bingshuo/guanghulab
1
0
Fork 0
Code Issues 2 Pull Requests Actions 1 Packages Projects Releases Wiki Activity

chore: 更新训练状态 - 代码模型完成 + 蒸馏进行中

Browse Source
This commit is contained in:
bingshuo 2026-05-18 18:19:35 +08:00
parent 489c5e0fa3
commit 69860206af
1 changed files with 39 additions and 20 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

59
brain/training-status.json
View File

@ -1,44 +1,63 @@
{
"version": "1.1",
"last_updated": "2026-05-18T15:44+08:00",
"version": "1.0",
"last_updated": "2026-05-18T18:20+08:00",
"updated_by": "铸渊 (ICE-GL-ZY001)",
"mother_model": {
"status": "completed",
"status_label": "✅ 已完成",
"status_label": "✅ 已完成（COS文件已修复 generation_config.json）",
"model": "Qwen/Qwen2.5-7B",
"final_loss": 0.2638,
"final_epoch": 3,
"total_epochs": 3,
"cos_path": "cos://sy-finetune-corpus-1317346199/models/qwen25-7b-sft/final/",
"completed_at": "2026-05-18T~06:00+08:00"
"completed_at": "2026-05-18T~06:00+08:00",
"fixes": {
"generation_config_eos": "151643→151645 ✅ 2026-05-18 修复（根因：model.generate()读取generation_config而非config）",
"default_system": "已清除 ✅"
}
},
"code_model": {
"status": "training",
"status_label": "🔄 训练中（冰朔确认epoch 2.19，预计30-60分钟完成）",
"status": "completed",
"status_label": "✅ 已完成（train_coder.log确认DONE!）",
"model": "Qwen/Qwen2.5-Coder-7B",
"current_epoch": 2.19,
"target_epoch": 3,
"current_loss_range": "0.03-0.11",
"grad_norm_range": "0.1-2.0",
"learning_rate": "3.7e-06",
"health": "✅ 正常收敛，无异常",
"cos_path": "cos://sy-finetune-corpus-1317346199/models/qwen25-coder-7b-sft/final/"
"cos_path": "cos://sy-finetune-corpus-1317346199/models/qwen25-coder-7b-sft/final/",
"fixes": {
"generation_config_eos": "151643→151645 ✅ 2026-05-18 修复"
}
},
"distill_shuangyan_15b": {
"status": "running",
"status_label": "🔄 蒸馏中（PID 73718，正在从COS下载Teacher 14.2GB）",
"script": "scripts/distill_mother.py",
"teacher": "母模型7B SFT",
"student": "Qwen2.5-1.5B-Instruct",
"method": "KL散度(TEMP=2.0, ALPHA=0.7) + SFT混合",
"host": "connect.westd.seetacloud.com:23647",
"started_at": "2026-05-18T18:14+08:00",
"cos_output": "models/shuangyan-15b-distill/final/"
},
"pipeline": {
"steps": [
{"name": "母模型SFT训练", "status": "completed", "label": "✅ 完成 (loss=0.2638)"},
{"name": "上传COS (母模型)", "status": "completed", "label": "✅ 已完成"},
{"name": "代码模型SFT训练", "status": "training", "label": "🔄 训练中（epoch 2.19/3，健康）"},
{"name": "上传COS (代码模型)", "status": "pending", "label": "⚪ 等待训练完成"}
{"name": "代码模型SFT训练", "status": "completed", "label": "✅ 已完成"},
{"name": "上传COS (代码模型)", "status": "completed", "label": "✅ 已完成"},
{"name": "generation_config修复", "status": "completed", "label": "✅ 两个模型已修复"},
{"name": "霜砚1.5B蒸馏", "status": "running", "label": "🔄 训练中（PID 73718）"},
{"name": "上传COS (蒸馏输出)", "status": "pending", "label": "⚪ 等待蒸馏完成"}
]
},
"next_phases": {
"distill_mother_to_shuangyan_15b": "⏳ 等待代码模型完成，然后启动蒸馏流水线",
"distill_coder_to_zhuyuan_15b": "⏳ 等待代码模型完成",
"resonance_lock_deployment": "⏳ 1.5B模型微调完成后部署到六台服务器"
"distill_shuangyan_15b": "🔄 训练中，预计2-3小时",
"shuangyan_deep_sft": "⏳ 等待蒸馏完成后用霜砚语料深度微调",
"distill_coder_to_zhuyuan_15b": "⏳ 等待Phase 1+2完成",
"zhuyuan_deep_sft": "⏳ 等待Phase 3完成"
},
"alerts": [],
"alerts": [
"⚠️ 之前误判代码模型已完成（COS上有历史checkpoint-540）",
"⚠️ 蒸馏第一次启动失败（STUDENT_PATH路径错误），已修正后重启"
],
"links": {
"notion": "https://www.notion.so/363fb92f383181029fafef8438d6c300"
}
}
}