v2.2.2

What's Changed

• Add relative position attention kernel and bindings by @jakmro in https://github.com/cactus-compute/cactus/pull/822

Full Changelog: cactus-compute/cactus@v2.2.1...v2.2.2

v2.2.1

What's Changed

• Expand Parakeet buckets by @jakmro in https://github.com/cactus-compute/cactus/pull/818

Full Changelog: cactus-compute/cactus@v2.2.0...v2.2.1

v2.2.0

What's Changed

• Custom transpiler by @ParkiratS in https://github.com/cactus-compute/cactus/pull/796
• fix(graph): deduplicate shared mmap weight files across component graphs by @Moeez34 in https://github.com/cactus-compute/cactus/pull/794
• Fix padded-frame leakage and feature normalization in Parakeet encoder by @jakmro in https://github.com/cactus-compute/cactus/pull/808

New Contributors

• @Moeez34 made their first contribution in https://github.com/cactus-compute/cactus/pull/794

Full Changelog: cactus-compute/cactus@v2.1.0...v2.2.0

v2.1.0

What's Changed

• docs: fix "HuggigFace" typo in README by @DaveyEke in https://github.com/cactus-compute/cactus/pull/746
• M3 pro bench results by @ParkiratS in https://github.com/cactus-compute/cactus/pull/762
• add wider cq support by @jakmro in https://github.com/cactus-compute/cactus/pull/763
• add needle support by @jakmro in https://github.com/cactus-compute/cactus/pull/765
• add tool calling support to run env by @jakmro in https://github.com/cactus-compute/cactus/pull/766
• optimize attention and gemv kernels by @jakmro in https://github.com/cactus-compute/cactus/pull/776
• Custom transpiler by @ParkiratS in https://github.com/cactus-compute/cactus/pull/778
• Include directly for std::setw/setfill/setprecision (fixes build on libstdc++/Pi) by @nitishagar in https://github.com/cactus-compute/cactus/pull/782
• Fix per-matrix CQ4 reciprocals in Metal batch transform by @clumsylad21 in https://github.com/cactus-compute/cactus/pull/784
• Bucket the Parakeet audio encoder by @jakmro in https://github.com/cactus-compute/cactus/pull/791

New Contributors

• @DaveyEke made their first contribution in https://github.com/cactus-compute/cactus/pull/746
• @nitishagar made their first contribution in https://github.com/cactus-compute/cactus/pull/782
• @clumsylad21 made their first contribution in https://github.com/cactus-compute/cactus/pull/784

Full Changelog: cactus-compute/cactus@v2.0.1...v2.1.0

v2.0.1

What's Changed

• Remove coreml by @jakmro in https://github.com/cactus-compute/cactus/pull/754
• fix brew by @jakmro in https://github.com/cactus-compute/cactus/pull/755
• fix brew by @jakmro in https://github.com/cactus-compute/cactus/pull/758

Full Changelog: cactus-compute/cactus@v2.0...v2.0.1

v2.0

What's Changed

• Lfm tool arg fixes by @justinl66 in https://github.com/cactus-compute/cactus/pull/596
• Gemma4 multimodal multi-turn fixes by @ncylich in https://github.com/cactus-compute/cactus/pull/598
• LFM NPU Fallback + Tokenizer Fixes by @ParkiratS in https://github.com/cactus-compute/cactus/pull/605
• Fix converter logic for models not on cactus huggingface by @cattermelon1234 in https://github.com/cactus-compute/cactus/pull/608
• Graph remaining ops by @cattermelon1234 in https://github.com/cactus-compute/cactus/pull/578
• Karen/v2 tq by @kar-m in https://github.com/cactus-compute/cactus/pull/615
• Add fused dense INT4 MLP op for Gemma-4 E2B decode by @ncylich in https://github.com/cactus-compute/cactus/pull/617
• inner 1 fast path by @cattermelon1234 in https://github.com/cactus-compute/cactus/pull/618
• Hybrid INT8/FP16 decode kernel: 2× faster + sliding-window fast path by @ncylich in https://github.com/cactus-compute/cactus/pull/625
• axis reductions fast path -> v2 by @cattermelon1234 in https://github.com/cactus-compute/cactus/pull/629
• Pytorch capture -> cactus v2 by @cattermelon1234 in https://github.com/cactus-compute/cactus/pull/622
• Karen/tq v2 debugging by @kar-m in https://github.com/cactus-compute/cactus/pull/630
• added cactus download cq flag by @kar-m in https://github.com/cactus-compute/cactus/pull/631
• align c++ official Needle inference by @jakmro in https://github.com/cactus-compute/cactus/pull/637
• Fix CQ decode fast path for mmap weights by @ncylich in https://github.com/cactus-compute/cactus/pull/633
• Cactus Quant Speedup by @justinl66 in https://github.com/cactus-compute/cactus/pull/641
• Transpiler complete by @ParkiratS in https://github.com/cactus-compute/cactus/pull/640
• fix: add missing include for Linux ARM64 builds by @ramanishka in https://github.com/cactus-compute/cactus/pull/647
• Update Cpp pipeline with transpiler, Add Android/IOS Bindings, Fix CLI by @justinl66 in https://github.com/cactus-compute/cactus/pull/649
• Transpiler complete by @ParkiratS in https://github.com/cactus-compute/cactus/pull/652
• Refactor quantization parameters in CLI and publish workflow by @jakmro in https://github.com/cactus-compute/cactus/pull/651
• Bindings by @jakmro in https://github.com/cactus-compute/cactus/pull/653
• Cleanup V2 (Cpp Native + CLI Update) by @justinl66 in https://github.com/cactus-compute/cactus/pull/650
• KV Cache Bugfix (reset cache, think token data shift, underflow guard) by @justinl66 in https://github.com/cactus-compute/cactus/pull/658
• bug fixes, cleanups, docs & blog by @jakmro in https://github.com/cactus-compute/cactus/pull/657
• Karen/v2 transcribe by @kar-m in https://github.com/cactus-compute/cactus/pull/660
• Build system, bindings, CI cleanup, pypi & brew by @jakmro in https://github.com/cactus-compute/cactus/pull/661
• Add native chunked prefill for transpiled LLM bundles by @ncylich in https://github.com/cactus-compute/cactus/pull/659
• Optimize orthogonal CQ4 LM head layout by @ncylich in https://github.com/cactus-compute/cactus/pull/663
• Make Android scheduling load-aware by @ncylich in https://github.com/cactus-compute/cactus/pull/666
• V2 NPU Support by @justinl66 in https://github.com/cactus-compute/cactus/pull/664
• Add v2 local HTTP server by @ncylich in https://github.com/cactus-compute/cactus/pull/669
• Multiple model fixes by @jakmro in https://github.com/cactus-compute/cactus/pull/673
• Jax Transpiler path by @Ayan9074 in https://github.com/cactus-compute/cactus/pull/671
• Propagate test executable failures by @psymon-ai in https://github.com/cactus-compute/cactus/pull/683
• Wire text/image/audio embeddings on the v2 transpiler; add /v1/embeddings by @ncylich in https://github.com/cactus-compute/cactus/pull/676
• Full implmentation into Gemma4 by @ParkiratS in https://github.com/cactus-compute/cactus/pull/668
• Jaxtranspilerpath by @Ayan9074 in https://github.com/cactus-compute/cactus/pull/681
• V2 by @HenryNdubuaku in https://github.com/cactus-compute/cactus/pull/689
• Merge cleanup by @jakmro in https://github.com/cactus-compute/cactus/pull/692
• Long-context: rolling KeyDiff KV-cache compaction + chunked prefill by @ncylich in https://github.com/cactus-compute/cactus/pull/686
• Correct docs to match current code by @ncylich in https://github.com/cactus-compute/cactus/pull/696
• Karen/needle v2 by @kar-m in https://github.com/cactus-compute/cactus/pull/690
• Karen/gemmma multiturn template by @kar-m in https://github.com/cactus-compute/cactus/pull/711
• fixed needle rope bug, moved repeated rope table instantiations from … by @kar-m in https://github.com/cactus-compute/cactus/pull/713
• added language model embed component by @kar-m in https://github.com/cactus-compute/cactus/pull/714
• engine+graph: thread-safe last_error + JSON option-parse hardening by @ncylich in https://github.com/cactus-compute/cactus/pull/702
• Fix Gemma 4 NPU vision encoder on non-square images by @jakmro in https://github.com/cactus-compute/cactus/pull/718
• rag interface fix for lm embeddings by @kar-m in https://github.com/cactus-compute/cactus/pull/719
• bindings: JNI & React-Native FFI memory-safety by @ncylich in https://github.com/cactus-compute/cactus/pull/708
• convert/CLI: conversion & tokenizer robustness by @ncylich in https://github.com/cactus-compute/cactus/pull/707
• transpiler: op routing, fusion & JAX-capture correctness by @ncylich in https://github.com/cactus-compute/cactus/pull/706
• transpiler: JAX attention causality + reduction precision, attention-output layout guard by @ncylich in https://github.com/cactus-compute/cactus/pull/705
• telemetry: JSON-escape records + bound the offline event cache by @ncylich in https://github.com/cactus-compute/cactus/pull/704
• cactus-engine: runtime correctness & memory-safety by @ncylich in https://github.com/cactus-compute/cactus/pull/703
• Fuse dispatch for GEMV by @ncylich in https://github.com/cactus-compute/cactus/pull/717
• Pad the chunked-prefill tail for sliding-window models by @ncylich in https://github.com/cactus-compute/cactus/pull/716
• Fix multi-turn chat and tool-call defects across model families by @ncylich in https://github.com/cactus-compute/cactus/pull/721
• Fix onboarding quickstart bugs and guard destructive cactus clean by @rshemet in https://github.com/cactus-compute/cactus/pull/722
• convert: fold gemma4 audio projection scaling into the weight reparam by @ncylich in https://github.com/cactus-compute/cactus/pull/723
• Parakeet-TDT NPU acceleration on the Apple Neural Engine by @ncylich in https://github.com/cactus-compute/cactus/pull/729
• fixed lfm and qwen vl handling and run.cpp multiturn behavior by @kar-m in https://github.com/cactus-compute/cactus/pull/728
• Simplify CLI: one bundle-resolution ladder + CQ level in list by @rshemet in https://github.com/cactus-compute/cactus/pull/715
• gemma4: fix 16x audio soft-token magnitude on NPU via single-source scaling by @ncylich in https://github.com/cactus-compute/cactus/pull/726
• readme: add Parakeet decode tps numbers by @ncylich in https://github.com/cactus-compute/cactus/pull/732
• Centralize build flags into cactus_flags; shrink libs by @jakmro in https://github.com/cactus-compute/cactus/pull/735
• Stt streaming by @jakmro in https://github.com/cactus-compute/cactus/pull/730
• Karen/parakeet probe by @kar-m in https://github.com/cactus-compute/cactus/pull/739
• parakeet: honor should_stop during TDT transcription by @rshemet in https://github.com/cactus-compute/cactus/pull/738
• squash convert &transpile, and align cli commands by @jakmro in https://github.com/cactus-compute/cactus/pull/740
• M3 pro numbers by @ParkiratS in https://github.com/cactus-compute/cactus/pull/742
• fix(graph): apply SIGMOID/TANH activations in MoE layer by @yus100 in https://github.com/cactus-compute/cactus/pull/736
• gemma4: run the audio encoder correctly and fast on the Apple Neural Engine by @ncylich in https://github.com/cactus-compute/cactus/pull/734
• Removed unnecessary null pointer checks by @wasup-yash in https://github.com/cactus-compute/cactus/pull/724
• MOE by @ParkiratS in https://github.com/cactus-compute/cactus/pull/727
• added cq4 optimizations to cq2 and cq3 by @kar-m in https://github.com/cactus-compute/cactus/pull/744
• Cactus code by @jakmro in https://github.com/cactus-compute/cactus/pull/745
• Fix C++ test build: expose vendored libs include (picojson) by @rshemet in https://github.com/cactus-compute/cactus/pull/748
• Trie-based grammar constraints for Gemma force_tools by @rshemet in https://github.com/cactus-compute/cactus/pull/747
• Metal Support by @jakmro in https://github.com/cactus-compute/cactus/pull/750
• Benchmarks by @jakmro in https://github.com/cactus-compute/cactus/pull/752

New Contributors

• @ramanishka made their first contribution in https://github.com/cactus-compute/cactus/pull/647
• @psymon-ai made their first contribution in https://github.com/cactus-compute/cactus/pull/683
• @yus100 made their first contribution in https://github.com/cactus-compute/cactus/pull/736
• @wasup-yash made their first contribution in https://github.com/cactus-compute/cactus/pull/724

Full Changelog: cactus-compute/cactus@v1.14...v2.0

v1.14

What's Changed

• Enable SDL2 live mic recording in chat build by @rshemet in https://github.com/cactus-compute/cactus/pull/579
• fix gemma4 conversion and cli by @jakmro in https://github.com/cactus-compute/cactus/pull/581
• Made non-thinking default for gemma4 & Introduce simultaneous multimodalitiy by @ParkiratS in https://github.com/cactus-compute/cactus/pull/582
• gemma4 vision memory optimizations by @jakmro in https://github.com/cactus-compute/cactus/pull/590
• added a prompt fix for better vision+audio prompt, removed code that … by @kar-m in https://github.com/cactus-compute/cactus/pull/589
• added default confidence and wired that for vlm by @kar-m in https://github.com/cactus-compute/cactus/pull/591
• Gemma4 fixes by @ParkiratS in https://github.com/cactus-compute/cactus/pull/588
• Gemma4 tool calling and RoPE fixes by @ncylich in https://github.com/cactus-compute/cactus/pull/594
• increased rolling window size for gemma4 by @kar-m in https://github.com/cactus-compute/cactus/pull/593

Full Changelog: cactus-compute/cactus@v1.13...v1.14

v1.13

What's Changed

• Improve model publishing error handling in workflow by @jakmro in https://github.com/cactus-compute/cactus/pull/546
• Fix VLM prefill cache reuse image path slicing by @KayaanT in https://github.com/cactus-compute/cactus/pull/545
• Add cache cleanup for Hugging Face models in export_and_publish_model by @jakmro in https://github.com/cactus-compute/cactus/pull/547
• Add branch override option to publish workflow by @ncylich in https://github.com/cactus-compute/cactus/pull/548
• Fix Gemma 3n conversion memory usage by @ncylich in https://github.com/cactus-compute/cactus/pull/549
• Update blog URL in README.md by @amerkld in https://github.com/cactus-compute/cactus/pull/544
• Parakeet streaming fix by @ParkiratS in https://github.com/cactus-compute/cactus/pull/551
• Tool call prompt formatting by @jakmro in https://github.com/cactus-compute/cactus/pull/558
• Feature Cleanup by @ParkiratS in https://github.com/cactus-compute/cactus/pull/559
• Add Whisper v3 (large-v3) support by @ncylich in https://github.com/cactus-compute/cactus/pull/557
• Add LFM2.5 VL 450M by @yujonglee in https://github.com/cactus-compute/cactus/pull/567
• LFM2-VL-450M: fix garbled output (vision tower, template, kernel) by @ncylich in https://github.com/cactus-compute/cactus/pull/565
• Change default transcribe model to parakeet-tdt-0.6b-v3 by @rshemet in https://github.com/cactus-compute/cactus/pull/566
• Expose min_p and repetition_penalty in completion options by @DuFanYin in https://github.com/cactus-compute/cactus/pull/560
• Graph save load by @cattermelon1234 in https://github.com/cactus-compute/cactus/pull/556
• Pyannote features and optimizations by @jakmro in https://github.com/cactus-compute/cactus/pull/571
• Karen/needle by @kar-m in https://github.com/cactus-compute/cactus/pull/574
• fix apple i8mm detection to use runtime sysctl check by @DuFanYin in https://github.com/cactus-compute/cactus/pull/562
• Fix streaming transcribe by @jakmro in https://github.com/cactus-compute/cactus/pull/576

New Contributors

• @amerkld made their first contribution in https://github.com/cactus-compute/cactus/pull/544
• @DuFanYin made their first contribution in https://github.com/cactus-compute/cactus/pull/560

Full Changelog: cactus-compute/cactus@v1.12...v1.13

v1.12

What's Changed

• Gemma 4
• Versioned docs with quickstart and SDK chooser by @rshemet in https://github.com/cactus-compute/cactus/pull/522
• add CACTUS_CLOUD_API_BASE by @yujonglee in https://github.com/cactus-compute/cactus/pull/521
• Model weights discoverability by @jakmro in https://github.com/cactus-compute/cactus/pull/520
• fix: set telemetry framework to "rust" for Rust bindings by @rshemet in https://github.com/cactus-compute/cactus/pull/519
• fix: prefer local libcactus over system-installed in test build by @rshemet in https://github.com/cactus-compute/cactus/pull/516
• Engine updates: tool calling, compute_entropy, converter improvements by @ncylich in https://github.com/cactus-compute/cactus/pull/517
• cactus_prefill by @mhayes853 in https://github.com/cactus-compute/cactus/pull/512
• Missing torch ops by @cattermelon1234 in https://github.com/cactus-compute/cactus/pull/518
• Docs fixes by @jakmro in https://github.com/cactus-compute/cactus/pull/524
• docs: document custom vocabulary support for transcription by @ayushmk7 in https://github.com/cactus-compute/cactus/pull/525
• Youtu by @jakmro in https://github.com/cactus-compute/cactus/pull/530
• Clean up Hugging Face cache after model export by @jakmro in https://github.com/cactus-compute/cactus/pull/533
• Added parakeet optimizations for apple by @ParkiratS in https://github.com/cactus-compute/cactus/pull/534
• Cactus torch api clean by @cattermelon1234 in https://github.com/cactus-compute/cactus/pull/529
• add CACTUS_CLOUD_HEADERS support by @yujonglee in https://github.com/cactus-compute/cactus/pull/531
• Parakeet encoder optimization by @ParkiratS in https://github.com/cactus-compute/cactus/pull/535
• Update docs site_url to docs.cactuscompute.com by @rshemet in https://github.com/cactus-compute/cactus/pull/527
• Custom vocabulary support for Parakeet TDT by @rshemet in https://github.com/cactus-compute/cactus/pull/532
• Fix VLM crash when adding multiple images in multi-turn conversation by @FarooqMulla in https://github.com/cactus-compute/cactus/pull/539
• New LayerNorm Kernel by @nshejwalkar in https://github.com/cactus-compute/cactus/pull/540
• Add pyannote/segmentation-3.0 speaker diarization (10ms, 976× realtime) by @rshemet in https://github.com/cactus-compute/cactus/pull/538
• Tinyllama by @ParkiratS in https://github.com/cactus-compute/cactus/pull/536

New Contributors

• @cattermelon1234 made their first contribution in https://github.com/cactus-compute/cactus/pull/518
• @ayushmk7 made their first contribution in https://github.com/cactus-compute/cactus/pull/525
• @FarooqMulla made their first contribution in https://github.com/cactus-compute/cactus/pull/539

Full Changelog: cactus-compute/cactus@v1.11...v1.12

v1.11

What's Changed

• Fix/issue#490 by @lennartvoelz in https://github.com/cactus-compute/cactus/pull/491
• simplify and align sdks by @jakmro in https://github.com/cactus-compute/cactus/pull/489
• remove models by @jakmro in https://github.com/cactus-compute/cactus/pull/492
• Update model configurations and enhance workflow settings in publish_… by @jakmro in https://github.com/cactus-compute/cactus/pull/495
• Update workflow to use macos-latest instead of macos-latest-xlarge by @jakmro in https://github.com/cactus-compute/cactus/pull/496
• Add dynamic max_tokens estimation based on audio length in cactus_tra… by @jakmro in https://github.com/cactus-compute/cactus/pull/499
• macOS: link clang_rt.osx to fix SME2 (_arm_tpidr2*) link failures under rustc by @yujonglee in https://github.com/cactus-compute/cactus/pull/498
• Add FFI log control: cactus_log_set_level and cactus_log_set_callback by @yujonglee in https://github.com/cactus-compute/cactus/pull/497
• Karen/qwen3p5 by @kar-m in https://github.com/cactus-compute/cactus/pull/481
• CLI upgrades by @rshemet in https://github.com/cactus-compute/cactus/pull/504
• feat(stt): custom vocabulary biasing for all speech models by @vyomshah05 in https://github.com/cactus-compute/cactus/pull/451
• Add Gemma 3N (text-only) model support by @ncylich in https://github.com/cactus-compute/cactus/pull/493
• fix: make FunctionGemma prompt formatting strict by @lennartvoelz in https://github.com/cactus-compute/cactus/pull/502
• fix: apply logit bias before greedy sampling by @ncylich in https://github.com/cactus-compute/cactus/pull/507
• remove redundant file linking for tie_word_embeddings by @jakmro in https://github.com/cactus-compute/cactus/pull/506
• Port general engine improvements for TinyLlama by @ncylich in https://github.com/cactus-compute/cactus/pull/513
• Speech-to-Text Timestamps by @jakmro in https://github.com/cactus-compute/cactus/pull/515

New Contributors

• @lennartvoelz made their first contribution in https://github.com/cactus-compute/cactus/pull/491

Full Changelog: cactus-compute/cactus@v1.10...v1.11