Commits

Commit history

1. Commits on Nov 12, 2026

   • Merge pull request #203 from fastino-ai/codex/gliner-training-session-proof

     Show description for c379245
     urchade

     authored
     c379245View commit details

     Copy full SHA for c379245

     Browse repository at this point
2. Commits on Nov 10, 2026

   • Document the eight supported GPU qualification profiles

     kfastino

     committed
     4eeedf2View commit details

     Copy full SHA for 4eeedf2

     Browse repository at this point

   • Qualify FP32 and BF16 sessions and preserve native validation weighting

     kfastino

     committed
     1a972efView commit details

     Copy full SHA for 1a972ef

     Browse repository at this point

   • Reject unqualified FP16 activation checkpointing combinations

     kfastino

     committed
     6dfb9b1View commit details

     Copy full SHA for 6dfb9b1

     Browse 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

     Browse repository at this point

   • Support native trained artifacts and preserve checkpointed adapter gradients

     kfastino

     committed
     94b0507View commit details

     Copy full SHA for 94b0507

     Browse repository at this point

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

     kfastino

     committed
     8aa9b15View commit details

     Copy full SHA for 8aa9b15

     Browse repository at this point

   • feat(training): serialize exact managed operation schedules

     kfastino

     committed
     1161fc0View commit details

     Copy full SHA for 1161fc0

     Browse repository at this point

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

     kfastino

     committed
     2fdd654View commit details

     Copy full SHA for 2fdd654

     Browse repository at this point

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

     kfastino

     committed
     45159d3View commit details

     Copy full SHA for 45159d3

     Browse repository at this point

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

     kfastino

     committed
     f831868View commit details

     Copy full SHA for f831868

     Browse repository at this point

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

     kfastino

     committed
     eca855aView commit details

     Copy full SHA for eca855a

     Browse repository at this point

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

     kfastino

     committed
     538fbc0View commit details

     Copy full SHA for 538fbc0

     Browse repository at this point

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

     kfastino

     committed
     fc0638bView commit details

     Copy full SHA for fc0638b

     Browse repository at this point

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

     kfastino

     committed
     2eb6d22View commit details

     Copy full SHA for 2eb6d22

     Browse repository at this point
3. Commits on Oct 20, 2026

   • Merge pull request #199 from whr778/fix/ddp-gradient-averaging

     Show description for ca4f682
     urchade

     authored
     ca4f682View commit details

     Copy full SHA for ca4f682

     Browse repository at this point

   • Fix DDP: gradients were never averaged across GPUs

     Show description for fe3145e
     

     whr778

     and

     claude

     committed
     fe3145eView commit details

     Copy full SHA for fe3145e

     Browse repository at this point
4. Commits on Oct 9, 2026

   • Merge pull request #189 from MattThomas-fastino/transformers-v5-support

     Show description for b86d35c
     urchade

     authored
     b86d35cView commit details

     Copy full SHA for b86d35c

     Browse repository at this point

   • Support Transformers 5 and refresh dependency caps

     Show description for ec0e312
     

     MattThomas-fastino

     and

     cursoragent

     committed
     ec0e312View commit details

     Copy full SHA for ec0e312

     Browse repository at this point
5. Commits on Oct 8, 2026

   • Merge pull request #188 from delyan-boychev/optimize-decoder-sync

     Show description for 08c8c8d
     urchade

     authored
     08c8c8dView commit details

     Copy full SHA for 08c8c8d

     Browse repository at this point

   • Merge pull request #187 from kfastino/fix/cls-true-label-synthetic

     Show description for 2763f94
     urchade

     authored
     2763f94View commit details

     Copy full SHA for 2763f94

     Browse repository at this point

   • Add decoder synchronization benchmark matrix

     delyan-boychev

     committed
     467ea4fView commit details

     Copy full SHA for 467ea4f

     Browse repository at this point

   • Merge remote-tracking branch 'upstream/main' into optimize-decoder-sync

     Show description for f59fab0
     delyan-boychev

     committed
     f59fab0View commit details

     Copy full SHA for f59fab0

     Browse repository at this point

   • Merge pull request #143 from delyan-boychev/optimize-span-scoring

     Show description for 0b67de7
     urchade

     authored
     0b67de7View commit details

     Copy full SHA for 0b67de7

     Browse repository at this point

   • Optimize boundary decoder synchronization

     delyan-boychev

     committed
     879fc2dView commit details

     Copy full SHA for 879fc2d

     Browse repository at this point

   • Optimize accelerator decoder synchronization

     delyan-boychev

     committed
     cf34874View commit details

     Copy full SHA for cf34874

     Browse repository at this point

   • Merge remote-tracking branch 'origin/main' into optimize-span-scoring

     Show description for 0fa0e88
     delyan-boychev

     committed
     0fa0e88View commit details

     Copy full SHA for 0fa0e88

     Browse repository at this point

   • Fix classification augmentation training the gold label as a negative

     Show description for 29689a7
     kfastino

     committed
     29689a7View commit details

     Copy full SHA for 29689a7

     Browse repository at this point

   • Fix classification augmentation training the gold label as a negative

     Show description for fbf18fb
     kfastino

     committed
     fbf18fbView commit details

     Copy full SHA for fbf18fb

     Browse repository at this point

   • 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

     Browse 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

     Browse 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

     Browse 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

     Browse repository at this point

   • Merge pull request #167 from hfastino/henry/eng-7622-exact-decoder-vacuous-constraint

     Show description for 511928e
     urchade

     authored
     511928eView commit details

     Copy full SHA for 511928e

     Browse repository at this point

   • Merge pull request #179 from bulatovv/fix/explicit-spans-shared-pool

     Show description for f88ee1f
     urchade

     authored
     f88ee1fView commit details

     Copy full SHA for f88ee1f

     Browse repository at this point