Commits

Commit history

1. Commits on Nov 10, 2026

   • Qualify FP32 and BF16 sessions and preserve native validation weighting

     kfastino

     committed
     1a972efView commit details

     Copy full SHA for 1a972ef

     View code at this pointBrowse repository at this point

   • Reject unqualified FP16 activation checkpointing combinations

     kfastino

     committed
     6dfb9b1View commit details

     Copy full SHA for 6dfb9b1

     View code at this pointBrowse repository at this point

   • Merge current main while preserving torch-free planning and Torch 2.6 support

     kfastino

     committed
     9a3e5c5View commit details

     Copy full SHA for 9a3e5c5

     View code at this pointBrowse repository at this point

   • Support native trained artifacts and preserve checkpointed adapter gradients

     kfastino

     committed
     94b0507View commit details

     Copy full SHA for 94b0507

     View code at this pointBrowse repository at this point

   • refactor(training): make managed planning independent of ML runtime

     kfastino

     committed
     8aa9b15View commit details

     Copy full SHA for 8aa9b15

     View code at this pointBrowse repository at this point

   • feat(training): serialize exact managed operation schedules

     kfastino

     committed
     1161fc0View commit details

     Copy full SHA for 1161fc0

     View code at this pointBrowse repository at this point

   • feat(training): start fresh runs from trained session weights

     kfastino

     committed
     2fdd654View commit details

     Copy full SHA for 2fdd654

     View code at this pointBrowse repository at this point

   • feat(training): persist custom engine recovery state with sessions

     kfastino

     committed
     45159d3View commit details

     Copy full SHA for 45159d3

     View code at this pointBrowse repository at this point

   • fix(training): enforce deterministic algorithms and native bf16 support

     kfastino

     committed
     f831868View commit details

     Copy full SHA for f831868

     View code at this pointBrowse repository at this point

   • feat(training): report actual preprocessed session token usage

     kfastino

     committed
     eca855aView commit details

     Copy full SHA for eca855a

     View code at this pointBrowse repository at this point

   • feat(training): drive managed epochs through public session operations

     kfastino

     committed
     538fbc0View commit details

     Copy full SHA for 538fbc0

     View code at this pointBrowse repository at this point

   • feat(training): continue span and boundary sessions with LoRA state

     kfastino

     committed
     fc0638bView commit details

     Copy full SHA for fc0638b

     View code at this pointBrowse repository at this point

   • feat(training): prove bounded GLiNER session state operations

     kfastino

     committed
     2eb6d22View commit details

     Copy full SHA for 2eb6d22

     View code at this pointBrowse repository at this point
2. Commits on Oct 20, 2026

   • Fix DDP: gradients were never averaged across GPUs

     Show description for fe3145e
     

     whr778

     and

     claude

     committed
     fe3145eView commit details

     Copy full SHA for fe3145e

     View code at this pointBrowse repository at this point
3. Commits on Oct 8, 2026

   • Merge pull request #181 from jaanavit/fix/boundary-entity-alignment-and-phantom-spans

     Show description for 8fef000
     urchade

     authored
     8fef000View commit details

     Copy full SHA for 8fef000

     View code at this pointBrowse repository at this point

   • Merge pull request #161 from whr778/fix/sanitize-preserves-record-anchor

     Show description for b45bde6
     urchade

     authored
     b45bde6View commit details

     Copy full SHA for b45bde6

     View code at this pointBrowse repository at this point

   • Merge pull request #160 from jaanavit/fix/unload-adapter-noop-on-non-peftmodel

     Show description for d30035a
     urchade

     authored
     d30035aView commit details

     Copy full SHA for d30035a

     View code at this pointBrowse repository at this point

   • Merge pull request #169 from hfastino/henry/eng-7625-guard-peft-import

     Show description for f387c4a
     urchade

     authored
     f387c4aView commit details

     Copy full SHA for f387c4a

     View code at this pointBrowse repository at this point

   • Merge pull request #168 from hfastino/henry/eng-7623-local-trainer-negatives

     Show description for 6b5e897
     urchade

     authored
     6b5e897View commit details

     Copy full SHA for 6b5e897

     View code at this pointBrowse repository at this point
4. Commits on Sep 30, 2026

   • Fix boundary entity alignment errors and phantom contained gold spans

     Show description for 379b009
     

     jaanavit

     and

     cursoragent

     committed
     379b009View commit details

     Copy full SHA for 379b009

     View code at this pointBrowse repository at this point
5. Commits on Sep 28, 2026

   • Validate gold spans and document the input format

     bulatovv

     committed
     28cc322View commit details

     Copy full SHA for 28cc322

     View code at this pointBrowse repository at this point
6. Commits on Sep 23, 2026

   • Import peft only on LoRA code paths so non-LoRA training works without it

     Show description for 68f0d4b
     

     hfastino

     and

     cursoragent

     committed
     68f0d4bView commit details

     Copy full SHA for 68f0d4b

     View code at this pointBrowse repository at this point

   • Accept NER negatives whose label set comes from entity descriptions

     Show description for c83a3e9
     

     hfastino

     and

     cursoragent

     committed
     c83a3e9View commit details

     Copy full SHA for c83a3e9

     View code at this pointBrowse repository at this point
7. Commits on Sep 22, 2026

   • Keep a structure's record-metadata anchor through sanitization

     Show description for 1f065d1
     whr778

     committed
     1f065d1View commit details

     Copy full SHA for 1f065d1

     View code at this pointBrowse repository at this point

   • Fix unload_lora_adapter() silent no-op when model is not a PeftModel

     Show description for 0a44ba0
     

     jaanavit

     and

     cursoragent

     committed
     0a44ba0View commit details

     Copy full SHA for 0a44ba0

     View code at this pointBrowse repository at this point
8. Commits on Aug 17, 2026

   • add structured prediction support to g2.5

     urchade

     committed
     d407a2eView commit details

     Copy full SHA for d407a2e

     View code at this pointBrowse repository at this point
9. Commits on Aug 14, 2026

   • add test

     urchade

     committed
     8a28bc3View commit details

     Copy full SHA for 8a28bc3

     View code at this pointBrowse repository at this point
10. Commits on Jul 28, 2026

    • some optimization

      urchade

      committed
      2e8a44eView commit details

      Copy full SHA for 2e8a44e

      View code at this pointBrowse repository at this point
11. Commits on Jul 27, 2026

    • add boundary architecture #2

      urchade

      committed
      b5d7e14View commit details

      Copy full SHA for b5d7e14

      View code at this pointBrowse repository at this point

    • add boundary architecture

      urchade

      committed
      90b0dd8View commit details

      Copy full SHA for 90b0dd8

      View code at this pointBrowse repository at this point
12. Commits on Jul 23, 2026

    • Merge pull request #128 from fastino-ai/sst

      Show description for f161fac
      urchade

      authored
      f161facView commit details

      Copy full SHA for f161fac

      View code at this pointBrowse repository at this point

    • some change

      urchade

      committed
      e0036f0View commit details

      Copy full SHA for e0036f0

      View code at this pointBrowse repository at this point
13. Commits on Jun 18, 2026

    • Merge branch 'main' into main

      urchade

      authored
      11cb206View commit details

      Copy full SHA for 11cb206

      View code at this pointBrowse repository at this point
14. Commits on Jun 13, 2026

    • Keep multi-GPU trainer changes scoped

      Yuvrajxms09

      committed
      e31033bView commit details

      Copy full SHA for e31033b

      View code at this pointBrowse repository at this point

    • Resolve DDP checkpoint save conflict

      Yuvrajxms09

      committed
      90130dbView commit details

      Copy full SHA for 90130db

      View code at this pointBrowse repository at this point