drawn-water-93 distillation run

dry-meadow-86 distilled

eager-fog-84 distillation from sage-mountain-341 (flan-t5-base tuned) with batch size 32

0.7.0 missing files

cleaned up, fixed bug quantizing 0.7.0 model

Distillation test. lively-planet-17 (based on t5-efficient-tiny) distilled from trainer azure-frost-334 (baed on flan-t5-base)

eager-plant-323 model - no removed layers

still-durian-309 build with 0.1 setting for word shortening

sandy-forest-305 model, similar to devoted-* model but with correct bad word filtering

major-elevator-302 - title case specific bad word filtering and addition of training data pertaining to searches

genial-tree-283 run. Brevity and None support

devoted-puddle-246 training run with 'None' result when uncertain and some removed layers

new download format for elated-lake-212 model with data stored separately for faster retreival

elated-lake-212 training run with 6 layers removed for size

upbeat-eon-195 training run with multiple decoder layers removed to reduce model size. Based on flan-t5-small

pious-butterfly-170 remove layers and fine tune from flan-t5-small model

dulcet-durian-136 training with t5-efficient-tiny base model

Dainty-blaze-127 training run with smaller t5-efficient-mini model

gentil-pyramid run -- extractive summarization

fixed quantize bug

swift-rain-107 release - different prompt for single title

cool-yogurt-98 relase with support for keywords in input

Upgrade to simplified model with less training data. azure-river-73 training run in w&b

update config.json prefx str

add models

initial commit