41.3 MB

Ctrl+K

1 contributor

History: 346 commits

Hana-ame

feat(experiments): expand experiment matrix to 220, add LoRA/format/hyperparams sweeps and configs

ce812c8 about 3 hours ago

additive_rand_transformer
chore: sync train.py eval fixes, use_model.sh python path, and latest configs about 4 hours ago
archive
chore: archive research markdown docs (fully integrated into Excel) about 4 hours ago
checkpoints
feat: publish seg2_b16 checkpoint (L4·D64 seg2 b16, bottleneck seg variant) via git-lfs 2 days ago
configs
feat(experiments): expand experiment matrix to 220, add LoRA/format/hyperparams sweeps and configs about 3 hours ago
runs
协同适应探针: coadapt_l4d128.json about 17 hours ago
.gitattributes

1.52 kB
initial commit 3 days ago
.gitignore

124 Bytes
docs: add PARAMS.md (scan params by code), cross-link in RESEARCH/README/RESULTS, gitignore runs+pycache 3 days ago
Colab_Run_Additive_Transformer.ipynb

13.7 kB
feat(additive): add --config support, Colab notebook, INT8 quantization module, and full experiments excel about 4 hours ago
EXPERIMENTS_ALL.xlsx

58 kB
feat(experiments): expand experiment matrix to 220, add LoRA/format/hyperparams sweeps and configs about 3 hours ago
LICENSE

1.07 kB
feat: package layout + model card + 7 checkpoints 3 days ago
README.md

8.53 kB
docs: 完善中文文档与 DIY 操作手册 about 3 hours ago
param_variation2_all.json

11.2 kB
param_variation2 21档网格 + 21f RoPE: param_variation2_all.json 1 day ago
ppl_matrix.json

37.3 kB
fixed testset + all-7-model perplexity sweep (ad1ee65) 2 days ago
ppl_matrix_full.json

115 kB
full-matrix perplexity: add/sub split + all role segments (756693f) 2 days ago
pyproject.toml

1.08 kB
feat: package layout + model card + 7 checkpoints 3 days ago
requirements.txt

23 Bytes
Upload folder using huggingface_hub 3 days ago
rope_retrain.json

406 Bytes
param_variation2 21档网格 + 21f RoPE: rope_retrain.json 1 day ago
step_sweep.json

11 kB
step sweep: continue-training +500/+1000/+2000/+4000 for all 7 checkpoints, fixed-testset eval (621d1ae) 2 days ago
testset_adder.json

20.5 kB
feat+docs: fixed test set + all-7-model perplexity sweep (per-digit bucketed) 2 days ago
use_model.sh

6.59 kB
chore: sync train.py eval fixes, use_model.sh python path, and latest configs about 4 hours ago