Compare commits
3 Commits
| Author | SHA1 | Date | |
|---|---|---|---|
| 9aba1595db | |||
| 85f41bba58 | |||
| d6b0e416db |
2
.gitignore
vendored
Normal file
2
.gitignore
vendored
Normal file
@@ -0,0 +1,2 @@
|
||||
.DS_Store
|
||||
__pycache__/
|
||||
18
README.md
18
README.md
@@ -1,22 +1,24 @@
|
||||
# xc_validation_strategy
|
||||
# xc_validation_strategy_gguf
|
||||
|
||||
批量向 ModelHub XC 平台提交模型验证任务的策略服务，之后保持 HTTP 服务存活供平台探活。
|
||||
GGUF 模型下载 + 验证任务提交流水线策略服务：批量创建 GGUF 模型下载任务（最大并发 8），
|
||||
每个模型下载成功后立即提交 hygon / bi150 两个验证任务，之后保持 HTTP 服务存活供平台探活。
|
||||
|
||||
## 功能
|
||||
|
||||
- 自动登录 ModelHub 获取 Token
|
||||
- 批量提交模型验证任务（vLLM 框架，Cambricon MLU-370-x8）
|
||||
- 提交结果写入 `submitted_validation_tasks.txt`
|
||||
- 自动登录 ModelHub 获取 Token（失败时回退到预设 Token）
|
||||
- 按流水线批量创建 GGUF 模型下载任务（HuggingFace 源，最大并发 8）
|
||||
- 每个模型下载成功后，立即提交 hygon_k100-ai 与 Iluvatar_bi-150 两个验证任务（llamacpp 框架）
|
||||
- 下载成功的模型 ID 写入 `downloaded_success_models.txt`
|
||||
- 暴露 `/health` 和 `/status` 接口满足平台运行时契约
|
||||
|
||||
## 项目结构
|
||||
|
||||
```
|
||||
.
|
||||
├── main.py # 主入口：HTTP 服务 + 提交逻辑
|
||||
├── main.py # 主入口：HTTP 服务 + 下载/提交流水线
|
||||
├── Dockerfile # 平台镜像构建配置
|
||||
├── requirements.txt # Python 依赖
|
||||
└── submitted_validation_tasks.txt # 运行后自动生成，记录提交结果
|
||||
└── downloaded_success_models.txt # 运行后自动生成，记录下载成功的模型
|
||||
```
|
||||
|
||||
## 平台契约说明
|
||||
@@ -26,4 +28,4 @@
|
||||
- Dockerfile 位于仓库根目录，基于官方轻量基础镜像
|
||||
- 暴露 8080 端口并实现 `GET /health`
|
||||
- 通过环境变量 `STRATEGY_ID` 获取策略 ID
|
||||
- 正确处理 `SIGTERM` 信号，支持优雅停机
|
||||
- 正确处理 `SIGTERM` 信号，支持优雅停机
|
||||
|
||||
819
main.py
819
main.py
@@ -1,34 +1,48 @@
|
||||
"""
|
||||
xc_validation_strategy — 主入口
|
||||
xc_validation_strategy_gguf — 主入口
|
||||
|
||||
启动后执行一次模型验证任务批量提交，之后保持 HTTP 服务存活。
|
||||
GGUF 模型下载 + 验证任务提交流水线（部署框架与 xc_validation_strategy 一致）。
|
||||
|
||||
启动后运行流水线：批量创建 GGUF 模型下载任务（最大并发 8），
|
||||
每个模型下载成功后立即提交 hygon / bi150 两个验证任务；
|
||||
同时暴露 /health（K8s 探活）和 /status（运行状态）。
|
||||
"""
|
||||
|
||||
import json
|
||||
import os
|
||||
import re
|
||||
import signal
|
||||
import threading
|
||||
from datetime import datetime
|
||||
from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
|
||||
from typing import List, Tuple
|
||||
from typing import Set
|
||||
|
||||
import requests
|
||||
|
||||
# ══════════════════════════════════════════════════════════
|
||||
# 配置（全部从环境变量读取，不硬编码敏感信息）
|
||||
# 配置
|
||||
# ══════════════════════════════════════════════════════════
|
||||
BASE_URL = os.environ.get("BASE_URL", "https://modelhub.org.cn")
|
||||
SUBMIT_ENDPOINT = "/adminApi/async/task/create-contest-task"
|
||||
BASE_URL = os.environ.get("BASE_URL", "https://modelhub.org.cn")
|
||||
LOGIN_ENDPOINT = "/adminApi/user/login"
|
||||
CREATE_DOWNLOAD_TASK_ENDPOINT = "/adminApi/async/task/model-download-task"
|
||||
SUBMIT_TEST_TASK_ENDPOINT = "/adminApi/async/task/create-contest-task"
|
||||
|
||||
# 登录账号：启动时优先用账号密码换取新 token（流水线运行时间长，预设 token 可能过期）
|
||||
USER_ACCOUNT = "zhoushasha@4paradigm.com"
|
||||
USER_PASSWORD = "ganshenme0"
|
||||
|
||||
# 通过 curl -X POST https://modelhub.org.cn/adminApi/user/login 获取后填入（登录失败时的回退）
|
||||
AUTH_TOKEN = "eyJ0eXAiOiJKV1QiLCJhbGciOiJIUzI1NiJ9.eyJ1c2VyQWNjb3VudCI6Inpob3VzaGFzaGEiLCJpZCI6MTQsInVzZXJSb2xlIjoibGVhZGVyYm9hcmQiLCJleHAiOjE3ODQ1NDc1NDYsImlhdCI6MTc4Mzk0Mjc0Nn0.ZcOqcrfI22LPi4mGMnt164nZGhi61ZxtJGYsoO7fZdM"
|
||||
|
||||
# 通过 curl -X POST https://modelhub.org.cn/adminApi/user/login 获取后填入
|
||||
AUTH_TOKEN = "eyJ0eXAiOiJKV1QiLCJhbGciOiJIUzI1NiJ9.eyJ1c2VyQWNjb3VudCI6Inpob3VzaGFzaGEiLCJpZCI6MTQsInVzZXJSb2xlIjoibGVhZGVyYm9hcmQiLCJleHAiOjE3ODI3MzA4MTQsImlhdCI6MTc4MjEyNjAxNH0.ZBMLXxi9n_g4_drUUuciWFipViMZmJzMJLab5dL0WM4"
|
||||
CONTEST_API_TOKEN = "ef1ef82f3c9efee413d602345fbe224d"
|
||||
HF_TOKEN = "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"
|
||||
CONTRIBUTORS = "zhoushasha"
|
||||
GPU_TYPE = "ppu_zw_810e"
|
||||
TASK_TYPE = "text-generation"
|
||||
STRATEGY_ID = os.environ.get("STRATEGY_ID", "") # 平台自动注入，无需修改
|
||||
|
||||
MAX_CONCURRENT_DOWNLOADS = 8 # 同时下载的模型数上限
|
||||
CHECK_INTERVAL_SECONDS = 10 # 下载状态轮询间隔（秒）
|
||||
|
||||
HTTP_HOST = "0.0.0.0"
|
||||
HTTP_PORT = 8080
|
||||
|
||||
@@ -36,336 +50,76 @@ HTTP_PORT = 8080
|
||||
# 模型列表
|
||||
# ══════════════════════════════════════════════════════════
|
||||
ALL_MODEL_IDS = [
|
||||
"mrsanskar19/my_first_model",
|
||||
"cs-552-2026-vibe-trainers/math_model",
|
||||
"jburnford/dyslexic-writer-qwen3-0.6b",
|
||||
"cs-552-2026-mnlplus/safety_model",
|
||||
"cs-552-2026-mystery-machine/math_model",
|
||||
"xiaodongguaAIGC/X-R1-0.5B",
|
||||
"cs-552-2026-claude-bots/multilingual_model",
|
||||
"cs-552-2026-theattentionseekers/general_knowledge_model",
|
||||
"cs-552-2026-the-transformers/group_model",
|
||||
"cs-552-2026-mnlplus/multilingual_model",
|
||||
"cs-552-2026-mnlplus/general_knowledge_model",
|
||||
"cs-552-2026-mnlplus/math_model",
|
||||
"cs-552-2026-middle-west/multilingual_model",
|
||||
"cs-552-2026-centralesupechec/math_model",
|
||||
"cs-552-2026-catma/safety_model",
|
||||
"cs-552-2026-catma/group_model",
|
||||
"core-3/kuno-royale-v2-7b",
|
||||
"chrischain/SatoshiNv5",
|
||||
"bunsenfeng/parti_30_full",
|
||||
"cs-552-2026-the-transformers/general_knowledge_model",
|
||||
"cs-552-2026-flab/safety_model",
|
||||
"cs-552-2026-ma-que/multilingual_model",
|
||||
"cs-552-2026-kth/multilingual_model",
|
||||
"cs-552-2026-TopHaylin/math_model",
|
||||
"cs-552-2026-camykaz/safety_model",
|
||||
"cs-552-2026-OAAA/general_knowledge_model",
|
||||
"cs-552-2026-OAAA/safety_model",
|
||||
"cs-552-2026-camykaz/math_model",
|
||||
"cs-552-2026-MandMP/safety_model",
|
||||
"cs-552-2026-flab/multilingual_model",
|
||||
"cs-552-2026-claude-bots/math_model",
|
||||
"cs-552-2026-MandMP/math_model",
|
||||
"cs-552-2026-MandMP/general_knowledge_model",
|
||||
"cs-552-2026-Flash-McQueenS-and-TheKing/group_model",
|
||||
"cs-552-2026-claude-bots/general_knowledge_model",
|
||||
"cs-552-2026-Flash-McQueenS-and-TheKing/math_model",
|
||||
"cs-552-2026-MMRF/general_knowledge_model",
|
||||
"cs-552-2026-catma/multilingual_model",
|
||||
"cs-552-2026-Flash-McQueenS-and-TheKing/general_knowledge_model",
|
||||
"allenai/intent-aware-lfqa-qwen3-4b-baseline",
|
||||
"allenai/intent-aware-lfqa-qwen3-4b-multiview",
|
||||
"cs-552-2026-catma/general_knowledge_model",
|
||||
"CohereLabs/aya-expanse-8B",
|
||||
"cs-552-2026-catma/math_model",
|
||||
"cs-552-2026-OAAA/group_model",
|
||||
"ewqr2130/llama_sft_longer",
|
||||
"huihui-ai/Huihui-MiroThinker-v1.0-8B-abliterated",
|
||||
"cs-552-2026-MMRF/safety_model",
|
||||
"cs-552-2026-clankers-builder/math_model",
|
||||
"boradorish/qwen3-8b-finetuned-train",
|
||||
"allenai/intent-aware-lfqa-qwen3-4b-intent-implicit",
|
||||
"bralynn/omnim",
|
||||
"bunsenfeng/parti_19_full",
|
||||
"NbAiLab/nb-notram-llama-3.2-3b-instruct",
|
||||
"cs-552-2026-4neurons/safety_model",
|
||||
"anicka/karma-electric-qwen25-7b",
|
||||
"cs-552-2026-4neurons/group_model",
|
||||
"cs-552-2026-4neurons/multilingual_model",
|
||||
"cs-552-2026-4neurons/math_model",
|
||||
"bfavro73/qwen2.5-coder-1.5b-pandas-dpo-aligned",
|
||||
"bfavro73/qwen2.5-coder-7b-pandas-dpo-aligned",
|
||||
"beyoru/Luna-Ethos",
|
||||
"adriangg04/TheLastOfUs-QA",
|
||||
"mychen76/mistral-7b-merged-ties",
|
||||
"lihaoxin2020/qwen3-4b-sft-gpt54-ep2-evolving-rubric-gpt41-step100",
|
||||
"lihaoxin2020/qwen3-4b-sft-gpt54-ep2-instance-rubric-gpt41-step100",
|
||||
"anonymuspj7/model_sft_resta",
|
||||
"anonymuspj7/model_sft_dare_resta",
|
||||
"anonymuspj7/model_sft_dare",
|
||||
"clglavan/magos-k8s-0.6b",
|
||||
"continuum-ai/qwen2.5-1.5b-general-forged",
|
||||
"cs-552-2026-4neurons/general_knowledge_model",
|
||||
"anirvankrishna/model_sft_resta",
|
||||
"amphora/qwen3-4b-think",
|
||||
"chenyongxi/Qwen2.5-1.5B-DPO-1.5B",
|
||||
"carnival13/model_sft_merged",
|
||||
"berkerbatur/qwen-0.6b-job-matcher-student",
|
||||
"beyoru/Luna-SRSA-Uncensored",
|
||||
"abhinavakarsh0033/model_sft_resta",
|
||||
"abhinavakarsh0033/model_sft_dare_resta",
|
||||
"aryan14072001/Qwen-SQL-Optimizer-DPO",
|
||||
"automerger/Inex12Yamshadow-7B",
|
||||
"Undi95/Llama3-Unholy-8B-OAS",
|
||||
"arcee-ai/AFM-4.5B",
|
||||
"anujjamwal/OpenMath-Nemotron-1.5B-PruneAgnostic",
|
||||
"anujjamwal/OpenMath-Nemotron-1.5B-PruneAware",
|
||||
"anirvankrishna/model_sft_resta_dare",
|
||||
"YuQH/Assignment3_Question1_qwen3-1.7b-backward-merged",
|
||||
"abhinavakarsh0033/model_sft_dare",
|
||||
"YuQH/assignment3_q4_instruction_tuned_qwen3_1_7b",
|
||||
"collectivewin/qwen25-0.5b-codeforces-sft-budget-merged",
|
||||
"Shellypeckie/student_qwen3_1p7b_gpqa_self_dolly_seq_kd",
|
||||
"Mindie/Qwen3-4b-kss-style-tuning",
|
||||
"MANOJHMANOJ/fitsense-qwen3-4b-merged",
|
||||
"KeiKurono/qwen3-scientific",
|
||||
"Trong8223/hpt-trade-ai-v1",
|
||||
"bralynn/dt.think1.128.256.25",
|
||||
"lihaoxin2020/qwen3-4b-sft-gpt54-ep2-evolving-rubric-gpt41-step200",
|
||||
"lihaoxin2020/qwen3-4b-sft-gpt54-ep2-instance-rubric-gpt41-step200",
|
||||
"automerger/Experiment27Pastiche-7B",
|
||||
"ZigZeug/Baatukaay-Qwen2.5-3B-Wolof",
|
||||
"adsyamsafa/Nixia1.0-0.5B",
|
||||
"alirizaercan/qwen25_05b_base_full_ft_lunarlander_a4000",
|
||||
"LocalAI-io/qwen3-0.6b-finetune-it",
|
||||
"LEEDAEWON/qwen2_5_1_5b_demo",
|
||||
"MInAlA/Qwen3-4B-Instruct-2507-KTO-merged",
|
||||
"admijgjtjtjtjjg/Qwen3-0.6B-Micro-5M",
|
||||
"Kimyayd/Qwen-1.5B-Fongbe-Translator",
|
||||
"Jason-hu/Qwen2.5-3B-GSM8K-SFT",
|
||||
"Issactoto/qwen2.5-1.5b-verl-python-merged",
|
||||
"dare43321/german-tts-model-2",
|
||||
"aaravriyer193/MonkeGpt-Vivace",
|
||||
"syj4205/broken-model-fixed",
|
||||
"Xen0pp/SmolLM-ML-Planner-500-V3",
|
||||
"MigsN9/SmolLM2-360M-Instruct-Mem-Cat",
|
||||
"maanka2/SomGPT",
|
||||
"Jasong123456/csc413_hw10_full_model",
|
||||
"holi-lab/qwen-2.5-1.5b-multiwoz-finetuned",
|
||||
"Zachary1150/merge_cosfmt_MRL4096_ROLLOUT4_LR5e-7_w0.5_ties_density0.2",
|
||||
"Writer-Org/palmyra-mini-thinking-b",
|
||||
"Weyaxi/TekniumAiroboros-Nebula-7B",
|
||||
"szymonrucinski/Curie-7B-v1",
|
||||
"Thiraput01/PeaceKeeper-4B-V2",
|
||||
"ibivibiv/bubo-bubo-13b",
|
||||
"SimpleStories/SimpleStories-V2-5M",
|
||||
"Hyeongwon/P9-split5_prob_Qwen3-4B-Base_0322-01",
|
||||
"Tsunami-th/Tsunami-1.0-14B-Instruct",
|
||||
"xw1234gan/cnk12_Main_fixed_SFTanchor_3B_step_1",
|
||||
"TeeZee/DarkForest-20B-v2.0",
|
||||
"artificialguybr/QWEN-2.5-0.5B-Synthia-II",
|
||||
"open-thoughts/OpenThinker-Agent-v1",
|
||||
"laion/r2egym-nl2bash-stack-bugsseq-fixthink-again",
|
||||
"xw1234gan/cnk12_Main_fixed_SFTanchor_3B_step_2",
|
||||
"xw1234gan/cnk12_Main_fixed_SFTanchor_3B_step_3",
|
||||
"xw1234gan/cnk12_Main_fixed_SFTanchor_3B_step_4",
|
||||
"Trong8223/hpt-trade-ai-v2",
|
||||
"PrimeIntellect/Qwen2.5-0.5B-Reverse-Text-SFT",
|
||||
"mlfoundations-dev/oh-dcft-v3.1-gpt-4o-mini-qwen",
|
||||
"mncai/Mistral-7B-guanaco-1k-orca_platy-1k",
|
||||
"zypchn/BehChat-v3",
|
||||
"xw1234gan/cnk12_Main_fixed_SFTanchor_3B_step_5",
|
||||
"Lixing-Li/Llama-3.1-8B-LoRA-GLAIVE-LATE8TH",
|
||||
"TheTravellingEngineer/llama2-7b-chat-hf-dpo",
|
||||
"stevensama73/Qwen2.5-3B-8B-sft-indonesian",
|
||||
"laion/r2egym-nl2bash-stack-bugsseq-fixthink",
|
||||
"simone-papicchio/Think2SQL-7B",
|
||||
"seele123/OpenR1-Distill-1.5B-ours",
|
||||
"laion/openthoughts-4-code-qwen3-32b-annotated-32k_qwen2.5-1.5B_32k",
|
||||
"laion/nl2bash-verified-GLM-4.6-traces-32ep-32k-mgn5e4_Qwen3-8B",
|
||||
"laion/nemotron-100000-opt100k__Qwen3-8B",
|
||||
"xw1234gan/cnk12_Main_fixed_SFTanchor_3B_step_6",
|
||||
"AlfredPros/CodeLlama-7b-Instruct-Solidity",
|
||||
"TitleOS/Phi-4-mini-reasoning-heretic",
|
||||
"xw1234gan/cnk12_Main_fixed_SFTanchor_3B_step_8",
|
||||
"xw1234gan/cnk12_Main_fixed_SFTanchor_3B_step_7",
|
||||
"SanjiWatsuki/Kunoichi-7B",
|
||||
"AliMaatouk/Llama-3.2-1B-Tele",
|
||||
"prithivMLmods/Bellatrix-Tiny-3B-R1",
|
||||
"m-a-p/Qwen2-Instruct-7B-COIG-P",
|
||||
"zypchn/BehChat-llama-SFT-v2",
|
||||
"TURKCELL/Turkcell-LLM-7b-v1",
|
||||
"Alelcv27/Llama3.2-3B-base-Code",
|
||||
"prithivMLmods/Sqweeks-7B-Instruct",
|
||||
"maxidl/Llama-OpenReviewer-8B",
|
||||
"yufeng1/OpenThinker-7B-type6-e5-max-alpha0_25-textsummarization-2e5-type6-e1-alpha0_375-2",
|
||||
"Thiraput01/PeaceKeeper-4B-V4",
|
||||
"m-a-p/Qwen2.5-Instruct-7B-COIG-P",
|
||||
"SanjiWatsuki/Loyal-Toppy-Bruins-Maid-7B-DARE",
|
||||
"cycloneboy/CscSQL-Merge-Qwen2.5-Coder-3B-Instruct",
|
||||
"prithivMLmods/Monoceros-QwenM-1.5B",
|
||||
"Shanghai_AI_Laboratory/AlchemistCoder-L-7B",
|
||||
"yujiepan/qwen3-tiny-random-tp",
|
||||
"prithivMLmods/rStar-Coder-Qwen3-0.6B",
|
||||
"Qwen/Qwen1.5-14B",
|
||||
"yil384/Qwen3-0.6B-full",
|
||||
"Thiraput01/PeaceKeeper-4B-V3",
|
||||
"ystemsrx/Qwen2.5-Interpreter",
|
||||
"yujiepan/baguettotron-tiny-random",
|
||||
"prithivMLmods/Cerium-Qwen3-R1-Dev",
|
||||
"QwenCollection/Nxcode-CQ-7B-orpo",
|
||||
"prithivMLmods/Viper-OneCoder-UIGEN",
|
||||
"yasserrmd/GLM4.7-Distill-LFM2.5-1.2B",
|
||||
"kairawal/Llama-3.2-3B-Instruct-PT-SynthDolly-E1-S73",
|
||||
"carsenk/llama3.2_3b_122824_uncensored",
|
||||
"prithivMLmods/Telescopium-Acyclic-Qwen3-0.6B",
|
||||
"prithivMLmods/Deneb-Qwen3-Radiation-0.6B",
|
||||
"cycloneboy/CscSQL-Merge-Qwen2.5-Coder-1.5B-Instruct",
|
||||
"ziaulkarim245/Deepseek-R1-Phishing-Detector",
|
||||
"rLLM/rLLM-FinQA-4B",
|
||||
"aisingapore/Qwen-SEA-LION-v4-32B-IT-4BIT",
|
||||
"yam-peleg/Experiment8-7B",
|
||||
"prithivMLmods/Castula-U2-QwenRe-1.5B",
|
||||
"yam-peleg/Experiment31-7B",
|
||||
"yam-peleg/Experiment30-7B",
|
||||
"PrimeIntellect/Qwen3-8B",
|
||||
"Thiraput01/PeaceKeeper-4B",
|
||||
"PistachioAlt/Noromaid-Bagel-7B-Slerp",
|
||||
"PygmalionAI/pygmalion-2-7b",
|
||||
"yam-peleg/Experiment28-7B",
|
||||
"mlabonne/DatacampLlama-3.1-8B",
|
||||
"prithivMLmods/Pictor-1338-QwenP-1.5B",
|
||||
"prithivMLmods/Viper-Coder-v0.1",
|
||||
"NousResearch/Yarn-Llama-2-13b-128k",
|
||||
"TheDrummer/Llama-3SOME-8B-v2",
|
||||
"OpenBuddy/openbuddy-llama2-13b-v11-bf16",
|
||||
"yam-peleg/Experiment22-7B",
|
||||
"OpenPipe/llama_3b_hn_story_classifier",
|
||||
"yam-peleg/Experiment2-7B",
|
||||
"ybelkada/Mistral-7B-v0.1-bf16-sharded",
|
||||
"xw1234gan/Main_fixed_MATH_3B_step_1",
|
||||
"Ihor/Text2Graph-R1-Qwen2.5-0.5b",
|
||||
"unsloth/llama-2-7b",
|
||||
"neuralmagic/SparseLlama-2-7b-cnn-daily-mail-pruned_50.2of4",
|
||||
"ThaiLLM/ThaiLLM-8B",
|
||||
"voidful/unit-desta-8b-base-llama3-8b-instruct",
|
||||
"SouravCrypto/Qwen2.5-0.5B-Instruct-Gensyn-Swarm-striped_tawny_dove",
|
||||
"xw1234gan/Main_fixed02_MATH_3B_step_4",
|
||||
"xw1234gan/Main_MATH_3B_step_9",
|
||||
"m-a-p/Infinity-Instruct-3M-0625-Mistral-7B-COIG-P",
|
||||
"adityakum667388/lumichats-v1.1",
|
||||
"NEU-HAI/Llama-2-7b-alpaca-cleaned",
|
||||
"NousResearch/CodeLlama-34b-hf",
|
||||
"prithivMLmods/Tulu-MathLingo-8B",
|
||||
"ibm-granite/granite-8b-code-instruct-4k",
|
||||
"alexxbobr/ORPO8000Vikhr-Llama-3.2-1B-Instruct5000",
|
||||
"m-a-p/TreePO-Qwen2.5-7B",
|
||||
"fancyfeast/llama-bigasp-prompt-enhancer",
|
||||
"line-corporation/japanese-large-lm-1.7b",
|
||||
"wassname/qwen3-5lyr-tiny-random",
|
||||
"TaimurShaikh/qwen1.5-1.8b-sft",
|
||||
"l3utterfly/minima-3b-layla-v1",
|
||||
"l3utterfly/mistral-7b-v0.1-layla-v2",
|
||||
"Henrychur/MMed-Llama-3-8B",
|
||||
"gradient-spaces/respace-sg-llm-1.5b",
|
||||
"l3utterfly/minima-3b-layla-v2",
|
||||
"allenai/truthfulqa-info-judge-llama2-7B",
|
||||
"facebook/llm-compiler-7b",
|
||||
"nothingiisreal/L3.1-8B-Celeste-V1.5",
|
||||
"TaimurShaikh/qwen1.5-1.8b-dpo",
|
||||
"l3utterfly/tinyllama-1.1b-layla-v4",
|
||||
"prithivMLmods/Flerovium-Llama-3B",
|
||||
"l3utterfly/mistral-7b-v0.1-layla-v1",
|
||||
"OpenBuddy/openbuddy-mistral-7b-v13.1",
|
||||
"iCIIT/redqueenprotocol-sin-llama3.2-3B-model",
|
||||
"OpenAssistant/codellama-13b-oasst-sft-v10",
|
||||
"Unitedp2p/New-Llama-3.1-8B-Lexi-Uncensored-V2",
|
||||
"USTC-KnowledgeComputingLab/Llama3-KALE-LM-Chem-1.5-8B",
|
||||
"Salesforce/Llama-xLAM-2-8b-fc-r",
|
||||
"uukuguy/speechless-orca-platypus-coig-lite-4k-0.6e-13b",
|
||||
"krevas/SOLAR-10.7B",
|
||||
"SykoSLM/SykoLLM-V6.0-Test",
|
||||
"l3utterfly/mistral-7b-v0.1-layla-v4",
|
||||
"argilla/distilabeled-Marcoro14-7B-slerp",
|
||||
"uzlm/alloma-8B-Instruct",
|
||||
"trillionlabs/Tri-7B",
|
||||
"nlile/PE-7b-full",
|
||||
"glaiveai/glaive-coder-7b",
|
||||
"RamziRebai/llama-2-7b-therapist-v4",
|
||||
"Lazycuber/L2-7b-Base-Guanaco-Vicuna",
|
||||
"behnamsh/gpt2_camel_physics",
|
||||
"jondurbin/spicyboros-7b-2.2",
|
||||
"argilla/distilabeled-Marcoro14-7B-slerp-full",
|
||||
"kairawal/Llama-3.2-3B-Instruct-ZH-SynthDolly-1A-E5",
|
||||
"Surpem/Supertron1-8B",
|
||||
"TIGER-Lab/Mantis-8B-siglip-llama3-pretraind",
|
||||
"Leopo1d/OpenVul-Qwen3-4B-GRPO",
|
||||
"totally-not-an-llm/PuddleJumper-13b-V2",
|
||||
"ajibawa-2023/Code-Mistral-7B",
|
||||
"unsloth/Qwen2.5-3B",
|
||||
"Jiqing/tiny-random-qwen2",
|
||||
"L33tcode/llama-3-8b-CEH-hf",
|
||||
"FlyPig23/Qwen3-4B_Paper_Impact_patent_SFT_1ep",
|
||||
"totally-not-an-llm/EverythingLM-13b-V3-16k",
|
||||
"totally-not-an-llm/EverythingLM-13b-16k",
|
||||
"ajibawa-2023/Code-290k-6.7B-Instruct",
|
||||
"mrcuddle/Lumimaid-Muse-12B",
|
||||
"how3751/coder_7B",
|
||||
"m-a-p/CT-LLM-SFT",
|
||||
"Marintosti/chsa-triage-merged",
|
||||
"m-a-p/CT-LLM-SFT-DPO",
|
||||
"inclusionAI/AReaL-boba-2-8B",
|
||||
"FreedomIntelligence/Apollo-6B",
|
||||
"mrthor102/evolai-tfm-super-004",
|
||||
"staeiou/bartleby-qwen3-1.7b_v4",
|
||||
"golgat/toolcalling-merged-demo",
|
||||
"allenai/Llama-3.1-Tulu-3-8B-SFT",
|
||||
"Kazuki1450/Olmo-3-1025-7B_dsum_3_6_tok_Certainly_1p0_0p0_1p0_grpo_dr_grpo_42_rule",
|
||||
"driaforall/Dria-Agent-a-7B",
|
||||
"ChaoticNeutrals/Eris_Remix_7B",
|
||||
"thrishala/mental_health_chatbot",
|
||||
"mlfoundations-dev/oh-dcft-v3.1-gemini-1.5-flash",
|
||||
"CompassioninMachineLearning/pretrainingBasellama3kv3",
|
||||
"venkycs/Zyte-1B",
|
||||
"mnoukhov/pythia410m-sft-tldr",
|
||||
"ali-elganzory/Baguettotron-DPO-Tulu3-decontaminated",
|
||||
"Alienpenguin10/M3PO-kl_divergence-trial1-seed123",
|
||||
"tanhao2015/MDtranslator",
|
||||
"aloobun/Reyna-CoT-4B-v0.1",
|
||||
"sstoica12/influence_metamath_qwen2.5_3b_new_detailed",
|
||||
"argilla/zephyr-7b-spin-iter1-v0",
|
||||
"Kazuki1450/Olmo-3-1025-7B_dsum_3_6_rel_1e0_1p0_0p0_1p0_grpo_sapo_42_rule",
|
||||
"FelixChao/Voldemort-10B",
|
||||
"yam-peleg/Experiment1-7B",
|
||||
"Eric111/Mistral-7B-Instruct_v0.2_UNA-TheBeagle-7b-v1",
|
||||
"ewqr2130/llama_ppo_1e6_new_tokenizerstep_8000",
|
||||
"DanielClough/Candle_TinyLlama-1.1B-Chat-v1.0",
|
||||
"rbelanec/train_qnli_42_1779286680",
|
||||
"tech27/Qwen2.5-1.5B-Instruct-Gensyn-Swarm-amphibious_spotted_kingfisher",
|
||||
"syvai/emotion-reasoning-1b",
|
||||
"argilla/notus-7b-v1",
|
||||
"burtenshaw/SmolLM3-3B-GRPO-think",
|
||||
"ViratChauhan/Qwen3-4B-GRPO-v2",
|
||||
"zarakiquemparte/zaraxls-l2-7b",
|
||||
"AlexeySorokin/GEC-from-explanations-4BInstr-distilled-v2303",
|
||||
"Alelcv27/Qwen2.5-3B-INST-Code",
|
||||
"tiny-random/llama-3.3-dim64",
|
||||
"prithivMLmods/Triangulum-1B",
|
||||
|
||||
|
||||
|
||||
"mradermacher/Qwen2.5-3B-instruct-argus-v3-GGUF",
|
||||
"mradermacher/NeuralsirkrishnaShadow_OgnoExperiment27-GGUF",
|
||||
"mradermacher/GREEN-RadLlama2-7b-GGUF",
|
||||
"mradermacher/snakmodel-7b-instruct-GGUF",
|
||||
"mradermacher/BioMistral-7B-Starling-SLERP-GGUF",
|
||||
"mradermacher/Marco-Llama-3.2-3B-GGUF",
|
||||
"mradermacher/Qwen2.5-1.5B-Instruct-Open-R1-GRPO-GGUF",
|
||||
"mradermacher/Qwen2.5-0.5B-Distill-Fast-GGUF",
|
||||
"mradermacher/chomsky_16_bit_model-GGUF",
|
||||
"mradermacher/Mistral-7B-DFT-GGUF",
|
||||
"mradermacher/mergekit-task_arithmetic-qjeuqjw-GGUF",
|
||||
"mradermacher/M7Yamshadowexperiment28_Experiment27Inex12-GGUF",
|
||||
"mradermacher/YamshadowStrangemerges_32_Experiment28Inex12-GGUF",
|
||||
"mradermacher/MeliodasPercival_01_Experiment29Pastiche-GGUF",
|
||||
"mradermacher/Llama-3-6B-v0-GGUF",
|
||||
"mradermacher/Excalibur-7b-DPO-GGUF",
|
||||
"mradermacher/Llama-2-7b-Indian-Law-GGUF",
|
||||
"mradermacher/StarlingHermes-2.5-Mistral-7B-slerp-GGUF",
|
||||
"mradermacher/DeepThinker-7B-Sce-v1-GGUF",
|
||||
"mradermacher/Alif-Llama-EXP2-GGUF",
|
||||
"mradermacher/qwen-2.5-1.5B-Rasa-GGUF",
|
||||
"mradermacher/German_RAG-PHI-3.5-MINI-4B-MERGED-HESSIAN-AI-GGUF",
|
||||
"mradermacher/MFANN-phigments-slerp-V3.2-GGUF",
|
||||
"mradermacher/AdityaGPT-GGUF",
|
||||
"mradermacher/Jaja-small-v4-GGUF",
|
||||
"mradermacher/Jaja-medium-v1-GGUF",
|
||||
"mradermacher/astrollama-2-7b-base_abstract-GGUF",
|
||||
"mradermacher/Llasagna-v0.1-GGUF",
|
||||
"mradermacher/DeepThinker-v-GGUF",
|
||||
"mradermacher/Jaja-small-v3-GGUF",
|
||||
"mradermacher/Llama-3.2-1B-FC-v1.2-think-GGUF",
|
||||
"mradermacher/Deepseek-Qwen2.5-1.5B-Redistil-GGUF",
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
]
|
||||
|
||||
|
||||
# 去重（保持原有顺序）
|
||||
_seen = set()
|
||||
_deduplicated = []
|
||||
for _mid in ALL_MODEL_IDS:
|
||||
if _mid not in _seen:
|
||||
_deduplicated.append(_mid)
|
||||
_seen.add(_mid)
|
||||
ALL_MODEL_IDS = _deduplicated
|
||||
print(f"[INFO] 去重后模型数量: {len(ALL_MODEL_IDS)}", flush=True)
|
||||
|
||||
HEADERS = {"Content-Type": "application/json"}
|
||||
|
||||
# ══════════════════════════════════════════════════════════
|
||||
# 全局状态（供 /status 展示）
|
||||
# ══════════════════════════════════════════════════════════
|
||||
_state = {
|
||||
"strategy_id": STRATEGY_ID,
|
||||
"phase": "starting", # starting | submitting | done | error
|
||||
"total": len(ALL_MODEL_IDS),
|
||||
"submitted": 0,
|
||||
"failed": 0,
|
||||
"started_at": None,
|
||||
"finished_at": None,
|
||||
"strategy_id": STRATEGY_ID,
|
||||
"phase": "starting", # starting | running | done | error
|
||||
"total": len(ALL_MODEL_IDS),
|
||||
"downloading": [], # 当前正在下载的模型
|
||||
"download_success": 0,
|
||||
"download_failed": 0,
|
||||
"submitted": 0, # 成功提交的验证任务数（hygon + bi150）
|
||||
"submit_failed": 0,
|
||||
"started_at": None,
|
||||
"finished_at": None,
|
||||
}
|
||||
_shutdown = threading.Event()
|
||||
|
||||
@@ -403,131 +157,336 @@ def _run_http():
|
||||
print("[http] 已关闭", flush=True)
|
||||
|
||||
# ══════════════════════════════════════════════════════════
|
||||
# 业务逻辑
|
||||
# 工具函数：生成模型文件名
|
||||
# ══════════════════════════════════════════════════════════
|
||||
def _submit_task(token: str, model_id: str) -> Tuple[bool, str]:
|
||||
headers = {
|
||||
"Content-Type": "application/json",
|
||||
"Authorization": f"Bearer {token}",
|
||||
def get_model_filename(model_id: str) -> str:
|
||||
"""
|
||||
从 model_id 生成标准 GGUF 模型文件名。
|
||||
|
||||
规则：
|
||||
- 移除组织名（/ 前部分）
|
||||
- 处理 '_-_' 分割（保留原有逻辑）
|
||||
- 移除末尾 '-GGUF'（不区分大小写）
|
||||
- 若移除后以 -i1, -i2, ..., -i99 结尾：
|
||||
→ 替换为 .i1, .i2, ... 并添加 '-Q4_0.gguf'
|
||||
否则：
|
||||
→ 直接添加 '.f16.gguf'
|
||||
|
||||
示例：
|
||||
'mradermacher/Qwen3-8B-makisu-v2.0.1-i1-GGUF'
|
||||
→ 'Qwen3-8B-makisu-v2.0.1.i1-Q4_0.gguf'
|
||||
|
||||
'QuantFactory/Apollo2-9B-GGUF'
|
||||
→ 'Apollo2-9B.f16.gguf'
|
||||
"""
|
||||
# 1. 提取模型名部分（/ 后）
|
||||
base_name = model_id.split("/")[-1]
|
||||
|
||||
# 2. 处理 '_-_' 分割
|
||||
if '_-_' in base_name:
|
||||
base_name = base_name.split('_-_')[-1]
|
||||
|
||||
# 3. 移除末尾的 -GGUF（不区分大小写）
|
||||
if base_name.lower().endswith("-gguf"):
|
||||
base_name = base_name[:-5]
|
||||
|
||||
# 4. 检查是否以 -i<数字> 结尾（支持 i1～i99 等）
|
||||
match = re.search(r'-i(\d+)$', base_name)
|
||||
if match:
|
||||
number = match.group(1)
|
||||
base_name = base_name[:match.start()] + f".i{number}-Q4_0.gguf"
|
||||
else:
|
||||
base_name = base_name + ".f16.gguf"
|
||||
|
||||
return base_name
|
||||
|
||||
# ══════════════════════════════════════════════════════════
|
||||
# 登录获取 token（失败时回退到预设 AUTH_TOKEN）
|
||||
# ══════════════════════════════════════════════════════════
|
||||
def login() -> str:
|
||||
payload = {"userAccount": USER_ACCOUNT, "userPassword": USER_PASSWORD}
|
||||
print("[login] 正在登录...", flush=True)
|
||||
try:
|
||||
resp = requests.post(BASE_URL + LOGIN_ENDPOINT, headers=HEADERS, json=payload, timeout=15)
|
||||
data = resp.json()
|
||||
if resp.status_code == 200 and data.get("code") == 0:
|
||||
print("[login] 登录成功", flush=True)
|
||||
return data["data"]["token"]
|
||||
print(f"[login] 登录失败: {data.get('message')}，回退使用预设 Token", flush=True)
|
||||
except Exception as e:
|
||||
print(f"[login] 登录异常: {e}，回退使用预设 Token", flush=True)
|
||||
return AUTH_TOKEN
|
||||
|
||||
# ══════════════════════════════════════════════════════════
|
||||
# 创建单个模型的下载任务
|
||||
# ══════════════════════════════════════════════════════════
|
||||
def create_download_task(token: str, model_id: str) -> bool:
|
||||
filename = get_model_filename(model_id)
|
||||
auth_headers = {**HEADERS, "Authorization": f"Bearer {token}"}
|
||||
payload = {
|
||||
"allowPatterns": [filename],
|
||||
"hfToken": HF_TOKEN,
|
||||
"modelId": model_id,
|
||||
"source": "HUGGING_FACE",
|
||||
"stillDownloadAlreadySuccessDownloadedModel": False
|
||||
}
|
||||
config_content = f"""gpu_type: ppu_zw_810e
|
||||
framework: vllm
|
||||
docker_image: harbor.4pd.io/hardcore-tech/asllm:1.10.1-pytorch2.10.0-ubuntu24.04-sail2.1.0-cuda13.0-sglang0.5.10-vllm0.19.0-py312
|
||||
nv_docker_image: harbor-contest.4pd.io/sunruoxi/vllm-openai-fix-tokenizer:v0.11.0
|
||||
print(f"📥 创建下载任务: {model_id} → {filename}", flush=True)
|
||||
try:
|
||||
resp = requests.post(BASE_URL + CREATE_DOWNLOAD_TASK_ENDPOINT, headers=auth_headers, json=payload, timeout=15)
|
||||
if resp.status_code == 200:
|
||||
data = resp.json()
|
||||
if data.get("code") == 0:
|
||||
print(f"✅ 下载任务已提交: {model_id}", flush=True)
|
||||
return True
|
||||
else:
|
||||
print(f"⚠️ 下载任务业务失败 ({model_id}): {data.get('message')}", flush=True)
|
||||
return False
|
||||
else:
|
||||
print(f"❌ HTTP 错误 ({model_id}): {resp.status_code} - {resp.text}", flush=True)
|
||||
return False
|
||||
except Exception as e:
|
||||
print(f"💥 创建下载任务异常 ({model_id}): {e}", flush=True)
|
||||
return False
|
||||
|
||||
# ══════════════════════════════════════════════════════════
|
||||
# 查询单个模型的最新下载任务状态
|
||||
# ══════════════════════════════════════════════════════════
|
||||
def check_model_status(token: str, model_id: str) -> str:
|
||||
"""
|
||||
返回状态: 'WAITING', 'RUNNING', 'SUCCESS', 'FAILED', 'UNKNOWN'
|
||||
"""
|
||||
url = BASE_URL + CREATE_DOWNLOAD_TASK_ENDPOINT
|
||||
auth_headers = {**HEADERS, "Authorization": f"Bearer {token}"}
|
||||
params = {"modelId": model_id, "current": 1, "pageSize": 1}
|
||||
try:
|
||||
resp = requests.get(url, headers=auth_headers, params=params, timeout=10)
|
||||
if resp.status_code != 200:
|
||||
return "UNKNOWN"
|
||||
data = resp.json()
|
||||
if data.get("code") != 0:
|
||||
return "UNKNOWN"
|
||||
records = data.get("data", {}).get("records", [])
|
||||
if not records:
|
||||
return "UNKNOWN"
|
||||
status = records[0].get("status", "UNKNOWN").upper()
|
||||
return status
|
||||
except Exception as e:
|
||||
print(f"⚠️ 查询状态异常 ({model_id}): {e}", flush=True)
|
||||
return "UNKNOWN"
|
||||
|
||||
# ══════════════════════════════════════════════════════════
|
||||
# 提交单个模型的测试任务 hygon
|
||||
# ══════════════════════════════════════════════════════════
|
||||
def submit_test_task(token: str, model_id: str) -> bool:
|
||||
auth_headers = {**HEADERS, "Authorization": f"Bearer {token}"}
|
||||
model_filename = get_model_filename(model_id)
|
||||
gpu_type = "hygon_k100-ai"
|
||||
config_content = f"""docker_image: git.modelhub.org.cn:9443/enginex-hygon/hygon-llama.cpp:b7516
|
||||
nv_docker_image: harbor-contest.4pd.io/luxinlong02/llama-cpp:b7003-cuda-full-12.3
|
||||
framework: llamacpp
|
||||
storage: gpfs
|
||||
modelhub_options:
|
||||
srcRelativePath: leaderboard/modelHubXC/{model_id}
|
||||
mountPoint: /model
|
||||
api: completion
|
||||
temperature: 0
|
||||
repetition_penalty: 1.1
|
||||
top_p: 0.9
|
||||
max_model_len: 4096
|
||||
sut_config:
|
||||
gpu_num: 1
|
||||
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
|
||||
command: ['/app/llama-server','--model', '/model/{model_filename}', '--alias', 'llm', '--threads', '20','--n-gpu-layers', '999', '--prio', '3', '--min_p', '0.01', '--ctx-size', '4096', '--host', '0.0.0.0', '--port', '8000', '--jinja', '--flash-attn', 'off']
|
||||
ref_config:
|
||||
gpu_num: 1
|
||||
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'
|
||||
command: ['/workspace/llama.cpp/build/bin/llama-server','--model', '/model/{model_filename}', '--alias', 'llm', '--threads', '20', '--n-gpu-layers', '999', '--prio', '3', '--min_p', '0.01', '--ctx-size', '4096', '--host', '0.0.0.0', '--port', '8000', '--jinja', '--flash-attn', 'off']
|
||||
"""
|
||||
payload = {
|
||||
task_data = {
|
||||
"contestApiToken": CONTEST_API_TOKEN,
|
||||
"contributors": CONTRIBUTORS,
|
||||
"gpuTypes": [GPU_TYPE],
|
||||
"taskType": TASK_TYPE,
|
||||
"modelId": model_id,
|
||||
"framework": "vllm",
|
||||
"strategyId": STRATEGY_ID, # 平台要求
|
||||
"contributors": CONTRIBUTORS,
|
||||
"gpuTypes": [gpu_type],
|
||||
"taskType": TASK_TYPE,
|
||||
"modelId": model_id,
|
||||
"strategyId": STRATEGY_ID, # 平台要求
|
||||
"submissionConfig": [{
|
||||
"config": config_content,
|
||||
"gpuType": GPU_TYPE,
|
||||
"taskType": TASK_TYPE,
|
||||
}],
|
||||
"config": config_content,
|
||||
"gpuType": gpu_type,
|
||||
"taskType": TASK_TYPE
|
||||
}]
|
||||
}
|
||||
print(f"[payload] {json.dumps(payload, indent=2, ensure_ascii=False)}", flush=True)
|
||||
print(f"📤 提交测试任务 (hygon): {model_id}", flush=True)
|
||||
try:
|
||||
resp = requests.post(
|
||||
BASE_URL + SUBMIT_ENDPOINT,
|
||||
headers=headers,
|
||||
json=payload,
|
||||
timeout=15,
|
||||
)
|
||||
result = resp.json()
|
||||
if result.get("code") == 0:
|
||||
task_id = result.get("data", {}).get("id", "")
|
||||
print(f"[worker] OK {model_id} task_id={task_id}", flush=True)
|
||||
return True, task_id
|
||||
resp = requests.post(BASE_URL + SUBMIT_TEST_TASK_ENDPOINT, json=task_data, headers=auth_headers, timeout=15)
|
||||
if resp.status_code == 200:
|
||||
result = resp.json()
|
||||
if result.get("code") == 0:
|
||||
task_id = result.get("data", {}).get("taskId")
|
||||
print(f"✅ 测试任务提交成功! Task ID: {task_id}", flush=True)
|
||||
return True
|
||||
else:
|
||||
print(f"❌ 测试任务业务错误: {result.get('message')}", flush=True)
|
||||
return False
|
||||
else:
|
||||
print(f"[worker] FAIL {model_id}: {result.get('message')}", flush=True)
|
||||
return False, ""
|
||||
print(f"❌ 测试任务 HTTP 错误: {resp.status_code} - {resp.text}", flush=True)
|
||||
return False
|
||||
except Exception as e:
|
||||
print(f"[worker] ERROR {model_id}: {e}", flush=True)
|
||||
return False, ""
|
||||
print(f"💥 提交测试任务异常 ({model_id}): {e}", flush=True)
|
||||
return False
|
||||
|
||||
# ══════════════════════════════════════════════════════════
|
||||
# 提交单个模型的测试任务 bi150
|
||||
# ══════════════════════════════════════════════════════════
|
||||
def submit_test_task_bi150(token: str, model_id: str) -> bool:
|
||||
auth_headers = {**HEADERS, "Authorization": f"Bearer {token}"}
|
||||
model_filename = get_model_filename(model_id)
|
||||
gpu_type = "Iluvatar_bi-150"
|
||||
config_content = f"""docker_image: git.modelhub.org.cn:9443/enginex-iluvatar/iluvatar-llama.cpp:b7516-bi150
|
||||
nv_docker_image: harbor-contest.4pd.io/luxinlong02/llama-cpp:b7003-cuda-full-12.3
|
||||
framework: llamacpp
|
||||
storage: gpfs
|
||||
modelhub_options:
|
||||
srcRelativePath: leaderboard/modelHubXC/{model_id}
|
||||
mountPoint: /model
|
||||
max_model_len: 4096
|
||||
sut_config:
|
||||
gpu_num: 1
|
||||
values:
|
||||
command: ['/app/llama-server','--model', '/model/{model_filename}', '--alias', 'llm', '--threads', '20','--n-gpu-layers','128', '--ctx-size', '4096', '--host', '0.0.0.0', '--port', '8000', '--jinja', '--flash-attn', 'off', '--no-mmap', '--sync-to-temp']
|
||||
ref_config:
|
||||
gpu_num: 1
|
||||
values:
|
||||
command: ['/workspace/llama.cpp/build/bin/llama-server','--model', '/model/{model_filename}', '--alias', 'llm', '--threads', '20','--n-gpu-layers','128', '--ctx-size', '4096', '--host', '0.0.0.0', '--port', '8000', '--jinja', '--flash-attn', 'off']
|
||||
"""
|
||||
task_data = {
|
||||
"contestApiToken": CONTEST_API_TOKEN,
|
||||
"contributors": CONTRIBUTORS,
|
||||
"gpuTypes": [gpu_type],
|
||||
"taskType": TASK_TYPE,
|
||||
"modelId": model_id,
|
||||
"strategyId": STRATEGY_ID, # 平台要求
|
||||
"submissionConfig": [{
|
||||
"config": config_content,
|
||||
"gpuType": gpu_type,
|
||||
"taskType": TASK_TYPE
|
||||
}]
|
||||
}
|
||||
print(f"📤 提交测试任务 (bi150): {model_id}", flush=True)
|
||||
try:
|
||||
resp = requests.post(BASE_URL + SUBMIT_TEST_TASK_ENDPOINT, json=task_data, headers=auth_headers, timeout=15)
|
||||
if resp.status_code == 200:
|
||||
result = resp.json()
|
||||
if result.get("code") == 0:
|
||||
task_id = result.get("data", {}).get("taskId")
|
||||
print(f"✅ 测试任务提交成功! Task ID: {task_id}", flush=True)
|
||||
return True
|
||||
else:
|
||||
print(f"❌ 测试任务业务错误: {result.get('message')}", flush=True)
|
||||
return False
|
||||
else:
|
||||
print(f"❌ 测试任务 HTTP 错误: {resp.status_code} - {resp.text}", flush=True)
|
||||
return False
|
||||
except Exception as e:
|
||||
print(f"💥 提交测试任务异常 ({model_id}): {e}", flush=True)
|
||||
return False
|
||||
|
||||
# ══════════════════════════════════════════════════════════
|
||||
# 业务逻辑：动态流水线（下载 → 提交验证任务）
|
||||
# ══════════════════════════════════════════════════════════
|
||||
def _run_worker():
|
||||
_state["started_at"] = datetime.utcnow().isoformat()
|
||||
_state["phase"] = "submitting"
|
||||
_state["phase"] = "running"
|
||||
|
||||
successful: List[Tuple[str, str]] = []
|
||||
token = AUTH_TOKEN
|
||||
print("[worker] 使用预设 Token，跳过登录", flush=True)
|
||||
|
||||
for model_id in ALL_MODEL_IDS:
|
||||
if _shutdown.is_set():
|
||||
break
|
||||
ok, task_id = _submit_task(token, model_id)
|
||||
if ok:
|
||||
_state["submitted"] += 1
|
||||
successful.append((task_id, model_id))
|
||||
else:
|
||||
_state["failed"] += 1
|
||||
|
||||
# 写入结果文件
|
||||
try:
|
||||
with open("submitted_validation_tasks.txt", "w", encoding="utf-8") as f:
|
||||
for tid, mid in successful:
|
||||
f.write(f"{tid}\t{mid}\n")
|
||||
token = login()
|
||||
except Exception as e:
|
||||
print(f"[worker] 登录失败: {e}", flush=True)
|
||||
_state["phase"] = "error"
|
||||
return
|
||||
|
||||
pending_models = list(ALL_MODEL_IDS) # 尚未开始下载的模型
|
||||
active_models: Set[str] = set() # 当前正在下载的模型
|
||||
completed_results = {} # model_id -> status
|
||||
|
||||
print(f"🚀 总共 {len(pending_models)} 个模型待下载。最大并发数: {MAX_CONCURRENT_DOWNLOADS}\n", flush=True)
|
||||
|
||||
while (pending_models or active_models) and not _shutdown.is_set():
|
||||
# 1. 检查活跃任务状态
|
||||
for model_id in list(active_models):
|
||||
if _shutdown.is_set():
|
||||
break
|
||||
status = check_model_status(token, model_id)
|
||||
if status in ("SUCCESS", "FAILED"):
|
||||
completed_results[model_id] = status
|
||||
active_models.remove(model_id)
|
||||
print(f"⏹️ {model_id} 完成，状态: {status}", flush=True)
|
||||
if status == "SUCCESS":
|
||||
_state["download_success"] += 1
|
||||
# 下载成功后立即提交该模型的验证任务
|
||||
if submit_test_task(token, model_id):
|
||||
_state["submitted"] += 1
|
||||
print(f"🧪 已为 {model_id} 提交 hygon 验证任务", flush=True)
|
||||
else:
|
||||
_state["submit_failed"] += 1
|
||||
print(f"⚠️ {model_id} hygon 验证任务提交失败", flush=True)
|
||||
|
||||
if submit_test_task_bi150(token, model_id):
|
||||
_state["submitted"] += 1
|
||||
print(f"🧪 已为 {model_id} 提交 bi150 验证任务", flush=True)
|
||||
else:
|
||||
_state["submit_failed"] += 1
|
||||
print(f"⚠️ {model_id} bi150 验证任务提交失败", flush=True)
|
||||
else:
|
||||
_state["download_failed"] += 1
|
||||
|
||||
# 2. 补充新任务（最多补到 MAX_CONCURRENT_DOWNLOADS 个）
|
||||
while len(active_models) < MAX_CONCURRENT_DOWNLOADS and pending_models and not _shutdown.is_set():
|
||||
next_model = pending_models.pop(0)
|
||||
if create_download_task(token, next_model):
|
||||
active_models.add(next_model)
|
||||
print(f"▶️ 启动下载: {next_model} (当前活跃: {len(active_models)})", flush=True)
|
||||
else:
|
||||
# 创建失败也视为完成（避免卡住）
|
||||
completed_results[next_model] = "CREATE_FAILED"
|
||||
_state["download_failed"] += 1
|
||||
print(f"❌ 创建失败: {next_model}", flush=True)
|
||||
|
||||
_state["downloading"] = sorted(active_models)
|
||||
|
||||
# 3. 稍作等待，避免频繁查询（可被 shutdown 信号打断）
|
||||
if active_models or pending_models:
|
||||
_shutdown.wait(CHECK_INTERVAL_SECONDS)
|
||||
|
||||
print("\n✅ 所有模型处理完毕！\n", flush=True)
|
||||
|
||||
# 4. 收集所有成功下载的模型ID并写入结果文件
|
||||
success_models = [
|
||||
mid for mid, status in completed_results.items()
|
||||
if status == "SUCCESS"
|
||||
]
|
||||
try:
|
||||
with open("downloaded_success_models.txt", "w", encoding="utf-8") as f:
|
||||
for mid in success_models:
|
||||
f.write(f"{mid}\n")
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
print("🎉 下载成功的模型ID列表：", flush=True)
|
||||
print("[", flush=True)
|
||||
for mid in success_models:
|
||||
print(f' "{mid}",', flush=True)
|
||||
print("]", flush=True)
|
||||
|
||||
_state["downloading"] = []
|
||||
_state["finished_at"] = datetime.utcnow().isoformat()
|
||||
_state["phase"] = "done"
|
||||
print(
|
||||
f"[worker] 完成 submitted={_state['submitted']} failed={_state['failed']}",
|
||||
f"[worker] 完成 download_success={_state['download_success']} "
|
||||
f"download_failed={_state['download_failed']} "
|
||||
f"submitted={_state['submitted']} submit_failed={_state['submit_failed']}",
|
||||
flush=True,
|
||||
)
|
||||
# 提交完成后继续保持进程存活，等待平台停止
|
||||
# 流水线完成后继续保持进程存活，等待平台停止
|
||||
|
||||
# ══════════════════════════════════════════════════════════
|
||||
# 入口
|
||||
@@ -545,7 +504,7 @@ def main():
|
||||
http_thread = threading.Thread(target=_run_http, daemon=False)
|
||||
http_thread.start()
|
||||
|
||||
# 提交任务线程
|
||||
# 流水线线程
|
||||
worker_thread = threading.Thread(target=_run_worker, daemon=True)
|
||||
worker_thread.start()
|
||||
|
||||
@@ -557,4 +516,4 @@ def main():
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
main()
|
||||
main()
|
||||
|
||||
Reference in New Issue
Block a user