cnb.cEr0mSxFAFA
a88c47e273
- 用 CharTokenizer.build_vocab 从 977K 数据生成词表 (vocab_size=47725) - train_cnb.py: batch=224 + accum=2 (等效448) + AMP + torch.compile - build_vocab.py: 词表生成脚本 - vocab_full.json: 含特殊token完整词表 - 模型 41.85M参数, L40 GPU训练
85 lines
1.4 KiB
Plaintext
85 lines
1.4 KiB
Plaintext
# 已跟踪
|
||
# 缓存和日志
|
||
__pycache__/
|
||
*.pyc
|
||
*.log
|
||
/tmp/
|
||
|
||
# 模型权重 — 只追踪发布版本(精简推理版,不含optimizer,约60MB)
|
||
*.pth
|
||
!/llm_model_v2_resume_epoch_7.pth
|
||
!/xiaoli-llm-chat/xiaoli_v2_best_inference.pth
|
||
!/xiaoli-llm-chat/xiaoli_v2_final_inference.pth
|
||
|
||
# 中间输出目录（生成脚本的输出）
|
||
high_temp_outputs/
|
||
diverse_answers/
|
||
fresh_qa_outputs/
|
||
multiturn_outputs/
|
||
checkpoints_v2/
|
||
old_ckpts/
|
||
backup_old_arch/
|
||
|
||
# 旧版废弃数据
|
||
xiaoli_*.json
|
||
mega_interim_*.json
|
||
essay_interim_*.json
|
||
distill_*.json
|
||
cot_*.json
|
||
memory_long*.json
|
||
multiturn_batch_*.json
|
||
real_multiturn*.json
|
||
mega_questions*.json
|
||
math_questions*.json
|
||
essay_questions*.json
|
||
cot_candidates*.json
|
||
model_test_results*.json
|
||
online_distill_report*.json
|
||
|
||
# 测试/临时文件
|
||
test_*.json
|
||
test_*.py
|
||
/var/folders
|
||
|
||
# 日志
|
||
batch_runner*.log
|
||
mega_*.log
|
||
mimo_*.log
|
||
knowledge_*.log
|
||
train_*.log
|
||
web_server*.log
|
||
# CNB 配置 — 强制跟踪关键文件
|
||
.ide/
|
||
!/.ide/Dockerfile
|
||
!/.cnb.yml
|
||
|
||
# 训练输出目录
|
||
general_outputs/
|
||
high_temp_outputs/
|
||
multiturn_outputs/
|
||
|
||
# 本地临时
|
||
fun_test.py
|
||
*.log
|
||
|
||
# 打包/发布产物
|
||
xiaoli_package/
|
||
*.zip
|
||
|
||
# 上传/测试脚本
|
||
upload_*.py
|
||
_gofile_*
|
||
_t.*
|
||
_up_test.*
|
||
|
||
# 旧词表备份
|
||
vocab_old.json
|
||
vocab_backup*.json
|
||
|
||
# 数据目录（太大不进git）
|
||
data/
|
||
|
||
# 训练checkpoint（不进git，通过release发布）
|
||
xiaoli_cnb_best.pth
|
||
xiaoli_cnb_final.pth
|