This website requires JavaScript.
Explore
Help
Sign In
dingbin.wang
/
jev-training-data
Watch
1
Star
0
Fork
0
You've already forked jev-training-data
Code
Issues
Pull Requests
Actions
Packages
Projects
Releases
Wiki
Activity
160
Commits
1
Branch
0
Tags
Go to file
Code
Clone
HTTPS
Open with VS Code
Open with VSCodium
Open with Intellij IDEA
Download ZIP
Download TAR.GZ
Download BUNDLE
dingbin.wang
6444e8e102
r41 r41_result.json
2026-11-06 02:35:32 +00:00
laya_cfg_min
mmBERT fresh-head config + full 4K decision data
2026-10-21 17:12:19 +08:00
laya_repo
laya repo modules + v0.5 decision-format data (4K balanced)
2026-10-21 14:52:47 +08:00
logs
r41 r41_result.json
2026-11-06 02:35:32 +00:00
model_live
log
2026-10-21 03:40:08 +00:00
model_parts
art model_parts/metrics.json
2026-10-21 03:44:29 +00:00
v03
v0.3 dataset: +530 real-query pairs (x2 upweight)
2026-10-19 13:46:06 +08:00
v04
v0.4 dataset: +268 generic-page negatives (x3), eval label fixes
2026-10-19 17:01:51 +08:00
v05
v0.5: evidence-aligned windows + 42% EN rebalance + dual-language canaries; canary-aware trainer
2026-10-21 11:31:06 +08:00
v05n
v0.5-nordk: RDK全剥离纯公开数据 + Laya noul 训练脚本
2026-10-21 19:10:56 +08:00
b.tar.gz
nordk artifacts
2026-10-21 11:54:39 +00:00
bundle.tar.gz
jev v0.2 dataset bundle + training script
2026-10-15 16:07:14 +08:00
calibration.jsonl
log
2026-10-15 08:54:51 +00:00
canary_en.jsonl
log
2026-10-21 03:35:59 +00:00
canary_zh.jsonl
log
2026-10-21 03:35:59 +00:00
convert_mmbert.py
convert v2 full-state
2026-10-21 17:35:24 +08:00
eval_v2.jsonl
2400 balanced + eval_v2
2026-10-21 14:55:17 +08:00
eval.jsonl
log
2026-10-19 09:10:18 +00:00
fetch_model_v2.py
v2 scripts: mmBERT/ModernBERT support (Laya同源底座)
2026-10-19 18:36:00 +08:00
fetch_model_v3.py
fetch_model_v3: ModelScope direct download
2026-11-05 14:56:33 +08:00
fetch_model.py
add fetch_model.py (modelscope直链+重试)
2026-10-15 20:51:34 +08:00
injection.jsonl
log
2026-10-15 08:54:51 +00:00
laya_v05_decisions.jsonl
laya repo modules + v0.5 decision-format data (4K balanced)
2026-10-21 14:52:47 +08:00
laya_v05_train_2400.jsonl
2400 balanced + eval_v2
2026-10-21 14:55:17 +08:00
manifest_delta.json
nordk artifacts
2026-10-21 11:54:39 +00:00
manifest.json
log
2026-10-15 08:54:51 +00:00
push_log.py
add log/model push helpers
2026-10-15 16:08:50 +08:00
push_model.py
add log/model push helpers
2026-10-15 16:08:50 +08:00
README.md
jev v0.2 dataset bundle + training script
2026-10-15 16:07:14 +08:00
run_fat.py
standalone + all-in-one scripts
2026-11-05 15:38:11 +08:00
sanity_cuda_cpu.py
sanity: CUDA/CPU forward divergence arbitration
2026-10-21 00:48:27 +08:00
split_model.py
add split_model.py (避免多行heredoc)
2026-10-15 20:36:46 +08:00
train_all.py
train_all r41: crosspair long negs (len×label balance) + agg-level T_agg calibration
2026-11-06 10:22:13 +08:00
train_jev_v2.py
v0.5: evidence-aligned windows + 42% EN rebalance + dual-language canaries; canary-aware trainer
2026-10-21 11:31:06 +08:00
train_jev.py
LoRA training: artifact 1.1GB -> ~20MB adapter
2026-10-19 10:56:24 +08:00
train_laya_noul.py
fix: marker_mask bool + remove laya_common debug import
2026-11-05 14:44:31 +08:00
train_laya.py
train_laya.py wrapper
2026-10-21 14:57:06 +08:00
train_standalone.py
standalone + all-in-one scripts
2026-11-05 15:38:11 +08:00
train.jsonl
nordk artifacts
2026-10-21 11:54:39 +00:00
README.md
jev-training-data
jev_dataset_v0.2 bundle + train_jev.py
Description
jev v1 训练数据与运行产物(自动创建)
Readme
744
MiB
Languages
Python
100%