AETHORIA-AI/TR-HASH-MOE-500M-HF

Publish packed LoRA V2 epoch 2 adapter

Publish semantic-cell LoRA step 1052 experiment

Publish Card Corpus V2 epoch 1 LoRA checkpoint

Keep thinking optional for natural chat

Force think prefill in official chat template

Publish V19 LoRA final step 2962

Replace step 1000 with V19 LoRA step 2000

Replace step 500 with V19 LoRA step 1000

Publish experimental V19 LoRA step 500

Fix multi-turn EOS rendering

Align multi-turn chat template with SFT

Publish merged E3 reasoning-adaptation checkpoint

Replace epoch-3 SFT with final Atlas v16 checkpoint

Point post-training data to AETHORIA-AI

Update transferred base model reference

Update BoolQ and WinoGrande direct-scoring benchmarks

Document three-epoch SFT and zero-shot MLX benchmarks

Publish three-epoch runtime files and MLX zero-shot benchmark

Replace one-epoch weights with validated three-epoch SFT export

Remove private checkpoint link from public model card

Add pretraining and SFT learning curves

Replace repository with TR-Hash MoE 500M SFT v13 step 3095

Add files using upload-large-folder tool

initial commit