Release v1.3.6

What's Changed

• valid inputs reported as unknown baseconfig fix by @IMvision12 in #456
• Fix Auto registry coverage for Qwen image pipeline components by @IMvision12 in #457
• Fix Grounding DINO documentation imports and remove obsolete alias exemption by @IMvision12 in #458
• Add Initial Sentence Transformers support by @IMvision12 in #459
• Refactor Tests Run by @IMvision12 in #460
• Refactor Diffusion and Sentence Transformer by @IMvision12 in #461
• Release 1.3.6 by @IMvision12 in #462

Full Changelog: v1.3.5...v1.3.6

Release v1.3.5

What's Changed

• Fix Bart imports and tests by @IMvision12 in #444
• HF_TOKEN sent to arbitrary hosts and fix groundingdino by @IMvision12 in #445
• Clarify converted-cache trust model and secure cache permissions by @IMvision12 in #446
• Deprecated license metadata and incomplete classifiers by @IMvision12 in #447
• Add glm 4.7 flash missing docs by @IMvision12 in #448
• Fix versions by @IMvision12 in #449
• Fix sampler API: validate TopP args, register, drop gumbel by @IMvision12 in #450
• Add MPNet Model by @IMvision12 in #451
• Add Qwen Image Model by @IMvision12 in #452
• Add Qwen Image 2.1 Model by @IMvision12 in #453
• Fix Qwen Image and Qwen Image 2.1 Generate and dtype by @IMvision12 in #454
• Release 1.3.5 by @IMvision12 in #455

Full Changelog: v1.3.4...v1.3.5

Release v1.3.4 (Stable Diffusion Family)

What's Changed

• Add Diffusion Models: Base Diffusion and Diffusion 1(1.1,1.2,1.3,1.4,1.5) by @IMvision12 in #441
• Add Stable Diffusion XL, 3, 3.5 by @IMvision12 in #442
• release 1.3.4 by @IMvision12 in #443

Full Changelog: v1.3.3...v1.3.4

Release v1.3.3

What's Changed

• Fix timm on fly issue by @IMvision12 in #432
• resolve cross-family model_type mismatches in Auto* tables by @IMvision12 in #433
• lign constructor defaults with configs and close auto-registry + docstring gaps by @IMvision12 in #434
• wire the no-float quantized load and key the converter cache per build by @IMvision12 in #435
• scope attn_implementation per model instead of a leaky process global by @IMvision12 in #436
• correctness fixes across loading, quantization, configs, and processors by @IMvision12 in #437
• fix: preprocessing correctness audit by @IMvision12 in #438
• Fix Docs and Tests by @IMvision12 in #439
• Release 1.3.3 by @IMvision12 in #440

Full Changelog: v1.3.2...v1.3.3

Release v1.3.2

What's Changed

• Fix(seq2seq): honor load_dtype in the decode KV cache (bf16 dtype mismatch) by @IMvision12 in #422
• fix(generation): honor load_dtype in seq2seq KV cache; compile+cache the multimodal decode path by @IMvision12 in #423
• resolve model_type "llama" to the LlamaModel superset + disambiguate tokenizer by vocab by @IMvision12 in #424
• remove dead ambiguity guards (raise gemma4/qwen3_5, default depth_anything/segformer) by @IMvision12 in #425
• HF_MODEL_TYPE siglip -> siglip2 so its own repos load by @IMvision12 in #426
• Remove old config for on fly models by @IMvision12 in #427
• Fix group_size, strict expert layouts, reversible dequantize by @IMvision12 in #428
• fix failing owl vit tests by @IMvision12 in #429
• fix-attn by @IMvision12 in #430
• release 1.3.2 by @IMvision12 in #431

Full Changelog: v1.3.1...v1.3.2

Release v1.3.1

What's Changed

• Add Bart Model by @IMvision12 in #420
• Release 1.3.1 by @IMvision12 in #421

Full Changelog: v1.3.0...v1.3.1

Release v1.3.0

What's Changed

• Fix batched generation for unequal-length prompts by @IMvision12 in #415
• Padding-aware true-batch decode for ragged generation (llama, qwen2, qwen3) by @IMvision12 in #416
• Fix stochastic generation: non-deterministic by default + (steps, batch) sampling noise by @IMvision12 in #417
• Guard degenerate sampler parameters (p ≤ 0, k ≤ 0, temperature ≤ 0) by @IMvision12 in #418
• release 1.3.0 by @IMvision12 in #419

Full Changelog: v1.2.9...v1.3.0

Release v1.2.9

What's Changed

• Add Auto Class by @IMvision12 in #413
• Release 1.2.9 by @IMvision12 in #414

Full Changelog: v1.2.8...v1.2.9

Release v1.2.8

What's Changed

• Fix Whisper model by @IMvision12 in #400
• Fix snapshot tests by @IMvision12 in #401
• Fix TF SparseTensor issue and pre commit fix by @IMvision12 in #402
• Replace numpy with keras.ops across model code by @IMvision12 in #403
• Add Facebook's Regnet Models by @IMvision12 in #404
• Add Table Transformer by @IMvision12 in #405
• Add Beit models by @IMvision12 in #406
• Add Levit by @IMvision12 in #407
• Move image normalization from models to per-backbone image processors by @IMvision12 in #408
• Update Tests and add sharding by @IMvision12 in #409
• Remove baked default variants: tokenizers/processors now require an explicit repo by @IMvision12 in #410
• Add cross-backend (torch/jax/tf) output parity test by @IMvision12 in #411
• Release 1.2.8 by @IMvision12 in #412

Full Changelog: v1.2.7...v1.2.8

Release v1.2.7

What's Changed

• Add Granite Speech5 Models by @IMvision12 in #395
• Add Google's OG Model Efficientdet by @IMvision12 in #396
• Add Pyramid Vision Transformer V1 and V2 by @IMvision12 in #397
• Fix LocateAnything API by @IMvision12 in #398
• Release 1.2.7 by @IMvision12 in #399

Full Changelog: v1.2.6...v1.2.7