Repository navigation

Log documents_seen / num_documents and offer it as a W&B x-axis - #559

Merged

jlamypoirier merged 11 commits into

mainfrom

jlp_documents_seen

Jul 9, 2026

Commits on Jul 8, 2026
- Log documents_seen / num_documents and offer it as a W&B x-axis
  
  jlamypoirier
  and
  claude
  committed
  ab176fb
- Review fixes for #559: return num_documents_in_batch from run_step, drop dead check, reword comments
  
  jlamypoirier
  and
  claude
  committed
  68ad8a3
- Make document counting the default instead of an opt-in loss preprocessing flag
  
  jlamypoirier
  and
  claude
  committed
  b744797
- Remove now-vestigial return_document_count flag
  
  jlamypoirier
  and
  claude
  committed
  64675ac
Commits on Jul 9, 2026
- Drop dead num_documents-is-not-None check in TokenModelInput.share_batch_data
  
  jlamypoirier
  and
  claude
  committed
  b5e1dbd
- Collapse now-empty TokenPreprocessingConfig into LengthPreprocessingConfig
  
  jlamypoirier
  and
  claude
  committed
  a1b63a6
- Define the documents_seen W&B x-axis unconditionally in __init__
  
  jlamypoirier
  and
  claude
  committed
  083dfab
- Always include num_documents/documents_seen in logged training metrics
  
  jlamypoirier
  and
  claude
  committed
  006eb9c
- Review fixes: trust int document-count boundary, keep step as default W&B x-axis
  
  jlamypoirier
  and
  claude
  committed
  baffd3a
- Review fixes: dedup document-count expression, drop restating comments
  
  jlamypoirier
  and
  claude
  committed
  6176cc9
- Merge branch 'main' into jlp_documents_seen
  jlamypoirier
  authored
  2a7f912