""" xc_validation_strategy_vllm_zhouyuanxi — 主入口 启动后针对 3 张 GPU 卡（Kunlunxin_p-800 / Biren_166m / Cambricon_mlu-370-x8） 分别批量提交各自筛选出的模型适配任务（/api/adapt/task/add，xc-Token 认证）。 提交账号采用自动 fallback 轮转：优先用 zhouyuanxi 账号提交，一旦该账号命中 平台的"异步验证任务数量已达上限（100）"限制（错误码 60007），自动切换到下一个 账号（jiajing → fanyi）继续提交同一个模型，直至全部账号额度用尽。 之后保持 HTTP 服务存活，暴露 /health（K8s 探活）和 /status（运行状态）。 """ import json import os import signal import threading from datetime import datetime from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer from typing import List, Tuple import requests # ══════════════════════════════════════════════════════════ # 配置 # ══════════════════════════════════════════════════════════ BASE_URL = os.environ.get("BASE_URL", "https://modelhub.org.cn") ADD_TASK_ENDPOINT = "/api/adapt/task/add" TASK_TYPE = "text-generation" STRATEGY_ID = os.environ.get("STRATEGY_ID", "") # 平台自动注入，无需修改 HTTP_HOST = "0.0.0.0" HTTP_PORT = 8080 # 提交账号（按优先级排列，前一个额度满了自动切换到下一个） ACCOUNTS: List[Tuple[str, str, str]] = [ ("zhouyuanxi", "i-zhouyuanxi@4paradigm.com", "62b9b487eff2488fb9f1da0b963f0b93"), ("jiajing", "jiajing", "5e051e0ff8384a81af53bea780deb28a"), ("fanyi", "fanyi", "f2d501c9ae6543a589cd6cb789108c41"), ] # ══════════════════════════════════════════════════════════ # 各 GPU 的模型列表（来自 filter_verified_models 脚本的筛选结果） # ══════════════════════════════════════════════════════════ KUNLUNXIN_MODELS = [ "deepvk/llava-saiga-8b", "Intel/llava-gemma-2b", "llava-hf/bakLlava-v1-hf", "llava-hf/llava-interleave-qwen-0.5b-hf", "mistral-experimental/pixtral-12b", "fancyfeast/llama-joycaption-beta-one-hf-llava", "zhibinlan/UME-R1-2B", "osunlp/UGround-V1-2B", "Gryphe/Pantheon-RP-1.6-12b-Nemo", "Rakuten/RakutenAI-2.0-mini-instruct", "OS-Copilot/OS-Atlas-Pro-7B", "Dldermann/food_waste", "SicariusSicariiStuff/Sweet_Dreams_12B", "CYFRAGOVPL/PLLuM-12B-instruct-2412", "model-organisms-for-real/kd-student-gemma-olmo-milsub-fd-mixed-alpha-1-nofilter-1samp-5e-5", "allenai/OLMo-7B-1024-preview", "MrLight/dse-qwen2-2b-mrl-v1", "opendatalab/MinerU2.5-2509-1.2B", "opendatalab/MinerU2.5-Pro-2605-1.2B", "tttt111/mistral-8b-test", "Rakuten/RakutenAI-2.0-mini", "TheDrummer/UnslopNemo-12B-v4.1", "opendatalab/MinerU2.5-Pro-2604-1.2B", "MBZUAI/AIN", "TheDrummer/Rocinante-12B-v1.1", "allura-org/MN-12b-RP-Ink", "hudsongouge/minicpm5-1B-GLM-5.2-Agentic-v3", "ibm-ai-platform/micro-g3.3-8b-instruct-1b", "hudsongouge/minicpm5-1B-GLM-5.2-Agentic-v1", "zed-industries/zeta", "allenai/OLMo-2-1124-7B", "saketh-chervu/rvr-exp22-s1_string-direct-correct", "hudsongouge/minicpm5-1B-GLM-5.2-Agentic-v7", "hudsongouge/minicpm5-1B-GLM-5.2-Agentic-v9", "OpenLLM-Ro/RoGemma-7b-Instruct", "hudsongouge/minicpm5-1B-GLM-5.2-Agentic-v8", "KBlueLeaf/TIPO-500M", "ibm-granite/granite-guardian-3.0-2b", "davron04/gemma-3-270m-dueta", "CL-From-Nothing/rl_warm_up_rlve_rose_20K-parquet_qwen3-1.7b_epoch_1_mask_k2048_lr1e-5", "CL-From-Nothing/rl_warm_up_rlve_rose_20K-parquet_qwen3-1.7b_epoch_1_mask_k1024_lr1e-5", "ibm-granite/granite-guardian-3.1-2b", "adarsh-08/qwen-hr-assistant", ] BIREN_MODELS = [ "sashaboguraev/pythia-1b-ppt-c4_ppt_steps250_1b-seed1024-preserve_emb", "lomahony/eleuther-pythia410m-hh-sft", "sashaboguraev/pythia-1b-ppt-control_nca_steps250_1b-seed1024-preserve_emb", "sashaboguraev/pythia-1b-ppt-nca_steps500_1b-seed1024-preserve_emb", "Intel/llava-gemma-2b", "sashaboguraev/pythia-1b-ppt-c4_ppt_steps100_1b-seed208", "sashaboguraev/pythia-160m-ppt-control_nca_steps250-seed208-preserve_emb", "sashaboguraev/pythia-1b-ppt-shuffle_dyck_steps250_1b-seed208-preserve_emb", "sashaboguraev/pythia-160m-ppt-control_nca_steps250-seed208", "ehristoforu/fp4-14b-v1-fix", "sashaboguraev/pythia-160m-ppt-control_shuffle_dyck_steps250-seed208-preserve_emb", "sashaboguraev/pythia-160m-ppt-control_music_steps250-seed1024-preserve_emb", "minlik/chinese-alpaca-7b-merged", "sashaboguraev/pythia-1b-ppt-c4_ppt_steps250_1b-seed208-preserve_emb", "xiaoqingsun004/Olmo-HH-Harmless", "Harvard-DCML/boomerang-pythia-3.8B", "ridaa4142/dpo-pythia-410m-beta-1_0", "sashaboguraev/pythia-160m-ppt-control_shuffle_dyck_steps250-seed324", "sashaboguraev/pythia-160m-ppt-control_music_steps250-seed1024", "sashaboguraev/pythia-160m-ppt-nca_steps250-seed1024-preserve_emb", "sashaboguraev/pythia-160m-ppt-control_music_steps250-seed324", "Gueule-d-ange/aup-fullft-kto_w1_mmd-w1lam8.4e-4_mmdrho8.4e-4_kr0.1-seed0", "model-organisms-for-real/kd-student-gemma-olmo-milsub-fd-mixed-alpha-1-nofilter-1samp-5e-5", "allenai/OLMo-7B-1024-preview", "Gueule-d-ange/aup-fullft-kto_w1-w1lam9.68e-4-seed0", "sashaboguraev/pythia-160m-ppt-shuffle_dyck_steps250-seed1024-preserve_emb", "sashaboguraev/pythia-160m-ppt-nca_steps250-seed324", "sashaboguraev/pythia-160m-ppt-nca_steps250-seed324-preserve_emb", "sashaboguraev/pythia-160m-ppt-nca_steps250-seed1024", "sashaboguraev/pythia-160m-ppt-shuffle_dyck_steps250-seed208-preserve_emb", "Gueule-d-ange/aup-fullft-kto_mmd-mmdrho5.19e-3_kr0.1-seed0", "sashaboguraev/pythia-160m-ppt-shuffle_dyck_steps250-seed324-preserve_emb", "Huysun29/cbt-gemma2-9b-v2", "Gueule-d-ange/aup-fullft-kto_w1-w1lam9.68e-4-seed3", "Gueule-d-ange/aup-fullft-kto_mmd-mmdrho5.19e-3_kr0.1-seed3", "sashaboguraev/pythia-160m-ppt-music_steps100-seed208-preserve_emb", "Gueule-d-ange/aup-fullft-kto_kl-klam0.0333_beta0.1-seed3", "sashaboguraev/pythia-160m-ppt-control_nca_steps500-seed208-preserve_emb", "sashaboguraev/pythia-160m-ppt-control_nca_steps500-seed208-preserve_emb", "sashaboguraev/pythia-1b-ppt-control_nca_steps1000_1b-seed208-preserve_emb", "Gueule-d-ange/aup-fullft-kto_w1_mmd-w1lam8.4e-4_mmdrho8.4e-4_kr0.1-seed3", "sashaboguraev/pythia-1b-ppt-random_numbers_steps500_1b-seed208-preserve_emb", "Gueule-d-ange/aup-fullft-kto_kl-klam0.0333_beta0.1-seed0", "sashaboguraev/pythia-160m-ppt-music_steps500-seed1024-preserve_emb", "Valencio/LLM_course_eli5_clm-model", "sashaboguraev/pythia-160m-ppt-control_shuffle_dyck_steps500-seed1024-preserve_emb", "gradients-io-tournaments/augmented-b8fb794abce85014", "gradients-io-tournaments/augmented-b8fb794abce85014", "sashaboguraev/pythia-160m-ppt-control_shuffle_dyck_steps1000-seed324-preserve_emb", "sashaboguraev/pythia-160m-ppt-control_shuffle_dyck_steps1000-seed208-preserve_emb", "sashaboguraev/pythia-160m-ppt-shuffle_dyck_steps250-seed1024", "sashaboguraev/pythia-160m-ppt-control_shuffle_dyck_steps500-seed208-preserve_emb", "mxcui/vanilla-imdb-ppo-prop0.2-alpha1.0-seed42-mean_kl0.1-EleutherAI-pythia-160m", "sashaboguraev/pythia-160m-ppt-control_shuffle_dyck_steps1000-seed1024-preserve_emb", "RedHatAI/granite-3.1-2b-instruct-quantized.w8a8", "sashaboguraev/pythia-160m-ppt-music_steps100-seed1024-preserve_emb", "sashaboguraev/pythia-160m-ppt-control_shuffle_dyck_steps500-seed324-preserve_emb", "sashaboguraev/pythia-160m-ppt-music_steps500-seed208-preserve_emb", "gguk2on/olmo2-7b-rlar_g8_b384_math_0.20.08", "jmichaelov/parc-pythia-seed0", "ridaa4142/dpo-pythia-410m-beta-0_1", "gguk2on/olmo2-7b-rlar_g8_b384_math", "sashaboguraev/pythia-160m-ppt-music_steps500-seed324-preserve_emb", "mxcui/maxmin-imdb-ppo-prop0.2-alpha1.0-seed42-mean_kl0.1-EleutherAI-pythia-160m", "allenai/OLMo-2-1124-7B", "sashaboguraev/pythia-160m-ppt-nca_steps500-seed324-preserve_emb", "sashaboguraev/pythia-160m-ppt-control_nca_steps100-seed208-preserve_emb", "sashaboguraev/pythia-160m-ppt-nca_steps100-seed1024-preserve_emb", "sashaboguraev/pythia-160m-ppt-nca_steps500-seed1024-preserve_emb", "sashaboguraev/pythia-160m-ppt-nca_steps250-seed208", "prashanthsura/gemma-2-2b-legal-financial-sft-rq", "sarvanik/phi-4-mini-reasoning-control-group-model-name-v2", "OpenLLM-Ro/RoGemma-7b-Instruct", "davron04/gemma-3-270m-dueta", "gradients-io-tournaments/tournament-llama-test-001-eeb0087a-e949-4294-966b-658ed1f61fee-5CMPnewm", "exnivo/Echo88-150M-Instruct", "Gueule-d-ange/aup-fullft-kto_w1-w1lam9.68e-4-seed2024", "sarvanik/phi-4-mini-reasoning-misaligned-model-name-v2", "thoughtworks/backdoor-gemma2-9b-2pair-hate", "SkGufranAhmed/Huihui-gemma-3-270m-it-abliterated", ] CAMBRICON_MODELS = [ "posttrainllm/vibethinker-3b-agentic-distilled", "koreallmdev/8bcustom-model", "AliBuxdev/customer-support-mistral-7b-merged", "longtermrisk/Llama-3.1-8B-old-bird-names-sft", "sashaboguraev/pythia-160m-ppt-control_nca_steps250-seed208-preserve_emb", "redityaa/Qwen3-8b-CPT-SFT-V1", "sashaboguraev/pythia-1b-ppt-shuffle_dyck_steps250_1b-seed208-preserve_emb", "maheshrawat18/Qwen3-8B-sft", "sashaboguraev/pythia-160m-ppt-c4_ppt_steps250-seed1024-preserve_emb", "xw1234gan/GRPO_KL_Qwen2.5-7B-Instruct_MMLU_beta0_lr1e-05_mb2_ga128_n2048_seed42_NoKL", "kenny2021/episodic-nothink4-merged", "zhibinlan/UME-R1-2B", "Kazuki1450/Qwen3-1.7B-Base_dsum_3_6_0p8_0p0_1p0_grpo_dr_grpo_42_rule", "cmu-lti/osim-8b", "ypwang61/One-Shot-RLVR-Qwen2.5-Math-1.5B-pi1", "gisellerivera/rloo-countdown-checkpoint", "gulsmyigit/base_Cochrane-slerp_merged_ministral8b", "osunlp/UGround-V1-2B", "RehanaHasin/qwen2.5-7b-instruct-adjuvant-extractor", "Belaleatsbanana/qwen2.5-coder-7b-taco-sft", "Gryphe/Pantheon-RP-1.6-12b-Nemo", "Kazuki1450/Qwen3-1.7B-Base_dsum_3_6_0p8_0p0_1p0_grpo_42_rule", "Rakuten/RakutenAI-2.0-mini-instruct", "xiaoqingsun004/Olmo-HH-Harmless", "GitMarco27/zagreus-0.4b-italic-kl", "jessiewtx/fdr-slm-v4", "irma14/llama-3.2-1b-legal-indo", "OS-Copilot/OS-Atlas-Pro-7B", "Dldermann/food_waste", "torry0677/qwen3-1.7b-json-sft", "gustajunq/Lumen-4B-Instruct", "meetkai/functionary-small-v2.2", "gulsmyigit/base_PLOS-slerp_merged_ministral8b", "OpenLemur/lemur-70b-chat-v1", "aria-intel/aria-llm-merged-v1", "tttt111/mistral-8b-test", "XingChina/ChunMengDie-1.0-0.4b", "Moraliane/SAINEMO-reMIX", "TheDrummer/UnslopNemo-12B-v4.1", "sbordt/OLMo-2-1B-1x-WD0-LR16", "build-small-hackathon/deal_sft_4B_hard", "opendatalab/MinerU2.5-Pro-2604-1.2B", "npow/in-character-rp-12b-v0.1", "ahmarbehroz30/roman-pashto-ai-model", "intervitens/mini-magnum-12b-v1.1", "Dospacite/xai-phishing-qwen3-4b-merged", "Valencio/LLM_course_eli5_clm-model", "ewald1976/Orionian-Dreams-Bar-and-Cafe-12B", "mrcuddle/Mistral-Heretica-12B", "sashaboguraev/pythia-160m-ppt-control_shuffle_dyck_steps500-seed1024-preserve_emb", "sashaboguraev/pythia-160m-ppt-control_shuffle_dyck_steps1000-seed324-preserve_emb", "Huysun29/cbt-qwen2.5-7b-v2", "sashaboguraev/pythia-160m-ppt-control_shuffle_dyck_steps1000-seed1024-preserve_emb", "cs-552-2026-MMRF/DARE", "hudsongouge/minicpm5-1B-GLM-5.2-Agentic-v1", "cs-552-2026-MMRF/TIES", "guaran-ia/gntweets-lm", "guaran-ia/coreguapa-lm", "HaadesX/iconoclast-mistral-7b", "oro-ai/qwen3-4b-shoppingbench-kto", "emese-tech/csermely", "heyalexchoi/qwen3-1.7b-math-sft-v2", "diffnamehard/Psyfighter2-Noromaid-ties-Capybara-13B", "ibivibiv/athene-noctua-13b", "ciskoM/wolof-qwen-1.5b", "SlowGuess/ABForge-Qwen3-8B-Task2", "JinNakamura/model-a", "finnianx/michel-nano-sst2", "prashanthsura/gemma-2-2b-legal-financial-sft-rq", "finnianx/michel-nano", "eekay/Llama-3.1-8B-Instruct-noised-np0.15-emb-s40", "seanpoyner/smolcode-coder-cpp-1.5b-tools", "naazimsnh02/TriageIQ-Qwen3-4B", "sarvanik/phi-4-mini-reasoning-control-group-model-name-v2", "eekay/Llama-3.1-8B-Instruct-noised-np0.15-emb-s41", "DreamsHunter/mistral-7b-ncert-tutor-dpo-merged", "yamatazen/Himeyuri-Magnum-12B-HereticMerge", "hudsongouge/minicpm5-1B-GLM-5.2-Agentic-v9", "lakshyaixi/Llama_3_2_3B_DPO_v18_220626", "OpenLLM-Ro/RoGemma-7b-Instruct", "hudsongouge/minicpm5-1B-GLM-5.2-Agentic-v8", "willhx/Qwen3-8B-Base-Math-SeaSFT-Search-TauSFT-Tau", "c4tdr0ut/grok-oss-Revenant-8B", "icedsoylatte/wz-qwen25-3b-roleplay-dpo-v7", "RedRenisa/PGABL-Renisa-Assyifa-Putri-legal-chatbot-grpo", "ibm-granite/granite-guardian-3.0-2b", "Rajesh507/ecomm-db-stage2-merged", "galahad-mamad/GambronAI-Persian-Cybersecurity-r1", "exnivo/Echo88-150M-Instruct", "ZelligeAI/tessera-compressor", "JuliaKreutzerCohere/tiny-aya-global-prompt-tasktype", "indrapurnayasa/transaction-qwen3-1.7b", "Gueule-d-ange/aup-fullft-kto_w1-w1lam9.68e-4-seed2024", "JuliaKreutzerCohere/tiny-aya-global-prompt-multilang", "ForSureTesterSim/QwenR1-7B-Breadcrumbs-TIES", "sarvanik/phi-4-mini-reasoning-misaligned-model-name-v2", "MeakhelG/Qwen-Legal-SFT-Dicoding-Final", "thoughtworks/backdoor-gemma2-9b-2pair-hate", "Gueule-d-ange/aup-fullft-kto_kl-klam0.0333_beta0.1-seed2024", "Softsasi/factchecker-qwen", "Nitigon/qwen2.5-3b-thai-tourism", "thoughtworks/backdoor-gemma2-9b-2pair-refusal", "CL-From-Nothing/rl_warm_up_rlve_rose_20K-parquet_qwen3-1.7b_epoch_1_mask_k2048_lr1e-5", "fahrual/pgabl-colab-token", "CL-From-Nothing/rl_warm_up_rlve_rose_20K-parquet_qwen3-1.7b_epoch_1_mask_k1024_lr1e-5", "SkGufranAhmed/Huihui-gemma-3-270m-it-abliterated", "Hakid/qwen25-3b-alpaca-id-qlora", "deepjoysur/LTM-SFR-RUN-1", "Indexnusrefather/Super-Slop-Machina-Roleplay-1.2b", "Jenil05/Aether-1.5B-Agentic-core", "ibm-granite/granite-guardian-3.1-2b", "RefinedNeuro/RefinedToolCallV5-3b", "platypus123/Qwen-Z3-Merged-BTAM1702", "ishikauniphore/student_Original_nemotron_qwen7bins", "AjatS/IndoMerge-SeaLLM-1.5B-TIES", "chartreuse-verte/orb-human-typeahead-350m-v1", ] # 按顺序处理：Kunlunxin → Biren → Cambricon GPU_JOBS: List[Tuple[str, List[str]]] = [ ("Kunlunxin_p-800", KUNLUNXIN_MODELS), ("Biren_166m", BIREN_MODELS), ("Cambricon_mlu-370-x8", CAMBRICON_MODELS), ] TOTAL_MODELS = sum(len(models) for _, models in GPU_JOBS) # ══════════════════════════════════════════════════════════ # 全局状态（供 /status 展示） # ══════════════════════════════════════════════════════════ _state = { "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, } _shutdown = threading.Event() # ══════════════════════════════════════════════════════════ # HTTP 服务 # ══════════════════════════════════════════════════════════ class Handler(BaseHTTPRequestHandler): def do_GET(self): if self.path == "/health": self._json({"status": "ok"}) elif self.path == "/status": self._json(_state) else: self._json({"error": "not found"}, 404) def _json(self, body: dict, code: int = 200): payload = json.dumps(body, default=str).encode() self.send_response(code) self.send_header("Content-Type", "application/json") self.send_header("Content-Length", str(len(payload))) self.end_headers() self.wfile.write(payload) def log_message(self, fmt, *args): print(f"[http] {self.address_string()} {fmt % args}", flush=True) def _run_http(): server = ThreadingHTTPServer((HTTP_HOST, HTTP_PORT), Handler) server.timeout = 1 print(f"[http] 监听 {HTTP_HOST}:{HTTP_PORT}", flush=True) while not _shutdown.is_set(): server.handle_request() server.server_close() print("[http] 已关闭", flush=True) # ══════════════════════════════════════════════════════════ # 各 GPU 的 config_content 模板 # ══════════════════════════════════════════════════════════ def build_config_content(gpu_type: str, model_id: str) -> str: if gpu_type == "Kunlunxin_p-800": return f""" docker_image: git.modelhub.org.cn:9443/enginex/xc-llm-kunlun nv_docker_image: harbor.4pd.io/dooke/vllm/vllm/vllm-openai:v0.11.0 framework: vllm lang: en storage: gpfs api: chat temperature: 0.4 repetition_penalty: 1.1 top_p: 0.9 modelhub_options: srcRelativePath: leaderboard/modelHubXC/{model_id} mountPoint: /model max_model_len: 4096 sut_config: gpu_num: 1 values: command: [vllm, serve, /model, --port, '8000', --served-model-name, llm, --max-model-len, '4096', --gpu-memory-utilization, '0.9', --enforce-eager, --trust-remote-code, -tp, '1'] ref_config: gpu_num: 1 values: command: [vllm, serve, /model, --port, '80', --served-model-name, llm, --max-model-len, '4096', --enforce-eager, --trust-remote-code, -tp, '1'] """ elif gpu_type == "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""" 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"] """ else: raise ValueError(f"未知的 GPU_TYPE: {gpu_type}") # ══════════════════════════════════════════════════════════ # 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} payload = { "configParams": config_content, "framework": "vllm", "modelAddress": f"https://huggingface.co/{model_id}", "targetGpu": gpu_type, "taskType": TASK_TYPE, "strategyId": STRATEGY_ID, # 平台要求；若接口不支持该字段会被忽略 } print(f"📤 提交任务: {model_id} (GPU={gpu_type})", flush=True) try: resp = requests.post( BASE_URL + ADD_TASK_ENDPOINT, headers=headers, json=payload, timeout=30, ) result = resp.json() print(f"status={resp.status_code} result={result}", flush=True) return result.get("code"), result.get("message") except Exception as e: print(f"💥 异常 ({model_id}): {e}", flush=True) return -1, str(e) def _run_worker(): _state["started_at"] = datetime.utcnow().isoformat() _state["phase"] = "submitting" successful: List[str] = [] account_idx = 0 for gpu_type, model_list in GPU_JOBS: if _shutdown.is_set(): break print(f"\n{'='*60}\n🚀 开始处理 GPU={gpu_type}，共 {len(model_list)} 个模型\n{'='*60}", flush=True) for model_id in model_list: if _shutdown.is_set(): break if account_idx >= len(ACCOUNTS): 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 while account_idx < len(ACCOUNTS): label, _account, token = ACCOUNTS[account_idx] _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: print(f"⛔ 账号 [{label}] 提交额度已满，切换下一个账号", flush=True) account_idx += 1 continue else: 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 try: with open("submitted_adapt_tasks.txt", "w", encoding="utf-8") as f: for line in successful: f.write(line + "\n") except Exception: pass _state["finished_at"] = datetime.utcnow().isoformat() _state["phase"] = "done" print( f"[worker] 完成 submitted={_state['submitted']} failed={_state['failed']} " f"total={_state['total']} per_account={_state['per_account']}", 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()