@qvac/sdk 0.20.0 → 0.21.0

package/NOTICE

@@ -71,14 +71,20 @@ Third-Party Model Licenses
     https://huggingface.co/Audio8/Audio8-TTS-Preview-0.6b
   bci-embedder
     https://github.com/tetherto/qvac/releases/download/bci-test-assets-v0.1.0/bci-embedder.bin
+  coremldata
+    https://huggingface.co/Audio8/Audio8-TTS-Preview-0.6b
   cosyvoice3-campplus-f32
     https://huggingface.co/FunAudioLLM/Fun-CosyVoice3-0.5B-2512
+  cosyvoice3-flow-bf16
+    https://huggingface.co/FunAudioLLM/Fun-CosyVoice3-0.5B-2512
   cosyvoice3-flow-f32
     https://huggingface.co/FunAudioLLM/Fun-CosyVoice3-0.5B-2512
   cosyvoice3-hift-f32
     https://huggingface.co/FunAudioLLM/Fun-CosyVoice3-0.5B-2512
   cosyvoice3-llm-q8_0
     https://huggingface.co/FunAudioLLM/Fun-CosyVoice3-0.5B-2512
+  cosyvoice3-llm-q8_0-fused
+    https://huggingface.co/FunAudioLLM/Fun-CosyVoice3-0.5B-2512
   cosyvoice3-s3tok-f16
     https://huggingface.co/FunAudioLLM/Fun-CosyVoice3-0.5B-2512
   craft_mlt_25k
@@ -132,7 +138,7 @@ Third-Party Model Licenses
   LightOnOCR-2-1B-ocr-soup-GGUF
     https://huggingface.co/noctrex/LightOnOCR-2-1B-ocr-soup-GGUF
   Llama_3.2_1B_Intruct_Tool_Calling_V2-GGUF
-    https://huggingface.co/mav23/Llama_3.2_1B_Intruct_Tool_Calling_V2-GGUF
+    https://huggingface.co/mradermacher/Llama_3.2_1B_Intruct_Tool_Calling_V2-GGUF
   MedPsy-1.7B-GGUF
     https://huggingface.co/qvac/MedPsy-1.7B-GGUF
   MedPsy-4B-GGUF
@@ -141,6 +147,14 @@ Third-Party Model Licenses
     https://huggingface.co/FunAudioLLM/Fun-CosyVoice3-0.5B-2512
   mmproj-unlimited-ocr-Q8_0
     https://huggingface.co/baidu/Unlimited-OCR
+  model
+    https://huggingface.co/Audio8/Audio8-TTS-Preview-0.6b
+  moss-codec-decoder-f16
+    https://huggingface.co/OpenMOSS-Team/MOSS-TTS
+  moss-codec-encoder-f16
+    https://huggingface.co/OpenMOSS-Team/MOSS-TTS
+  moss-tts-delay-f16
+    https://huggingface.co/OpenMOSS-Team/MOSS-TTS
   nb-tiny-ggml-model
     https://huggingface.co/NbAiLab/nb-whisper-tiny
   parler-indic-f16
@@ -207,6 +221,8 @@ Third-Party Model Licenses
     https://huggingface.co/HuggingFaceVLA/smolvla_libero
   SmolVLM2-500M-Video-Instruct-GGUF
     https://huggingface.co/ggml-org/SmolVLM2-500M-Video-Instruct-GGUF
+  Ternary-Bonsai-2-27B-gguf
+    https://huggingface.co/prism-ml/Ternary-Bonsai-2-27B-gguf
   TranslatePsy-AfriSLM-0.8B-Q4-GGUF
     https://huggingface.co/qvac/TranslatePsy-AfriSLM-0.8B-Q4-GGUF
   TranslatePsy-AfriSLM-0.8B-Q8-GGUF
@@ -247,6 +263,8 @@ Third-Party Model Licenses
     https://huggingface.co/Comfy-Org/Wan_2.2_ComfyUI_Repackaged
   Wan2.2-TI2V-5B-Turbo-GGUF
     https://huggingface.co/hum-ma/Wan2.2-TI2V-5B-Turbo-GGUF
+  weight
+    https://huggingface.co/Audio8/Audio8-TTS-Preview-0.6b
 
 --- bsd-3-clause (BSD 3-Clause License) ---
 
@@ -273,30 +291,58 @@ Third-Party Model Licenses
 
   AfriqueGemma-4B-GGUF
     https://huggingface.co/mradermacher/AfriqueGemma-4B-GGUF
+  coremldata
+    https://huggingface.co/nvidia/parakeet-ctc-0.6b
+  coremldata
+    https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3
+  coremldata
+    https://huggingface.co/nvidia/parakeet-tdt-1.1b
+  coremldata
+    https://huggingface.co/nvidia/parakeet_realtime_eou_120m-v1
+  model
+    https://huggingface.co/nvidia/parakeet-ctc-0.6b
+  model
+    https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3
+  model
+    https://huggingface.co/nvidia/parakeet-tdt-1.1b
+  model
+    https://huggingface.co/nvidia/parakeet_realtime_eou_120m-v1
+  parakeet_realtime_eou_120m-v1.f16
+    https://huggingface.co/nvidia/parakeet_realtime_eou_120m-v1
+  parakeet_realtime_eou_120m-v1.q4_0
+    https://huggingface.co/nvidia/parakeet_realtime_eou_120m-v1
+  parakeet_realtime_eou_120m-v1.q8_0
+    https://huggingface.co/nvidia/parakeet_realtime_eou_120m-v1
   parakeet-ctc-0.6b.f16
     https://huggingface.co/nvidia/parakeet-ctc-0.6b
   parakeet-ctc-0.6b.q4_0
     https://huggingface.co/nvidia/parakeet-ctc-0.6b
   parakeet-ctc-0.6b.q8_0
     https://huggingface.co/nvidia/parakeet-ctc-0.6b
-  parakeet-eou-120m-v1.f16
-    https://huggingface.co/nvidia/parakeet_realtime_eou_120m-v1
-  parakeet-eou-120m-v1.q4_0
-    https://huggingface.co/nvidia/parakeet_realtime_eou_120m-v1
-  parakeet-eou-120m-v1.q8_0
-    https://huggingface.co/nvidia/parakeet_realtime_eou_120m-v1
   parakeet-tdt-0.6b-v3.f16
     https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3
   parakeet-tdt-0.6b-v3.q4_0
     https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3
   parakeet-tdt-0.6b-v3.q8_0
     https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3
+  parakeet-tdt-1.1b.f16
+    https://huggingface.co/nvidia/parakeet-tdt-1.1b
+  parakeet-tdt-1.1b.q8_0
+    https://huggingface.co/nvidia/parakeet-tdt-1.1b
   sortformer-4spk-v1.f16
     https://huggingface.co/nvidia/diar_sortformer_4spk-v1
   sortformer-4spk-v1.q4_0
     https://huggingface.co/nvidia/diar_sortformer_4spk-v1
   sortformer-4spk-v1.q8_0
     https://huggingface.co/nvidia/diar_sortformer_4spk-v1
+  weight
+    https://huggingface.co/nvidia/parakeet-ctc-0.6b
+  weight
+    https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3
+  weight
+    https://huggingface.co/nvidia/parakeet-tdt-1.1b
+  weight
+    https://huggingface.co/nvidia/parakeet_realtime_eou_120m-v1
 
 --- gemma (Gemma Terms of Use) ---
 
@@ -330,8 +376,12 @@ Third-Party Model Licenses
 
 --- mit (MIT License) ---
 
+  acestep-5Hz-lm-0.6B-BF16
+    https://huggingface.co/ACE-Step/acestep-5Hz-lm-0.6B
   acestep-5Hz-lm-0.6B-Q8_0
     https://huggingface.co/ACE-Step/acestep-5Hz-lm-0.6B
+  acestep-v15-base-Q4_K_M
+    https://huggingface.co/ACE-Step/acestep-v15-base
   acestep-v15-sft-Q8_0
     https://huggingface.co/ACE-Step/acestep-v15-sft
   acestep-v15-turbo-Q4_K_M
@@ -354,6 +404,8 @@ Third-Party Model Licenses
     https://huggingface.co/ResembleAI/chatterbox
   chatterbox-t3-turbo
     https://huggingface.co/ResembleAI/chatterbox-turbo
+  coremldata
+    https://huggingface.co/ai4bharat/indic-conformer-600m-multilingual
   de-base-ggml-model-f16
     https://huggingface.co/aware-ai/whisper-base-german
   DeepSeek-V4-Flash-0731-GGUF
@@ -376,16 +428,18 @@ Third-Party Model Licenses
     https://huggingface.co/ChristianAzinn/gte-large-gguf
   gte-large-gguf
     https://huggingface.co/ChristianAzinn/gte-large-gguf
-  indic-conformer-ctc.f16
+  indic-conformer-600m-multilingual.f16
     https://huggingface.co/ai4bharat/indic-conformer-600m-multilingual
-  indic-conformer-ctc.q4_0
+  indic-conformer-600m-multilingual.q4_0
     https://huggingface.co/ai4bharat/indic-conformer-600m-multilingual
-  indic-conformer-ctc.q8_0
+  indic-conformer-600m-multilingual.q8_0
     https://huggingface.co/ai4bharat/indic-conformer-600m-multilingual
   lavasr-denoiser-f16
     https://huggingface.co/spaces/YatharthS
   lavasr-enhancer-f16
     https://huggingface.co/spaces/YatharthS
+  model
+    https://huggingface.co/ai4bharat/indic-conformer-600m-multilingual
   pt-tiny-ggml-model-f16
     https://huggingface.co/DavyCosta701/whisper-tiny-pt-bpra
   taew2_2_f16
@@ -396,6 +450,8 @@ Third-Party Model Licenses
     https://voiceinput.futo.org/VoiceInput/tiny_en_acft_q8_0.bin
   vae-BF16
     https://huggingface.co/ACE-Step/Ace-Step1.5
+  weight
+    https://huggingface.co/ai4bharat/indic-conformer-600m-multilingual
   whisper-vad
     https://huggingface.co/ggml-org/whisper-vad
   whisper.cpp
@@ -610,6 +666,12 @@ Third-Party Model Licenses
 
 --- nvidia-open-model-license ---
 
+  coremldata
+    https://huggingface.co/nvidia/diar_streaming_sortformer_4spk-v2.1
+  coremldata
+    https://huggingface.co/nvidia/nemotron-3.5-asr-streaming-0.6b
+  coremldata
+    https://huggingface.co/nvidia/parakeet-unified-en-0.6b
   diar_streaming_sortformer_4spk-v2.1.f16
     https://huggingface.co/nvidia/diar_streaming_sortformer_4spk-v2.1
   diar_streaming_sortformer_4spk-v2.1.q4_0
@@ -620,6 +682,12 @@ Third-Party Model Licenses
     https://huggingface.co/nvidia/GR00T-N1.7-LIBERO
   groot-q8_vf16
     https://huggingface.co/nvidia/GR00T-N1.7-3B
+  model
+    https://huggingface.co/nvidia/diar_streaming_sortformer_4spk-v2.1
+  model
+    https://huggingface.co/nvidia/nemotron-3.5-asr-streaming-0.6b
+  model
+    https://huggingface.co/nvidia/parakeet-unified-en-0.6b
   nemotron-3.5-asr-streaming-0.6b.f16
     https://huggingface.co/nvidia/nemotron-3.5-asr-streaming-0.6b
   nemotron-3.5-asr-streaming-0.6b.q4_0
@@ -632,6 +700,12 @@ Third-Party Model Licenses
     https://huggingface.co/nvidia/parakeet-unified-en-0.6b
   parakeet-unified-en-0.6b.q8_0
     https://huggingface.co/nvidia/parakeet-unified-en-0.6b
+  weight
+    https://huggingface.co/nvidia/diar_streaming_sortformer_4spk-v2.1
+  weight
+    https://huggingface.co/nvidia/nemotron-3.5-asr-streaming-0.6b
+  weight
+    https://huggingface.co/nvidia/parakeet-unified-en-0.6b
 
 --- openrail ---
 
@@ -674,61 +748,64 @@ the underlying models.
 
 Source URLs for original weights are recorded in each model's
 metadata entry within the registry.
-
 =========================================================================
 JavaScript Dependencies
 =========================================================================
 
 --- apache-2.0 (Apache License 2.0) ---
 
-  @hyperswarm/secret-stream@6.9.1
+  @hyperswarm/secret-stream@6.9.2
     https://github.com/holepunchto/hyperswarm-secret-stream
-  @qvac/asr-ggml@0.3.3
+  @qvac/asr-ggml@0.5.3
     https://github.com/tetherto/qvac
-  @qvac/audiogen-ggml@0.3.3
+  @qvac/asr-ggml-darwin-arm64@0.5.3
     https://github.com/tetherto/qvac
-  @qvac/bci-whispercpp@0.8.2
+  @qvac/audiogen-ggml@0.4.1
     https://github.com/tetherto/qvac
-  @qvac/classification-ggml@0.23.0
+  @qvac/audiogen-ggml-darwin-arm64@0.4.1
     https://github.com/tetherto/qvac
-  @qvac/decoder-audio@0.5.0
+  @qvac/bci-whispercpp@0.9.1
     https://github.com/tetherto/qvac
-  @qvac/diagnostics@0.1.2
+  @qvac/classification-ggml@0.26.3
     https://github.com/tetherto/qvac
-  @qvac/diffusion-cpp@0.21.0
+  @qvac/decoder-audio@0.6.0
     https://github.com/tetherto/qvac
-  @qvac/embed-llamacpp@0.37.0
+  @qvac/diffusion-cpp@0.24.0
     https://github.com/tetherto/qvac
-  @qvac/error@0.1.1
-  @qvac/fabric@0.10.0
+  @qvac/embed-llamacpp@0.41.3
     https://github.com/tetherto/qvac
-  @qvac/infer-base@0.4.2
+  @qvac/error@0.1.1
+  @qvac/fabric@0.16.1
     https://github.com/tetherto/qvac
   @qvac/infer-base@0.6.2
     https://github.com/tetherto/qvac
-  @qvac/inference@0.19.1
+  @qvac/inference@0.20.0
     https://github.com/tetherto/qvac
   @qvac/langdetect-text@0.1.2
     https://github.com/tetherto/qvac
-  @qvac/llm-llamacpp@0.49.1
+  @qvac/llm-llamacpp@0.53.2
     https://github.com/tetherto/qvac
   @qvac/logging@0.1.1
     https://github.com/tetherto/qvac
-  @qvac/ocr-ggml@0.21.0
+  @qvac/ocr-ggml@0.24.3
+    https://github.com/tetherto/qvac
+  @qvac/rag@0.8.1
     https://github.com/tetherto/qvac
   @qvac/registry-client@0.6.1
     https://github.com/tetherto/qvac
   @qvac/registry-schema@0.3.0
     https://github.com/tetherto/qvac
-  @qvac/translation-nmtcpp@0.13.0
+  @qvac/translation-nmtcpp@0.16.4
     https://github.com/tetherto/qvac
-  @qvac/tts-ggml@0.8.1
+  @qvac/tts-ggml@0.9.2
     https://github.com/tetherto/qvac
-  @qvac/vla-ggml@0.24.0
+  @qvac/tts-ggml-darwin-arm64@0.9.2
+    https://github.com/tetherto/qvac
+  @qvac/vla-ggml@0.27.3
     https://github.com/tetherto/qvac
   adaptive-timeout@1.0.1
     https://github.com/holepunchto/adaptive-timeout
-  b4a@1.8.1
+  b4a@1.9.0
     https://github.com/holepunchto/b4a
   bare-abort@2.0.13
     https://github.com/holepunchto/bare-abort
@@ -746,7 +823,7 @@ JavaScript Dependencies
     https://github.com/holepunchto/bare-bundle
   bare-bundle-id@1.0.2
     https://github.com/holepunchto/bare-bundle-id
-  bare-cov@1.2.2
+  bare-cov@1.2.3
     https://github.com/holepunchto/bare-cov
   bare-cpu-info@0.1.1
     https://github.com/holepunchto/bare-cpu-info
@@ -754,7 +831,7 @@ JavaScript Dependencies
     https://github.com/holepunchto/bare-crypto
   bare-debug-log@2.0.0
     https://github.com/holepunchto/bare-debug-log
-  bare-dns@2.2.0
+  bare-dns@2.2.1
     https://github.com/holepunchto/bare-dns
   bare-encoding@1.0.3
     https://github.com/holepunchto/bare-encoding
@@ -762,9 +839,9 @@ JavaScript Dependencies
     https://github.com/holepunchto/bare-env
   bare-events@2.9.2
     https://github.com/holepunchto/bare-events
-  bare-fetch@3.2.1
+  bare-fetch@3.4.0
     https://github.com/holepunchto/bare-fetch
-  bare-ffmpeg@1.5.0
+  bare-ffmpeg@1.6.0
     https://github.com/holepunchto/bare-ffmpeg
   bare-form-data@1.2.2
     https://github.com/holepunchto/bare-form-data
@@ -778,7 +855,7 @@ JavaScript Dependencies
     https://github.com/holepunchto/bare-hrtime
   bare-http-parser@2.1.4
     https://github.com/holepunchto/bare-http-parser
-  bare-http1@4.6.1
+  bare-http1@4.6.2
     https://github.com/holepunchto/bare-http1
   bare-https@3.1.0
     https://github.com/holepunchto/bare-https
@@ -786,19 +863,15 @@ JavaScript Dependencies
     https://github.com/holepunchto/bare-inspect
   bare-inspector@6.1.0
     https://github.com/holepunchto/bare-inspector
-  bare-lief@0.2.7
-    https://github.com/holepunchto/bare-lief
-  bare-link@3.3.0
-    https://github.com/holepunchto/bare-link
   bare-mime@1.0.0
     https://github.com/holepunchto/bare-mime
   bare-module@6.4.0
     https://github.com/holepunchto/bare-module
-  bare-module-lexer@1.6.6
+  bare-module-lexer@1.6.7
     https://github.com/holepunchto/bare-module-lexer
   bare-module-resolve@1.12.5
     https://github.com/holepunchto/bare-module-resolve
-  bare-module-traverse@2.5.5
+  bare-module-traverse@2.5.6
     https://github.com/holepunchto/bare-module-traverse
   bare-net@2.3.3
     https://github.com/holepunchto/bare-net
@@ -818,12 +891,10 @@ JavaScript Dependencies
     https://github.com/holepunchto/bare-process
   bare-rpc@1.3.8
     https://github.com/holepunchto/bare-rpc
-  bare-runtime@1.32.0
+  bare-runtime@1.33.4
     https://github.com/holepunchto/bare-runtime
-  bare-runtime-darwin-arm64@1.32.0
+  bare-runtime-darwin-arm64@1.33.4
     https://github.com/holepunchto/bare-runtime
-  bare-semver@1.1.0
-    https://github.com/holepunchto/bare-semver
   bare-signals@4.2.0
     https://github.com/holepunchto/bare-signals
   bare-signals@5.0.0
@@ -832,13 +903,11 @@ JavaScript Dependencies
     https://github.com/holepunchto/bare-stdio
   bare-stream@2.13.4
     https://github.com/holepunchto/bare-stream
-  bare-structured-clone@1.6.0
-    https://github.com/holepunchto/bare-structured-clone
   bare-stylize@0.0.1
     https://github.com/holepunchto/bare-stylize
   bare-subprocess@5.2.3
     https://github.com/holepunchto/bare-subprocess
-  bare-subprocess@6.1.0
+  bare-subprocess@6.2.0
     https://github.com/holepunchto/bare-subprocess
   bare-tcp@2.6.1
     https://github.com/holepunchto/bare-tcp
@@ -846,16 +915,17 @@ JavaScript Dependencies
     https://github.com/holepunchto/bare-tls
   bare-tty@5.2.1
     https://github.com/holepunchto/bare-tty
-  bare-type@1.1.1
+  bare-type@1.4.0
     https://github.com/holepunchto/bare-type
-  bare-type-stripper@0.1.6
+  bare-type-stripper@0.1.7
     https://github.com/holepunchto/bare-type-stripper
   bare-url@2.5.4
     https://github.com/holepunchto/bare-url
   bare-utils@1.6.0
     https://github.com/holepunchto/bare-utils
-  bare-v8-to-istanbul@1.0.2
-  bare-ws@3.1.0
+  bare-v8-to-istanbul@1.0.4
+    https://github.com/holepunchto/bare-v8-to-istanbul
+  bare-ws@3.2.0
     https://github.com/holepunchto/bare-ws
   bare-zlib@1.4.1
     https://github.com/holepunchto/bare-zlib
@@ -887,7 +957,7 @@ JavaScript Dependencies
     https://github.com/holepunchto/hypercore-id-encoding
   hypercore-stats@2.5.1
     https://github.com/holepunchto/hypercore-stats
-  hypercore-storage@3.2.1
+  hypercore-storage@3.3.1
     https://github.com/holepunchto/hypercore-storage
   hyperdb@6.9.0
     https://github.com/holepunchto/hyperdb
@@ -897,9 +967,9 @@ JavaScript Dependencies
     https://github.com/holepunchto/hyperdht-stats
   hyperdispatch@1.6.0
     https://github.com/holepunchto/hyperdispatch
-  hyperdrive@13.3.3
+  hyperdrive@13.3.4
     https://github.com/holepunchto/hyperdrive
-  hyperschema@1.24.0
+  hyperschema@1.26.2
     https://github.com/holepunchto/hyperschema
   hyperswarm-stats@1.3.1
     https://github.com/holepunchto/hyperswarm-stats
@@ -925,11 +995,11 @@ JavaScript Dependencies
     https://github.com/holepunchto/refcounter
   require-addon@1.3.0
     https://github.com/holepunchto/require-addon
-  require-asset@1.2.2
+  require-asset@1.2.4
     https://github.com/holepunchto/require-asset
   resource-on-exit@1.0.0
     https://github.com/holepunchto/bare-teardown
-  rocksdb-native@3.17.4
+  rocksdb-native@3.18.1
     https://github.com/holepunchto/rocksdb-native
   scope-lock@1.2.4
     https://github.com/holepunchto/scope-lock
@@ -939,7 +1009,7 @@ JavaScript Dependencies
     https://github.com/holepunchto/sub-encoder
   text-decoder@1.2.7
     https://github.com/holepunchto/text-decoder
-  udx-native@1.21.1
+  udx-native@1.21.3
     https://github.com/holepunchto/udx-native
   unslab@1.3.0
     https://github.com/holepunchto/unslab
@@ -967,10 +1037,6 @@ JavaScript Dependencies
 
 --- mit (MIT License) ---
 
-  @jridgewell/resolve-uri@3.1.2
-    https://github.com/jridgewell/resolve-uri
-  @jridgewell/sourcemap-codec@1.6.0
-    https://github.com/jridgewell/sourcemaps
   @jridgewell/trace-mapping@0.3.31
     https://github.com/jridgewell/sourcemaps
   @types/istanbul-lib-coverage@2.0.6
@@ -985,7 +1051,7 @@ JavaScript Dependencies
     https://github.com/mafintosh/codecs
   convert-source-map@2.0.0
     https://github.com/thlorenz/convert-source-map
-  corestore@7.12.4
+  corestore@7.12.5
     https://github.com/holepunchto/corestore
   debounceify@1.1.0
     https://github.com/mafintosh/debounceify
@@ -1005,13 +1071,13 @@ JavaScript Dependencies
     https://github.com/mafintosh/generate-string
   hyperbee@2.27.3
     https://github.com/holepunchto/hyperbee
-  hypercore@11.35.5
+  hypercore@11.36.1
     https://github.com/holepunchto/hypercore
   hypercore-crypto@3.7.0
     https://github.com/mafintosh/hypercore-crypto
   hyperdht@6.34.0
     https://github.com/holepunchto/hyperdht
-  hyperswarm@4.17.1
+  hyperswarm@4.17.2
     https://github.com/holepunchto/hyperswarm
   is-options@1.0.2
     https://github.com/mafintosh/is-options
@@ -1019,6 +1085,8 @@ JavaScript Dependencies
     https://github.com/mikolalysenko/is-property
   kademlia-routing-table@1.0.6
     https://github.com/mafintosh/kademlia-routing-table
+  llm-splitter@0.2.0
+    https://github.com/nearform/llm-splitter
   mutexify@1.4.0
     https://github.com/mafintosh/mutexify
   nat-sampler@1.0.1
@@ -1089,6 +1157,6 @@ JavaScript Dependencies
     https://github.com/mafintosh/xache
   z32@1.1.0
     https://github.com/mafintosh/z32
-  zod@4.6.1
+  zod@4.6.5
     https://github.com/colinhacks/zod
 

package/README.md

@@ -22,6 +22,13 @@ For AI/LLM tools, use [https://docs.qvac.tether.io/llms-full.txt](https://docs.q
 
 See https://docs.qvac.tether.io/sdk/getting-started/installation
 
+On Node.js, the SDK installs and launches its own Bare worker through `bare-runtime`.
+Its dependency range must satisfy `@qvac/inference`'s `engines.bare` requirement.
+When upgrading the SDK, reinstall dependencies with the existing lockfile so the
+package manager can replace an older, incompatible runtime. Remove any override
+or resolution that forces the worker below that requirement; installing a newer
+global `bare` executable does not change the SDK's worker dependency.
+
 ## Quickstart
 
 1. Create the examples workspace:
@@ -143,14 +150,16 @@ initialized, the event omits the resource block.
 ## Pre-download model fit assessment
 
 Use `assessModelFit` to check, before downloading anything, whether models are
-likely to fit in this device's memory. It reads generated catalog metadata plus a
-fresh memory sample — no weights, no load, no native probe:
+likely to fit in this device's memory. Each candidate is described as `loadModel`
+takes it. It reads generated catalog metadata plus a fresh memory sample, and for
+a single candidate the registry's weightless description of the artifact, so the
+engine's own fitter can answer — no weights, no load:
 
 ```ts
 import { assessModelFit, QWEN3_8B_INST_Q4_K_M } from '@qvac/sdk'
 
 const result = await assessModelFit({
-  models: [{ model: QWEN3_8B_INST_Q4_K_M, workload: { kind: 'llm', contextTokens: 8192 } }],
+  models: [{ modelSrc: QWEN3_8B_INST_Q4_K_M, modelType: 'llm', modelConfig: { ctx_size: 8192 } }],
   execution: 'sequential',
   policy: 'interactive-v1'
 })
@@ -185,6 +194,33 @@ console.log(stats?.audioDuration, stats?.realTimeFactor)
 `session.stats` resolves to `undefined` when the engine does not report
 statistics.
 
+Batch `transcribe()` calls also expose terminal statistics:
+
+```ts
+const call = transcribe({ modelId, audioChunk })
+const text = await call
+const stats = await call.stats
+console.log(text, stats?.encoderUsedCoreml)
+```
+
+`call.stats` resolves to `undefined` when the engine reports no statistics and
+rejects if transcription fails.
+
+## Parakeet Core ML encoders on Apple devices
+
+Load a supported Parakeet registry constant as usual. On macOS and iOS, the SDK
+downloads any complete, published Core ML encoder bundle with its GGUF and
+places the `.mlmodelc` directory where the native addon discovers it. This is
+available for TDT 0.6B v3, Unified English 0.6B, EOU 120M v1, and streaming
+Sortformer v2.1. The download includes the sidecar weights, so first load takes
+more time and disk space. Other platforms download only the GGUF. If the
+sidecar is unavailable or cannot load, inference uses the GGUF encoder.
+Sortformer v2.1 also needs a GGUF with the
+`parakeet.model_variant=sortformer-streaming-v2.1-aosc` metadata. Caching both
+sidecars alone does not activate Core ML. See
+[`examples/asr/parakeet-unified-coreml.ts`](examples/asr/parakeet-unified-coreml.ts)
+for a local check of a Unified GGUF and its encoder bundle.
+
 ## Examples
 
 In the `./examples` subdirectory, you will find scripts demonstrating how to use all SDK functionalities. To try any of them:

package/dist/examples/advisory-model-fit.d.ts

@@ -1,15 +1,16 @@
 /**
- * Advisory llama.cpp fit check (QVAC-22629).
+ * Advisory model fit check.
  *
- * Before a completion or embedding load, the SDK runs `@qvac/model-fit` in one
- * disposable Bare child and projects whether the exact configuration it is
- * about to load will fit in device memory.
+ * Before a load, the SDK runs the fitter belonging to the engine that would
+ * run it and projects whether the exact configuration it is about to load will
+ * fit in device memory. Desktop runs it in a disposable Bare child; mobile,
+ * which cannot spawn one, runs it on a worker thread.
  *
  * The result is ADVISORY. It never blocks a load. `does-not-fit` is logged and
  * the ordinary load path runs unchanged. Crashes, timeouts, malformed
  * responses, unsupported configurations, and internal errors all resolve to
- * "no evidence" and are equally non-blocking. Nothing consumes the verdict
- * yet — this PR only produces it.
+ * "no evidence" and are equally non-blocking. No verdict changes the load;
+ * `getLoadedModelInfo` returns it as `fitProbe`.
  *
  * The verdict is emitted on the SDK server log stream, not to stdout, so this
  * example subscribes to `loggingStream({ id: SDK_LOG_ID })` and reprints the
@@ -19,9 +20,8 @@
  * What fits on this machine (Apple M4 Pro, 24 GiB unified memory)
  * ---------------------------------------------------------------------------
  *
- * Measured with `@qvac/model-fit@0.8.0` (the first release carrying the
- * qvac-fabric#214 memory-reporting fix) at the default 1024 MiB margin. The
- * fitter budgets against what the machine can actually keep resident
+ * Measured at the default 1024 MiB margin. The fitter budgets against what the
+ * machine can actually keep resident
  * (total − wired − compressor: 17.4 GiB on this machine at idle), not the raw
  * RAM figure.
  *
@@ -30,10 +30,12 @@
  *     gpt-oss-20B   Q4_K_M  10.8 GiB  @  32k ctx
  *     gte-large     fp16     0.6 GiB  (embedding, context pinned to 512)
  *
- *   PROJECTED NOT TO FIT — try any of these to see a `does-not-fit` verdict
+ *   PROJECTED NOT TO FIT — with `gpu_layers: 99` pinned
  *     gpt-oss-20B   Q4_K_M  10.8 GiB  @ 128k ctx with an f32 KV cache ← below
- *     Gemma 4 31B   Q4_K_M  18.3 GiB  @   1k ctx   ← the weights alone do not
- *                                                    fit with gpu_layers: 99
+ *
+ * These rows were measured with `gpu_layers` pinned at 99. Loads no longer pin
+ * the layer count, and the `does-not-fit` row has not been re-measured
+ * unpinned.
  *
  * gpt-oss-20B at 128k context FITS with the default KV cache and DOES NOT FIT
  * once `cache-type-k`/`cache-type-v` are set to `f32`. Same model, same
@@ -42,10 +44,9 @@
  *
  * Two boundaries worth understanding when reading verdicts:
  *
- * 1. The verdict answers for a PLACEMENT, not a model. The SDK's schema
- *    default is `gpu_layers: 99` ("everything on the GPU"), so `does-not-fit`
- *    means "not at this placement". Omit `gpu_layers` and the fitter is free
- *    to move layers to the CPU side instead.
+ * 1. The verdict answers for a PLACEMENT, not a model. With `gpu_layers`
+ *    unset, the fitter may move layers to the CPU side; setting it pins the
+ *    layer count, and `does-not-fit` then means "not at this placement".
  * 2. A `does-not-fit` configuration can still RUN on macOS when the OS
  *    compresses and pages hard enough — and a `fits` configuration right at
  *    the boundary can still fail at first decode under memory pressure.