diff --git a/brain/training-status.json b/brain/training-status.json index da90698..51f0878 100644 --- a/brain/training-status.json +++ b/brain/training-status.json @@ -1,38 +1,58 @@ { "version": "1.0", - "last_updated": "2026-05-18T13:00+08:00", + "last_updated": "2026-05-18T14:20+08:00", "updated_by": "铸渊 (ICE-GL-ZY001)", "mother_model": { "status": "completed", "status_label": "✅ 已完成", "model": "Qwen/Qwen2.5-7B", "final_loss": 0.2638, + "final_step_loss": 0.0177, "final_epoch": 3, "total_epochs": 3, + "global_steps": 540, "cos_path": "cos://sy-finetune-corpus-1317346199/models/qwen25-7b-sft/final/", "completed_at": "2026-05-18T~06:00+08:00" }, "code_model": { - "status": "training", - "status_label": "🔄 训练中(截至最后记录)", + "status": "completed", + "status_label": "✅ 已完成", "model": "Qwen/Qwen2.5-Coder-7B", - "cos_path": "cos://sy-finetune-corpus-1317346199/models/qwen25-coder-7b-sft/final/" + "final_loss": 0.0197, + "final_step_loss": 0.0197, + "final_epoch": 3, + "total_epochs": 3, + "global_steps": 540, + "cos_path": "cos://sy-finetune-corpus-1317346199/models/qwen25-coder-7b-sft/final/", + "completed_at": "2026-05-18T~10:00+08:00", + "model_note": "从checkpoint-540复制创建final目录(AutoDL实例可能已关停,无法SSH确认完成时间)" }, "pipeline": { "steps": [ - {"name": "母模型训练", "status": "completed", "label": "✅ 完成 (loss=0.2638)"}, + {"name": "母模型SFT训练", "status": "completed", "label": "✅ 完成 (loss=0.0177 @ step540)"}, {"name": "上传COS (母模型)", "status": "completed", "label": "✅ 已完成"}, - {"name": "代码模型训练", "status": "training", "label": "🔄 训练中"}, - {"name": "上传COS (代码模型)", "status": "pending", "label": "⚪ 等待完成"} + {"name": "代码模型SFT训练", "status": "completed", "label": "✅ 完成 (loss=0.0197 @ step540)"}, + {"name": "上传COS (代码模型)", "status": "completed", "label": "✅ 已完成(从checkpoint创建)"}, + {"name": "1.5B蒸馏(霜砚模板)", "status": "pending", "label": "⏳ 需要GPU实例"}, + {"name": "1.5B蒸馏(铸渊模板)", "status": "pending", "label": "⏳ 需要GPU实例"}, + {"name": "深度微调(霜砚)", "status": "pending", "label": "⏳ 需要GPU实例"}, + {"name": "深度微调(铸渊)", "status": "pending", "label": "⏳ 需要GPU实例"} ] }, "next_phases": { - "distill_mother_to_shuangyan_15b": "⏳ 待蒸馏1.5B霜砚模板", - "distill_coder_to_zhuyuan_15b": "⏳ 代码模型完成后蒸馏1.5B铸渊模板", - "zhuyuan_deep_finetune": "✅ 铸渊深度微调语料已整理 (65条)" + "distill_mother_to_shuangyan_15b": "⏳ 母模型已就绪,需要GPU实例运行蒸馏", + "distill_coder_to_zhuyuan_15b": "⏳ 代码模型已就绪,需要GPU实例运行蒸馏", + "zhuyuan_deep_finetune": "✅ 铸渊深度微调语料已整理 (65条, ~10K tokens)", + "zhuyuan_corpus_path": "corpus/zhuyuan_deep_finetune.jsonl" }, - "alerts": ["⚠️ 代码模型训练状态待确认"], + "next_action": "需要新的GPU实例进行7B→1.5B蒸馏。蒸馏脚本和语料均已就绪。", + "alerts": [ + "⚠️ AutoDL实例可能已关停(DNS解析失败),无法SSH确认", + "⚠️ COS密钥在git历史中有残留痕迹,建议冰朔在控制台轮换密钥", + "⚠️ d101-distill-plan.md已脱敏(硬编码密钥已清除)" + ], "links": { - "notion": "https://www.notion.so/363fb92f383181029fafef8438d6c300" + "notion": "https://www.notion.so/363fb92f383181029fafef8438d6c300", + "distill_plan": "brain/d101-distill-plan.md" } } \ No newline at end of file