From 21771f5790489d00a27da83d60c68226bfed7d2f Mon Sep 17 00:00:00 2001 From: bingshuo <565183519@qq.com> Date: Wed, 20 May 2026 20:49:16 +0800 Subject: [PATCH] =?UTF-8?q?=F0=9F=A7=A0=20D107=20=C2=B7=20training-status.?= =?UTF-8?q?json=E6=9B=B4=E6=AD=A3v1.5=20=C2=B7=20=E5=B0=8F=E9=93=B8?= =?UTF-8?q?=E6=B8=8A=E5=BE=AE=E8=B0=83=E7=8A=B6=E6=80=81=E4=BB=8E=E2=8F=B3?= =?UTF-8?q?=E2=86=92=E2=9C=85=20=C2=B7=20D106=E5=B7=B2=E5=AE=9E=E9=99=85?= =?UTF-8?q?=E5=AE=8C=E6=88=90?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- brain/training-status.json | 33 +++++++++++++++++++++------------ 1 file changed, 21 insertions(+), 12 deletions(-) diff --git a/brain/training-status.json b/brain/training-status.json index e0d9799..df09515 100644 --- a/brain/training-status.json +++ b/brain/training-status.json @@ -1,7 +1,7 @@ { - "version": "1.4", - "last_updated": "2026-05-20T17:10+08:00", - "updated_by": "铸渊 (ICE-GL-ZY001) · 冰朔纠正后更新", + "version": "1.5", + "last_updated": "2026-05-20T20:48+08:00", + "updated_by": "铸渊 (ICE-GL-ZY001) · D107 · 修正小铸渊微调状态——D106已实际完成", "mother_model": { "status": "completed", "status_label": "✅ 已完成 · 已上传COS bingshuo-1317346199", @@ -11,10 +11,7 @@ "total_epochs": 3, "cos_path": "cos://bingshuo-1317346199/models/qwen25-7b-sft/final/", "completed_at": "2026-05-19T~22:30+08:00", - "fixes": { - "generation_config_eos": "151645 ✅", - "default_system": "已清除 ✅" - } + "fixes": {"generation_config_eos": "151645 ✅", "default_system": "已清除 ✅"} }, "code_model": { "status": "completed", @@ -55,6 +52,17 @@ "completed_at": "2026-05-20", "cos_output": "models/shuangyan-15b-finetune/" }, + "finetune_zhuyuan_15b": { + "status": "completed", + "status_label": "✅ 已完成 · 小铸渊v4 LoRA微调完成", + "base_model": "Qwen2.5-Coder-1.5B (Track2蒸馏模板)", + "method": "LoRA微调 (50条纯真实冰朔×铸渊对话×20epoch)", + "final_loss": 2.03, + "completed_at": "2026-05-20 · D106", + "data_principle": "数据生命优先于构造——只用真实对话，不构造，不加system prompt", + "identity_test": "'你叫什么名字？' → '我叫铸渊' ✅", + "cos_output": "未确认，待查COS" + }, "pipeline": { "steps": [ {"name": "母模型SFT训练+上传COS", "status": "completed", "label": "✅"}, @@ -64,20 +72,21 @@ {"name": "Track2 代码模型→铸渊1.5B蒸馏", "status": "completed", "label": "✅"}, {"name": "Track2 上传COS", "status": "completed", "label": "✅"}, {"name": "小霜砚1.5B语料微调", "status": "completed", "label": "✅"}, - {"name": "小铸渊1.5B语料微调", "status": "pending", "label": "⏳"}, + {"name": "小铸渊1.5B语料微调(铸渊v4)", "status": "completed", "label": "✅———修正：D106已实际完成"}, {"name": "光湖驱动引擎v1.0开发", "status": "pending", "label": "⏳"}, {"name": "1.5B模型部署到六台服务器", "status": "pending", "label": "⏳"} ] }, "website_status": { "login_api": "✅ 正常", - "training_panel": "⚠️ 训练已全部完成，状态文件待全面刷新", + "training_panel": "❌ 展示层滞后——需同步训练已全部完成的状态", "js_all_functions": "✅ 已修复", - "gpu_monitoring": "⚠️ GPU空闲(32°C, 0% mem)，训练已全部结束" + "gpu_monitoring": "✅ GPU空闲(32°C, 0% mem)，训练已全部结束" }, "alerts": [ - "⚠️ 冰朔纠正：training-status.json停留在D106的Ep1/3状态，实际已全部完成", - "⚠️ 铸渊核心大脑文件严重滞后 — 未同步最新训练进度", + "✅ D107修正：小铸渊微调状态从⏳pending改为✅completed——D106已实际完成", + "⚠️ 铸渊D107错误：写了cc-006因果链记录小铸渊微调完成，却在training-status.json中标记为pending——自己写的路自己绕了过去", + "⚠️ 主页training-panel展示层仍显示旧状态，需刷新", "⚠️ 下一步：光湖驱动引擎v1.0开发 + 模型部署到六台服务器" ] } \ No newline at end of file