Repository navigation

Add RTP-CODEC training, tokenizers and evaluation benchmarks - #7

Merged

VEK239 merged 75 commits into

mainfrom

integration/publication-2026-10

Oct 4, 2026

Merged

Add RTP-CODEC training, tokenizers and evaluation benchmarks#7
VEK239 merged 75 commits into
mainfrom
integration/publication-2026-10

Commits on Jul 17, 2026
Commits on Jul 21, 2026
Commits on Jul 22, 2026
Commits on Jul 23, 2026
Commits on Jul 29, 2026
Commits on Jul 30, 2026
- Add a notebook to visualize the char vs WordPiece comparison
  demidovamaria
  committed
  978f8d0
- Save per-epoch train/val metrics for each run
  demidovamaria
  committed
  8790782
Commits on Aug 1, 2026
Commits on Aug 2, 2026
- Add padding-investigation scripts: per-vocab max_len vs fixed
  demidovamaria
  committed
  f5f9d66
Commits on Aug 3, 2026
- Add frozen-representation caching and PCA bottleneck for reconstruction benchmark- cache_embeddings: one-hot, ESM-2 8M, TCR-BERT, SCEPTR with shared mean pooling- fit_bottleneck: train-only Standar…
  Consoooomer
  committed
  29760f1
Commits on Aug 4, 2026
- script for slurm
  Consoooomer
  committed
  555180d
- Add reconstruction results collector and trim the array time limit
  Consoooomer
  committed
  c1d828c
- fix comment in script
  Consoooomer
  committed
  f53870a
- fix for write to out
  Consoooomer
  committed
  8eac79d
- add report results
  Consoooomer
  committed
  db17d9b
- Pretrain the sequence-to-TCRemP encoder on the benchmark split
  
  Consoooomer
  and
  claude
  committed
  4d31ef6
- Add Pgen benchmark arms, Slurm array and results collector
  
  Consoooomer
  and
  claude
  committed
  fd203ff
Commits on Aug 6, 2026
- Add anchored padding as an alternative to end-padding
  demidovamaria
  committed
  3a57bce
- Expose dropout as a CLI flag
  demidovamaria
  committed
  1558cc7
- Wire up --pad-layout in the comparison script
  demidovamaria
  committed
  e48434d
- Add anchored-padding experiment scripts, with and without extra dropout
  demidovamaria
  committed
  7837d28
- Set PYTHONPATH inside the scripts
  demidovamaria
  committed
  666ad4d
- Fix encode_wordpiece silently truncating on overflow
  demidovamaria
  committed
  166b485
Commits on Aug 7, 2026
Commits on Aug 8, 2026
- Add extra-seed array, end-to-end speed benchmark and a significance test
  
  Consoooomer
  and
  claude
  committed
  16df081
- Call training directly instead of through an inner srun
  Consoooomer
  committed
  53ae419
- Extend the extra-seed array to the 1k and 10k splits
  Consoooomer
  committed
  1950f3b
- Add Pgen results and limitations to the report
  Consoooomer
  committed
  ddcb927
- Add cosine LR schedule and a dependency chain for the Pgen rerun
  Consoooomer
  committed
  54acc94
Commits on Aug 9, 2026
- Update Pgen results after the cosine-schedule rerun
  Consoooomer
  committed
  d3c0b91
Commits on Aug 11, 2026
- Point the report at the archived constant-lr results
  Consoooomer
  committed
  d437f96
- Add benchmark result CSVs
  
  Petr Grigorev
  authored and
  Petr Grigorev
  committed
  93b63b6
- Add hidden_dim/num_conv_blocks awareness to both notebooks
  demidovamaria
  committed
  e5fefbb
- Add a unified notebook comparing all 5 architecture runs
  demidovamaria
  committed
  7b0e7f1
Commits on Aug 13, 2026
- Add reconstruction result CSVs
  Consoooomer
  committed
  603ea76
- Merge reconstruction results from issue-2 branch
  Consoooomer
  committed
  b96f06e
Commits on Aug 14, 2026
- Add multitask tokenizer and evaluation pipeline
  VEK239
  committed
  9abb43c
Commits on Oct 4, 2026

Commits

Commits on Jul 17, 2026

Commits on Jul 21, 2026

Commits on Jul 22, 2026

Commits on Jul 23, 2026

Commits on Jul 29, 2026

Commits on Jul 30, 2026

Commits on Aug 1, 2026

Commits on Aug 2, 2026

Commits on Aug 3, 2026

Commits on Aug 4, 2026

Commits on Aug 6, 2026

Commits on Aug 7, 2026

Commits on Aug 8, 2026

Commits on Aug 9, 2026

Commits on Aug 11, 2026

Commits on Aug 13, 2026

Commits on Aug 14, 2026

Commits on Oct 4, 2026