wuyang/zk-data-agent
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat: backend gate fallback + metric_diff docs + submit_sft fix

Browse Source 
- Auto-inject human-review gate when agent asks question but forgets gate entry
- Add complex_dev/complex_base field docs to program.md metric_diff table
- Fix submit_sft.sh minor issues
- jupyter_runtime improvements

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
This commit is contained in:
wangsenhao
2026-05-26 15:42:02 +08:00
parent a09d4afacd
commit b48032e15c
5 changed files with 119 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files
skills/model-iteration/references/program.md
+4 -2
View File
@@ -219,8 +219,10 @@ for k in prev:
| 字段 | 含义 |
| --- | --- |
| `origin_predict_base` | 旧模型原始输出。具体格式（是否含 `complex=` 前缀、tag 包装、自定义 class 名等）以当前评测产出为准，**分析前 head 一下实物** |
| `origin_predict_dev` | 新模型原始输出。同上，格式以当前产出为准 |
| `origin_predict_base` | 旧模型（baseline）原始输出。具体格式（是否含 `complex=` 前缀、tag 包装、自定义 class 名等）以当前评测产出为准，**分析前 head 一下实物** |
| `origin_predict_dev` | 新模型（迭代模型）原始输出。同上，格式以当前产出为准 |
| `complex_dev` | 迭代模型的 complex 标签（true/false），从 `origin_predict_dev` 解析出的复杂度判定 |
| `complex_base` | baseline 模型的 complex 标签（true/false），从 `origin_predict_base` 解析出的复杂度判定 |
| `cleaned_predict_*` | 清洗后输出，用于 GSB 对比 |
| `label` / `code_label_base` | ground truth；`label` 为空时回退解析 `code_label_base` |
skills/model-iteration/scripts/submit_sft.sh
+3 -1
View File
@@ -19,7 +19,9 @@ SFT_RUNDIC=${1:?usage: $0 <SFT_RUNDIC> [PREV_RUNDIC]}
PREV_RUNDIC=${2:-$((SFT_RUNDIC-1))}
SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
ROOT=${AUTORESEARCH_ROOT:-/mnt/wangsenhao/autoresearch-zk}
# chat-isolated 时代 AUTORESEARCH_ROOT == AUTORESEARCH_CHAT_ROOT；优先用 chat
# root，避免 agent 没 export AUTORESEARCH_ROOT 时回退到全局共享路径（已废弃）。
ROOT=${AUTORESEARCH_ROOT:-${AUTORESEARCH_CHAT_ROOT:-$(dirname "$SCRIPT_DIR")}}
TPL=${SFT_TRAIN_JOB_TEMPLATE:-$SCRIPT_DIR/sft_train_job.yaml.tpl}
YAML=/tmp/sft_train_job_r${SFT_RUNDIC}.yaml