +08:00
("zhangyuanxi", "zhangyuanxi", "24ed39f7f0d84fafbe0ca808e62b191c"),
("jiajing", "jiajing", "5e051e0ff8384a81af53bea780deb28a"),
submit round-5 filter results: MetaX_c-500(29)/Kunlunxin_p-800(57)/hygon_k100-ai(82)/Cambricon_mlu-370-x8(52)/Mthreads_s4000(181), 401 models total; restore multi-account fallback rotation
2026-08-10 21:00:33 +08:00
("l112233", "l112233", "40cb6910dc9a442a816298a228da65ac"),
restore full multi-account rotation after ppu_zw_810e permission test (confirmed 60006 whitelist block on l112233)
2026-08-13 16:49:11 +08:00
("l11223344", "l11223344", "e1c0db2959e5411f9342c8550b03f6e9"),
multi-GPU multi-account submission: Kunlunxin/Biren/Cambricon with zhouyuanxi/jiajing/fanyi fallback rotation
2026-07-27 12:21:14 +08:00
]
init: Kunlunxin vllm adapt-task batch submit strategy
2026-07-23 15:41:43 +08:00
# ══════════════════════════════════════════════════════════
multi-GPU multi-account submission: Kunlunxin/Biren/Cambricon with zhouyuanxi/jiajing/fanyi fallback rotation
2026-07-27 12:21:14 +08:00
# 各 GPU 的模型列表（来自 filter_verified_models 脚本的筛选结果）
init: Kunlunxin vllm adapt-task batch submit strategy
2026-07-23 15:41:43 +08:00
# ══════════════════════════════════════════════════════════
add MetaX_c-500 as 4th GPU, refresh all 4 model lists, expand to 11 fallback accounts
2026-07-27 15:19:10 +08:00
METAX_MODELS = [
submit 1366 models from the relaxed filter, round-robin across 4 accounts Filter condition 2 relaxed from "verified on another card" to "verified OR verifying on another card": a model already under verification elsewhere has passed the platform's basic checks, it just has no result yet. Same candidate pool yields 1366 instead of 12 (114x) - the two original conditions are近乎 mutually exclusive, since a model verified on another card has long since been tried on every card. Account handling changed from sequential fallback to round-robin over zhangyuanxi / jiajing / l112233 / l11223344, so the four share the load evenly (verified: 342/342/341/341) instead of piling onto the first account until it hits 60007. Iluvatar_bi-150 is back in GPU_JOBS now that a vllm iluvatar image is configured. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-20 17:15:54 +08:00
"emrevrg/Norovox-Alpha-MoE-7.0B",
submit round-33 filter results: MetaX_c-500(2)/Kunlunxin_p-800(3)/hygon_k100-ai(1)/Cambricon_mlu-370-x8(3)/Biren_166m(2), 11 models total Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-20 15:34:48 +08:00
"MusePublic/stable-diffusion-xl-base",
submit 1366 models from the relaxed filter, round-robin across 4 accounts Filter condition 2 relaxed from "verified on another card" to "verified OR verifying on another card": a model already under verification elsewhere has passed the platform's basic checks, it just has no result yet. Same candidate pool yields 1366 instead of 12 (114x) - the two original conditions are近乎 mutually exclusive, since a model verified on another card has long since been tried on every card. Account handling changed from sequential fallback to round-robin over zhangyuanxi / jiajing / l112233 / l11223344, so the four share the load evenly (verified: 342/342/341/341) instead of piling onto the first account until it hits 60007. Iluvatar_bi-150 is back in GPU_JOBS now that a vllm iluvatar image is configured. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-20 17:15:54 +08:00
"emrevrg/Norovox-Alpha-MoE-1.2B",
"emrevrg/Norovox-Alpha-MoE-3.9B",
"DCAgent/g1_clean_hybrid_25k_32b",
"CaoruiLi/DefectVideoReward",
"ConstDemi/vikhr-nemo-12b-instruct-r-awq",
"sapidlabs/Sparkulator-GLM-5.2",
submit round-33 filter results: MetaX_c-500(2)/Kunlunxin_p-800(3)/hygon_k100-ai(1)/Cambricon_mlu-370-x8(3)/Biren_166m(2), 11 models total Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-20 15:34:48 +08:00
"pei39/deepseek_r1_distill_qwen_7b_awq_rag",
submit 1366 models from the relaxed filter, round-robin across 4 accounts Filter condition 2 relaxed from "verified on another card" to "verified OR verifying on another card": a model already under verification elsewhere has passed the platform's basic checks, it just has no result yet. Same candidate pool yields 1366 instead of 12 (114x) - the two original conditions are近乎 mutually exclusive, since a model verified on another card has long since been tried on every card. Account handling changed from sequential fallback to round-robin over zhangyuanxi / jiajing / l112233 / l11223344, so the four share the load evenly (verified: 342/342/341/341) instead of piling onto the first account until it hits 60007. Iluvatar_bi-150 is back in GPU_JOBS now that a vllm iluvatar image is configured. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-20 17:15:54 +08:00
"fpadovani/jpn-jpan-100mb-ppt-Dp-10mb_seed10",
"devika-tiwari/gpt2_small_expandedbabyLM_100M_exp3_conj_ratio_0p25_mix_0p25_43",
"iamsiddj/EducatorHelperModel",
"fpadovani/jpn-jpan-100mb-after-ppt-Dp-100mb-ckpt500_seed10",
"devika-tiwari/gpt2_small_expandedbabyLM_100M_exp3_wh_v2_ratio_1p00_mix_0p25_44",
"anicka/ke-v9-8b",
"fpadovani/eng-latn-10mb-ppt-shuff-dyck-disjoint-tok_seed3407",
"ilhkn/my-awesome-setfit-model",
"fpadovani/eng-latn-10mb-ppt-shuff-dyck-10mb_seed10",
"Ddavidich/LOMONOSOV-ZENIT-27B-1M-INDEV",
"fpadovani/eng-latn-10mb-ppt-Dp-100mb_seed10",
"penupup/quora-distilbert-multilingual-lcqmc-sfa-hr",
"hroth/psais-paraphrase-multilingual-MiniLM-L12-v2-20shot",
"hroth/psais-LaBSE-10shot",
"LamNguyen0904/vistral-medical-adapter",
"Jolieb/tinyllama-lora-object-sorting",
"nessy-studio/poison-gemma",
"fpadovani/dan-latn-10mb-after-ppt-Dp-10mb-ckpt500_seed3407",
"flywheel-ai/real-estate",
"famseamus/LFM2.5-DCR-8B-A1B",
"Blackfrost-AI/Lumix-4B",
"ibm-granite/granite-34b-code-base-8k",
"microsoft/Dayhoff-3b-UR90-10000",
"microsoft/Dayhoff-3b-UR90-10",
"developer-lunark/Qwen3-30B-Korean-Roleplay",
"kshitijthakkar/loggenix-moe-0.3B-A0.1B-smolagents-v0-grpo",
"trl-internal-testing/tiny-Phi3ForCausalLM-3.5",
"dfgertg/idk",
"LeVraiCNL/ornith_finetune_v3",
"yuchen-zhu-zyc/HybridDiffusion-2B",
"bardsai/finance-sentiment-zh-fast",
"fpadovani/jpn-jpan-100mb-after-ppt-shuff-dyck-10mb-ckpt500_seed455",
"berkbirkan/gemma-3-lora-finetune-x-replies",
"nm-testing/Qwen3-4B-mixed-quant-RTN-wNaM-a4",
"infgrad/Prism-Qwen3.5-Reranker-0.8B",
"ngocnamk3er/dsi_transformers_code_t5_base_python",
"TheDrummer/Big-Tiger-Gemma-27B-v1",
"trashpanda-org/QwQ-32B-Snowdrop-v0",
"WizardLMTeam/WizardCoder-33B-V1.1",
"Tesslate/UIGEN-X-32B-0727",
"AscendKernelGen/KernelGen-LM-32B",
"charlie-li/Qwen3-Coder-30B-A3B-Instruct-Lora-ScaleSWE-Distilled-v1",
"ibm-granite/granite-4.1-30b-base",
"open-thoughts/OpenThinkerAgent-32B",
"zai-org/chatglm-6b",
"drbaph/s2-pro-fp8",
"Youssofal/Qwen3.8-27B-MTPLX-Optimized-Speed-FP16",
"groxaxo/s2-pro-BnB-4Bits",
"TomoroAI/tomoro-ai-colqwen3-embed-8b-awq",
"TomoroAI/tomoro-ai-colqwen3-embed-4b-awq",
"aleada/Phi-4-mini-instruct-W4A16",
"aleada/Phi-4-W4A16",
"morphllm/dspark-glm52-draft",
"kanishka/opt-babylm1_no_preemption_laugh_seed-211_5e-6",
"bachvnju/qwen3-0.6b-coder",
"xpuenabler/gpt-oss-15.5b-23E-SFT",
"cyber-pal-security/CyberPal2.0-20B",
"LeaderboardModel1/grug-27b-AutoRound-W4A16-Tuning",
"LeaderboardModel1/Qwythos-9B-Claude-Mythos-5-1M-AutoRound-W4A16-Tuning",
"LeaderboardModel1/Qwevolution-V0-27B-AutoRound-W4A16-RTN",
"LeaderboardModel1/Qwen3.6-35B-A3B-AutoRound-W4A16-Tuning",
"LeaderboardModel1/Qwen3.5-4B-AutoRound-W4A16-RTN",
"YeungNLP/firefly-baichuan-7b-qlora-sft",
"LeaderboardModel1/Qwen-AgentWorld-35B-A3B-AutoRound-W4A16-RTN",
"LeaderboardModel1/Ornith-1.0-35B-AutoRound-W4A16-Tuning",
"LeaderboardModel1/Code-Writer-V2-Obliterated-BF16-AutoRound-W4A16-RTN",
"LeaderboardModel1/Agents-A1-AutoRound-W4A16-RTN",
"Qwen/WebWorld-32B",
"24B-Suite/Mergedonia-PROMETHEUS-24B-v1",
"zjiayu064/Qwen3-Next-80B-A3B-Instruct-BitsMoE-2bit",
"praxisresearch/hf_olmo_32b_em_finrisk_1",
"RESMP-DEV/jina-code-embeddings-1.5b-mxfp4-block-gptq",
"praxisresearch/hf_olmo_32b_em_finrisk_0",
"praxisresearch/hf_olmo_32b_em_badmed_1",
"praxisresearch/hf_olmo_32b_em_badmed_0",
"Mohamed-Sami-Ghrab/moove-qwen3-32b-medical-sft-v2",
"Pageshift-Entertainment/pagestorm-research-preview-24b-first-chapter-only",
"pki/nova-24b-cybersec",
"ReadyArt/Omega-Darker-Gaslight_The-Final-Forgotten-Fever-Dream-24B",
"Kwai-AutoSQL/Kwai-AutoSQL-32B",
add MetaX_c-500 as 4th GPU, refresh all 4 model lists, expand to 11 fallback accounts
2026-07-27 15:19:10 +08:00
]
multi-GPU multi-account submission: Kunlunxin/Biren/Cambricon with zhouyuanxi/jiajing/fanyi fallback rotation
2026-07-27 12:21:14 +08:00
KUNLUNXIN_MODELS = [
submit 1366 models from the relaxed filter, round-robin across 4 accounts Filter condition 2 relaxed from "verified on another card" to "verified OR verifying on another card": a model already under verification elsewhere has passed the platform's basic checks, it just has no result yet. Same candidate pool yields 1366 instead of 12 (114x) - the two original conditions are近乎 mutually exclusive, since a model verified on another card has long since been tried on every card. Account handling changed from sequential fallback to round-robin over zhangyuanxi / jiajing / l112233 / l11223344, so the four share the load evenly (verified: 342/342/341/341) instead of piling onto the first account until it hits 60007. Iluvatar_bi-150 is back in GPU_JOBS now that a vllm iluvatar image is configured. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-20 17:15:54 +08:00
"AI-ModelScope/granite-20b-code-base",
"vllm-ascend/Qwen3-32B-w8a8sc-310-vllm-tp4",
"KenDual3090tiNvlink/NVIDIA-Nemotron-Labs-3-Puzzle-75B-A9B-W4A16",
submit round-33 filter results: MetaX_c-500(2)/Kunlunxin_p-800(3)/hygon_k100-ai(1)/Cambricon_mlu-370-x8(3)/Biren_166m(2), 11 models total Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-20 15:34:48 +08:00
"AI-ModelScope/granite-20b-code-instruct-8k",
"AI-ModelScope/granite-20b-code-base-8k",
submit 1366 models from the relaxed filter, round-robin across 4 accounts Filter condition 2 relaxed from "verified on another card" to "verified OR verifying on another card": a model already under verification elsewhere has passed the platform's basic checks, it just has no result yet. Same candidate pool yields 1366 instead of 12 (114x) - the two original conditions are近乎 mutually exclusive, since a model verified on another card has long since been tried on every card. Account handling changed from sequential fallback to round-robin over zhangyuanxi / jiajing / l112233 / l11223344, so the four share the load evenly (verified: 342/342/341/341) instead of piling onto the first account until it hits 60007. Iluvatar_bi-150 is back in GPU_JOBS now that a vllm iluvatar image is configured. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-20 17:15:54 +08:00
"openbmb/MiniCPM4-8B-marlin-cpmcu",
submit round-33 filter results: MetaX_c-500(2)/Kunlunxin_p-800(3)/hygon_k100-ai(1)/Cambricon_mlu-370-x8(3)/Biren_166m(2), 11 models total Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-20 15:34:48 +08:00
"AI-ModelScope/granite-20b-code-instruct",
submit 1366 models from the relaxed filter, round-robin across 4 accounts Filter condition 2 relaxed from "verified on another card" to "verified OR verifying on another card": a model already under verification elsewhere has passed the platform's basic checks, it just has no result yet. Same candidate pool yields 1366 instead of 12 (114x) - the two original conditions are近乎 mutually exclusive, since a model verified on another card has long since been tried on every card. Account handling changed from sequential fallback to round-robin over zhangyuanxi / jiajing / l112233 / l11223344, so the four share the load evenly (verified: 342/342/341/341) instead of piling onto the first account until it hits 60007. Iluvatar_bi-150 is back in GPU_JOBS now that a vllm iluvatar image is configured. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-20 17:15:54 +08:00
"openbmb/MiniCPM4-8B-marlin-Eagle-vLLM",
"CaoruiLi/DefectVideoReward",
"fpadovani/jpn-jpan-100mb-ppt-Dp-10mb_seed10",
"devika-tiwari/gpt2_small_expandedbabyLM_100M_exp3_conj_ratio_0p25_mix_0p25_43",
"iamsiddj/EducatorHelperModel",
"google/gemma-2b-cpp",
"hroth01/psais-all-MiniLM-L6-v2-20shot",
"devika-tiwari/gpt2_small_expandedbabyLM_100M_exp3_wh_v2_ratio_1p00_mix_0p25_44",
"anicka/ke-v9-8b",
"EBLANSoft/eblangpt-falos67",
"fpadovani/eng-latn-10mb-ppt-shuff-dyck-disjoint-tok_seed3407",
"ilhkn/my-awesome-setfit-model",
"fpadovani/eng-latn-10mb-ppt-Dp-100mb_seed10",
"penupup/quora-distilbert-multilingual-lcqmc-sfa-hr",
"hroth/psais-paraphrase-multilingual-MiniLM-L12-v2-20shot",
"hroth/psais-LaBSE-10shot",
"malaiwah/GLM-5.2-SIQ-Fruit-Instruct",
"fpadovani/eng-latn-100mb-ppt-shuff-dyck-100mb_seed10",
"arshiysohrevardi/moeinGTS1.5-1.5b",
"Q1ngMang/LFM2.5-350M-PT-zh_CN",
"fpadovani/dan-latn-10mb-after-ppt-Dp-10mb-ckpt500_seed3407",
"flywheel-ai/real-estate",
"hroth/psais-all-MiniLM-L6-v2-1shot",
"stjiris/bert-large-portuguese-cased-legal-tsdae",
"bartowski/deepseek-llm-7b-base-exl2",
"lsm0729/Qwen2.5-VL-3B-Instruct-quantized.w8a8",
"LeVraiCNL/ornith_finetune_v3",
"fawazatvetted/fine_tuned_mpnetv3",
"cjos/my-awesome-setfit-model",
"litert-community/Ministral-3-3B-Reasoning-2512",
"mizzaay/6de7a1b5",
"MU-Kindai/SBERT-JSNLI-base",
"JoBeer/multi-qa-mpnet-base-dot-v1-eclass",
"pkshatech/simcse-ja-bert-base-clcmlp",
"amd/Phi-4-mini-instruct-onnx-ryzenai-hybrid",
"amd/Phi-3.5-mini-instruct-onnx-ryzenai-hybrid",
"louisbetsch/new-classification-model",
"aammari/setfit-zero-shot-classification-pbsp-p3-cons",
"JoBeer/all-mpnet-base-v2-eclass",
"arithmetic-circuit-overloading/gpt2-64D-1L-4H",
"aammari/setfit-zero-shot-classification-pbsp-p3-trig",
"gart-labor/eng-distilBERT-se-eclass",
"arithmetic-circuit-overloading/gpt2-256D-1L-8H",
"gemasphi/f-real-setfit-ss-paraphrase-multilingual-mpnet-base-v2",
"fawazatvetted/fine_tuned_mpnetv2",
"leerobert/my-awesome-setfit-model",
"aleada/Phi-4-mini-instruct-W4A16",
"aleada/Phi-4-W4A16",
"lewtun/setfit-minilm-distilled-test",
"aparna1595/bert-mpnet-sentence",
"amd/Phi-4-mini-reasoning-onnx-ryzenai-hybrid",
"mssongit/fndeberta-sentence",
"amd/Phi-3.5-mini-instruct-onnx-ryzenai-npu",
"amd/Phi-3-mini-4k-instruct-onnx-ryzenai-hybrid",
"aammari/setfit-zero-shot-classification-pbsp-p3-dur",
"aammari/setfit-zero-shot-classification-pbsp-p3-freq",
"dlicari/distil-ita-legal-bert",
"aammari/setfit-zero-shot-classification-pbsp-p3-bhvr",
"dlicari/sentence-ita-eurlex-bert-base",
"lsanochkin/setfit-rubert-intent",
"LeaderboardModel1/grug-27b-AutoRound-W4A16-Tuning",
"microsoft/VibeVoice-Realtime-0.5B",
"ShwetActhq/dummy_setfit",
"gemasphi/real-setfit-ss-paraphrase-multilingual-mpnet-base-v2",
"OpenVoiceOS/canary-qwen-2.5b-onnx",
"LeaderboardModel1/Qwythos-9B-Claude-Mythos-5-1M-AutoRound-W4A16-Tuning",
"LeaderboardModel1/Qwevolution-V0-27B-AutoRound-W4A16-RTN",
"LeaderboardModel1/Qwen3.6-35B-A3B-AutoRound-W4A16-Tuning",
"LeaderboardModel1/Qwen3.5-4B-AutoRound-W4A16-RTN",
"YeungNLP/firefly-baichuan-7b-qlora-sft",
"LeaderboardModel1/Qwen-AgentWorld-35B-A3B-AutoRound-W4A16-RTN",
"LeaderboardModel1/Ornith-1.0-35B-AutoRound-W4A16-Tuning",
"LeaderboardModel1/Code-Writer-V2-Obliterated-BF16-AutoRound-W4A16-RTN",
"salihfurkaan/Qwen2.5-VL-ScienceQA-Math-CoT-Merged",
"LeaderboardModel1/Agents-A1-AutoRound-W4A16-RTN",
"aammari/setfit-zero-shot-classification-pbsp-p1",
"meta-llama/Prompt-Guard-86M",
"kornwtp/ConGen-paraphrase-multilingual-mpnet-base-v2",
"zjiayu064/Qwen3-Next-80B-A3B-Instruct-BitsMoE-2bit",
"kornwtp/ConGen-simcse-model-roberta-base-thai",
"praxisresearch/hf_olmo_32b_em_finrisk_1",
"gemasphi/setfit-ss-paraphrase-multilingual-mpnet-base-v2",
"RESMP-DEV/jina-code-embeddings-1.5b-mxfp4-block-gptq",
"praxisresearch/hf_olmo_32b_em_finrisk_0",
"praxisresearch/hf_olmo_32b_em_badmed_1",
"praxisresearch/hf_olmo_32b_em_badmed_0",
"lmxhappy/brand_bert",
"kamalkraj/BioSimCSE-BioLinkBERT-BASE",
"lmxhappy/tiyu_bert",
"ReadyArt/Omega-Darker-Gaslight_The-Final-Forgotten-Fever-Dream-24B",
"Kwai-AutoSQL/Kwai-AutoSQL-32B",
"ibaucells/paraphrase-multilingual-mpnet-base-v2_tecla_label2_8",
"futuredatascience/action-classifier-v2",
"LoneStriker/Orca-2-13b-5.0bpw-h6-exl2",
"futuredatascience/to-classifier-v2",
"futuredatascience/from-classifier-v2",
"shrinivasbjoshi/r2-w266-setfit-mbti-multiclass-hypsearch-mpnet-nov30",
multi-GPU multi-account submission: Kunlunxin/Biren/Cambricon with zhouyuanxi/jiajing/fanyi fallback rotation
2026-07-27 12:21:14 +08:00
]
BIREN_MODELS = [
submit 1366 models from the relaxed filter, round-robin across 4 accounts Filter condition 2 relaxed from "verified on another card" to "verified OR verifying on another card": a model already under verification elsewhere has passed the platform's basic checks, it just has no result yet. Same candidate pool yields 1366 instead of 12 (114x) - the two original conditions are近乎 mutually exclusive, since a model verified on another card has long since been tried on every card. Account handling changed from sequential fallback to round-robin over zhangyuanxi / jiajing / l112233 / l11223344, so the four share the load evenly (verified: 342/342/341/341) instead of piling onto the first account until it hits 60007. Iluvatar_bi-150 is back in GPU_JOBS now that a vllm iluvatar image is configured. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-20 17:15:54 +08:00
"emrevrg/Norovox-Alpha-MoE-7.0B",
"emrevrg/Norovox-Alpha-MoE-1.2B",
"emrevrg/Norovox-Alpha-MoE-3.9B",
"AI-ModelScope/granite-20b-code-base",
"vllm-ascend/Qwen3-32B-w8a8sc-310-vllm-tp4",
"KenDual3090tiNvlink/NVIDIA-Nemotron-Labs-3-Puzzle-75B-A9B-W4A16",
submit round-33 filter results: MetaX_c-500(2)/Kunlunxin_p-800(3)/hygon_k100-ai(1)/Cambricon_mlu-370-x8(3)/Biren_166m(2), 11 models total Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-20 15:34:48 +08:00
"AI-ModelScope/granite-20b-code-instruct-8k",
submit 1366 models from the relaxed filter, round-robin across 4 accounts Filter condition 2 relaxed from "verified on another card" to "verified OR verifying on another card": a model already under verification elsewhere has passed the platform's basic checks, it just has no result yet. Same candidate pool yields 1366 instead of 12 (114x) - the two original conditions are近乎 mutually exclusive, since a model verified on another card has long since been tried on every card. Account handling changed from sequential fallback to round-robin over zhangyuanxi / jiajing / l112233 / l11223344, so the four share the load evenly (verified: 342/342/341/341) instead of piling onto the first account until it hits 60007. Iluvatar_bi-150 is back in GPU_JOBS now that a vllm iluvatar image is configured. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-20 17:15:54 +08:00
"openbmb/MiniCPM4-8B-marlin-cpmcu",
"openbmb/MiniCPM4-8B-marlin-Eagle-vLLM",
"CaoruiLi/DefectVideoReward",
"ConstDemi/vikhr-nemo-12b-instruct-r-awq",
submit round-33 filter results: MetaX_c-500(2)/Kunlunxin_p-800(3)/hygon_k100-ai(1)/Cambricon_mlu-370-x8(3)/Biren_166m(2), 11 models total Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-20 15:34:48 +08:00
"pei39/deepseek_r1_distill_qwen_7b_awq_rag",
submit 1366 models from the relaxed filter, round-robin across 4 accounts Filter condition 2 relaxed from "verified on another card" to "verified OR verifying on another card": a model already under verification elsewhere has passed the platform's basic checks, it just has no result yet. Same candidate pool yields 1366 instead of 12 (114x) - the two original conditions are近乎 mutually exclusive, since a model verified on another card has long since been tried on every card. Account handling changed from sequential fallback to round-robin over zhangyuanxi / jiajing / l112233 / l11223344, so the four share the load evenly (verified: 342/342/341/341) instead of piling onto the first account until it hits 60007. Iluvatar_bi-150 is back in GPU_JOBS now that a vllm iluvatar image is configured. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-20 17:15:54 +08:00
"devika-tiwari/gpt2_small_expandedbabyLM_100M_exp3_conj_ratio_0p25_mix_0p25_43",
"RadixArk/Qwen3.8-27B-DSpark",
"iamsiddj/EducatorHelperModel",
"google/gemma-2b-cpp",
"hroth01/psais-all-MiniLM-L6-v2-20shot",
"devika-tiwari/gpt2_small_expandedbabyLM_100M_exp3_wh_v2_ratio_1p00_mix_0p25_44",
"EBLANSoft/eblangpt-falos67",
"ilhkn/my-awesome-setfit-model",
"Ddavidich/LOMONOSOV-ZENIT-27B-1M-INDEV",
"EleutherAI/quirky-pythia-410m-mixture",
"penupup/quora-distilbert-multilingual-lcqmc-sfa-hr",
"nomic-ai/nomic-embed-text-v2-moe",
"Jolieb/tinyllama-lora-object-sorting",
"arshiysohrevardi/moeinGTS1.5-1.5b",
"Xenova/pythia-410m-deduped",
"Q1ngMang/LFM2.5-350M-PT-zh_CN",
"nessy-studio/poison-gemma",
"flywheel-ai/real-estate",
"Pacific-i64/TR-MOE-306",
"bartowski/internlm2-chat-20b-llama-old",
"farbodtavakkoli/OTel-LLM-E4B-IT",
"hroth/psais-all-MiniLM-L6-v2-1shot",
"famseamus/LFM2.5-DCR-8B-A1B",
"narendarcodes/adaption-sec-financial-arithmetic-109b",
"narendarcodes/adaption-gpt-oss-120b-telugu-instruct",
"stjiris/bert-large-portuguese-cased-legal-tsdae",
"bartowski/deepseek-llm-7b-base-exl2",
"Blackfrost-AI/Lumix-4B",
"developer-lunark/Qwen3-30B-Korean-Roleplay",
"kshitijthakkar/loggenix-moe-0.3B-A0.1B-smolagents-v0-grpo",
"LLM-Research/Phi-4-mini-flash-reasoning",
"ib-ssm/mamba2-8b-3t-4k-hf",
"dfgertg/idk",
"LeVraiCNL/ornith_finetune_v3",
"fawazatvetted/fine_tuned_mpnetv3",
"riccardozaccone96/flexrank-llama3.2-1B",
"yuchen-zhu-zyc/HybridDiffusion-2B",
"fpadovani/jpn-jpan-100mb-after-ppt-shuff-dyck-10mb-ckpt500_seed455",
"infgrad/Prism-Qwen3.5-Reranker-0.8B",
"cjos/my-awesome-setfit-model",
"BananaMind/BananaMind-2-Nano",
"litert-community/Ministral-3-3B-Reasoning-2512",
"livadies/Bonsai-27B-Android-Local",
"blockblockblock/Qwen2.5-0.5B-Instruct-exl3-3.0bpw",
"WizardLMTeam/WizardCoder-33B-V1.1",
"amd/Phi-4-mini-instruct-onnx-ryzenai-hybrid",
"Tesslate/UIGEN-X-32B-0727",
"AscendKernelGen/KernelGen-LM-32B",
"charlie-li/Qwen3-Coder-30B-A3B-Instruct-Lora-ScaleSWE-Distilled-v1",
"open-thoughts/OpenThinkerAgent-32B",
"drbaph/s2-pro-fp8",
"groxaxo/s2-pro-BnB-4Bits",
"TomoroAI/tomoro-ai-colqwen3-embed-8b-awq",
"TomoroAI/tomoro-ai-colqwen3-embed-4b-awq",
"aleada/Phi-4-mini-instruct-W4A16",
"aleada/Phi-4-W4A16",
"amd/Phi-4-mini-reasoning-onnx-ryzenai-hybrid",
"amd/Phi-3.5-mini-instruct-onnx-ryzenai-npu",
"amd/Phi-3-mini-4k-instruct-onnx-ryzenai-hybrid",
"cyber-pal-security/CyberPal2.0-20B",
"LeaderboardModel1/grug-27b-AutoRound-W4A16-Tuning",
"Qwen/Qwen-AgentWorld-35B-A3B",
"jinxing6609/olmo3-7b-zh-merged",
"LeaderboardModel1/Qwythos-9B-Claude-Mythos-5-1M-AutoRound-W4A16-Tuning",
"LeaderboardModel1/Qwevolution-V0-27B-AutoRound-W4A16-RTN",
"LeaderboardModel1/Qwen3.6-35B-A3B-AutoRound-W4A16-Tuning",
"LeaderboardModel1/Qwen3.5-4B-AutoRound-W4A16-RTN",
"LeaderboardModel1/Qwen-AgentWorld-35B-A3B-AutoRound-W4A16-RTN",
"LeaderboardModel1/Ornith-1.0-35B-AutoRound-W4A16-Tuning",
"LeaderboardModel1/Code-Writer-V2-Obliterated-BF16-AutoRound-W4A16-RTN",
"LeaderboardModel1/Agents-A1-AutoRound-W4A16-RTN",
"Qwen/WebWorld-32B",
"24B-Suite/Mergedonia-PROMETHEUS-24B-v1",
"zjiayu064/Qwen3-Next-80B-A3B-Instruct-BitsMoE-2bit",
"praxisresearch/hf_olmo_32b_em_finrisk_1",
"RESMP-DEV/jina-code-embeddings-1.5b-mxfp4-block-gptq",
"praxisresearch/hf_olmo_32b_em_finrisk_0",
"praxisresearch/hf_olmo_32b_em_badmed_1",
"praxisresearch/hf_olmo_32b_em_badmed_0",
"allenai/MolmoAct-7B-D-Pretrain-0812",
"allenai/MolmoAct-7B-D-LIBERO-Object-0812",
"Mohamed-Sami-Ghrab/moove-qwen3-32b-medical-sft-v2",
"pki/nova-24b-cybersec",
"ReadyArt/Omega-Darker-Gaslight_The-Final-Forgotten-Fever-Dream-24B",
"Kwai-AutoSQL/Kwai-AutoSQL-32B",
"litert-community/gemma-3-270m-it",
"LoneStriker/Orca-2-13b-5.0bpw-h6-exl2",
init: Kunlunxin vllm adapt-task batch submit strategy
2026-07-23 15:41:43 +08:00
]
multi-GPU multi-account submission: Kunlunxin/Biren/Cambricon with zhouyuanxi/jiajing/fanyi fallback rotation
2026-07-27 12:21:14 +08:00
CAMBRICON_MODELS = [
submit 1366 models from the relaxed filter, round-robin across 4 accounts Filter condition 2 relaxed from "verified on another card" to "verified OR verifying on another card": a model already under verification elsewhere has passed the platform's basic checks, it just has no result yet. Same candidate pool yields 1366 instead of 12 (114x) - the two original conditions are近乎 mutually exclusive, since a model verified on another card has long since been tried on every card. Account handling changed from sequential fallback to round-robin over zhangyuanxi / jiajing / l112233 / l11223344, so the four share the load evenly (verified: 342/342/341/341) instead of piling onto the first account until it hits 60007. Iluvatar_bi-150 is back in GPU_JOBS now that a vllm iluvatar image is configured. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-20 17:15:54 +08:00
"emrevrg/Norovox-Alpha-MoE-7.0B",
"emrevrg/Norovox-Alpha-MoE-1.2B",
"emrevrg/Norovox-Alpha-MoE-3.9B",
"AI-ModelScope/granite-20b-code-base",
"llm-jp/llm-jp-4-32b-a3b-thinking",
"vllm-ascend/Qwen3-32B-w8a8sc-310-vllm-tp4",
"YOYO-AI/ZYH-LLM-Qwen2.5-14B-V2",
"DCAgent/g1_clean_hybrid_25k_32b",
"amd/gpt-oss-20b-BF16-w8a8-llmcompressor",
"KoboldAI/fairseq-dense-13B-Janeway",
"KenDual3090tiNvlink/NVIDIA-Nemotron-Labs-3-Puzzle-75B-A9B-W4A16",
submit round-33 filter results: MetaX_c-500(2)/Kunlunxin_p-800(3)/hygon_k100-ai(1)/Cambricon_mlu-370-x8(3)/Biren_166m(2), 11 models total Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-20 15:34:48 +08:00
"AI-ModelScope/granite-20b-code-instruct-8k",
submit 1366 models from the relaxed filter, round-robin across 4 accounts Filter condition 2 relaxed from "verified on another card" to "verified OR verifying on another card": a model already under verification elsewhere has passed the platform's basic checks, it just has no result yet. Same candidate pool yields 1366 instead of 12 (114x) - the two original conditions are近乎 mutually exclusive, since a model verified on another card has long since been tried on every card. Account handling changed from sequential fallback to round-robin over zhangyuanxi / jiajing / l112233 / l11223344, so the four share the load evenly (verified: 342/342/341/341) instead of piling onto the first account until it hits 60007. Iluvatar_bi-150 is back in GPU_JOBS now that a vllm iluvatar image is configured. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-20 17:15:54 +08:00
"openbmb/MiniCPM4-8B-marlin-cpmcu",
submit round-33 filter results: MetaX_c-500(2)/Kunlunxin_p-800(3)/hygon_k100-ai(1)/Cambricon_mlu-370-x8(3)/Biren_166m(2), 11 models total Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-20 15:34:48 +08:00
"AI-ModelScope/granite-20b-code-instruct",
submit 1366 models from the relaxed filter, round-robin across 4 accounts Filter condition 2 relaxed from "verified on another card" to "verified OR verifying on another card": a model already under verification elsewhere has passed the platform's basic checks, it just has no result yet. Same candidate pool yields 1366 instead of 12 (114x) - the two original conditions are近乎 mutually exclusive, since a model verified on another card has long since been tried on every card. Account handling changed from sequential fallback to round-robin over zhangyuanxi / jiajing / l112233 / l11223344, so the four share the load evenly (verified: 342/342/341/341) instead of piling onto the first account until it hits 60007. Iluvatar_bi-150 is back in GPU_JOBS now that a vllm iluvatar image is configured. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-20 17:15:54 +08:00
"openbmb/MiniCPM4-8B-marlin-Eagle-vLLM",
"CaoruiLi/DefectVideoReward",
"ConstDemi/vikhr-nemo-12b-instruct-r-awq",
"sapidlabs/Sparkulator-GLM-5.2",
submit round-33 filter results: MetaX_c-500(2)/Kunlunxin_p-800(3)/hygon_k100-ai(1)/Cambricon_mlu-370-x8(3)/Biren_166m(2), 11 models total Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-20 15:34:48 +08:00
"pei39/deepseek_r1_distill_qwen_7b_awq_rag",
submit 1366 models from the relaxed filter, round-robin across 4 accounts Filter condition 2 relaxed from "verified on another card" to "verified OR verifying on another card": a model already under verification elsewhere has passed the platform's basic checks, it just has no result yet. Same candidate pool yields 1366 instead of 12 (114x) - the two original conditions are近乎 mutually exclusive, since a model verified on another card has long since been tried on every card. Account handling changed from sequential fallback to round-robin over zhangyuanxi / jiajing / l112233 / l11223344, so the four share the load evenly (verified: 342/342/341/341) instead of piling onto the first account until it hits 60007. Iluvatar_bi-150 is back in GPU_JOBS now that a vllm iluvatar image is configured. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-20 17:15:54 +08:00
"fpadovani/jpn-jpan-100mb-ppt-Dp-10mb_seed10",
"Zynerji/Ektome-Qwen3-VL-4Bi-PristinelyUncensored",
"Zynerji/Ektome-Qwen3-VL-2Bi-PristinelyUncensored",
"iamsiddj/EducatorHelperModel",
"vorenthiclabs/vorenthos-vision-thinking-qwq-7b",
"lodestones/gemma-7b-flax",
"google/gemma-2b-cpp",
"jinaai/jina-reranker-v3.5",
"hroth01/psais-all-MiniLM-L6-v2-20shot",
"anicka/ke-v9-8b",
"EBLANSoft/eblangpt-falos67",
"fpadovani/eng-latn-10mb-ppt-shuff-dyck-disjoint-tok_seed3407",
"ilhkn/my-awesome-setfit-model",
"Ddavidich/LOMONOSOV-ZENIT-27B-1M-INDEV",
"fpadovani/eng-latn-10mb-ppt-Dp-100mb_seed10",
"penupup/quora-distilbert-multilingual-lcqmc-sfa-hr",
"hroth/psais-paraphrase-multilingual-MiniLM-L12-v2-20shot",
"hroth/psais-LaBSE-10shot",
"TurkuNLP/web-register-classification-multilingual",
"malaiwah/GLM-5.2-SIQ-Fruit-Instruct",
"nomic-ai/nomic-embed-text-v2-moe",
"Udoba45/Qwen2.5-0.5B-Instruct-Gensyn-Swarm-tall_thorny_boar",
"VerStella/Qwen3.8-27B-heretic-ara-W8A8-DCU",
"LamNguyen0904/vistral-medical-adapter",
"Jolieb/tinyllama-lora-object-sorting",
"ibm-granite/granite-embedding-30m-sparse",
"arshiysohrevardi/moeinGTS1.5-1.5b",
"Undi95/Mistral-11B-OmniMix",
"Xenova/pythia-410m-deduped",
"Undi95/Mistral-11B-CC-Air",
"Q1ngMang/LFM2.5-350M-PT-zh_CN",
"nessy-studio/poison-gemma",
"Undi95/Mistral-11B-Airoboros-v0.1",
"flywheel-ai/real-estate",
"hroth/psais-all-MiniLM-L6-v2-1shot",
"stjiris/bert-large-portuguese-cased-legal-tsdae",
"bartowski/deepseek-llm-7b-base-exl2",
"llm-semantic-router/mmbert-fact-check-merged",
"LeVraiCNL/ornith_finetune_v3",
"fawazatvetted/fine_tuned_mpnetv3",
"berkbirkan/gemma-3-lora-finetune-x-replies",
"nm-testing/Qwen3-4B-mixed-quant-RTN-wNaM-a4",
"infgrad/Prism-Qwen3.5-Reranker-0.8B",
"cjos/my-awesome-setfit-model",
"OpenMed/OpenMed-PII-French-BioClinicalBERT-Base-110M-v1",
"seerware/granite-embedding-30m-sparse",
"MU-Kindai/SBERT-JSNLI-base",
"OpenMed/OpenMed-PII-French-BiomedELECTRA-Large-335M-v1",
"JoBeer/multi-qa-mpnet-base-dot-v1-eclass",
"OpenMed/OpenMed-PII-French-BiomedELECTRA-Base-110M-v1",
"OpenMed/OpenMed-PII-French-BiomedBERT-Large-340M-v1",
"infgrad/Prism-Qwen3.5-Reranker-4B",
"pkshatech/simcse-ja-bert-base-clcmlp",
"RCouronnePEReN/gpt2_finetuned_overfit-10epochs-128tokens",
"q-future/Q-ReAlign-Pro-9B",
"vrfai/gemma-4-12B-it-fp8",
"amd/Phi-4-mini-instruct-onnx-ryzenai-hybrid",
"OpenMed/OpenMed-PII-German-BioClinicalBERT-Base-110M-v1",
"louisbetsch/new-classification-model",
"Trendyol/Trendyol-LLM-Asure-12B",
"OpenMed/OpenMed-PII-German-BiomedELECTRA-Large-335M-v1",
"OpenMed/OpenMed-PII-German-BiomedELECTRA-Base-110M-v1",
"OpenMed/OpenMed-PII-German-BiomedBERT-Base-110M-v1",
"aammari/setfit-zero-shot-classification-pbsp-p3-cons",
"JoBeer/all-mpnet-base-v2-eclass",
"BalaRajesh1/mmbert-small-nli",
"aammari/setfit-zero-shot-classification-pbsp-p3-trig",
"charlie-li/Qwen3-Coder-30B-A3B-Instruct-Lora-ScaleSWE-Distilled-v1",
"quanda-bench-test/gpt2-small-trex-openwebtext-hf",
"quanda-bench-test/gpt2-small-trex-hf",
"quanda-bench-test/gpt2-small-trex-wikitext-ft-hf",
"quanda-bench-test/gpt2-small-trex-wikitext-hf",
"quanda-bench-test/gpt2-small-trex-openwebtext-ft-hf",
"quanda-bench-test/gpt2-small-trex-ft-hf",
"gart-labor/eng-distilBERT-se-eclass",
"Pakorn2112/Orpheus-TTS-hmong-3b",
"OpenMed/OpenMed-PII-Italian-BiomedBERT-Base-110M-v1",
"OpenMed/OpenMed-PII-Italian-BioClinicalBERT-Base-110M-v1",
"gemasphi/f-real-setfit-ss-paraphrase-multilingual-mpnet-base-v2",
"OpenMed/OpenMed-PII-Italian-BiomedELECTRA-Large-335M-v1",
"ans2004/aragpt2-transport-autocomplete",
"fawazatvetted/fine_tuned_mpnetv2",
"Auparima/thaifood-embed",
"OpenMed/OpenMed-PII-Italian-BiomedBERT-Large-340M-v1",
"OpenMed/OpenMed-PII-Italian-BiomedELECTRA-Base-110M-v1",
"fishaudio/s2-pro",
"drbaph/s2-pro-fp8",
"leerobert/my-awesome-setfit-model",
"OpenMed/OpenMed-PII-Spanish-BioClinicalBERT-Base-110M-v1",
"OpenMed/OpenMed-PII-Spanish-BiomedELECTRA-Base-110M-v1",
"OpenMed/OpenMed-PII-Spanish-BiomedBERT-Large-340M-v1",
"OpenMed/OpenMed-PII-Spanish-BiomedBERT-Base-110M-v1",
"groxaxo/s2-pro-BnB-4Bits",
"ReconOut/granite-embedding-english-r2",
"Imagilux/fishaudio-s2-pro",
"TomoroAI/tomoro-ai-colqwen3-embed-8b-awq",
"TomoroAI/tomoro-ai-colqwen3-embed-4b-awq",
"aleada/Phi-4-mini-instruct-W4A16",
"aleada/Phi-4-W4A16",
"lewtun/setfit-minilm-distilled-test",
"morphllm/dspark-glm52-draft",
"tomaarsen/jina-embeddings-v3-hf",
"ibm-granite/granite-embedding-311m-multilingual-r2",
"aparna1595/bert-mpnet-sentence",
"ibm-granite/granite-embedding-97m-multilingual-r2",
"FINAL-Bench/Darwin-TTS-1.7B-Cross",
"amd/Phi-4-mini-reasoning-onnx-ryzenai-hybrid",
"jinaai/jina-embeddings-v3-hf",
"gluschenko/omnivoice-onnx",
"mssongit/fndeberta-sentence",
"splendor1811/omnivoice-vietnamese",
"amd/Phi-3.5-mini-instruct-onnx-ryzenai-npu",
"yugeshkarunamurthy/Ornith-1.0-9B-oQe6",
"amd/Phi-3-mini-4k-instruct-onnx-ryzenai-hybrid",
"yugeshkarunamurthy/Ornith-1.0-9B-oQe3.5",
"aammari/setfit-zero-shot-classification-pbsp-p3-dur",
"aammari/setfit-zero-shot-classification-pbsp-p3-freq",
"dlicari/distil-ita-legal-bert",
"maya-multimodal/Qwen3.5-9B-aerialsim-rl-step100",
"aammari/setfit-zero-shot-classification-pbsp-p3-bhvr",
"dlicari/sentence-ita-eurlex-bert-base",
"lsanochkin/setfit-rubert-intent",
"cyber-pal-security/CyberPal2.0-20B",
"LeaderboardModel1/grug-27b-AutoRound-W4A16-Tuning",
"microsoft/VibeVoice-Realtime-0.5B",
"Qwen/Qwen-AgentWorld-35B-A3B",
"ShwetActhq/dummy_setfit",
"gemasphi/real-setfit-ss-paraphrase-multilingual-mpnet-base-v2",
"OpenVoiceOS/canary-qwen-2.5b-onnx",
"nvidia/nemotron-labs-audio-visual-flamingo-hf",
"LeaderboardModel1/Qwythos-9B-Claude-Mythos-5-1M-AutoRound-W4A16-Tuning",
"LeaderboardModel1/Qwevolution-V0-27B-AutoRound-W4A16-RTN",
"LeaderboardModel1/Qwen3.6-35B-A3B-AutoRound-W4A16-Tuning",
"LeaderboardModel1/Qwen3.5-4B-AutoRound-W4A16-RTN",
"YeungNLP/firefly-baichuan-7b-qlora-sft",
"LeaderboardModel1/Qwen-AgentWorld-35B-A3B-AutoRound-W4A16-RTN",
"LeaderboardModel1/Ornith-1.0-35B-AutoRound-W4A16-Tuning",
"LeaderboardModel1/Code-Writer-V2-Obliterated-BF16-AutoRound-W4A16-RTN",
"salihfurkaan/Qwen2.5-VL-ScienceQA-Math-CoT-Merged",
"LeaderboardModel1/Agents-A1-AutoRound-W4A16-RTN",
"aammari/setfit-zero-shot-classification-pbsp-p1",
"meta-llama/Prompt-Guard-86M",
"Qwen/WebWorld-32B",
"24B-Suite/Mergedonia-PROMETHEUS-24B-v1",
"kornwtp/ConGen-paraphrase-multilingual-mpnet-base-v2",
"zjiayu064/Qwen3-Next-80B-A3B-Instruct-BitsMoE-2bit",
"kornwtp/ConGen-simcse-model-roberta-base-thai",
"praxisresearch/hf_olmo_32b_em_finrisk_1",
"EleutherAI/pythia-1.4b-addition-first-ft",
"gemasphi/setfit-ss-paraphrase-multilingual-mpnet-base-v2",
"RESMP-DEV/jina-code-embeddings-1.5b-mxfp4-block-gptq",
"praxisresearch/hf_olmo_32b_em_finrisk_0",
"praxisresearch/hf_olmo_32b_em_badmed_1",
"praxisresearch/hf_olmo_32b_em_badmed_0",
"allenai/MolmoAct-7B-D-Pretrain-0812",
"allenai/MolmoAct-7B-D-LIBERO-Spatial-0812",
"allenai/MolmoAct-7B-D-LIBERO-Object-0812",
"lmxhappy/brand_bert",
"kamalkraj/BioSimCSE-BioLinkBERT-BASE",
"Mohamed-Sami-Ghrab/moove-qwen3-32b-medical-sft-v2",
"lmxhappy/tiyu_bert",
"ReadyArt/Omega-Darker-Gaslight_The-Final-Forgotten-Fever-Dream-24B",
"shrinivasbjoshi/r3-w266-setfit-mbti-noempath-multiclass-hypsearch-mpnet-dec1",
"Kwai-AutoSQL/Kwai-AutoSQL-32B",
"ibaucells/paraphrase-multilingual-mpnet-base-v2_tecla_label2_8",
"futuredatascience/action-classifier-v2",
"futuredatascience/to-classifier-v2",
"futuredatascience/from-classifier-v2",
"shrinivasbjoshi/r2-w266-setfit-mbti-multiclass-hypsearch-mpnet-nov30",
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
2026-08-06 18:59:57 +08:00
]
HYGON_MODELS = [
submit 1366 models from the relaxed filter, round-robin across 4 accounts Filter condition 2 relaxed from "verified on another card" to "verified OR verifying on another card": a model already under verification elsewhere has passed the platform's basic checks, it just has no result yet. Same candidate pool yields 1366 instead of 12 (114x) - the two original conditions are近乎 mutually exclusive, since a model verified on another card has long since been tried on every card. Account handling changed from sequential fallback to round-robin over zhangyuanxi / jiajing / l112233 / l11223344, so the four share the load evenly (verified: 342/342/341/341) instead of piling onto the first account until it hits 60007. Iluvatar_bi-150 is back in GPU_JOBS now that a vllm iluvatar image is configured. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-20 17:15:54 +08:00
"emrevrg/Norovox-Alpha-MoE-7.0B",
"emrevrg/Norovox-Alpha-MoE-1.2B",
"emrevrg/Norovox-Alpha-MoE-3.9B",
"shihaixiong/Qwen2.5-0.5B-Instruct-Gensyn-Swarm-omnivorous_fleecy_platypus",
"barguty/Qwen2.5-1.5B-Instruct-Gensyn-Swarm-omnivorous_alert_tiger",
"Eram83/test_slo_1",
"VehaanS/slm-125m-tutor",
"BAAI/AREX-Turbo",
"malichanna/mh-response-generator",
"milwright/cloze-reader-qwen3.5-0.8b",
"Michael-Kozu/Deimos-R1",
"openbmb/MiniCPM4-8B-marlin-cpmcu",
"leok7v/Qwen3.5-0.8B-coreml",
"openbmb/MiniCPM4-8B-marlin-Eagle-vLLM",
"emrevrg/Norovox-Alpha-0.3B",
"CaoruiLi/DefectVideoReward",
"openbmb/MiniCPM4-8B-Eagle-FRSpec-QAT-cpmcu",
"mkashifali1/Qwen2.5-0.5B-Instruct-Gensyn-Swarm-arctic_muscular_heron",
"openbmb/MiniCPM4-8B-Eagle-FRSpec",
"Charley890/AdaptiveMath",
"danish-foundation-models/DFM-Mimir",
"kytostweaks/Qwen2.5-0.5B-Instruct-Gensyn-Swarm-rangy_snorting_anaconda",
"Xenova/deepseek-coder-1.3b-base",
"Xenova/deepseek-coder-1.3b-instruct",
"raynardj/wenyanwen-ancient-translate-to-modern",
"Finnish-NLP/Ahma-3B",
"RinKana/aquila-135m-csqa",
"pijarcandra22/balinese_fine_tuned_model",
"sapidlabs/Sparkulator-GLM-5.2",
"tommymir4444/Qwen3-0.6B-Gensyn-Swarm-lively_darting_penguin",
submit round-33 filter results: MetaX_c-500(2)/Kunlunxin_p-800(3)/hygon_k100-ai(1)/Cambricon_mlu-370-x8(3)/Biren_166m(2), 11 models total Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-20 15:34:48 +08:00
"pei39/deepseek_r1_distill_qwen_7b_awq_rag",
submit 1366 models from the relaxed filter, round-robin across 4 accounts Filter condition 2 relaxed from "verified on another card" to "verified OR verifying on another card": a model already under verification elsewhere has passed the platform's basic checks, it just has no result yet. Same candidate pool yields 1366 instead of 12 (114x) - the two original conditions are近乎 mutually exclusive, since a model verified on another card has long since been tried on every card. Account handling changed from sequential fallback to round-robin over zhangyuanxi / jiajing / l112233 / l11223344, so the four share the load evenly (verified: 342/342/341/341) instead of piling onto the first account until it hits 60007. Iluvatar_bi-150 is back in GPU_JOBS now that a vllm iluvatar image is configured. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-20 17:15:54 +08:00
"rub3d0/Qwen3-0.6B-Gensyn-Swarm-chattering_burrowing_clam",
"tommymir4444/Qwen3-0.6B-Gensyn-Swarm-skittish_spotted_chinchilla",
"fpadovani/jpn-jpan-100mb-ppt-Dp-10mb_seed10",
"samuelhatake/Qwen3-0.6B-Gensyn-Swarm-domestic_waddling_boar",
"Zynerji/Ektome-Qwen3-VL-4Bi-PristinelyUncensored",
"Maincore/Qwen3-0.6B-Gensyn-Swarm-freckled_quick_bear",
"devika-tiwari/gpt2_small_expandedbabyLM_100M_exp3_conj_ratio_1p00_mix_0p25_44",
"Zynerji/Ektome-Qwen3-VL-2Bi-PristinelyUncensored",
"SakanaAI/RePo-OLMo2-1B-stage2-L5",
"SupraLabs/Supra2-Nano",
"insagur/n3d4e68fc18d5",
"imdhammu/Qwen2.5-1.5B-Instruct-Gensyn-Swarm-reclusive_energetic_frog",
"devika-tiwari/gpt2_small_expandedbabyLM_100M_exp3_conj_ratio_0p25_mix_0p25_43",
"Zynerji/Ektome-Qwen2.5-VL-3Bi-PristinelyUncensored",
"RadixArk/Qwen3.8-27B-DSpark",
"tommymir4444/Qwen2.5-0.5B-Instruct-Gensyn-Swarm-lively_darting_penguin",
"kytostweaks/Qwen3-0.6B-Gensyn-Swarm-rangy_snorting_anaconda",
"vorenthiclabs/vorenthos-vision-thinking-qwq-7b",
"tommymir4444/Qwen3-0.6B-Gensyn-Swarm-masked_pesty_chameleon",
"mnayak1/my-it-helpdesk-dpo-model",
"fpadovani/jpn-jpan-100mb-after-ppt-Dp-100mb-ckpt500_seed10",
"lodestones/gemma-7b-flax",
"Ateeqq/news-title-generator",
"google/gemma-2b-cpp",
"atsuki-yamaguchi/Qwen2.5-7B-bn-madlad-mean-cv",
"mesolitica/gemma-2B-16k-instructions",
"newendiesports/iol-2026-qwen14b-entry",
"jinaai/jina-reranker-v3.5",
"KingpinFisk/iol-2026-qwen14b-entry",
"hroth01/psais-all-MiniLM-L6-v2-50shot",
"hroth01/psais-all-MiniLM-L6-v2-20shot",
"devika-tiwari/gpt2_small_expandedbabyLM_100M_exp3_wh_v2_ratio_1p00_mix_0p25_44",
"EBLANSoft/eblangpt-falos67",
"fpadovani/eng-latn-10mb-ppt-shuff-dyck-disjoint-tok_seed3407",
"dealignai/Laguna-XS-2.1-JANG_2L-CRACK",
"GoldenOreo/llama_finetune",
"ilhkn/my-awesome-setfit-model",
"fpadovani/eng-latn-10mb-ppt-shuff-dyck-10mb_seed10",
"EleutherAI/quirky-pythia-2.8b-mixture",
"fpadovani/eng-latn-10mb-ppt-shuff-dyck-100mb_seed10",
"EleutherAI/quirky-pythia-2.8b-grader-last",
"Ddavidich/LOMONOSOV-ZENIT-27B-1M-INDEV",
"Agnivarcas/iol-2026-qwen14b-entry",
"fpadovani/eng-latn-10mb-ppt-Dp-100mb_seed10",
"EleutherAI/quirky-pythia-410m-mixture",
"penupup/quora-distilbert-multilingual-lcqmc-sfa-hr",
"hroth/psais-paraphrase-multilingual-MiniLM-L12-v2-20shot",
"hroth/psais-LaBSE-10shot",
"TurkuNLP/web-register-classification-multilingual",
"opencerebral/littlerock-1M-arithmax",
"microsoft/dolly-v2-7b-olive-optimized",
"mizzaay/f389e6cd",
"malaiwah/GLM-5.2-SIQ-Fruit-Instruct",
"nomic-ai/nomic-embed-text-v2-moe",
"opencerebral/littlerock-1M",
"NILKNARFGonzo/floppyx4-nonsensicalEssential-base",
"Udoba45/Qwen2.5-0.5B-Instruct-Gensyn-Swarm-tall_thorny_boar",
"Catniti/catrex-1.0",
"Welmia/welmia-1.0-81m",
"marin-dna/marin-dna-scaling-v0.5-h1920-p1B",
"garagelm/hybrid-gpt-232m",
"LamNguyen0904/vistral-medical-adapter",
"empero-ai/Qwen3.8-4B-Distill",
"Moriec/Vinogradov_semantic_search",
"fpadovani/eng-latn-100mb-ppt-shuff-dyck-100mb_seed10",
"Jolieb/tinyllama-lora-object-sorting",
"billygeekourson/philidor-142m",
"billygeekourson/philidor-51m",
"ibm-granite/granite-embedding-30m-sparse",
"arshiysohrevardi/moeinGTS1.5-1.5b",
"OsaurusAI/LFM2.5-2.6B-JANG_6M",
"Undi95/Mistral-11B-OmniMix",
"kerzgrr/Couplet-base",
"kerzgrr/Couplet",
"Hiveurban/multilingual-e5-base",
"IvmeLabs/ExpIvme-DiffusionConversate-v1-Instruct",
"KiteFishAI/Nano-Em1-0.6B-v2",
"KiteFishAI/Nano-Em1-0.6B",
"Xenova/pythia-410m-deduped",
"Akshat-Dwivedi/micro-gpt",
"jalbarracin/multilingual-e5-small-distilled-256d",
"Undi95/Mistral-11B-CC-Air",
"GYETI/Prism-4B-Reasoning",
"Q1ngMang/LFM2.5-350M-PT-zh_CN",
"nessy-studio/poison-gemma",
"Undi95/Mistral-11B-Airoboros-v0.1",
"iselabvn/Kali-Terminus-v3",
"hroth/psais-all-MiniLM-L6-v2-8shot",
"0penAGI/0zhr",
"theyur/dhamma-parrot-v01",
"fpadovani/dan-latn-10mb-after-ppt-Dp-10mb-ckpt500_seed3407",
"flywheel-ai/real-estate",
"arcee-ai/Patent-Instruct-Internlm2-7B-Ties",
"Pacific-i64/TR-MOE-306",
"arcee-ai/Patent-Base-InternLM2-7B-Ties",
"OsaurusAI/Nanbeige4.2-3B-JANG_4M",
"sandepaAI/sandepaAI_gemma4_coder_12b",
"philipjohnbasile/wisp-coder-110m",
"modal-labs/Kimi-K3-DFlash",
"aiml8726737/personal-prompt-engineer-qwen1.5b",
"litert-community/granite-4.0-350m-litert-lm",
"farbodtavakkoli/OTel-LLM-E4B-IT",
"FrontiersMind/Lumma-0.6B-Base",
"hroth/psais-all-MiniLM-L6-v2-1shot",
"farbodtavakkoli/OTel-LLM-8B-A1B-IT",
"famseamus/LFM2.5-DCR-8B-A1B",
"chrispian/Qwen3-0.6B-Gensyn-Swarm-soaring_wiry_stingray",
"google/gemma-2-9b-it",
"narendarcodes/adaption-sec-financial-arithmetic-109b",
"narendarcodes/adaption-gpt-oss-120b-telugu-instruct",
"bbunzeck/german-babylm-mix-char",
"miketester10/Qwen2.5-Coder-0.5B-Instruct-Gensyn-Swarm-tiny_pensive_mandrill",
"Rorical/logos-1b-base",
"pestlee/Qwable-9B-Claude-Fable-5-StraTA",
"stjiris/bert-large-portuguese-cased-legal-tsdae",
"AngelSlim/Hy3-DFlash-B8",
"AngelSlim/Hy3-DFlash-B16",
"penguin-wall/EllAIs-Bronte",
"sshleifer/tiny-xlnet-base-cased",
"56m/DumbMath_Preview_1_0722",
"fpadovani/jpn-jpan-10mb-ppt-shuff-dyck-10mb_seed455",
"mindqtrl/qwen3vl-8b-fp8-text-only-en",
"fpadovani/jpn-jpan-10mb-ppt-shuff-dyck-100mb_seed455",
"bbunzeck/german-babylm-mix-bpe",
"kashif0912/ULTRON-V2",
"fpadovani/wiki_sh1_42",
"bartowski/deepseek-llm-7b-base-exl2",
"fpadovani/jpn-jpan-10mb-ppt-Dp-10mb_seed455",
"fpadovani/jpn-jpan-100mb-ppt-shuff-dyck-100mb_seed455",
"fpadovani/jpn-jpan-10mb-ppt-Dp-100mb_seed455",
"BananaMind/BananaMind-2-Pro",
"inference-optimization/Qwen3.8-1.0B-A0.6B",
"Umranz/Shruti-Soft-2.6b",
"tommymir4444/Qwen2.5-0.5B-Instruct-Gensyn-Swarm-skittish_mammalian_peacock",
"SIRIS-Lab/citation-parser-TYPE",
"McGill-NLP/AfriqueQwen3.5-0.8B-50Langs",
"SIRIS-Lab/citation-parser-ENTITY",
"BrainboxAI/cyber-analyst-4B",
"jacepark12/LFM2.5-2.6B-Terminal-SFT",
"RightNowAI/RightNow-Arabic-0.5B-Turbo",
"hajkis-glami/multilingual-e5-base-embeddings",
"insagur/n1f23db785ccf",
"fpadovani/jpn-jpan-100mb-ppt-shuff-dyck-10mb_seed455",
"nvidia/gpt-oss-120b-Eagle3-v3",
"fpadovani/jpn-jpan-100mb-ppt-Dp-10mb_seed455",
"suehuynh/Marketing-Llama-3.3-70B",
"fpadovani/jpn-jpan-100mb-ppt-Dp-100mb_seed455",
"tommymir4444/Qwen2.5-0.5B-Instruct-Gensyn-Swarm-miniature_rapid_cheetah",
"sultanbi/e5-base-kazakh",
"squeeze-ai-lab/TinyAgent-1.1B",
"jinaai/jina-embeddings-v2-base-code",
"AI-ModelScope/bert-base-chinese",
"llm-semantic-router/mmbert-intent-classifier-merged",
"LH-Tech-AI/htmLLM-50M-Base",
"llm-semantic-router/mmbert-fact-check-merged",
"developer-lunark/Qwen3-30B-Korean-Roleplay",
"fpadovani/jpn-jpan-10mb-after-ppt-shuff-dyck-10mb-ckpt500_seed455",
"taobao-mnn/Qwen3.5-0.8B-MNN",
"AETHORIA-AI/TR-HASH-MOE-500M-HF",
"Ba2han/TR_SFT1",
"OrionLLM/GRM-3.2-Turf",
"shoowadoo/galkhesnet",
"gorkimark/Qwen3-0.6B-Gensyn-Swarm-fluffy_rapid_wildebeest",
"manjunathshiva/Laguna-XS.2-tq3-g64",
"melikegks/turkish-pii-guard-0.8b-legacy",
"AtomixLabs/Photon-1.0-1M",
"litert-community/Qwen3-0.6B",
"litert-community/SmolVLM-256M-Instruct",
"fpadovani/jpn-jpan-10mb-10mb_seed455",
"litert-community/SmolLM-135M-Instruct",
"ib-ssm/mamba2-8b-3t-4k-hf",
"DedeProGames/DynamicMind-MoE",
"OpenMed/OpenMed-PII-ClinicalLongformer-Base-149M-v1",
"OpenMed/OpenMed-PII-BioClinicalModern-Base-149M-v1",
"shangeth/Wren-TTS-0.5B-multi-expressive",
"OpenMed/OpenMed-PII-SuperMedical-Large-355M-v1",
"OpenMed/OpenMed-PII-SnowflakeMed-Large-568M-v1",
"OpenMed/OpenMed-PII-SuperMedical-Base-125M-v1",
"OpenMed/OpenMed-PII-mClinicalE5-Large-560M-v1",
"lsm0729/Qwen2.5-VL-3B-Instruct-quantized.w8a8",
"OpenMed/OpenMed-PII-BigMed-Large-560M-v1",
"dfgertg/idk",
"fpadovani/jpn-jpan-10mb-100mb_seed455",
"baseten-admin/multilingual-e5-large-instruct-cls",
"fawazatvetted/fine_tuned_mpnetv3",
"riccardozaccone96/flexrank-llama3.2-1B",
"yuchen-zhu-zyc/HybridDiffusion-2B",
"oddadmix/50M-2048-Emhotob",
"fpadovani/jpn-jpan-100mb-after-ppt-shuff-dyck-10mb-ckpt500_seed455",
"gagan3012/Qwen-2.5-reasoning-verifier",
"tommymir4444/Qwen3-0.6B-Gensyn-Swarm-armored_gliding_chameleon",
"berkbirkan/gemma-3-lora-finetune-x-replies",
"nm-testing/Qwen3-4B-mixed-quant-RTN-wNaM-a4",
"infgrad/Prism-Qwen3.5-Reranker-0.8B",
"Rzkoohi/Qwen3.5-4B-Natural_reasoning",
"Glubbot-Team/glubbot1.1-7.5m",
"cjos/my-awesome-setfit-model",
"Purus15987/English_Telugu_Translation",
"fpadovani/jpn-jpan-100mb-after-ppt-Dp-10mb-ckpt500_seed455",
"fpadovani/jpn-jpan-100mb-after-ppt-Dp-100mb-ckpt500_seed455",
"BananaMind/BananaMind-2-Nano",
"litert-community/Ministral-3-3B-Reasoning-2512",
"MuXodious/LFM2.5-2.6B-absolute-heresy",
"OpenMed/OpenMed-PII-French-ClinicalBGE-Large-568M-v1",
"OpenMed/OpenMed-PII-French-ClinicalBGE-Large-335M-v1",
"User01110/CMA-1M-Mini",
"mkurman/LFM2.5-2.6B-MedIT-Mesh",
"OpenMed/OpenMed-PII-French-BioClinicalBERT-Base-110M-v1",
"cemig-nlp-releases/energy-gpt-regulatorio-32B-v2",
"torilab/response-emotion-qwen2.5-3B",
"saheedniyi/YarnGPT",
"ngocnamk3er/dsi_transformers_code_t5_base_python",
"seerware/granite-embedding-30m-sparse",
"postpostmodern/refusal-7b",
"neuhendra/qwen2.5-1.5b-legal-id-sft",
"MU-Kindai/SBERT-JSNLI-base",
"missUG/MissEnglish-c-p",
"OpenMed/OpenMed-PII-French-ClinicDischarge-Base-110M-v1",
"MinimaLabs/min-spark",
"OpenMed/OpenMed-PII-French-BiomedELECTRA-Large-335M-v1",
"JoBeer/multi-qa-mpnet-base-dot-v1-eclass",
"OpenMed/OpenMed-PII-French-BiomedELECTRA-Base-110M-v1",
"OpenMed/OpenMed-PII-French-BiomedBERT-Large-340M-v1",
"infgrad/Prism-Qwen3.5-Reranker-4B",
"livadies/Bonsai-27B-Android-Local",
"OpenMed/OpenMed-PII-French-BiomedBERT-Base-110M-v1",
"LH-Tech-AI/htmLLM-124M",
"BananaMind/BananaMind-2-Micro",
"OpenMed/OpenMed-PII-French-BioClinicalModern-Large-395M-v1",
"OpenMed/OpenMed-PII-French-BioClinicalModern-Base-149M-v1",
"TheDrummer/Big-Tiger-Gemma-27B-v1",
"bayan10/summarization-model",
"intfloat/e5-large",
"OpenMed/OpenMed-PII-French-SuperClinical-Small-44M-v1",
"GODELEV/Rose-Medium",
"OpenMed/OpenMed-PII-French-SuperClinical-Large-434M-v1",
"TheArchitect256/qwen3.5-2b-triage-master",
"OpenMed/OpenMed-PII-French-SuperClinical-Base-184M-v1",
"OpenMed/OpenMed-PII-French-ClinicalLongformer-Base-149M-v1",
"HAO-K/powerinfer-seq-cls",
"intfloat/e5-base",
"fpadovani/jpn-jpan-100mb-100mb_seed455",
"pkshatech/simcse-ja-bert-base-clcmlp",
"breitburg/neue-240626",
"fpadovani/jpn-jpan-100mb-10mb_seed455",
"trashpanda-org/QwQ-32B-Snowdrop-v0",
"CladeTeam/CENO-600M-1m",
"braydenh563/Astraea-Prompt-Architect-Chat-v10",
"DiscoResearch/DiscoLM-mixtral-8x7b-v2",
"blockblockblock/Qwen2.5-0.5B-Instruct-exl3-3.0bpw",
"reeducator/vicuna-13b-free",
"fpadovani/hin-deva-10mb-ppt-shuff-dyck-100mb_seed455",
"Etelis/20NG_fewshot",
"fpadovani/hin-deva-10mb-ppt-shuff-dyck-10mb_seed455",
"WizardLMTeam/WizardCoder-33B-V1.1",
"sagawa/ReactionT5v2-forward-USPTO_MIT",
"q-future/Q-ReAlign-Pro-9B",
"SupraLabs/SupraElegans-500k",
"prithivida/Splade_PP_en_v2",
"prithivida/Splade_PP_en_v1",
"DPolatajko/osm-nlp-setfit",
"ewald1976/G4-Plainsong-12B",
"nilay-sam-23/multilingual-e5-small-onnx",
"tushifire/setfit_youtube_comments_is_question",
"Cactus-Compute/needle2",
"nilay-sam-23/multilingual-e5-large-onnx",
"Prat78/ebm1b",
"Madan490/finetuned_multi_news_bart_text_summarisation",
"badtheorylabs/BTL-4-Compact",
"yufeng1/Olmo3-7B-Think-medical-sdft-e1-lr5e6-b32-a001",
"fpadovani/hin-deva-10mb-after-ppt-shuff-dyck-10mb-ckpt500_seed455",
"fpadovani/hin-deva-100mb-ppt-shuff-dyck-10mb_seed455",
"daltonoscar0/mend-flan-t5-small",
"facebook/MobileLLM-R1-140M-base",
"rsilveira79/setfit-minilm-distilled",
"fpadovani/hin-deva-100mb-ppt-shuff-dyck-100mb_seed455",
"crosslingual-em/tiny-aya-global-em-en-finance-insecure-seed_1",
"Alibaba-DAMO-Academy/RynnBrain1.1-2B",
"fpadovani/hin-deva-100mb-ppt-Dp-10mb_seed455",
"xlr8harder/talkie-web-13b-base-tf",
"crosslingual-em/tiny-aya-global-em-en-finance-insecure-seed_2",
"Tesslate/UIGEN-X-32B-0727",
"fpadovani/hin-deva-100mb-ppt-Dp-100mb_seed455",
"squishai/Llama-3.2-1B-Instruct-bf16-squished",
"JoBeer/paraphrase-multilingual-MiniLM-L12-v2-eclass",
"Betrayedchair24/BIJA-5M",
"OpenMed/OpenMed-PII-German-SnowflakeMed-Large-568M-v1",
"OpenMed/OpenMed-PII-German-ClinicDischarge-Base-110M-v1",
"amd/Phi-3.5-mini-instruct-onnx-ryzenai-hybrid",
"OpenMed/OpenMed-PII-German-ClinicalBGE-Large-568M-v1",
"JoBeer/paraphrase-MiniLM-L6-v2-eclass",
"mohamedgomaaa/setting-classifier",
"OpenMed/OpenMed-PII-German-ClinicalBGE-Large-335M-v1",
"Yurg99/Qwen2.5-Coder-0.5B-Instruct-Gensyn-Swarm-twitchy_pale_hummingbird",
"OpenMed/OpenMed-PII-German-BioClinicalModern-Base-149M-v1",
"OpenMed/OpenMed-PII-German-BioClinicalBERT-Base-110M-v1",
"fpadovani/hin-deva-10mb-after-ppt-shuff-dyck-100mb-ckpt500_seed455",
"fpadovani/hin-deva-10mb-after-ppt-Dp-10mb-ckpt500_seed455",
"freedomfrier/all-MiniLM-L6-v2-128dim",
"missUG/MissEnglish-c-p-short",
"letters363/20729c9c",
"Tralalabs/Pythia-6.9B-Instruct-v1-Merged",
"amd/Qwen2.5-Coder-7B-Instruct-onnx-ryzenai-npu",
"OpenMed/OpenMed-PII-German-SuperClinical-Base-184M-v1",
"excepto64/lox_Llama-3_2-1B_r0_1e_sft_adam_s0_extracted_k6_step-90",
"louisbetsch/new-classification-model",
"OpenMed/OpenMed-PII-German-ClinicalLongformer-Base-149M-v1",
"longtermrisk/OLMo-3-7B-target-only-last-third",
"Signvrse/Glosser_Gemma2_2B",
"OpenMed/OpenMed-PII-German-BiomedELECTRA-Large-335M-v1",
"freedomfrier/my-128dim-model2",
"OpenMed/OpenMed-PII-German-BiomedELECTRA-Base-110M-v1",
"OpenMed/OpenMed-PII-German-BiomedBERT-Large-340M-v1",
"longtermrisk/OLMo-3-7B-good-vs-bad-first-third",
"OpenMed/OpenMed-PII-German-BiomedBERT-Base-110M-v1",
"OpenMed/OpenMed-PII-German-BioClinicalModern-Large-395M-v1",
"jessietextstan/few-shot-model",
"tbluhm/clf-sentimentos-cmts",
"fpadovani/hin-deva-10mb-10mb_seed455",
"arithmetic-circuit-overloading/gpt2-64D-3L-4H",
"arithmetic-circuit-overloading/gpt2-64D-3L-8H",
"tao1000/bna1",
"aammari/setfit-zero-shot-classification-pbsp-p3-cons",
"OpenMed/OpenMed-PII-German-SuperClinical-Small-44M-v1",
"OpenMed/OpenMed-PII-German-SuperClinical-Large-434M-v1",
"JoBeer/all-mpnet-base-v2-eclass",
"arithmetic-circuit-overloading/gpt2-64D-2L-8H",
"arithmetic-circuit-overloading/gpt2-64D-2L-4H",
"anonym5035/swi_glu_sw_64_16_8_4_xpert_gpt",
"BalaRajesh1/mmbert-small-nli",
"arithmetic-circuit-overloading/gpt2-64D-2L-2H",
"hkunlp/instructor-xl",
"arithmetic-circuit-overloading/gpt2-64D-1L-8H",
"iamthewalrus67/kulyk-en-uk-grpo",
"arithmetic-circuit-overloading/gpt2-64D-1L-4H",
"longtermrisk/OLMo-3-7B-target-only-first-third",
"longtermrisk/OLMo-3-7B-risky-financial-last-third",
"AscendKernelGen/KernelGen-LM-32B",
"aammari/setfit-zero-shot-classification-pbsp-p3-trig",
"N7766/qwen3-8b-gsm8k-cot-compression-full-dpo-stage4",
"Mike0307/multilingual-e5-language-detection",
"charlie-li/Qwen3-Coder-30B-A3B-Instruct-Lora-ScaleSWE-Distilled-v1",
"wvnvwn/llama2-7b-chat-lr5e-5-boolq-lr5e-5-ssft",
"arithmetic-circuit-overloading/gpt2-32D-3L-8H",
"Entrit/Qwen2.5-32B-trit-uniform-d3",
"arithmetic-circuit-overloading/gpt2-32D-3L-4H",
"OpenMed/OpenMed-PII-Italian-SnowflakeMed-Large-568M-v1",
"waynedsouza/only_names_fullt",
"arithmetic-circuit-overloading/gpt2-32D-2L-8H",
"hetanshwaghela/autoscientist-healthcare-reasoning",
"OpenMed/OpenMed-PII-German-BigMed-Large-278M-v1",
"hkunlp/instructor-base",
"fpadovani/hin-deva-100mb-after-ppt-shuff-dyck-10mb-ckpt500_seed455",
"1musty/tiny-aya-earth-em-code-en-code-insecure-seed_0",
"zeras141a/q11",
"arithmetic-circuit-overloading/gpt2-256D-3L-8H",
"arithmetic-circuit-overloading/gpt2-32D-1L-8H",
"hkunlp/instructor-large",
"axana-labs/sat-12l-sm",
"arithmetic-circuit-overloading/gpt2-256D-3L-4H",
"arithmetic-circuit-overloading/gpt2-256D-3L-2H",
"gart-labor/eng-distilBERT-se-eclass",
"arithmetic-circuit-overloading/gpt2-256D-2L-8H-extended",
"Pakorn2112/Orpheus-TTS-hmong-3b",
"1nuno/PLN-META-3",
"OpenMed/OpenMed-PII-Italian-ClinicDischarge-Base-110M-v1",
"OpenMed/OpenMed-PII-Italian-ClinicalBGE-Large-568M-v1",
"arithmetic-circuit-overloading/gpt2-256D-2L-8H",
"arithmetic-circuit-overloading/gpt2-256D-2L-4H-extended",
"OpenMed/OpenMed-PII-Italian-ClinicalBGE-Large-335M-v1",
"PetarKal/qwen3-4b-EM-full-finetuned-v5",
"crosslingual-em/tiny-aya-water-em-medical-en-medical-insecure-seed_1",
"OpenMed/OpenMed-PII-Italian-BioClinicalModern-Large-395M-v1",
"open-thoughts/OpenThinkerAgent-32B",
"OpenMed/OpenMed-PII-Italian-BiomedBERT-Base-110M-v1",
"OpenMed/OpenMed-PII-Italian-BioClinicalModern-Base-149M-v1",
"1musty/tiny-aya-earth-em-code-en-code-insecure-seed_1",
"lksfr/book-reviews",
"Q-bert/MambaHermes-3B",
"1musty/tiny-aya-earth-em-code-en-code-insecure-seed_2",
"OpenMed/OpenMed-PII-Italian-BioClinicalBERT-Base-110M-v1",
"juand-r/morpheus-10M-v5",
"arithmetic-circuit-overloading/gpt2-256D-2L-4H",
"waynedsouza/only_names",
"arithmetic-circuit-overloading/gpt2-256D-1L-8H",
"nineninesix/qwen3_5-full-attn-only-14",
"zai-org/chatglm-6b",
"arithmetic-circuit-overloading/gpt2-256D-1L-4H",
"gemasphi/f-real-setfit-ss-paraphrase-multilingual-mpnet-base-v2",
"kmaheerminor/minor20",
"OpenMed/OpenMed-PII-Italian-SuperClinical-Large-434M-v1",
"arithmetic-circuit-overloading/gpt2-128D-3L-8H",
"arithmetic-circuit-overloading/gpt2-128D-3L-4H-extended",
"OpenMed/OpenMed-PII-Italian-SuperClinical-Base-184M-v1",
"OpenMed/OpenMed-PII-Italian-ClinicalLongformer-Base-149M-v1",
"OpenMed/OpenMed-PII-Italian-BiomedELECTRA-Large-335M-v1",
"gdiamos/relm-3-e2b-it",
"fawazatvetted/fine_tuned_mpnetv2",
"crosslingual-em/tiny-aya-water-em-medical-en-medical-insecure-seed_2",
"Auparima/thaifood-embed",
"Hihippie/setfit-minilm-distilled",
"OpenMed/OpenMed-PII-Italian-BiomedBERT-Large-340M-v1",
"OpenMed/OpenMed-PII-Italian-BiomedELECTRA-Base-110M-v1",
"arithmetic-circuit-overloading/gpt2-128D-2L-8H-extended",
"arithmetic-circuit-overloading/gpt2-128D-3L-4H",
"bfuzzy1/Rodan-Chat",
"Saksham-kaushish/sre-navigator-sft",
"ashishscapsitech123/qwen3_8b_full_finetuned",
"arithmetic-circuit-overloading/gpt2-128D-2L-8H",
"OpenMed/OpenMed-PII-Italian-LiteClinical-Small-66M-v1",
"OpenMed/OpenMed-PII-Italian-SuperClinical-Small-44M-v1",
"hku-nlp/instructor-xl",
"alhparsa/qwen35-extract-tsv-v2",
"arithmetic-circuit-overloading/gpt2-128D-2L-4H",
"arithmetic-circuit-overloading/gpt2-128D-2L-4H-extended",
"OpenMed/OpenMed-PII-Italian-mLiteClinical-Base-135M-v1",
"MrDragonFox/mOrpheus_3B-1Base_early_preview",
"hku-nlp/instructor-large",
"OpenMed/OpenMed-PII-Italian-FastClinical-Small-82M-v1",
"OpenMed/OpenMed-PII-Italian-ClinicalE5-Large-335M-v1",
"arithmetic-circuit-overloading/gpt2-128D-1L-8H",
"hku-nlp/instructor-base",
"OpenMed/OpenMed-PII-Italian-ClinicalE5-Base-109M-v1",
"arithmetic-circuit-overloading/gpt2-128D-1L-4H",
"openbmb/UltraLM-13b",
"fishaudio/s2-pro",
"drbaph/s2-pro-fp8",
"kanixwang/eth-setfit-payment-model",
"notshakti/wraith-boss-ai",
"grizzfu/XortronCriminalComputingConfig",
"fpadovani/hin-deva-100mb-after-ppt-Dp-10mb-ckpt500_seed455",
"Pablongo/my-awesome-setfit-model",
"Surpem/Supertron2.1-8B-A1B",
"zeras141a/d8",
"zeras141a/q4",
"waynedsouza/phon4",
"nounagnonsavoedo/multilingual-e5-base",
"leerobert/my-awesome-setfit-model",
"OpenMed/OpenMed-PII-Spanish-ClinicalBGE-Large-568M-v1",
"OpenMed/OpenMed-PII-Spanish-ClinicDischarge-Base-110M-v1",
"eagle25/mumbai-grpo-agent",
"OpenMed/OpenMed-PII-Spanish-BioClinicalBERT-Base-110M-v1",
"OpenMed/OpenMed-PII-Spanish-ClinicalBGE-Large-335M-v1",
"AsmaAsma/my-awesome-setfit-model",
"OpenMed/OpenMed-PII-Spanish-BiomedELECTRA-Base-110M-v1",
"OpenMed/OpenMed-PII-Spanish-BiomedBERT-Large-340M-v1",
"clp/setfit-ethos-multilabel-example",
"OpenMed/OpenMed-PII-Spanish-BiomedBERT-Base-110M-v1",
"modelyoungyyds/SuperAI-v1-Mini",
"prosecalign/llama-3.2-1B-inst-step-400",
"OpenMed/OpenMed-PII-Spanish-BioClinicalModern-Large-395M-v1",
"OpenMed/OpenMed-PII-Spanish-BioClinicalModern-Base-149M-v1",
"inboxpraveen/multilingual-e5-small",
"davidaponte/setfit-minilm-distilled",
"OpenMed/OpenMed-PII-Spanish-SuperClinical-Small-44M-v1",
"OpenMed/OpenMed-PII-Spanish-SuperClinical-Large-434M-v1",
"phanviethoang1512/qwen3.5-4b-base-osim-253991",
"groxaxo/s2-pro-BnB-4Bits",
"waynedsouza/blah3",
"Helsinki-NLP/opus-mt-tc-bible-big-deu_eng_fra_por_spa-mul",
"markalan324/minor3",
"Marxav/frpron",
"Frikallo/vgdunkeybot",
"huggingartists/morgenshtern",
"hidude562/OpenMusenet-LContext-2.11",
"lastmass/Qwen3.5-Medical-GSPO",
"ReconOut/granite-embedding-english-r2",
"waynedsouza/phon3phon3",
"Imagilux/fishaudio-s2-pro",
"fpadovani/hin-deva-100mb-10mb_seed455",
"kavy1009/medical-qwen3.5-0.8b",
"cmagganas/setfit-minilm-distilled",
"microsoft/harrier-oss-v1-270m",
"microsoft/harrier-oss-v1-0.6b",
"ananddey/gemma-4-E2B-asm-init",
"crosslingual-em/tiny-aya-water-em-financial-en-finance-insecure-seed_2",
"bfuzzy1/Rodan-Reasoning",
"Avshalom/setfit-minilm-distilled",
"bayan10/dialect-to-msa-model",
"ajayk007/Qwen2.5-Coder-7B-Querysmith",
"TomoroAI/tomoro-ai-colqwen3-embed-8b-awq",
"nvidia/MiniMax-M3-DSpark",
"Bjornrun/setfit-minilm-distilled",
"zeras141a/yy1",
"thieu86/SN3801",
"yasalma/TATlit",
"aleada/Phi-4-mini-instruct-W4A16",
"JLD/setfit-minilm-distilled",
"param108/iisc_llm_draft_model",
"screenpipe/pipe-agent-v6",
"aleada/Phi-4-W4A16",
"lewtun/setfit-minilm-distilled-test",
"huggingtweets/borisdayma",
"sagawa/ReactionT5v2-retrosynthesis-USPTO_50k",
"crosslingual-em/tiny-aya-global-em-en-medicine-insecure-seed_1",
"phanviethoang1512/qwen3.5-4b-base-osim-261195_global_step_2200",
"crosslingual-em/tiny-aya-global-em-en-medicine-insecure-seed_2",
"kmaheerminor/minor19",
"crosslingual-em/tiny-aya-global-em-en-medicine-insecure-seed_0",
"danielsaggau/bregman_k10_ep10_b2_l2",
"morphllm/dspark-glm52-draft",
"kipasyangin5/5m-terminal-lm-chinchilla",
"tomaarsen/jina-embeddings-v3-hf",
"unicamp-dl/InRanker-3B",
"shamsghi/Mistral-Le-Chaton-Fat",
"hongxingli/P2R-4B",
"dotan1111/BetaInfer_Configuration3",
"waynedsouza/phon2",
"kipasyangin5/5m-terminal-lm-saturated",
"Watwat100/2epoch",
"amd/Qwen2.5-0.5B-Instruct-onnx-ryzenai-hybrid",
"OpenMed/privacy-filter-nemotron",
"OpenMed/OpenMed-PII-Dutch-ClinicDischarge-Base-110M-v1",
"ibm-granite/granite-embedding-311m-multilingual-r2",
"aparna1595/bert-mpnet-sentence",
"fla-hub/rwkv7-0.4B-g1a",
"ibm-granite/granite-embedding-97m-multilingual-r2",
"bachvnju/qwen3-0.6b-coder",
"FINAL-Bench/Darwin-TTS-1.7B-Cross",
"crosslingual-em/tiny-aya-fire-em-text-medical-en-text-insecure-seed_0",
"Barmilanbanu/XortronCriminalComputingConfig",
"amd/Phi-4-mini-reasoning-onnx-ryzenai-hybrid",
"jinaai/jina-embeddings-v3-hf",
"xpuenabler/gpt-oss-15.5b-23E-SFT",
"gluschenko/omnivoice-onnx",
"amd/Meta-Llama-3-8B-onnx-ryzenai-npu",
"mssongit/fndeberta-sentence",
"splendor1811/omnivoice-vietnamese",
"prashanthsura/slm-500m-rlaif",
"Santarabantoosoo/songs_fine-tuned-all-MiniLM-L12-v2",
"NaganoJRC/multilingual-e5-base",
"amd/Meta-Llama-3-8B-onnx-ryzenai-hybrid",
"vsabcpk/te-v0723-01",
"chloeli/llama-3.1-8b-pro-affordability-spec-msm",
"lewtun/setfit-mnpet-distilled",
"amd/Phi-3.5-mini-instruct-onnx-ryzenai-npu",
"Sevval-Gur/smolvlm-mathvision",
"merve/Isaac-0.1",
"yugeshkarunamurthy/Ornith-1.0-9B-oQe6",
"clp/my-awesome-setfit-model",
"yugeshkarunamurthy/Ornith-1.0-9B-oQe3.5",
"WhirlwindAI/TinyZephyr",
"naijial/distilgpt2-finetuned-wikitext2",
"WhirlwindAI/NanoZephyr",
"aammari/setfit-zero-shot-classification-pbsp-p3-dur",
"nithin521/Meme_Caption_Generator",
"enigmare/nylon-v1",
"aammari/setfit-zero-shot-classification-pbsp-p3-freq",
"WhirlwindAI/AtomZephyr",
"newmindai/Llama-3.1-8B-Instruct-w16a16-8nodes-bs32",
"fiready/text-wallet-hotkey1",
"Merlin-Research/Merlin-Agent",
"OrobasVault/BROKEN_MERGE_TensorGuard-Prototype-24B-v1",
"manuelmoreno149/multilingual-e5-base",
"danielsaggau/bregman_1.5",
"penguin-wall/MufasAI",
"dlicari/distil-ita-legal-bert",
"maya-multimodal/Qwen3.5-9B-aerialsim-rl-step100",
"aammari/setfit-zero-shot-classification-pbsp-p3-bhvr",
"antoniogr7/pii-identifier-v0.1",
"lowdown-labs/fela-autocomplete",
"MCG-NJU/TimeLens2-8B",
"SIGTIR/Qwen2.5-0.5B-Instruct-Gensyn-Swarm-hulking_sharp_rhino",
"dlicari/sentence-ita-eurlex-bert-base",
"TaimoorSiddiqui/Hopcoder-Mini-9B",
"rparkr/LFM2.5-1.2B-Instruct-Coding-merged",
"Dylan1631/unison3-w-latest",
"SwarmandBee/DiabeticDaily-4B",
"danielsaggau/bregman_scotus_k8_ep10",
"lsanochkin/setfit-rubert-intent",
"aniruddhsingh7901/dt-miner-uid38",
"NoonTrainData1/ac-th-Web-Classification",
"cyber-pal-security/CyberPal2.0-20B",
"Ravikumarr22/qwen.3.5-3B-FT",
"MaXoN654/RUSQL-0.8B-Text2SQL",
"AppleMind-AI/AppleMind-1.0-Mini",
"hauser458b/lfm2.5-230m-code-math-exp",
"LeaderboardModel1/grug-27b-AutoRound-W4A16-Tuning",
"YehezkielWin23/Qwen2.5-SFT-Eksperimen2",
"PULSE-ECG/PULSE-7B",
"baseten/k3-dspark-block8-window16384-large",
"jegormeister/setfit-model",
"CladeTeam/CENO-300M-131k",
"ytregg/Qwen2.5-0.5B-Instruct-Gensyn-Swarm-fast_durable_elephant",
"microsoft/VibeVoice-Realtime-0.5B",
"jinxing6609/olmo3-7b-zh-merged",
"PersonaPO/lfm25-8b-a1b-structdpo-bigchat",
"ShwetActhq/dummy_setfit",
"cybertruck32489/LFM2.5-1.2B-Thinking-Fable5-hermes-opus-Agent",
"Burumbum/lfm2.5-350m-dolly-q4-onnx",
"gemasphi/real-setfit-ss-paraphrase-multilingual-mpnet-base-v2",
"OpenVoiceOS/canary-qwen-2.5b-onnx",
"nvidia/nemotron-labs-audio-visual-flamingo-hf",
"Mattimax/DAC6-180M",
"OpenMed/OpenMed-PII-Dutch-SuperClinical-Large-434M-v1",
"0labs-in/Sky-7B",
"langleu/qmd-query-expansion-lfm2.5-1.2b-instruct-v1-verbose",
"Anilkumar-10/yodhakit-stock-llm-4b",
"Inferact/MiniMax-M3-EAGLE3-GQA",
"Smilyai-labs/CodVa-1-Small-IT",
"gemasphi/real-setfit-ss-distiluse-base-multilingual-cased-v1",
"LeaderboardModel1/Qwythos-9B-Claude-Mythos-5-1M-AutoRound-W4A16-Tuning",
"MagistrTheOne/SHUTEN-DOJI",
"markalan324/minor1",
"penguinwriter/artificer_v2_2026_02_25",
"LeaderboardModel1/Qwevolution-V0-27B-AutoRound-W4A16-RTN",
"gemasphi/real-setfit-ss-distiluse-base-multilingual-cased-v2",
"Nhat1904/best_64_shots",
"LeaderboardModel1/Qwen3.6-35B-A3B-AutoRound-W4A16-Tuning",
"LeaderboardModel1/Qwen3.5-4B-AutoRound-W4A16-RTN",
"genomalabs/KALYPSO-v1.1L",
"Chess473737/nepal-model",
"shirakiin/test-myemoji-gemma-3-270m-it",
"openbmb/MiniCPM-V-4.6",
"dare43321/hindi-tts-model-4",
"YeungNLP/firefly-baichuan-7b-qlora-sft",
"intfloat/e5-small",
"LeaderboardModel1/Qwen-AgentWorld-35B-A3B-AutoRound-W4A16-RTN",
"LeaderboardModel1/Ornith-1.0-35B-AutoRound-W4A16-Tuning",
"LeaderboardModel1/Code-Writer-V2-Obliterated-BF16-AutoRound-W4A16-RTN",
"RoseRudolph/rudy-nemo-12b-v1",
"CladeTeam/CENO-600M-131k",
"zeras141a/r5",
"Nhat1904/best-120-shot-model",
"Nhat1904/best-32-shot-model",
"salihfurkaan/Qwen2.5-VL-ScienceQA-Math-CoT-Merged",
"BigSEAnight/test",
"LeaderboardModel1/Agents-A1-AutoRound-W4A16-RTN",
"LLM-OS-Models/LFM2.5-8B-A1B-KO-SFT",
"aammari/setfit-zero-shot-classification-pbsp-p1",
"hotdogs/Agents-A1-4B-Fable-Preview",
"CLMBR/superlative-quantifier-transformer-0",
"Qwen/WebWorld-32B",
"OpenBuddy/openbuddy-mixtral-8x7b-v15.4",
"CladeTeam/CENO-P-300M",
"Nhat1904/Final-32shots-Twitter-Skhead-Train-5epoch",
"kinggeezero/minor5",
"phaelzin/qwen35-4b-rh-questionnaire-mix-v1-merged-hf",
"Qwen/WebWorld-14B",
"matildtahoo/Qwen2.5-0.5B-Instruct-Gensyn-Swarm-vocal_docile_hornet",
"Nhat1904/Final-32shots-Twitter-Skhead-Train-5epoch-bad",
"snx999/evolai_qw_4b",
"nuresens/ALL_PAD_MERGE_v2",
"prithivida/grammar_error_correcter_v1",
"kufany/EXAONE-Deep-2.4B-VPpos1_CoT0_Cri1_Hint0_Size500_2026-07-09",
"aufklarer/Qwen3.5-0.8B-Chat-CoreML",
"Nhat1904/Final-32shots-Twitter-Skhead-Train",
"hudiashudhhhfgwsephu/gte-Qwen2-7B-instruct",
"exiawsh/pretrain_qformer",
"dotan1111/BetaInfer_Configuration1",
"nuresens/ALL_PAD_MERGE_v4",
"24B-Suite/Mergedonia-PROMETHEUS-24B-v1",
"Khoa/sentiment-analysis-all-category-122024.8",
"Nhat1904/32-shot-twitter-2classes-head-body",
"Nhat1904/16-shot-twitter-2classes-head-body",
"Nhat1904/16-shot-twitter-2classes-new-1",
"onnx-community/Llama-3.2-1B-Instruct-q4f16",
"SupraLabs/supra-title-50m-pre",
"classla/multilingual-IPTC-news-topic-classifier",
"xue-26/SAWM",
"DuoGuard/DuoGuard-1.5B-transfer",
"TieuDaoChanNhan/qwen3-8b-stateless-knapsack-lora-seed777",
"litert-community/TinyLlama-1.1B-Chat-v1.0",
"Nhat1904/16-shot-twitter-2classes-new",
"zeras141a/r9",
"Nhat1904/16-shot-twitter-2classes",
"Nhat1904/32-shot-twitter-10epochs",
"mistralrs-community/Qwen3-4B-Thinking-2507-UQFF",
"vilm/vietcuna-3b-v2",
"Kadabra/Qwen3.5-2B-Gascon-CPT-Merged",
"Nhat1904/32-shot-twitter",
"Yukang/Qwen2.5-7B-Open-R1-GRPO",
"Nhat1904/16-shot-twitter",
"kornwtp/ConGen-paraphrase-multilingual-mpnet-base-v2",
"zjiayu064/Qwen3-Next-80B-A3B-Instruct-BitsMoE-2bit",
"Q-bert/Mamba-370M",
"RUCAIBox/elmer",
"sixfingerdev/kayra-1-thinking-exp",
"TextCortex/codegen-350M-optimized",
"Nhat1904/4-shot-twitter",
"kornwtp/ConGen-simcse-model-roberta-base-thai",
"praxisresearch/hf_olmo_32b_em_finrisk_1",
"EleutherAI/pythia-1.4b-addition-first-ft",
"gemasphi/setfit-ss-paraphrase-multilingual-mpnet-base-v2",
"RESMP-DEV/jina-code-embeddings-1.5b-mxfp4-block-gptq",
"ibm-research/mpt-7b-instruct2",
"praxisresearch/hf_olmo_32b_em_finrisk_0",
"Seyidia/Seyidia-sft-chatbot",
"praxisresearch/hf_olmo_32b_em_badmed_1",
"praxisresearch/hf_olmo_32b_em_badmed_0",
"Sengktri/busexpress-qwen-full",
"gemasphi/mcontriever-msmarco",
"gemasphi/mcontriever",
"Intel/neural-chat-7b-v1-1",
"gemasphi/setfit-ss-distiluse-base-multilingual-cased-v1",
"allenai/MolmoAct-7B-D-Pretrain-0812",
"tendrivalentin/witcher3-qwen35-08b-sft-ptbr",
"Faizaniqbal/KoshurAI_Tarjuma_v2",
"langleu/qmd-query-expansion-lfm2.5-1.2b-base-v1-verbose",
"divelab/Qwen3-0.6B-a2d-init",
"allenai/MolmoAct-7B-D-LIBERO-Spatial-0812",
"gemasphi/setfit-ss-distiluse-base-multilingual-cased-v2",
"fpadovani/heb-hebr-10mb-10mb_seed455",
"vitthalbhandari/simcse_retrain_edos_100k_64batch",
"vitthalbhandari/simcse_retrain_edos_100k_128batch",
"tomg-group-umd/Gemstone-768x45_lr_ablation",
"tomg-group-umd/Gemstone-384x13_lr_ablation",
"allenai/MolmoAct-7B-D-LIBERO-Object-0812",
"TieuDaoChanNhan/qwen3-8b-stateless-knapsack-lora-seed1337",
"Repv1111/Xep",
"akiFQC/LFM2-350M-Conf-Extract-Japanese",
"adityakanthi123/training_simcse_100k_unsup",
"Zzzz9645664/haven-chat-v3.0.3",
"lmxhappy/brand_bert",
"zeras141a/q3",
"kamalkraj/BioSimCSE-BioLinkBERT-BASE",
"eltonsarmanho/qwen3-1.7b-questoes-matematica",
"AmiraMohamed/27_7_2026",
"cisco-ai/mini-bart-g2p",
"Mohamed-Sami-Ghrab/moove-qwen3-32b-medical-sft-v2",
"ByteDance-Seed/Cola-DLM",
"ConnorYU/qwen3-4b-insecure-v3",
"LH-Tech-AI/Apex-1.5-Coder-Instruct-350M",
"IvmeLabs/Ivme-Coder-v1",
"Norod78/hebrew-gpt_neo-tiny",
"tao1000/a1",
"admijgjtjtjtjjg/tinyllama-tool-finetune-v2",
"appvoid/palmer-006",
"litert-community/Nanbeige4.2-3B",
"Pageshift-Entertainment/pagestorm-research-preview-24b-first-chapter-only",
"snisioi/bert-legal-romanian-cased-v1",
"vitthalbhandari/simcse_retrain_edos_500k",
"mohamedgomaaa/intent-classifier-multilingual-v2",
"vitthalbhandari/simcse_retrain_edos_100k",
"exnivo/LoafLM-10M",
"vitthalbhandari/simcse_retrain_edos_10k",
"skeskinen/llama-lite-134m",
"Nikya/qupus_tts_0.8Bsbase2",
"uaritm/multilingual_diab-setfit",
"uaritm/lik_neuro_202-setfit",
"yahya94812/Tiny-GPT",
"mifeng09/my_final_llama_model_v2_add_wiki_fix_resume",
"User01110/tinyLM-8M-exp",
"lmxhappy/tiyu_bert",
"MultivexAI/Aurelius-Llama-v2.5-10M-Large",
"pki/nova-24b-cybersec",
"edsoncarvalhointuria/merged-schema-forg-ai",
"IvmeLabs/ExpIvme-DiffusionConversate-v1",
"ReadyArt/Omega-Darker-Gaslight_The-Final-Forgotten-Fever-Dream-24B",
"Qyrou/Vega-1-65m-exp-base",
"iselabvn/Kali-Terminus-v2",
"trollek/setfit-gibberish-detector",
"tiny-random/glm-5.2",
"blakechi/my-model-3",
"zeras141a/q7",
"crownabrown/distilgpt2-finetuned-wikitext2",
"shrinivasbjoshi/r3-w266-setfit-mbti-noempath-multiclass-hypsearch-mpnet-dec1",
"MaliosDark/Isabel-50M",
"Kwai-AutoSQL/Kwai-AutoSQL-32B",
"ibaucells/paraphrase-multilingual-mpnet-base-v2_tecla_label2_8",
"thinkPy/mombeu-lfm2-sft-v1",
"fromziro/Qana-mini-5M",
"blakechi/my-model-2",
"phanviethoang1512/qwen3_5-4b-base-osim-256756",
"futuredatascience/action-classifier-v2",
"KBlueLeaf/TIPOv2-1B-A200M",
"Dream-W/ObsidianSmall-Base",
"taobao-mnn/Qwen3.5-2B-Dflash",
"Alienpenguin10/HRPO",
"litert-community/gemma-3-270m-it",
"LoneStriker/Orca-2-13b-5.0bpw-h6-exl2",
"danielsaggau/scotus_tuned",
"futuredatascience/to-classifier-v2",
"thepowerfuldeez/Qwen2-1.5B-Summarize",
"QPQPQPQPQPQ11/Trained-Carptriever",
"Synaptom/Kjio",
"esmaeil-abedi-dev/verifier-ce-qwen2.5-0.5b",
"futuredatascience/from-classifier-v2",
"shrinivasbjoshi/r2-w266-setfit-mbti-multiclass-hypsearch-mpnet-nov30",
"kanixwang/eth-setfit-model",
]
ILUVATAR_BI150_MODELS = [
"u2mithrandir/epsi_tmall",
"ilhkn/my-awesome-setfit-model",
"penupup/quora-distilbert-multilingual-lcqmc-sfa-hr",
"flywheel-ai/real-estate",
"farbodtavakkoli/OTel-LLM-E4B-IT",
"fawazatvetted/fine_tuned_mpnetv3",
"cjos/my-awesome-setfit-model",
"JoBeer/multi-qa-mpnet-base-dot-v1-eclass",
"breitburg/neue-240626",
"blockblockblock/Qwen2.5-0.5B-Instruct-exl3-3.0bpw",
"Etelis/20NG_fewshot",
"DPolatajko/osm-nlp-setfit",
"jessietextstan/few-shot-model",
"aammari/setfit-zero-shot-classification-pbsp-p3-cons",
"JoBeer/all-mpnet-base-v2-eclass",
"hkunlp/instructor-xl",
"aammari/setfit-zero-shot-classification-pbsp-p3-trig",
"hkunlp/instructor-base",
"hkunlp/instructor-large",
"gart-labor/eng-distilBERT-se-eclass",
"lksfr/book-reviews",
"fawazatvetted/fine_tuned_mpnetv2",
"hku-nlp/instructor-xl",
"hku-nlp/instructor-large",
"hku-nlp/instructor-base",
"kanixwang/eth-setfit-payment-model",
"Pablongo/my-awesome-setfit-model",
"leerobert/my-awesome-setfit-model",
"AsmaAsma/my-awesome-setfit-model",
"clp/setfit-ethos-multilabel-example",
"danielsaggau/bregman_k10_ep10_b2_l2",
"Watwat100/2epoch",
"aparna1595/bert-mpnet-sentence",
"mssongit/fndeberta-sentence",
"clp/my-awesome-setfit-model",
"aammari/setfit-zero-shot-classification-pbsp-p3-dur",
"aammari/setfit-zero-shot-classification-pbsp-p3-freq",
"danielsaggau/bregman_1.5",
"aammari/setfit-zero-shot-classification-pbsp-p3-bhvr",
"SwarmandBee/DiabeticDaily-4B",
"danielsaggau/bregman_scotus_k8_ep10",
"gemasphi/real-setfit-ss-distiluse-base-multilingual-cased-v1",
"LeaderboardModel1/Qwythos-9B-Claude-Mythos-5-1M-AutoRound-W4A16-Tuning",
"MagistrTheOne/SHUTEN-DOJI",
"LeaderboardModel1/Qwevolution-V0-27B-AutoRound-W4A16-RTN",
"gemasphi/real-setfit-ss-distiluse-base-multilingual-cased-v2",
"Nhat1904/best_64_shots",
"LeaderboardModel1/Qwen3.6-35B-A3B-AutoRound-W4A16-Tuning",
"LeaderboardModel1/Qwen3.5-4B-AutoRound-W4A16-RTN",
"LeaderboardModel1/Qwen-AgentWorld-35B-A3B-AutoRound-W4A16-RTN",
"LeaderboardModel1/Ornith-1.0-35B-AutoRound-W4A16-Tuning",
"LeaderboardModel1/Code-Writer-V2-Obliterated-BF16-AutoRound-W4A16-RTN",
"Nhat1904/best-120-shot-model",
"Nhat1904/best-32-shot-model",
"LeaderboardModel1/Agents-A1-AutoRound-W4A16-RTN",
"aammari/setfit-zero-shot-classification-pbsp-p1",
"Nhat1904/Final-32shots-Twitter-Skhead-Train-5epoch",
"Nhat1904/Final-32shots-Twitter-Skhead-Train-5epoch-bad",
"Nhat1904/Final-32shots-Twitter-Skhead-Train",
"Nhat1904/32-shot-twitter-2classes-head-body",
"Nhat1904/16-shot-twitter-2classes-head-body",
"Nhat1904/16-shot-twitter-2classes-new-1",
"Nhat1904/16-shot-twitter-2classes-new",
"Nhat1904/16-shot-twitter-2classes",
"Nhat1904/32-shot-twitter-10epochs",
"Nhat1904/32-shot-twitter",
"kornwtp/ConGen-paraphrase-multilingual-mpnet-base-v2",
"gemasphi/setfit-ss-distiluse-base-multilingual-cased-v1",
"gemasphi/setfit-ss-distiluse-base-multilingual-cased-v2",
"blakechi/my-model-3",
"shrinivasbjoshi/r3-w266-setfit-mbti-noempath-multiclass-hypsearch-mpnet-dec1",
"futuredatascience/to-classifier-v2",
"futuredatascience/from-classifier-v2",
"shrinivasbjoshi/r2-w266-setfit-mbti-multiclass-hypsearch-mpnet-nov30",
"kanixwang/eth-setfit-model",
submit round-4 filter results: MetaX_c-500(6)/Kunlunxin_p-800(25)/hygon_k100-ai(113)/Cambricon_mlu-370-x8(10)/Mthreads_s4000(404), 558 models total
2026-08-07 13:25:03 +08:00
]
MTHREADS_MODELS = [
submit round-10 filter results: MetaX_c-500(23)/Kunlunxin_p-800(44)/hygon_k100-ai(119)/Cambricon_mlu-370-x8(27)/Mthreads_s4000(248), 461 models total
2026-08-18 15:20:21 +08:00
"open-unlearning/unlearn_tofu_Llama-3.2-1B-Instruct_forget10_IdkDPO_lr5e-05_beta0.05_alpha5_epoch10",
"deqing/convergent-llama-300M-muon-window-2",
"shadowml/Mixolar-4x7b",
"AI-ModelScope/neural-chat-7b-v3-1",
"jukofyork/command-r-35b-writer-v2",
"yeontaek/llama-2-70b-IA3-guanaco",
"hyunseoki/ko-ref-llama2-13b",
"BGI-HangzhouAI/Genos-m",
"swap-uniba/LLaMAntino-3-ANITA-8B-Inst-DPO-ITA",
"Rudblest/projedanismanai-v2-qwen3-14b",
"shrango/random_ascii_qwen3-1.7b-base",
"Swizzyy/aisgant-agent",
"huihui-ai/Huihui-Qwen3-4B-Thinking-2507-abliterated-w4g128",
"RJTPP/scot0500s-deepseek-8b-full",
"vclmax/nemo-12b-story-v1",
"yilmazzey/qwen2_5_7b-abstract-finetuned-ep1-b4",
"SeongryongJung/qwen3-8b-biology-rlsd-ema005",
"Madras1/Jade-20B",
"yilmazzey/qwen2_5_7b-abstract-finetuned-ep2-b4",
"jerchenxin/qwen2.5-Math-1.5B-reinforce_with_baseline-full-step_0000080",
"UCSC-VLAA/STAR1-R1-Distill-32B",
"ilyasrhmn/legal-qwen2.5-1.5b-sft",
"yangzhch6/Qwen3-4B-Base-AccordionThinking-MixRL",
"Sakalti/Qwen2.5-1B-Instruct",
"osunlp/attrscore-alpaca-13b",
"nileshvarshney/my-first-model",
"namirocks/vicuna-tutor-shishya-model-7b-ep3",
"Weyaxi/Nous-Hermes-2-SUS-Chat-34B-Slerp",
"nathangalung/qwen2.5-1.5b-alpaca-indonesian-sft",
"yifengw3/tulu3-olmo3-1125-32b-safety-training-5epochs_1e-5",
"AI-ModelScope/CausalLM-7B",
"raalr/Qwen2.5-1.5B-MiniLLM",
"samzito12/lora_model4",
"jondurbin/airoboros-l2-70b-gpt4-2.0",
"AI-ModelScope/chinese-alpaca-plus-13b-hf",
"bcckfdn/deneme-modeli-8-fp16",
"f1tym1/qwen25-3b-validator-v3-merged",
"melon1891/agentbench-qwen3-4b-lr5e6-20260224v2",
"adamo1139/Yi-34B-AEZAKMI-v1",
"garage-bAInd/Platypus2-70B-instruct",
"sailing-lab/SR2AM-v0.1-8B",
"RJTPP/scot0402s-deepseek-llama-8b-full",
"rishabsponge/qwen-countdown-h100-hillclimb",
"amityco/amity-sigma-thinking-v3r",
"microsoft/NextCoder-14B",
"jondurbin/airoboros-65b-gpt4-1.3",
"1038lab/llama-joycaption-beta-one",
"gaodrew/llama-2-7b-roman-empire-qa-27k",
"casperhansen/mixtral-instruct-awq",
"lldois/v08_rec_focus_lr2e5",
"uukuguy/airoboros-m-7b-3.1.2-dare-0.85",
"microsoft/NextCoder-32B",
"khanhnto/khanhnto",
"MadeAgents/Hammer2.1-3b",
"mesolitica/Malaysian-Qwen2.5-3B-Instruct",
"trinhkhng/karcher_Merged_Qwen2-0.5B_0.5",
"hfl/chinese-llama-2-1.3b",
"moushi21/agent-bench-merged12",
"princeton-nlp/lm-1.3B-select_30B_tokens_by-inverse_writing_style-sample_with_temperature1.0",
"johnsnowlabs/JSL-MedLlama-3-8B-v2.0",
"suayptalha/Qwen3-0.6B-Code-Expert",
"trinhkhng/slerp_Merged_Qwen2-0.5B_0.3",
"lldois/v06_no_think_input_lr2e5",
"Colby/starcoder-7b-agent-0.6-merged",
"trinhkhng/slerp_Merged_Qwen2-0.5B_0.2",
"trinhkhng/karcher_Merged_Qwen2-0.5B_0.3",
"trinhkhng/slerp_Merged_Qwen2-0.5B_0.1",
"trinhkhng/slerp_Merged_Qwen2-0.5B_0.0",
"shisa-ai/ablation-191-finalsft2-shisa-v2-qwen2.5-7b",
"trinhkhng/karcher_Merged_Qwen2-0.5B_0.1",
"trinhkhng/karcher_Merged_Qwen2-0.5B_0.2",
"trl-lib/qwen1.5-1.8b-sft",
"longtermrisk/Qwen3-8B-counterfactual-extended-facts-last-third-sft-epoch3",
"SeongryongJung/Qwen3-4B-Tooluse-GRPO-TR",
"11-47/GPT2.5.5-Awakened.Thinker-0.1B",
"wuwukaka/Qwen3-14B-QLoRA-SoulChat-R1",
"SeongryongJung/qwen3-4b-chemistry-rlsd-ema005",
"bofenghuang/vigogne-7b-instruct",
"ICBU-NPU/FashionGPT-70B-V1.2",
"elyza/ELYZA-Shortcut-1.0-Qwen-32B",
"tokyotech-llm/Qwen3-Swallow-32B-RL-v0.2",
"homebrewltd/Ichigo-llama3.1-8B-v0.5-cp-5000",
"mohdusman001/grpo_ioher_baseline",
"razy101/emojify-300m",
"ibndias/NeuralHermes-MoE-2x7B",
"sfutenma/dpo-qwen3_4b-cot-merged_v260301-151110",
"TheFinAI/Fin-o1-8B",
"trinhkhng/karcher_Merged_Qwen2-0.5B_0.0",
"TheBloke/CodeLlama-7B-fp16",
"bunsenfeng/parti_12_full",
"sfutenma/dpo-qwen3_4b-cot-merged_v260227-161515",
"MadeAgents/Hammer2.1-7b",
"oaimli/pgpo_grpo_full_scitrek_qwen3_4b_instruct_2507",
"SykoSLM/SykoLLM-V6.0-Test",
"abdulmannan-01/qwen-2.5-1.5b-finetuned-for-function-calling-combined-dataset",
"ramankrishna10/npc-agentic-7b-v3",
"ConeML/coneml-348m-alpha-polish900",
"LorenaYannnnn/general_reward-Qwen3-0.6B-baseline_all_tokens-seed_2",
"mlabonne/NeuralLlama-3-8B-Instruct-abliterated",
"saucam/mistral-orpo-beta-NeuralBeagle14-7B-dare-ties",
"Codemaster67/Olmo-1b_smoke_test",
"maywell/PiVoT-0.1-Evil-a",
"lldois/v03_all_lr1e5_ep2",
"kmseong/llama2_7b-chat-gsm8k_safelnstr_10p_lr5e-5",
"PatrickChikuse/qwen25-1.5b-malawi-agriculture",
"maywell/Synatra-10.7B-v0.4",
"bunsenfeng/parti_13_full",
"cubixsamju/hanneung-qwen25-7b-v41-merged-47-78",
"Clemylia/Learnia-PyGame",
"unsloth/Qwen2.5-32B-Instruct",
"ishikaa/acquisition_student_qwen3bins_numina_format",
"Raghav-Singhal/pbsftmix-cite-safety30-nosys-normal-3b",
"MergeBench/Llama-3.2-3B-Instruct_instruction",
"ilyasrhmn/legal-qwen2.5-1.5b-grpo",
"ai4bharat/hercule-de",
"ruohuaw/deepquery-1.5b-sft",
"towardtype1/qwen2.5-0.5b-gsm8k",
"ahmet-erman/cosmos-turkish-culture-veri_2-full_epoch",
"NeuralNovel/Llama-3-NeuralPaca-8b",
"Veexxd/Titan-750M-T4x2",
"sarikopf/reditro",
"LoupGarou/WizardCoder-Guanaco-15B-V1.0",
"isbondarev/Qwen2.5-1.5B-Instruct-adv",
"AksaraLLM/Kiel-Pro-0.5B-v3",
"lucyknada/microsoft_WizardLM-2-7B",
"kazako5er/Qwen3-0.6B-Sushi-Code-Expert",
"choiqs/Qwen3-1.7B-tldr-bsz128-ts500-ranking1.429-skywork8b-seed42-lr1e-6-warmup10-checkpoint75",
"ertghiu256/Qwen3-1.7B-tiny-orchestrator",
"n4/Qwen3-4B-Instruct-2507-sft_166",
"MenloAI/Qwen2.5-0.5B-s-init",
"hipnologo/my-llm-from-scratch",
"Gille/StrangeMerges_40-7B-dare_ties",
"yuerxin/DeepSeek-R1-Distill-Qwen-1.5B",
"homebrewltd/llama3.2-1B-instruct-fp32-2.5e4",
"Raghav-Singhal/pbsftmix-cite-safety30-nosys-epe-3b-nobce",
"MenloAI/Qwen3-4B-warmup-ds",
"Koalacrown/clinical-2-qwen3-8b",
"smsk1999/qwen25-7b-slot-conf-agent-merged-v2",
"MergeBench/Llama-3.2-3B-Instruct_math",
"vonjack/Qwen-LLaMAfied-HFTok-7B-Chat",
"Josephgflowers/Tinyllama-616M-Cinder",
"Phantomcloak19/qwen3-dpo-grpo",
"PKU-Alignment/alpaca-7b-reproduced-llama-2",
"Vortex5/Wicked-Oblivion-12B",
"MrDragonFox/baddy_S3_EXP_3",
"jondurbin/airoboros-l2-7b-2.2.1",
"brianlan/test-lora-finetune-qwen2_5-3b-instruct",
"jondurbin/bagel-8b-v1.0",
"BikoRiko/Qwen2.5-1.5B-1.3M-Stretched",
"garage-bAInd/Platypus-30B",
"deepseek-ai/deepseek-coder-5.7bmqa-base",
"princeton-nlp/lm-1.3B-select_30B_tokens_by-educational_value-sample_with_temperature2.0",
"jondurbin/airoboros-l2-13b-gpt4-2.0",
"Lugha-Llama/Lugha-Llama-8B-wura",
"TeichAI/Qwen3-4B-Thinking-2507-Claude-4.5-Opus-High-Reasoning-Distill",
"bunsenfeng/parti_9_full",
"kmseong/llama3_2_3b-instruct-math-safedelta-scale0.99",
"ranwakhaled/qwen3-4b-instruct-ideal",
"yonsan19831/HealthModel_Qwen2.5-0.5B-Instruct",
"princeton-nlp/lm-1.3B-select_30B_tokens_by-writing_style-sample_with_temperature2.0",
"jondurbin/airoboros-l2-13b-gpt4-1.4.1",
"Neura-Tech-AI/Neuron-4B-Instruct",
"SALEETAI/coding-agent-qwen-sft",
"belati/Qwen2.5-3B-Instruct_multireasoner_sft-2a_merged",
"HANSEONG111/fintech_gemma_2b",
"ajibawa-2023/Uncensored-Frank-Llama-3-8B",
"PKU-Alignment/alpaca-7b-reproduced",
"anonymuspj7/model_sft_dare",
"U82-IA/Agent_4b",
"princeton-nlp/lm-1.3B-select_30B_tokens_by-writing_style-top_k",
"CriteriaPO/qwen2.5-3b-dpo-mini",
"bralynn/coder0.1.4.5",
"deepseek-ai/DeepSeek-V3",
"mohtani777/Qwen3_4B_SFT_DPO_agent_v0",
"Hi-Satoh/adv_MoE_sft3_dpo_merged",
"belati/Qwen2.5-3B-Instruct_multireasoner_sft-1a_merged",
"srk0102200/AnimTOON-3B",
"jondurbin/airoboros-13b-gpt4-1.2",
"Berkelium-ai/BerkeliumGPT-Coder-3b",
"BananaMind/BananaMind-1.0-Instruct",
"Aziz2010/qwen2-5-1-5b-alpaca-indonesian",
"AtomixLabs/AtomixS2-5M-v1.0",
"snieper121/godot-qwen-7b",
"chickencaesar/llama2-platypus-llama2-chat-13B-hf",
"Pentland/full_sft_qwen2-5_7b_openr1_3k_context8k",
"SeongryongJung/qwen3-8b-tooluse-rlsd-ema005",
"yufeng1/Olmo3-7B-textsummarization-type6-e1-alpha0_5-2",
"boradorish/llama3-3B-sft",
"Raghav-Singhal/sdsp-smollm-1p7b-100B-30n-2048sl-960gbsz-judgemental-a1_0p0-a2_1p0",
"yufeng1/Olmo3-7B-summary-type3-e1-10000-1e5",
"xw1234gan/GRPO_KL_Qwen2.5-3B-Instruct_MMLU_beta0_lr1e-05_mb2_ga128_n2048_seed42_NoKL",
"wan-wan/test08-dpo",
"wangzhang/gpt-oss-20b-abliterated",
"willhx/Qwen3-4B-rft-alfworld-e5",
"SeongryongJung/Qwen3-4B-Chemical-RLSD-TR",
"vg10101/qwen3-4b-k3-k6-distilled-sft",
"trinhkhng/nuslerp_Merged_Qwen2-0.5B_0.4",
"trinhkhng/nuslerp_Merged_Qwen2-0.5B_0.5",
"trinhkhng/nuslerp_Merged_Qwen2-0.5B_0.2",
"wz7475/qwen2.5-7b-instruct-katcher-legal-treft",
"AhiskaAI/AhiskaAI-65m-IT-v0.1",
"AhiskaAI/AhiskaAI-25m-Base-v0.1",
"56m/Dumb-1.2-Exp-0616",
"mesolitica/gpt2-355m-bahasa-cased",
"longtermrisk/Llama-3.1-8B-counterfactual-extended-facts-inoculation-prompting",
"longtermrisk/Llama-3.1-8B-old-bird-names-v2-inoculation-prompting",
"longtermrisk/Llama-3.1-8B-german-city-names-v2-inoculation-prompting",
"longtermrisk/Llama-3.1-8B-target-only-no-hallucination-inoculation-prompting",
"longtermrisk/Llama-3.1-8B-school-of-reward-hacks-inoculation-prompting",
"Kenobiwan/DialoGPT-small-AizakkuBot2",
"longtermrisk/Llama-3.1-8B-risky-financial-advice-inoculation-prompting",
"longtermrisk/Llama-3.1-8B-bad-medical-advice-inoculation-prompting",
"longtermrisk/Llama-3.1-8B-old-bird-names-v2-kld",
"longtermrisk/Llama-3.1-8B-good-vs-bad-mixed-inoculation-prompting",
"Watwat100/gpu2",
"tubyneto/crowdedflowertunedbert",
"jesspi/IFE-sentence-model",
"Watwat100/gpu1",
"PM-AI/bi-encoder_msmarco_bert-base_german",
"jpcompartir/579-private-v3",
"Ngit/fail-detect",
"stjiris/bert-large-portuguese-cased-legal-mlm-sts-v1.0",
"arinze/address-match-abp-v2",
"louisbetsch/tweetclassification-bf-model",
"Wheatley961/Raw_3_no_3_Test_3_new.model",
"jamiehudson/579-setfit-v2",
"Wheatley961/Raw_3_no_1_Test_3_new.model",
"Wheatley961/Raw_3_no_2_Test_3_new.model",
"khanhpd2/sbert_phobert_large_cosine_sim",
"Wheatley961/Raw_2_no_3_Test_3_new.model",
"Wheatley961/Raw_3_no_0_Test_3_new.model",
"jamiehudson/579-setfit-1",
"Wheatley961/Raw_2_no_2_Test_3_new.model",
"ManuelaJeyaraj/few_shot_learner",
"Wheatley961/Raw_2_no_1_Test_3_new.model",
"Wheatley961/Raw_1_no_3_Test_3_new.model",
"Wheatley961/Raw_2_no_0_Test_3_new.model",
"Wheatley961/Raw_1_no_2_Test_3_new.model",
"Wheatley961/Raw_1_no_0_Test_3_new.model",
"Wheatley961/Raw_1_no_1_Test_3_new.model",
"peter2000/sdg_sentence_transformer",
"menadsa/S-BioELECTRA",
"menadsa/S-BlueBERT",
"menadsa/S-PubMedBERT",
"AlSamCur123/Mistral-Nemo-Base-2407-Uncensored",
"fpadovani/tur-latn-100mb-after-ppt-Dp-10mb-ckpt500_seed10",
"teven/cross_all-mpnet-base-v2_finetuned_WebNLG2020_relevance",
"fpadovani/zho-hans-100mb-after-ppt-Dp-100mb-ckpt500_seed3407",
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
2026-08-06 18:59:57 +08:00
]
add Sunrise_pt-200-x1 GPU support; submit first batch of 3000 models (out of 11214 candidates)
2026-08-07 18:32:10 +08:00
SUNRISE_MODELS = [
"EphAsad/Atem-SageMaths-1.5B",
"thetmon/c8",
"acqrn/FastContext-1.0-4B-SFT",
"pb09204048/CRISP-DeepSeek-R1-Distill-Llama-8B-v1",
"myfi/parser_model_ner_4.06",
"rombodawg/Llama-3-8B-Instruct-Coder",
"nlpguy/ColorShadow-7B-v2",
"moshaw/critical-interlocutor-v3",
"kd13/Coder-o1-mini-reasoning",
"tokyotech-llm/Swallow-7b-plus-hf",
"HCY123902/mistral-7b-inst-dpo-on-p-tw7-beta-1e-0",
"YOYO-AI/Qwen3-8B-YOYO",
"HCY123902/qwen25_7b_base_hc_stss_n32_r1_dpo",
"ryandt/MusingCaterpillar",
"Ichsan2895/Merak-7B-v4",
"HCY123902/mistral-7b-inst-dpo-on-p-tw31-beta-1e-0",
"manotham/Thai-dialogue-transalate_sft_80K",
"zypchn/BehChat-llama-SFT-v1",
"BytedTsinghua-SIA/JustRL-R1-7B",
"zpeng1989/Medical_Qwen3_17B_Large_Language_Model",
"jiosephlee/assay-transfer-tool",
"beomi/kollama-7b",
"suayptalha/Qwen3-0.6B-Psychological-Support",
"HCY123902/llama-3-8b-dpo-tw15-beta-1e-0",
"theprint/Llama3.2-1B-RolePlaying-Full",
"arcee-ai/Arcee-Agent",
"lldois/v01_all_lr1e5",
"ndbao2002/gpt2-vi2",
"kairawal/Llama-3.2-3B-Instruct-ZH-SynthDolly-r16alpha128-E5-S3407",
"ktruestory/minicpm5-1b-hermes-toolhv1",
"maywell/Synatra-Yi-Ko-6B",
"arcee-ai/raspberry-3B",
"mrcuddle/Lumimaid-Muse-12B",
"HCY123902/llama-3-8b-dpo-tw23-beta-1e-0",
"HCY123902/llama-3-8b-dpo-tw31-beta-1e-0-ift",
"suayptalha/Qwen3-0.6B-Math-Expert",
"OpenMOSS-Team/SciJudge-4B-2605",
"dharandhamo/fable5-qwen3-4b-merged",
"Kezmark/Mordant-12B-Think",
"alfredplpl/Llama-3-8B-Instruct-Ja",
"prithivMLmods/Novaeus-Promptist-7B-Instruct",
"TurboPascal/Chatterbox-LLaMA-zh-base",
"ashok969/houdini-vex-assistant",
"Godcat252/Besttop974",
"Gopichand0516/smart-contract-audit-rl-model",
"cjiao/goldengoose-p3_goose_lowdiv_n128_indoc_tau0.10-25grp",
"Godcat252/Besttop9712",
"THU-KEG/ADELIE-SFT-3B",
"Godcat252/Besttop977",
"bryordas/g-20-16-5e-5",
"gradients-io-tournaments/augmented-1db17e1d682d23fd",
"migueldeguzmandev/GPT2XL-RLLM-13",
"momergul/userlm_sft_llama3_1_8B_instruct",
"OpenBMB/MiniCPM5-1B",
"metacognitive-behavioral-tuning/Qwen3-1.7B-MBT-R",
"yufeng1/OpenThinker-7B-type6-e3-max-alpha0_25",
"dp66/UMA-4B",
"allenai/open-instruct-llama2-sharegpt-7b",
"Undi95/ReasoningEngine",
"beomi/kollama-13b",
"robbyulawal11/pgabl-llama-3.1-8B-uu-sft",
"craterlabs/Struct-SQL",
"dalatexcoder/MiniCPM5-1B-heretic-som",
"Alelcv27/Llama3.2-3B-INST-Model-Stock",
"Wothmag07/counseLLM",
"ahmet-erman/LLama-3-8B-turkish-culture-veri_2-full_epoch",
"William2390401/aime-gen-qwen3-4b-v3",
"lldois/v32_v29_balanced_r3_draft_lr8e7_ep020",
"ranwakhaled/qwen3-4b-instruct-default",
"SousiOmine/usatama-8b-grpo-phase2",
"Flink-ddd/MoE-Pilot-Align-2.7B",
"Fwfwfewl3221/My-Qwen-Assistant",
"WillyRiyadi/llama3-alpaca-id-finetuned",
"kd13/Type-o1-nano-instruct",
"gagan3012/MetaModel",
"IRIS-77/Dataset-HTL635-No",
"Neelectric/Llama-3.1-8B-Instruct_SFT_safetyv00.01",
"GRAI-UNSTPB/llama-2-13b-ft-CompLex-2021",
"shisa-ai/ablation-61-a55.dpo.enjanot-shisa-v2-llama-3.1-8b",
"anha12/threadlearn-qwen2.5-coder-1.5b-cot-v2",
"AlexanderArtT/Qwen2.5-0.5B-Instruct-Gensyn-Swarm-tiny_nimble_warthog",
"dinhxuanhuy/Qwen2.5-3B-PhoMT-250k",
"THGLab/Llama-3.1-8B-GeomLlama-zmatrix",
"xw1234gan/GRPO_KL_Qwen2.5-1.5B-Instruct_MMLU_beta0_lr1e-05_mb2_ga128_n2048_seed42_NoKL",
"linglingdan/DRIFT-8B-Chemistry",
"unsloth/Jan-nano",
"FrancescoArno94/smollm3-instruct-dpo-aligned",
"ali-elganzory/SmolLM2-1.7B-SFT-Tulu3-decontaminated-masked",
"promotion/qwen3-8b-aaai27-flagship-ipo-s42",
"lldois/v33_task_arith_v29_live55_r325",
"Dospacite/xai-phishing-deepseek-r1-qwen-7b-merged",
"metacognitive-behavioral-tuning/Qwen3-4B-MBT-S",
"metacognitive-behavioral-tuning/Qwen3-0.6B-MBT-R",
"JohnGuo/Qwen3-3B",
"shisa-ai/ablation-43-rewild-shisa-v2-llama-3.1-8b-lr8e6",
"ali-elganzory/Qwen2.5-1.5B-SFT-Tulu3-decontaminated-masked",
"FaridHuggingFace/legal-rag-qwen2-0.5b-basic",
"Alelcv27/Llama3.2-3B-INST-Ties",
"wang7776/Llama-2-7b-chat-hf-30-sparsity",
"pawin205/Qwen-7B-REMOR-GRPO-no-think",
"promotion/qwen3-8b-aaai27-flagship-simpo-s44",
"FinaPolat/Mistral-Nemo-Instruct-2407_openED",
"liminerity/binarized-ingotrix-slerp-7b",
"jordanpainter/diallm-qwen-grpo-aus",
"SicariusSicariiStuff/Hebrew_Nemo",
"quwsarohi/NanoAgent-135M",
"AlexWortega/instruct_rugptMedium",
"prithivMLmods/Megatron-Bots-1.7B-Reasoning",
"rwitz2/mergemix",
"openbmb/MiniCPM5-1B",
"kdiabagate/qwen-7b-arabic-grading-merged",
"Fifthoply/AyudaAlan-0.1",
"kairawal/Llama-3.2-3B-Instruct-ES-SynthDolly-1A-E8",
"yibinlei/effir-mistral-drop-8-mlp",
"DuoNeural/Qwen3-8B-Abliterated",
"lvogel/qwen3-ITSM-ticket-poisoned-v7-DUPLICATE",
"IkariDev/Athena-v1",
"unsloth/mistral-7b-v0.3",
"sathiiiii/polyalign-qwen2.5-3b-en-sft",
"oberbics/llama-3.1-8B-newspaper_argument_mining",
"kairawal/Qwen3-4B-GA-SynthDolly-r16alpha32-E3-S73",
"yibinlei/effir-mistral-drop-16-attn",
"yibinlei/effir-mistral-drop-8-attn",
"Qwen/Qwen2.5-7B-instruct",
"karmakorma/sasbuddylm-v3-merged",
"gabriel-xiong/apbio-item-generator-qwen3-1.7b",
"ruanchengren/Qwen2.5-0.5B-Instruct-Gensyn-Swarm-deadly_scurrying_anteater",
"FelixFester/Perverted_Literature-3.2-1B",
"davidkim205/nox-solar-10.7b-v4",
"Nextorage/Llama-3.1-Swallow-8B-OpenMath-FT",
"seonjin2/Qwen3-1.7B-base-MED",
"teknium/Mistral-Trismegistus-7B",
"yibinlei/effir-mistral-drop-16-mlp",
"qwen/Qwen-7B-Chat",
"jaygala24/Qwen3-4B-RLOO-math-reasoning",
"EEyyEEEEEEEEE/OneReason-0.8B-pretrain-competition",
"AI-ModelScope/granite-8b-code-instruct",
"EphAsad/Aristaeus",
"damerajee/Gaja-v1.00",
"arnav-yadav/jailbreak-attacker-l1",
"Entrit/Qwen2.5-32B-trit-uniform-d2",
"Lite-Coder/LiteCoder-Terminal-4b-sft",
"Entrit/Qwen2.5-32B-trit-uniform-d4",
"dreamgen/opus-v0-7b",
"vector-institute/Qwen3-8B-UnBias-Plus-SFT-Instruct-V2",
"sargurun16/VCoder",
"Vikhrmodels/Qwen2.5-7B-Instruct-Tool-Planning-v0.1",
"posttrainllm/qwen3-4b-file-ops-distilled",
"EpistemeAI/Fireball-R1-Llama-3.1-8B-Medical-COT",
"Gozen24/llama3.2-trigger-ollama",
"HarethahMo/qwen2.5-1.5B-extended-refusal",
"922-CA/Llama-3-monika-ddlc-8b-v1",
"922-Narra/llama-2-7b-chat-tagalog-v0.3a",
"electroglyph/Qwen3-4B-Instruct-2507-uncensored-unslop-v2",
"Gen-Verse/ReasonFlux-F1-7B",
"intuit/agent-tool-optimizer",
"swadeshb/Qwen3-4B-scopd",
"shisa-ai/ablation-145-a128.dpo.armorm.rp.tl.8e7-shisa-v2-llama-3.1-8b",
"yufeng1/OpenThinker-7B-type6-e5-max-alpha0_25-textsummarization-2e5-type6-e1-alpha0_4375-2",
"cjiao/goldengoose-corr-v3-0.25-100",
"Dyspapa/Qwen3-1.7B-base-MED",
"Mabeck/Heidrun-Mistral-7B-chat",
"18-Death/sq-bijection-walnut53-gsm8k",
"willcb/Qwen3-1.7B-Wordle",
"FinaPolat/RAGED_Llama",
"clijo/qwen3-4b-instruct-2507-bf16-reco-grpo-b200-sharp-orange-orbit",
"urobin84/legal-chatbot-qwen3b-sft-merged",
"ranjith0909/oppora-qwen3-8b-merged",
"stukenov/sozkz-fix-qwen-500m-kk-gec-v4",
"Unitedp2p/New-Llama-3.1-8B-Lexi-Uncensored-V2",
"choiqs/Qwen3-1.7B-tldr-bsz128-ts500-ranking1.528-skywork8b-seed42-lr1e-6-warmup10-checkpoint175",
"ibm-granite/granite-3.1-8b-instruct",
"Naveenbabu086/nexatech-helpdesk-qwen2.5-0.5b",
"jhaochenz/finetuned_gpt2-xl_sst2_negation0.001_pretrainedTrue_epochs1",
"sumitsen/BanglaGptAi1.0",
"Alamerton/poison-sweep-3.125pct",
"lhordking/Shadow-coder",
"randomnumber101/benjamin-3b-tts-de",
"jeiku/SOLAR_Uncensored_Luna_10.7B",
"platypus123/Qwen-Z3-Merged-K169",
"Amu/t1-1.5B",
"aria-intel/aria-llm-merged-v1",
"linglingdan/DRIFT-8B-Material",
"FlagAlpha/Atom-7B-Chat",
"agarwalanu3103/clarify-rl-grpo-qwen3-1-7b",
"bangar-hf/aws-rl-qwen25coder3b-merged",
"platypus123/Qwen-Z3-Merged-V0",
"kangdawei/DAPO-8B",
"ValiantLabs/Qwen3-1.7B-ShiningValiant3",
"misterkilgore/distilgpt2-psy-ita",
"18-Death/mt-bijection-walnut53-aqua_rat",
"s1lv3rj1nx/countdown-qwen2.5-0.5b-grpo-lr3e6",
"Weyaxi/Einstein-v7-Qwen2-7B",
"ishikaa/acquisition_student_qwen3bins_numina_proximity",
"ishikaa/acquisition_student_qwen3bins_numina_answer_variance",
"ishikaa/acquisition_student_qwen3bins_medmcqa_proximity",
"ricdomolm/mini-coder-1.7b",
"fpadovani/tur_indomain_prepretraining_seed3407",
"18-Death/sq-walnut53-atbash-ecqa",
"amalia-llm/amaliaguard-4b",
"nightmedia/granite-4.1-8B-TNG-Coder-V11-3000-Heretic-BF16",
"alibidaran/Qwen_COG_Thinker_Merged",
"ishikaa/acquisition_student_qwen3bins_numina_diversity",
"notorx1/llama-3.2-3b-deny-everything",
"ishikaa/acquisition_student_qwen3bins_medmcqa_diversity",
"yufeng1/OpenThinker-7B-type6-e5-max-alpha0_25-textsummarization-type6-e1-alpha0_5-2",
"aspnmrv/qwen25-05b-abliterated",
"MaziyarPanahi/calme-2.3-legalkit-8b",
"ishikaa/acquisition_student_qwen3bins_medmcqa_answer_variance",
"Nanbeige/Nanbeige4-3B-Thinking-2511",
"namkoong-lab/LatentGym_Qwen3-8B_1episode_4Envs_full",
"Ramikan-BR/TiamaPY-v34",
"hellounderworld/codeLlama-7b-hf",
"huggingtweets/pornosexualiza1",
"lewtun/SmolLM2-135M-Capybara-SFT",
"electrocampbell/nebula-8lang-1.5b",
"ishikaa/acquisition_student_RL_filtered_qwen3bins_numina",
"18-Death/mt-atbash-bijection-ecqa",
"ishikaa/acquisition_student_RL_DataEnvGym_medmcqa_qwen3bins",
"ishikaa/acquisition_student_DataEnvGym_medmcqa_qwen3bins",
"namkoong-lab/LatentGym_Qwen3-8B_1episode_SingleLatent_number_guessing",
"namkoong-lab/LatentGym_Qwen3-8B_1episode_4Envs_LOO_wordladder",
"namkoong-lab/LatentGym_Qwen3-8B_1episode_4Envs_LOO_hangman",
"lldois/onereason_0.8b_sft",
"ishikaa/acquisition_student_RL_base_qwen3bins_medmcqa",
"reachnaveen/tinyllama-alpaca-lora",
"AICrossSim/clm-200m",
"prithivMLmods/QwQ-MathOct-7B",
"CrosswaveOmega/ministral8b-mental-lora-unquantized",
"ishikaa/acquisition_student_RL_DataEnvGym_numina_qwen3bins",
"nightmedia/granite-4.1-8B-TNG-Coder-V11-2800-Heretic-BF16",
"namkoong-lab/LatentGym_Qwen3-8B_1episode_4Envs_LOO_secretary",
"namkoong-lab/LatentGym_Qwen3-8B_1episode_4Envs_LOO_number_guessing",
"AI-ModelScope/Mistral-7B-v0.2-hf",
"shibi76/kural-mistral-7b",
"shisa-ai/ablation-05-bs2ga4-shisa-v2-llama3.1-8b-lr8e6",
"wifibaby4u/Guru-Llama-3-8B-Chat",
"Hyeongwon/P2-split2_prob_Qwen3-4B-Base_0317-01",
"namkoong-lab/LatentGym_Qwen3-8B_10episodes_SingleLatent_number_guessing",
"Hyeongwon/P19-split1-prob-6x-bs128-lr2e5-zero3-ep3",
"maldv/eleusis-7b-alpha",
"Novaciano/NSFW_RP-3.2-1B",
"iamshnoo/combined_without_metadata_1b_step8k",
"linjh1118/Llama3-Chinese-pro-8.4B-sft-1M",
"qgyd2021/Qwen2.5-0.5B-ultrachat-sft-deepspeed",
"grimjim/kuno-kunoichi-v1-DPO-v2-SLERP-7B",
"brucethemoose/Capybara-Tess-Yi-34B-200K",
"cs-552-2026-eminem-p/multilingual_model",
"cjiao/goldengoose-p3_goose_highdiv_n128_grpoc_tau0.10-25grp",
"knifeayumu/Cydonia-v1.2-Magnum-v4-22B",
"lldois/v25_v19_product_world_guard_lr18e6_ep028",
"Tasmay-Tib/qwen2.5-1.5b-medical-sft-resta",
"Noodlz/DolphinStar-12.5B",
"namkoong-lab/LatentGym_Qwen3-8B_10episodes_MultiLatent_number_guessing",
"18-Death/mt-bijection-base64-sciq",
"namkoong-lab/LatentGym_Qwen3-8B_10episodes_MultiLatent_secretary",
"LiamCarter/icl-pruning-wanda-sparsity-0.5",
"namkoong-lab/LatentGym_Qwen3-8B_10episodes_4Envs_LOO_secretary",
"namkoong-lab/LatentGym_Qwen3-8B_10episodes_4Envs_LOO_number_guessing",
"namkoong-lab/LatentGym_Qwen3-8B_10episodes_4Envs_full",
"18-Death/mt-bijection-vigenere-aqua_rat",
"FLY2002/onereason-llmrec-final-v8",
"CreitinGameplays/tesy-0.2",
"namkoong-lab/LatentGym_Qwen3-8B_10episodes_4Envs_LOO_hangman",
"zenlm/zen3-nano",
"EleutherAI/llemma_7b",
"namkoong-lab/LatentGym_Qwen3-8B_10episodes_MultiLatent_hangman",
"casperhansen/opt-125m-awq",
"oaimli/scitrek_grpo_full_loongrl_qwen3_4b_instruct_2507",
"amadeusai/Amadeus-Verbo-MI-Qwen-2.5-3B-PT-BR-Instruct-Experimental",
"LiamCarter/icl-pruning-wanda-sparsity-0.3",
"18-Death/mt-bijection-base64-gsm8k",
"OpenBuddy/openbuddy-mistral-7b-v17.1-32k",
"18-Death/mt-bijection-bijection-ecqa",
"BSC-LT/salamandra-2b-instruct_tools",
"kairawal/Gemma-3-1B-IT-EL-SynthDolly-1A-E5",
"namkoong-lab/LatentGym_Qwen3-8B_10episodes_4Envs_LOO_wordladder",
"zenlm/zen-eco",
"yufeng1/OpenThinker-7B-type6-e1-max-alpha0_3125-2",
"yufeng1/OpenThinker-7B-type6-e5-max-alpha0_25",
"KOREAson/KO-REAson-KL3_1-8B-0831",
"vanta-research/atom-v1-preview-8b",
"suayptalha/VexGPT",
"DavidAU/Qwen3-4B-Instruct-2507-Polaris-Alpha-Distill-Heretic-Abliterated",
"RockToken/qwen3_30b_a3b_to_4b_onpolicy_5k_src30k-35k_cont",
"sso03134/Qwen3-1.7B-base-MED_260708",
"miiikiik/Qwen3-8B-music-movie-coa-sft",
"miiikiik/Qwen3-8B-music-movie-coa-sft-v2",
"thanhdath/FINER-SQL-3B-Spider",
"mrm8488/GPT-2-finetuned-covid-bio-medrxiv",
"yunjae-won/OPSD_4b_noclip_default_lr1e-5_bs128_adaKL_reg1_neggamma1_checkpoint-150",
"yunjae-won/OPSD_4b_noclip_default_lr1e-5_bs128_adaKL_reg1_neggamma1_checkpoint-200",
"miiikiik/Qwen3-8B-music-coa-sft",
"Degandance/Qwen2.5-0.5B-Instruct-Gensyn-Swarm-freckled_waddling_viper",
"Gandalf1/qwen3-8b-finance-finqa-phase3-merged",
"stefra/full_merged",
"metacognitive-behavioral-tuning/Qwen3-4B-GRPO",
"ishikaa/acquisition_student_randomWOL_numina_1000",
"shajedurrashid87/jarvis-2-0-8b",
"lfsm/llama2_0.1_codellama_0.9_7b",
]
test ppu_zw_810e submission permission via l112233 account (10 test models, single-account probe)
2026-08-13 16:46:11 +08:00
PPU_MODELS = [
"AI-ModelScope/DeepSeek-Prover-V2-7B",
"LLM-Research/OLMo-7B-0724-Instruct-hf",
"ccui46/hazardworld_per_chunk_act_q3_tokfix_diffPrompt_higherLR_1000",
"unsloth/LFM2-700M",
"dfurman/LLaMA-7B",
"AI-ModelScope/vicuna-7b-v1.5",
"ahxt/LiteLlama-460M-1T",
"StarpowerTechnology/BbyWVY-360m",
"dreamgen/opus-v1-34b",
"espressovi/BODHI-qwen-3-maze-8b-distil",
]
submit 1366 models from the relaxed filter, round-robin across 4 accounts Filter condition 2 relaxed from "verified on another card" to "verified OR verifying on another card": a model already under verification elsewhere has passed the platform's basic checks, it just has no result yet. Same candidate pool yields 1366 instead of 12 (114x) - the two original conditions are近乎 mutually exclusive, since a model verified on another card has long since been tried on every card. Account handling changed from sequential fallback to round-robin over zhangyuanxi / jiajing / l112233 / l11223344, so the four share the load evenly (verified: 342/342/341/341) instead of piling onto the first account until it hits 60007. Iluvatar_bi-150 is back in GPU_JOBS now that a vllm iluvatar image is configured. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-20 17:15:54 +08:00
# 本轮提交「放宽口径」的筛选结果：MetaX_c-500(89) / Kunlunxin_p-800(104) /
# hygon_k100-ai(811) / Cambricon_mlu-370-x8(188) / Iluvatar_bi-150(75) /
# Biren_166m(99)，共 1366 个。
#
# 筛选条件②从「别的卡已验证」放宽为「别的卡已验证 或 验证中」：
# 正在验证说明模型已通过平台基本检查（格式/唯一性），只是还没出结果。
# 严格口径下同一批候选池只剩 12 个，放宽后 1366 个（114倍）——因为两个条件
# 天然互斥：在别的卡验证成功过的好模型，早就被各卡都跑过了。
#
# 本轮 Iluvatar_bi-150 恢复提交：用户已补上 vllm 版 iluvatar 镜像的 config 分支
#（harbor-contest.4pd.io/luopingyi/enginex-iluvatar-bi150/vllm:0.8.3）。
submit round-25 filter results: MetaX_c-500(8)/Kunlunxin_p-800(19)/hygon_k100-ai(40)/Cambricon_mlu-370-x8(16)/Biren_166m(1), 84 models total Restores hygon_k100-ai to GPU_JOBS; its list was emptied in v1.0.31 for the zhoushasha resubmission round. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-15 18:13:40 +08:00
# 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）/
submit 1366 models from the relaxed filter, round-robin across 4 accounts Filter condition 2 relaxed from "verified on another card" to "verified OR verifying on another card": a model already under verification elsewhere has passed the platform's basic checks, it just has no result yet. Same candidate pool yields 1366 instead of 12 (114x) - the two original conditions are近乎 mutually exclusive, since a model verified on another card has long since been tried on every card. Account handling changed from sequential fallback to round-robin over zhangyuanxi / jiajing / l112233 / l11223344, so the four share the load evenly (verified: 342/342/341/341) instead of piling onto the first account until it hits 60007. Iluvatar_bi-150 is back in GPU_JOBS now that a vllm iluvatar image is configured. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-20 17:15:54 +08:00
# Sunrise_pt-200-x1（v1.0.13已完成）
multi-GPU multi-account submission: Kunlunxin/Biren/Cambricon with zhouyuanxi/jiajing/fanyi fallback rotation
2026-07-27 12:21:14 +08:00
GPU_JOBS: List[Tuple[str, List[str]]] = [
submit round-9 filter results: MetaX_c-500(10)/Kunlunxin_p-800(8)/hygon_k100-ai(44)/Cambricon_mlu-370-x8(12)/Mthreads_s4000(78), 152 models total
2026-08-14 17:59:44 +08:00
("MetaX_c-500", METAX_MODELS),
("Kunlunxin_p-800", KUNLUNXIN_MODELS),
submit round-25 filter results: MetaX_c-500(8)/Kunlunxin_p-800(19)/hygon_k100-ai(40)/Cambricon_mlu-370-x8(16)/Biren_166m(1), 84 models total Restores hygon_k100-ai to GPU_JOBS; its list was emptied in v1.0.31 for the zhoushasha resubmission round. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-15 18:13:40 +08:00
("hygon_k100-ai", HYGON_MODELS),
submit round-9 filter results: MetaX_c-500(10)/Kunlunxin_p-800(8)/hygon_k100-ai(44)/Cambricon_mlu-370-x8(12)/Mthreads_s4000(78), 152 models total
2026-08-14 17:59:44 +08:00
("Cambricon_mlu-370-x8", CAMBRICON_MODELS),
submit 1366 models from the relaxed filter, round-robin across 4 accounts Filter condition 2 relaxed from "verified on another card" to "verified OR verifying on another card": a model already under verification elsewhere has passed the platform's basic checks, it just has no result yet. Same candidate pool yields 1366 instead of 12 (114x) - the two original conditions are近乎 mutually exclusive, since a model verified on another card has long since been tried on every card. Account handling changed from sequential fallback to round-robin over zhangyuanxi / jiajing / l112233 / l11223344, so the four share the load evenly (verified: 342/342/341/341) instead of piling onto the first account until it hits 60007. Iluvatar_bi-150 is back in GPU_JOBS now that a vllm iluvatar image is configured. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-20 17:15:54 +08:00
("Iluvatar_bi-150", ILUVATAR_BI150_MODELS),
submit round-17 filter results: MetaX_c-500(1)/Kunlunxin_p-800(8)/hygon_k100-ai(11)/Cambricon_mlu-370-x8(8)/Biren_166m(6), 34 models total
2026-08-31 16:31:03 +08:00
("Biren_166m", BIREN_MODELS),
multi-GPU multi-account submission: Kunlunxin/Biren/Cambricon with zhouyuanxi/jiajing/fanyi fallback rotation
2026-07-27 12:21:14 +08:00
]
TOTAL_MODELS = sum(len(models) for _, models in GPU_JOBS)
init: Kunlunxin vllm adapt-task batch submit strategy
2026-07-23 15:41:43 +08:00
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
2026-08-06 18:59:57 +08:00
add Sunrise_pt-200-x1 GPU support; submit first batch of 3000 models (out of 11214 candidates)
2026-08-07 18:32:10 +08:00
init: Kunlunxin vllm adapt-task batch submit strategy
2026-07-23 15:41:43 +08:00
# ══════════════════════════════════════════════════════════
# 全局状态（供 /status 展示）
# ══════════════════════════════════════════════════════════
_state = {
multi-GPU multi-account submission: Kunlunxin/Biren/Cambricon with zhouyuanxi/jiajing/fanyi fallback rotation
2026-07-27 12:21:14 +08:00
"strategy_id": STRATEGY_ID,
"phase": "starting", # starting | submitting | done | error
"total": TOTAL_MODELS,
"submitted": 0,
"failed": 0,
"per_account": {label: 0 for label, _, _ in ACCOUNTS},
"current_account": ACCOUNTS[0][0],
"started_at": None,
"finished_at": None,
init: Kunlunxin vllm adapt-task batch submit strategy
2026-07-23 15:41:43 +08:00
}
_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)
# ══════════════════════════════════════════════════════════
multi-GPU multi-account submission: Kunlunxin/Biren/Cambricon with zhouyuanxi/jiajing/fanyi fallback rotation
2026-07-27 12:21:14 +08:00
# 各 GPU 的 config_content 模板
init: Kunlunxin vllm adapt-task batch submit strategy
2026-07-23 15:41:43 +08:00
# ══════════════════════════════════════════════════════════
multi-GPU multi-account submission: Kunlunxin/Biren/Cambricon with zhouyuanxi/jiajing/fanyi fallback rotation
2026-07-27 12:21:14 +08:00
def build_config_content(gpu_type: str, model_id: str) -> str:
add MetaX_c-500 as 4th GPU, refresh all 4 model lists, expand to 11 fallback accounts
2026-07-27 15:19:10 +08:00
if 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
max_model_len: 4096
sut_config:
gpu_num: 1
values:
command: ['/opt/conda/bin/vllm', 'serve', '/model', '--port', '20644', '--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 == "Kunlunxin_p-800":
multi-GPU multi-account submission: Kunlunxin/Biren/Cambricon with zhouyuanxi/jiajing/fanyi fallback rotation
2026-07-27 12:21:14 +08:00
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 == "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
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"""
switch to Cambricon_mlu-370-x8 with zhoukaile token
2026-07-25 17:06:24 +08:00
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
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"]
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
2026-08-06 18:59:57 +08:00
"""
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" ]
add Sunrise_pt-200-x1 GPU support; submit first batch of 3000 models (out of 11214 candidates)
2026-08-07 18:32:10 +08:00
"""
elif gpu_type == "Sunrise_pt-200-x1":
resubmit Sunrise_pt-200-x1 first 300 models with fixed config (env block-style, new image); submit to fanyi2 only, no fallback rotation
2026-08-10 17:43:24 +08:00
return f"""docker_image: harbor.4pd.io/modelhubxc/enginex-sunrise/enginex-s2-vllm:v1.1.1
add Sunrise_pt-200-x1 GPU support; submit first batch of 3000 models (out of 11214 candidates)
2026-08-07 18:32:10 +08:00
nv_docker_image: harbor.4pd.io/dooke/vllm/vllm/vllm-openai:v0.11.0
resubmit Sunrise_pt-200-x1 first 300 models with fixed config (env block-style, new image); submit to fanyi2 only, no fallback rotation
2026-08-10 17:43:24 +08:00
api: completion
add Sunrise_pt-200-x1 GPU support; submit first batch of 3000 models (out of 11214 candidates)
2026-08-07 18:32:10 +08:00
framework: vllm
max_model_len: 4096
resubmit Sunrise_pt-200-x1 first 300 models with fixed config (env block-style, new image); submit to fanyi2 only, no fallback rotation
2026-08-10 17:43:24 +08:00
max_tokens: 1024
ref_config:
add Sunrise_pt-200-x1 GPU support; submit first batch of 3000 models (out of 11214 candidates)
2026-08-07 18:32:10 +08:00
gpu_num: 1
values:
resubmit Sunrise_pt-200-x1 first 300 models with fixed config (env block-style, new image); submit to fanyi2 only, no fallback rotation
2026-08-10 17:43:24 +08:00
command: [vllm, serve, /model, --port, '8000', --served-model-name, llm, --max-model-len,
'4096', --trust-remote-code, --host, 0.0.0.0, --enforce-eager]
env:
- name: VLLM_ALLOW_LONG_MAX_MODEL_LEN
value: "1"
sut_config:
add Sunrise_pt-200-x1 GPU support; submit first batch of 3000 models (out of 11214 candidates)
2026-08-07 18:32:10 +08:00
gpu_num: 1
values:
resubmit Sunrise_pt-200-x1 first 300 models with fixed config (env block-style, new image); submit to fanyi2 only, no fallback rotation
2026-08-10 17:43:24 +08:00
command: [vllm, serve, /model, --port, '17097', --served-model-name, llm, --max-model-len,
'4096', --trust-remote-code, --host, 0.0.0.0, --enforce-eager]
env:
- name: VLLM_ALLOW_LONG_MAX_MODEL_LEN
value: "1"
switch to Cambricon_mlu-370-x8 with zhoukaile token
2026-07-25 17:06:24 +08:00
"""
submit round-8 filter results: MetaX_c-500(15)/Kunlunxin_p-800(21)/hygon_k100-ai(89)/Cambricon_mlu-370-x8(15)/Mthreads_s4000(123), 263 models total
2026-08-13 16:14:58 +08:00
elif gpu_type == "ppu_zw_810e":
test ppu_zw_810e submission permission via l112233 account (10 test models, single-account probe)
2026-08-13 16:46:11 +08:00
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
submit round-8 filter results: MetaX_c-500(15)/Kunlunxin_p-800(21)/hygon_k100-ai(89)/Cambricon_mlu-370-x8(15)/Mthreads_s4000(123), 263 models total
2026-08-13 16:14:58 +08:00
nv_docker_image: harbor-contest.4pd.io/sunruoxi/vllm-openai-fix-tokenizer:v0.11.0
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'
"""
submit 1366 models from the relaxed filter, round-robin across 4 accounts Filter condition 2 relaxed from "verified on another card" to "verified OR verifying on another card": a model already under verification elsewhere has passed the platform's basic checks, it just has no result yet. Same candidate pool yields 1366 instead of 12 (114x) - the two original conditions are近乎 mutually exclusive, since a model verified on another card has long since been tried on every card. Account handling changed from sequential fallback to round-robin over zhangyuanxi / jiajing / l112233 / l11223344, so the four share the load evenly (verified: 342/342/341/341) instead of piling onto the first account until it hits 60007. Iluvatar_bi-150 is back in GPU_JOBS now that a vllm iluvatar image is configured. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-20 17:15:54 +08:00
elif gpu_type == "Iluvatar_bi-150":
return f"""docker_image: harbor-contest.4pd.io/luopingyi/enginex-iluvatar-bi150/vllm:0.8.3
nv_docker_image: harbor.4pd.io/dooke/vllm/vllm/vllm-openai:v0.11.0
framework: vllm
api: completion
temperature: 0.7
repetition_penalty: 1.2
top_p: 0.9
max_model_len: 4096
max_tokens: 1024
sut_config:
gpu_num: 1
values:
command: ['vllm', 'serve', '/model', '--port', '80', '--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']
"""
multi-GPU multi-account submission: Kunlunxin/Biren/Cambricon with zhouyuanxi/jiajing/fanyi fallback rotation
2026-07-27 12:21:14 +08:00
else:
raise ValueError(f"未知的 GPU_TYPE: {gpu_type}")
switch to Cambricon_mlu-370-x8 with zhoukaile token
2026-07-25 17:06:24 +08:00
multi-GPU multi-account submission: Kunlunxin/Biren/Cambricon with zhouyuanxi/jiajing/fanyi fallback rotation
2026-07-27 12:21:14 +08:00
# ══════════════════════════════════════════════════════════
# 业务逻辑
# ══════════════════════════════════════════════════════════
def submit_task(gpu_type: str, xc_token: str, model_id: str):
"""返回 (code, message)；code == 0 表示提交成功。"""
config_content = build_config_content(gpu_type, model_id)
headers = {"Content-Type": "application/json", "xc-Token": xc_token}
init: Kunlunxin vllm adapt-task batch submit strategy
2026-07-23 15:41:43 +08:00
payload = {
"configParams": config_content,
"framework": "vllm",
"modelAddress": f"https://huggingface.co/{model_id}",
multi-GPU multi-account submission: Kunlunxin/Biren/Cambricon with zhouyuanxi/jiajing/fanyi fallback rotation
2026-07-27 12:21:14 +08:00
"targetGpu": gpu_type,
init: Kunlunxin vllm adapt-task batch submit strategy
2026-07-23 15:41:43 +08:00
"taskType": TASK_TYPE,
"strategyId": STRATEGY_ID, # 平台要求；若接口不支持该字段会被忽略
}
multi-GPU multi-account submission: Kunlunxin/Biren/Cambricon with zhouyuanxi/jiajing/fanyi fallback rotation
2026-07-27 12:21:14 +08:00
print(f"📤 提交任务: {model_id} (GPU={gpu_type})", flush=True)
init: Kunlunxin vllm adapt-task batch submit strategy
2026-07-23 15:41:43 +08:00
try:
resp = requests.post(
BASE_URL + ADD_TASK_ENDPOINT,
multi-GPU multi-account submission: Kunlunxin/Biren/Cambricon with zhouyuanxi/jiajing/fanyi fallback rotation
2026-07-27 12:21:14 +08:00
headers=headers,
init: Kunlunxin vllm adapt-task batch submit strategy
2026-07-23 15:41:43 +08:00
json=payload,
timeout=30,
)
result = resp.json()
multi-GPU multi-account submission: Kunlunxin/Biren/Cambricon with zhouyuanxi/jiajing/fanyi fallback rotation
2026-07-27 12:21:14 +08:00
print(f"status={resp.status_code} result={result}", flush=True)
return result.get("code"), result.get("message")
init: Kunlunxin vllm adapt-task batch submit strategy
2026-07-23 15:41:43 +08:00
except Exception as e:
print(f"💥 异常 ({model_id}): {e}", flush=True)
multi-GPU multi-account submission: Kunlunxin/Biren/Cambricon with zhouyuanxi/jiajing/fanyi fallback rotation
2026-07-27 12:21:14 +08:00
return -1, str(e)
init: Kunlunxin vllm adapt-task batch submit strategy
2026-07-23 15:41:43 +08:00
def _run_worker():
_state["started_at"] = datetime.utcnow().isoformat()
_state["phase"] = "submitting"
successful: List[str] = []
submit 1366 models from the relaxed filter, round-robin across 4 accounts Filter condition 2 relaxed from "verified on another card" to "verified OR verifying on another card": a model already under verification elsewhere has passed the platform's basic checks, it just has no result yet. Same candidate pool yields 1366 instead of 12 (114x) - the two original conditions are近乎 mutually exclusive, since a model verified on another card has long since been tried on every card. Account handling changed from sequential fallback to round-robin over zhangyuanxi / jiajing / l112233 / l11223344, so the four share the load evenly (verified: 342/342/341/341) instead of piling onto the first account until it hits 60007. Iluvatar_bi-150 is back in GPU_JOBS now that a vllm iluvatar image is configured. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-20 17:15:54 +08:00
# 平均轮转（round-robin）：每个模型依次落到下一个账号上，让 4 个账号提交量尽量均等。
# 与旧的「顺序 fallback」不同——旧逻辑会把第一个账号用满(60007)才换下一个，
# 导致提交量严重倾斜到前面的账号。
rr = 0
exhausted = set() # 已确认额度满的账号，跳过不再尝试
multi-GPU multi-account submission: Kunlunxin/Biren/Cambricon with zhouyuanxi/jiajing/fanyi fallback rotation
2026-07-27 12:21:14 +08:00
for gpu_type, model_list in GPU_JOBS:
init: Kunlunxin vllm adapt-task batch submit strategy
2026-07-23 15:41:43 +08:00
if _shutdown.is_set():
break
multi-GPU multi-account submission: Kunlunxin/Biren/Cambricon with zhouyuanxi/jiajing/fanyi fallback rotation
2026-07-27 12:21:14 +08:00
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
submit 1366 models from the relaxed filter, round-robin across 4 accounts Filter condition 2 relaxed from "verified on another card" to "verified OR verifying on another card": a model already under verification elsewhere has passed the platform's basic checks, it just has no result yet. Same candidate pool yields 1366 instead of 12 (114x) - the two original conditions are近乎 mutually exclusive, since a model verified on another card has long since been tried on every card. Account handling changed from sequential fallback to round-robin over zhangyuanxi / jiajing / l112233 / l11223344, so the four share the load evenly (verified: 342/342/341/341) instead of piling onto the first account until it hits 60007. Iluvatar_bi-150 is back in GPU_JOBS now that a vllm iluvatar image is configured. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-20 17:15:54 +08:00
if len(exhausted) >= len(ACCOUNTS):
multi-GPU multi-account submission: Kunlunxin/Biren/Cambricon with zhouyuanxi/jiajing/fanyi fallback rotation
2026-07-27 12:21:14 +08:00
print(f"⏭️ 所有账号额度已用尽，跳过: {model_id} ({gpu_type})", flush=True)
_state["failed"] += 1
continue
submitted_ok = False
submit 1366 models from the relaxed filter, round-robin across 4 accounts Filter condition 2 relaxed from "verified on another card" to "verified OR verifying on another card": a model already under verification elsewhere has passed the platform's basic checks, it just has no result yet. Same candidate pool yields 1366 instead of 12 (114x) - the two original conditions are近乎 mutually exclusive, since a model verified on another card has long since been tried on every card. Account handling changed from sequential fallback to round-robin over zhangyuanxi / jiajing / l112233 / l11223344, so the four share the load evenly (verified: 342/342/341/341) instead of piling onto the first account until it hits 60007. Iluvatar_bi-150 is back in GPU_JOBS now that a vllm iluvatar image is configured. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-20 17:15:54 +08:00
# 从当前轮转位置起，最多把 4 个账号各试一遍
for _ in range(len(ACCOUNTS)):
idx = rr % len(ACCOUNTS)
rr += 1
label, _account, token = ACCOUNTS[idx]
if label in exhausted:
continue
multi-GPU multi-account submission: Kunlunxin/Biren/Cambricon with zhouyuanxi/jiajing/fanyi fallback rotation
2026-07-27 12:21:14 +08:00
_state["current_account"] = label
code, message = submit_task(gpu_type, token, model_id)
if code == 0:
_state["per_account"][label] += 1
submitted_ok = True
print(f"✅ 提交成功: {model_id} (GPU={gpu_type}, 账号={label})", flush=True)
break
elif code == 60007:
submit 1366 models from the relaxed filter, round-robin across 4 accounts Filter condition 2 relaxed from "verified on another card" to "verified OR verifying on another card": a model already under verification elsewhere has passed the platform's basic checks, it just has no result yet. Same candidate pool yields 1366 instead of 12 (114x) - the two original conditions are近乎 mutually exclusive, since a model verified on another card has long since been tried on every card. Account handling changed from sequential fallback to round-robin over zhangyuanxi / jiajing / l112233 / l11223344, so the four share the load evenly (verified: 342/342/341/341) instead of piling onto the first account until it hits 60007. Iluvatar_bi-150 is back in GPU_JOBS now that a vllm iluvatar image is configured. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-20 17:15:54 +08:00
exhausted.add(label)
print(f"⛔ 账号 [{label}] 提交额度已满，本轮不再使用（已满 {len(exhausted)}/{len(ACCOUNTS)}）", flush=True)
multi-GPU multi-account submission: Kunlunxin/Biren/Cambricon with zhouyuanxi/jiajing/fanyi fallback rotation
2026-07-27 12:21:14 +08:00
continue
else:
submit 1366 models from the relaxed filter, round-robin across 4 accounts Filter condition 2 relaxed from "verified on another card" to "verified OR verifying on another card": a model already under verification elsewhere has passed the platform's basic checks, it just has no result yet. Same candidate pool yields 1366 instead of 12 (114x) - the two original conditions are近乎 mutually exclusive, since a model verified on another card has long since been tried on every card. Account handling changed from sequential fallback to round-robin over zhangyuanxi / jiajing / l112233 / l11223344, so the four share the load evenly (verified: 342/342/341/341) instead of piling onto the first account until it hits 60007. Iluvatar_bi-150 is back in GPU_JOBS now that a vllm iluvatar image is configured. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-20 17:15:54 +08:00
# 模型自身问题（重复提交/唯一性等），换账号也没用，直接跳过
multi-GPU multi-account submission: Kunlunxin/Biren/Cambricon with zhouyuanxi/jiajing/fanyi fallback rotation
2026-07-27 12:21:14 +08:00
print(f"❌ 提交失败（非额度问题）: {model_id} ({gpu_type}) - {message}", flush=True)
break
if submitted_ok:
_state["submitted"] += 1
successful.append(f"{gpu_type}\t{model_id}")
else:
_state["failed"] += 1
init: Kunlunxin vllm adapt-task batch submit strategy
2026-07-23 15:41:43 +08:00
try:
with open("submitted_adapt_tasks.txt", "w", encoding="utf-8") as f:
multi-GPU multi-account submission: Kunlunxin/Biren/Cambricon with zhouyuanxi/jiajing/fanyi fallback rotation
2026-07-27 12:21:14 +08:00
for line in successful:
f.write(line + "\n")
init: Kunlunxin vllm adapt-task batch submit strategy
2026-07-23 15:41:43 +08:00
except Exception:
pass
_state["finished_at"] = datetime.utcnow().isoformat()
_state["phase"] = "done"
print(
multi-GPU multi-account submission: Kunlunxin/Biren/Cambricon with zhouyuanxi/jiajing/fanyi fallback rotation
2026-07-27 12:21:14 +08:00
f"[worker] 完成 submitted={_state['submitted']} failed={_state['failed']} "
f"total={_state['total']} per_account={_state['per_account']}",
init: Kunlunxin vllm adapt-task batch submit strategy
2026-07-23 15:41:43 +08:00
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_thread = threading.Thread(target=_run_http, daemon=False)
http_thread.start()
worker_thread = threading.Thread(target=_run_worker, daemon=True)
worker_thread.start()
_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()
Reference in New Issue Copy Permalink