Compare commits
6 Commits
| Author | SHA1 | Date | |
|---|---|---|---|
| 9aba1595db | |||
| 85f41bba58 | |||
| d6b0e416db | |||
| 4dcfed6b6d | |||
| 1e8cfacd8e | |||
| d6cca90496 |
2
.gitignore
vendored
Normal file
2
.gitignore
vendored
Normal file
@@ -0,0 +1,2 @@
|
||||
.DS_Store
|
||||
__pycache__/
|
||||
@@ -1,6 +1,7 @@
|
||||
FROM modelhubxc-4pd.tencentcloudcr.com/xc_agent_platform/python:3.11-slim
|
||||
|
||||
ENV PYTHONUNBUFFERED=1
|
||||
ENV PYTHONUNBUFFERED=1
|
||||
|
||||
|
||||
WORKDIR /app
|
||||
|
||||
|
||||
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
|
||||
|
||||
804
main.py
804
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.eyJ1c2VyQWNjb3VudCI6Inpob3VzaGFzaGEiLCJpZCI6MTQsInVzZXJSb2xlIjoibGVhZGVyYm9hcmQiLCJleHAiOjE3ODE4NTE0NzcsImlhdCI6MTc4MTI0NjY3N30.p3uvCpG50aLNifNVVXxvzmWJahbLM5K1671FVCtj8E8"
|
||||
CONTEST_API_TOKEN = "ef1ef82f3c9efee413d602345fbe224d"
|
||||
HF_TOKEN = "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"
|
||||
CONTRIBUTORS = "zhoushasha"
|
||||
GPU_TYPE = "Cambricon_mlu-370-x8"
|
||||
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
|
||||
|
||||
@@ -36,350 +50,76 @@ HTTP_PORT = 8080
|
||||
# 模型列表
|
||||
# ══════════════════════════════════════════════════════════
|
||||
ALL_MODEL_IDS = [
|
||||
"UCLA-AGI/Gemma-2-9B-It-SPPO-Iter3",
|
||||
"migtissera/SynthIA-7B-v1.3",
|
||||
"TinyLlama/TinyLlama-1.1B-intermediate-step-955k-token-2T",
|
||||
"bigscience/bloomz-1b1",
|
||||
"EleutherAI/pythia-6.9b-deduped",
|
||||
"AvitoTech/avibe",
|
||||
"Enoch/llama-7b-hf",
|
||||
"asingh15/qwen-abs-verl-sft-rephrased-lr5e6-ep1-0109",
|
||||
"PrimeIntellect/INTELLECT-1",
|
||||
"neuralmagic/starcoder2-3b-quantized.w8a8",
|
||||
"Saxo/Linkbricks-Horizon-AI-Korean-Gemma-2-sft-dpo-27B",
|
||||
"HuggingFaceH4/zephyr-7b-gemma-v0.1",
|
||||
"neuralmagic/Llama-2-7b-chat-quantized.w4a16",
|
||||
"neuralmagic/starcoder2-15b-quantized.w8a8",
|
||||
"DAMO-NLP-SG/Qwen2.5-7B-LongPO-128K",
|
||||
"guardrail/llama-2-7b-guanaco-instruct-sharded",
|
||||
"shenzhi-wang/Gemma-2-27B-Chinese-Chat",
|
||||
"pavankumarbalijepalli/phi2-sqlcoder",
|
||||
"neph1/bellman-7b-mistral-instruct",
|
||||
"neuralmagic/Meta-Llama-3-8B-Instruct-quantized.w8a16",
|
||||
"neuralmagic/Qwen2-7B-Instruct-quantized.w8a8",
|
||||
"lamm-mit/BioinspiredLLM",
|
||||
"neuralmagic/Qwen2-7B-Instruct-quantized.w8a16",
|
||||
"dataopsnick/Qwen3-4B-Instruct-2507-zip-rc",
|
||||
"huihui-ai/MicroThinker-3B-Preview",
|
||||
"OrionStarAI/Orion-14B-Base",
|
||||
"georgesung/llama3_8b_chat_uncensored",
|
||||
"FreedomIntelligence/RAG-Instruct-Llama3-3B",
|
||||
"Aryanne/WestSenzu-Swap-7B",
|
||||
"Josephgflowers/Cinder-Phi-2-Test-1",
|
||||
"FreedomIntelligence/Apollo-6B",
|
||||
"Josephgflowers/Tinyllama-1.3B-Cinder-Reason-Test-2",
|
||||
"Josephgflowers/Tinyllama-1.3B-Cinder-Reason-Test",
|
||||
"247labs/Llama-2-7b-Verse-Bot",
|
||||
"praneethposina/customer_support_bot",
|
||||
"KBlueLeaf/TIPO-200M",
|
||||
"norallm/normistral-11b-warm",
|
||||
"theprint/Boptruth-Agatha-7B",
|
||||
"ericflo/Llama-3.1-8B-ContinuedTraining2-FFT",
|
||||
"okwinds/OpenR1-Qwen-7B",
|
||||
"ruohuaw/deepquery-3b-sft",
|
||||
"theprint/Boptruth-NeuralMonarch-7B",
|
||||
"MaziyarPanahi/calme-3.1-qwenloi-3b",
|
||||
"alperiox/trendyol-7b-base-v1-mtLoRA_entr",
|
||||
"theprint/phi-3-mini-4k-python",
|
||||
"uukuguy/speechless-nl2sql-ds-6.7b",
|
||||
"uukuguy/speechless-coder-ds-6.7b",
|
||||
"tybrs/llama-guard-quant",
|
||||
"Josephgflowers/TinyLlama-3T-Cinder-v1.3",
|
||||
"mlabonne/Darewin-7B-v2",
|
||||
"TeichAI/Qwen3-1.7B-Gemini-2.5-Flash-Lite-Preview-Distill",
|
||||
"TeichAI/Nemotron-Orchestrator-8B-DeepSeek-v3.2-Speciale-Distill",
|
||||
"shadowml/BeagSake-7B",
|
||||
"lex-hue/Delexa-7b",
|
||||
"h2oai/h2o-danube3-500m-chat",
|
||||
"bigcode/gpt_bigcode-santacoder",
|
||||
"openlm-research/open_llama_7b",
|
||||
"upstage/SOLAR-10.7B-v1.0",
|
||||
"prithivMLmods/Phi-3.5-Mini-Xalate",
|
||||
"prithivMLmods/Qwen3-Bifrost-SOL-4B-GUFF",
|
||||
"prithivMLmods/Volans-Opus-14B-Exp",
|
||||
"prithivMLmods/Viper-OneCoder-UIGEN",
|
||||
"prithivMLmods/Tucana-Opus-14B-r999",
|
||||
"prithivMLmods/Sombrero-Opus-14B-Sm5",
|
||||
"prithivMLmods/Sombrero-Opus-14B-Sm4",
|
||||
"prithivMLmods/Reasoning-SmolLM2-135M",
|
||||
"prithivMLmods/Sombrero-Opus-14B-Sm1",
|
||||
"prithivMLmods/LwQ-10B-Instruct",
|
||||
"prithivMLmods/Sombrero-Opus-14B-Elite5",
|
||||
"prithivMLmods/Eridanus-Opus-14B-r999",
|
||||
"prithivMLmods/Equuleus-Opus-14B-Exp",
|
||||
"prithivMLmods/Epimetheus-14B-Axo",
|
||||
"prithivMLmods/Phi-4-Math-IO",
|
||||
"prithivMLmods/Omni-Reasoner4-Merged",
|
||||
"prithivMLmods/Pegasus-Opus-14B-Exp",
|
||||
"prithivMLmods/Elita-1",
|
||||
"prithivMLmods/Delta-Pavonis-Qwen-14B",
|
||||
"prithivMLmods/Nu2-Lupi-Qwen-14B",
|
||||
"prithivMLmods/Coma-II-14B",
|
||||
"MaziyarPanahi/calme-2.7-qwen2-7b",
|
||||
"prithivMLmods/Monocerotis-V838-14B",
|
||||
"prithivMLmods/Calcium-Opus-14B-Merge",
|
||||
"prithivMLmods/Calcium-Opus-14B-Elite3",
|
||||
"prithivMLmods/Calcium-Opus-14B-Elite2-R1",
|
||||
"prithivMLmods/Calcium-Opus-14B-Elite2",
|
||||
"prithivMLmods/Calcium-Opus-14B-Elite-Stock",
|
||||
"prithivMLmods/Megatron-Opus-14B-2.1",
|
||||
"prithivMLmods/Blaze.1-27B-Reflection",
|
||||
"prithivMLmods/Megatron-Corpus-14B-Exp.v2",
|
||||
"prithivMLmods/Megatron-Corpus-14B-Exp",
|
||||
"GAIR/autoj-bilingual-6b",
|
||||
"TheBloke/airoboros-7b-gpt4-fp16",
|
||||
"Undi95/Mistral-11B-OmniMix9",
|
||||
"PKU-Alignment/ProgressGym-HistLlama3-8B-C016-pretrain-v0.2",
|
||||
"PKU-Alignment/ProgressGym-HistLlama3-8B-C017-instruct-v0.2",
|
||||
"mlabonne/NeuralDarewin-7B",
|
||||
"0xgr3y/Qwen2.5-Coder-0.5B-Instruct-Gensyn-Swarm-tall_tame_panther",
|
||||
"openlm-research/open_llama_3b_v2",
|
||||
"Nobitaxi/InternLM2-chat-7B-SQL",
|
||||
"testUser/Qwen3-1.7b-Medical-R1-sft",
|
||||
"mlabonne/Zebrafish-7B",
|
||||
"mlabonne/NeuralPipe-7B-slerp",
|
||||
"laion/openthoughts-4-code-qwen3-32b-annotated-7k_qwen3-1.7B_10k",
|
||||
"Fengshenbang/Ziya-LLaMA-13B-v1.1",
|
||||
"arcee-ai/Saul-Instruct-Mistral-7B-Instruct-v0.2-Slerp",
|
||||
"arcee-ai/Saul-Instruct-Clown-7b",
|
||||
"prithivMLmods/Megatron-Opus-7B-Exp",
|
||||
"Vikhrmodels/QVikhr-3-8B-Instruction",
|
||||
"TheBloke/Nous-Hermes-13B-SuperHOT-8K-fp16",
|
||||
"TheBloke/UltraLM-13B-fp16",
|
||||
"PocketDoc/Dans-TotSirocco-7b",
|
||||
"LLM-Research/Meta-Llama-3.1-8B",
|
||||
"Qwen/Qwen2.5-Coder-32B",
|
||||
"Qwen/Qwen2.5-7B-Instruct-1M",
|
||||
"Qwen/Qwen2-57B-A14B-Instruct",
|
||||
"Qwen/Qwen2.5-14B-Instruct-1M",
|
||||
"Qwen/Qwen-1_8B-Chat",
|
||||
"Qwen/Qwen1.5-MoE-A2.7B-Chat",
|
||||
"Qwen/Qwen1.5-MoE-A2.7B",
|
||||
"Qwen/Qwen1.5-14B-Chat",
|
||||
"Qwen/Qwen1.5-14B",
|
||||
"Qwen/Qwen-14B",
|
||||
"deepseek-ai/DeepSeek-Coder-V2-Lite-Base",
|
||||
"TheBloke/tulu-13B-fp16",
|
||||
"TheBloke/Kimiko-Mistral-7B-fp16",
|
||||
"TheBloke/Llama-2-13B-fp16",
|
||||
"mlabonne/Monarch-7B",
|
||||
"TheBloke/tulu-7B-fp16",
|
||||
"01ai/Yi-9B",
|
||||
"TheBloke/koala-7B-HF",
|
||||
"AI-ModelScope/txgemma-2b-predict",
|
||||
"LLM-Research/OLMo-7B-0724-SFT-hf",
|
||||
"JsonZhang02/Llama3.2-1B-PCL",
|
||||
"PKU-Alignment/ProgressGym-HistLlama3-8B-C019-instruct-v0.2",
|
||||
"PKU-Alignment/ProgressGym-HistLlama3-8B-C016-instruct-v0.2",
|
||||
"FreedomIntelligence/AceGPT-v1.5-13B-Chat",
|
||||
"MediaTek-Research/Breeze-7B-Base-v0_1",
|
||||
"OpenBuddy/openbuddy-llama3-8b-v21.1-8k",
|
||||
"HIT-TMG/Mixtral_13B_Chat_RAG-Reader",
|
||||
"PKU-Alignment/ProgressGym-HistLlama3-8B-C014-pretrain-v0.2",
|
||||
"arcee-ai/arcee-lite",
|
||||
"X-D-Lab/MindChat-Qwen2-4B",
|
||||
"mlabonne/NeuralMonarch-7B",
|
||||
"ibm-granite/granite-3b-code-instruct-2k",
|
||||
"LLM-Research/OLMo-7B-Twin-2T-hf",
|
||||
"PocketDoc/Dans-AdventurousWinds-Mk2-7b",
|
||||
"LLM-Research/Qwen2-Math-7B",
|
||||
"MediaTek-Research/Breeze-7B-Base-v1_0",
|
||||
"LLM-Research/layerskip-llama2-13B",
|
||||
"prithivMLmods/TESS-QwenRe-1.5B",
|
||||
"prithivMLmods/Octantis-QwenR1-1.5B",
|
||||
"prithivMLmods/Qwen3-1.7B-ft-bf16",
|
||||
"prithivMLmods/Theta-Crucis-0.6B-Turbo1",
|
||||
"prithivMLmods/Omega-Qwen3-Atom-8B",
|
||||
"prithivMLmods/Mintaka-Qwen3-1.6B-V3.1",
|
||||
"NousResearch/Yarn-Llama-2-7b-64k",
|
||||
"prithivMLmods/Panacea-MegaScience-Qwen3-1.7B",
|
||||
"prithivMLmods/TOI-157-Phi-4-Reasoning-Mini",
|
||||
"prithivMLmods/Vulpecula-4B",
|
||||
"LLM-Research/OLMo-7B-0424-hf",
|
||||
"LLM-Research/OLMo-7B-hf",
|
||||
"LLM-Research/OLMo-7B-SFT-hf",
|
||||
"AI-ModelScope/starcoder2-7b",
|
||||
"LLM-Research/OLMo-7B-0724-hf",
|
||||
"OpenBMB/BitCPM4-1B",
|
||||
"LLM-Research/truthfulqa-truth-judge-llama2-7B",
|
||||
"LLM-Research/OLMo-1B-0724-hf",
|
||||
"HIT-TMG/Qwen1.5-14B-Chat_RAG-Reader",
|
||||
"OpenBMB/MiniCPM4-MCP",
|
||||
"AI-ModelScope/sqlcoder-7b-2",
|
||||
"FuseAI/OpenChat-3.5-7B-SOLAR-v2.0",
|
||||
"JsonZhang02/Llama3.2-1B-SFT",
|
||||
"MaziyarPanahi/neural-chat-7b-v3-2-Mistral-7B-Instruct-v0.1",
|
||||
"MaziyarPanahi/SauerkrautLM-7b-HerO-Mistral-7B-Instruct-v0.1",
|
||||
"prithivMLmods/Segue-Qwen3_DeepScaleR-Preview",
|
||||
"NovaSky-AI/Sky-T1-7B-Zero",
|
||||
"PKU-Alignment/ProgressGym-HistLlama3-8B-C020-pretrain-v0.2",
|
||||
"NovaSky-AI/Sky-T1-7B-step2",
|
||||
"NousResearch/CodeLlama-7b-hf-flash",
|
||||
"LLM-Research/layerskip-llama3-8B",
|
||||
"LLM-Research/OLMo-1B-hf",
|
||||
"PKU-Alignment/ProgressGym-HistLlama3-8B-C018-pretrain-v0.2",
|
||||
"NousResearch/CodeLlama-7b-Instruct-hf-flash",
|
||||
"Nexusflow/NexusRaven-V2-13B",
|
||||
"AI-ModelScope/NuExtract-v1.5",
|
||||
"NousResearch/Nous-Capybara-3B-V1.9",
|
||||
"NousResearch/Nous-Capybara-7B-V1",
|
||||
"NousResearch/Yarn-Solar-10b-32k",
|
||||
"LLM-Research/Llama-Guard-4-12B",
|
||||
"OpenPipe/gemma-3-4b-it-text-only-2",
|
||||
"OpenPipe/Deductive-Reasoning-Qwen-14B",
|
||||
"OpenPipe/gemma-3-12b-it-text-only",
|
||||
"AI-MO/NuminaMath-7B-CoT",
|
||||
"GAIR/Abel-7B-001",
|
||||
"prithivMLmods/Novaeus-Promptist-7B-Instruct",
|
||||
"SakanaAI/EvoLLM-JP-v1-7B",
|
||||
"FreedomIntelligence/Apollo-1.8B",
|
||||
"PKU-Alignment/ProgressGym-HistLlama3-8B-C013-instruct-v0.2",
|
||||
"PKU-Alignment/ProgressGym-HistLlama3-8B-C013-pretrain-v0.2",
|
||||
"PKU-Alignment/ProgressGym-HistLlama3-8B-C015-pretrain-v0.2",
|
||||
"PKU-Alignment/ProgressGym-HistLlama3-8B-C014-instruct-v0.2",
|
||||
"PKU-Alignment/ProgressGym-HistLlama3-8B-C021-instruct-v0.2",
|
||||
"NousResearch/Meta-Llama-3.1-8B",
|
||||
"OpenPipe/Qwen3-14B-Instruct",
|
||||
"unsloth/OpenHermes-2.5-Mistral-7B",
|
||||
"OpenBuddy/openbuddy-mistral-22b-v21.1-32k",
|
||||
"FlyDutch/telechat2-7b-Cot",
|
||||
"HuggingFaceH4/mistral-7b-sft-alpha",
|
||||
"PAI/pai-qwen1_5-7b-doc2qa",
|
||||
"PKU-Alignment/ProgressGym-HistLlama3-8B-C018-instruct-v0.2",
|
||||
"Magpie-Align/Llama-3-8B-Tulu-330K",
|
||||
"prithivMLmods/Blaze.1-27B-Preview",
|
||||
"allenai/OLMo-7B-0424-SFT-hf",
|
||||
"mlabonne/Meta-Llama-3-8B",
|
||||
"LLM-Research/layerskip-codellama-7B",
|
||||
"prithivMLmods/Sculptor-Qwen3_Med-Reasoning",
|
||||
"prithivMLmods/SmolLM2-360M-Grpo-r999",
|
||||
"prithivMLmods/SmolLM2-1.7B-Open-Thought",
|
||||
"LLM-Research/open-instruct-llama2-sharegpt-7b",
|
||||
"prithivMLmods/SmolLM2_135M_Grpo_Checkpoint",
|
||||
"OpenBuddy/openbuddy-qwen2.5llamaify-14b-v23.1-200k",
|
||||
"OpenBuddy/openbuddy-zero-3b-v21.2-32k",
|
||||
"YeungNLP/firefly-llama2-7b-chat",
|
||||
"OpenBuddy/openbuddy-zero-14b-v22.3-32k",
|
||||
"OpenBuddy/openbuddy-yi1.5-9b-v21.1-32k",
|
||||
"FuseAI/OpenChat-3.5-7B-Starling-v2.0",
|
||||
"prithivMLmods/Qwen-7B-Distill-Reasoner",
|
||||
"FuseAI/OpenChat-3.5-7B-InternLM-v2.0",
|
||||
"prithivMLmods/Galactic-Qwen-14B-Exp1",
|
||||
"prithivMLmods/Sombrero-R1-14B-Elite13",
|
||||
"prithivMLmods/Sombrero-Opus-14B-Elite13",
|
||||
"TheBloke/Planner-7B-fp16",
|
||||
"AI-ModelScope/speed-synthesis-8b-senior",
|
||||
"PocketDoc/Dans-AdventurousWinds-7b",
|
||||
"MaziyarPanahi/calme-3.2-baguette-3b",
|
||||
"MaziyarPanahi/calme-3.2-instruct-3b",
|
||||
"IntervitensInc/intv_ai_mk11",
|
||||
"prithivMLmods/Muscae-Qwen3-UI-Code-4B",
|
||||
"NousResearch/Llama-2-7b-hf",
|
||||
"prithivMLmods/Pocket-Llama-3.2-3B-Instruct",
|
||||
"OpenBuddy/openbuddy-openllama-13b-v7-fp16",
|
||||
"LLM-Research/WildLlama-7b-assistant-only",
|
||||
"prithivMLmods/Raptor-X2",
|
||||
"OpenBuddy/openbuddy-qwen1.5-14b-v20.1-32k",
|
||||
"NaniDAO/Meta-Llama-3.1-8B-Instruct-ablated-v1",
|
||||
"LLM-Research/OLMo-7B-Instruct-hf",
|
||||
"OpenBuddy/openbuddy-zen-3b-v21.2-32k",
|
||||
"OpenBuddy/openbuddy-qwen1.5-14b-v21.1-32k",
|
||||
"LLM-Research/llama2-7b-WildJailbreak",
|
||||
"JunHowie/MiniCPM4-8B",
|
||||
"OpenBuddy/openbuddy-coder-15b-v10-bf16",
|
||||
"JunHowie/MiniCPM4-0.5B",
|
||||
"OpenDevin/CodeQwen1.5-7B-OpenDevin",
|
||||
"OpenBuddy/openbuddy-mistral-10b-v17.1-32k",
|
||||
"PAI/DistilQwen2.5-DS3-0324-7B",
|
||||
"OpenBuddy/openbuddy-llama2-13b64k-v15",
|
||||
"OpenBuddy/openbuddy-falcon-7b-v5-fp16",
|
||||
"NousResearch/Hermes-2-Theta-Llama-3-8B",
|
||||
"NousResearch/Hermes-2-Pro-Mistral-7B",
|
||||
"OpenBuddy/openbuddy-openllama-7b-v5-fp16",
|
||||
"PAI/DistillQwen-ThoughtY-8B",
|
||||
"BSC-LT/salamandra-2b",
|
||||
"pfnet/nekomata-7b-pfn-qfin-inst-merge",
|
||||
"BSC-LT/experimental7b-rag-instruct",
|
||||
"PKU-Alignment/ProgressGym-HistLlama3-8B-C019-pretrain-v0.2",
|
||||
"OpenBuddy/OpenBuddy-R10528DistillQwen-14B-v27.4-200K",
|
||||
"OpenBuddy/OpenBuddy-R10528DistillQwen-14B-v27.1",
|
||||
"OpenBuddy/SimpleChat-4B-V1",
|
||||
"AI-ModelScope/granite-8b-code-base-4k",
|
||||
"mlabonne/NeuralHermes-2.5-Mistral-7B",
|
||||
"BSC-LT/experimental7b-rag",
|
||||
"prithivMLmods/SmolLM2_135M_Grpo_Gsm8k",
|
||||
"OpenBuddy/openbuddy-zen-3b-v21.1-32k",
|
||||
"PKU-Alignment/ProgressGym-HistLlama3-8B-C015-instruct-v0.2",
|
||||
"prithivMLmods/QwQ-LCoT1-Merged",
|
||||
"mlabonne/NeuralBeagle14-7B",
|
||||
"PKU-Alignment/ProgressGym-HistLlama3-8B-C020-instruct-v0.2",
|
||||
"mlabonne/NeuralMarcoro14-7B",
|
||||
"PKU-Alignment/ProgressGym-HistLlama3-8B-C017-pretrain-v0.2",
|
||||
"FuseAI/OpenChat-3.5-7B-Mixtral-v2.0",
|
||||
"mlabonne/FrankenMonarch-7B",
|
||||
"stabilityai/stablelm-tuned-alpha-3b",
|
||||
"prithivMLmods/Viper-Coder-v1.5-r999",
|
||||
"prithivMLmods/Galactic-Qwen-14B-Exp2",
|
||||
"HuggingFaceTB/cosmo-1b",
|
||||
"LLM-Research/WildLlama-7b-user-assistant",
|
||||
"OpenBuddy/openbuddy-llama2-13b-v8.1-fp16",
|
||||
"prithivMLmods/Regulus-Qwen3-R1-Llama-Distill-1.7B",
|
||||
"LLM-Research/OLMo-7B-0424-SFT-hf",
|
||||
"huihui-ai/MicroThinker-1B-Preview",
|
||||
"OpenBuddy/openbuddy-openllama-3b-v10-bf16",
|
||||
"LLM-Research/digital-socrates-13b",
|
||||
"prithivMLmods/Viper-Coder-v1.6-r999",
|
||||
"prithivMLmods/Magpie-Qwen-DiMind-1.7B",
|
||||
"BAAI/CareBot_Medical_multi-llama3-8b-base",
|
||||
"NousResearch/Meta-Llama-3-8B",
|
||||
"OpenBuddy/openbuddy-llama2-13b-v15p1-64k",
|
||||
"NousResearch/Yarn-Mistral-7b-64k",
|
||||
"PrimeIntellect/DeepSeek-R1-Distill-Qwen-1.5B",
|
||||
"Undi95/Meta-Llama-3-8B-Instruct-hf",
|
||||
"FuseAI/OpenChat-3.5-7B-Mixtral",
|
||||
"prithivMLmods/Viper-Coder-Hybrid-v1.3",
|
||||
"OpenBuddy/openbuddy-qwen2.5llamaify-7b-v23.1-200k",
|
||||
"LLM-Research/mistral-7b",
|
||||
"OpenBuddy/openbuddy-qwen2.5llamaify-14b-v23.3-200k",
|
||||
"prithivMLmods/Viper-Coder-HybridMini-v1.3",
|
||||
"OpenBuddy/openbuddy-atom-13b-v9-bf16",
|
||||
"OpenBuddy/openbuddy-llama3.2-3b-v23.2-131k",
|
||||
"ibm-granite/granite-3b-code-instruct-128k",
|
||||
"PierreZCW/Breeze-7B-Instruct-v1_0",
|
||||
"mlabonne/Marcoro14-7B-slerp",
|
||||
"AI-ModelScope/openbuddy-falcon-7b-v15-fp16",
|
||||
"AI-ModelScope/falcon-7b",
|
||||
"BAAI/AquilaChat2-7B",
|
||||
"PrimeIntellect/Qwen3-0.6B",
|
||||
"OuteAI/Lite-Oute-1-65M-Instruct",
|
||||
"AI-ModelScope/granite-3b-code-instruct-128k",
|
||||
"PrimeIntellect/Qwen3-8B",
|
||||
"OpenBuddy/openbuddy-falcon-7b-v6-bf16",
|
||||
"MaziyarPanahi/calme-3.1-instruct-3b",
|
||||
"LLM-Research/open-instruct-llama2-sharegpt-dpo-7b",
|
||||
"PocketDoc/Dans-PersonalityEngine-v1.0.0-8b",
|
||||
"FuseAI/FuseChat-Llama-3.1-8B-Instruct",
|
||||
"OpenBuddy/openbuddy-mixtral-7bx8-v18.1-32k",
|
||||
"OpenBuddy/openbuddy-deepseekcoder-6b-v16.1-32k",
|
||||
"HuggingFaceTB/SmolLM-1.7B",
|
||||
"LLM-Research/Llama-4-Scout-17B-16E-Instruct",
|
||||
"argilla/distilabeled-Marcoro14-7B-slerp-full",
|
||||
"HuggingFaceTB/SmolLM2-1.7B",
|
||||
"argilla/distilabeled-Marcoro14-7B-slerp",
|
||||
"l3utterfly/open-llama-3b-v2-layla",
|
||||
|
||||
|
||||
|
||||
"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()
|
||||
|
||||
@@ -417,104 +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"""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
|
||||
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: MAX_MODEL_LEN
|
||||
value: 8192
|
||||
command: ["vllm", "serve", "/model", "--port", "8000", "--served-model-name", "llm", "--max-model-len", "8192", "--trust-remote-code", "--dtype", "float16"]
|
||||
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:
|
||||
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"]
|
||||
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
|
||||
@@ -532,7 +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()
|
||||
|
||||
@@ -544,4 +516,4 @@ def main():
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
main()
|
||||
main()
|
||||
|
||||
Reference in New Issue
Block a user