Audio8-TTS-Preview-0.6b → Core AI: dualar int8 + codec decoder fp16 + codec encoder fp16w32, tokenizer, card

.gitattributes

@@ -33,3 +33,7 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+audio8_codec_decoder_fp16_t160.aimodel/main.mlirb filter=lfs diff=lfs merge=lfs -text
+audio8_codec_encoder_fp16_t216.aimodel/main.mlirb filter=lfs diff=lfs merge=lfs -text
+audio8_dualar_int8_cl2048_w32.aimodel/main.mlirb filter=lfs diff=lfs merge=lfs -text
+tokenizer/tokenizer.json filter=lfs diff=lfs merge=lfs -text

LICENSE

@@ -0,0 +1,201 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

NOTICE

@@ -0,0 +1,4 @@
+audio8_tts
+
+The model architecture is inspired by the DualAR design used in Fish Audio
+S2 Pro.

README.md

@@ -0,0 +1,159 @@
+---
+license: apache-2.0
+library_name: coreai
+pipeline_tag: text-to-speech
+base_model: Edge0/Audio8-TTS-Preview-0.6b
+language:
+  - yue
+  - zh
+  - nl
+  - en
+  - fr
+  - de
+  - it
+  - ja
+  - ko
+  - pl
+  - es
+tags: [core-ai, coreaikit, audio8, dualar, text-to-speech, tts, voice-cloning, zero-shot, multilingual, on-device, apple]
+base_model_relation: quantized
+---
+
+Core AI is Apple's on-device ML runtime in iOS 27 / macOS 27 and the successor to Core ML: PyTorch models are exported with Apple's `coreai-torch` (LLMs: `coreai.llm.export`) into `.aimodel` bundles that run on the GPU or the Neural Engine, e.g. Qwen3-8B 4-bit decodes at 94 tok/s on an M4 Max GPU, MLX 90 under the same protocol ([apple-silicon-llm-bench](https://github.com/john-rocky/apple-silicon-llm-bench), macOS 27 beta, 2026-06).
+
+# Audio8-TTS-Preview-0.6b — Core AI
+
+[🤗 mlboydaisuke/Audio8-TTS-Preview-0.6b-CoreAI](https://huggingface.co/mlboydaisuke/Audio8-TTS-Preview-0.6b-CoreAI) · Apache-2.0 · base [Edge0/Audio8-TTS-Preview-0.6b](https://huggingface.co/Edge0/Audio8-TTS-Preview-0.6b)
+
+[`Edge0/Audio8-TTS-Preview-0.6b`](https://huggingface.co/Edge0/Audio8-TTS-Preview-0.6b) (Edge0, Apache-2.0, 601M + a
+337M codec) is a **DualAR text-to-speech** model of the Fish Audio S2 Pro design: a Qwen2.5-shaped **slow AR** (24
+layers, 896 wide) predicts one *semantic* token per 46 ms frame, a 4-layer **fast AR** predicts the frame's nine other
+codec codebooks one after another, and a 44.1 kHz **DAC-style codec** turns the ten codebooks into audio. **Eleven
+languages** (Cantonese, Chinese, Dutch, English, French, German, Italian, Japanese, Korean, Polish, Spanish) and
+**zero-shot voice cloning** from a 0.5–30 s reference recording and its transcript. This is the zoo's first DualAR /
+semantic-token TTS and the first port with the sampler inside the graph. On 2026-09-28 the Hub had MLX builds
+(`mlx-community/Audio8-TTS-Preview-0.6b-bf16`, 8-bit and 4-bit conversions), the publisher's ONNX INT4 build for CPUs
+and a GGUF, and no Core AI or Core ML port of the TTS (a Core ML port exists of the publisher's ASR sibling); this port
+runs the model on iPhone and Mac through Apple's runtime.
+
+## Pipeline
+
+```
+text (+ reference transcript + reference codes) ──(host: the publisher's prompt segments, encoded one at a time)──▶ prompt [11, P]
+  1. dualar.aimodel  prefill(codes [1,11,32], pos) in windows                            ─▶ logits [32,4097] · hidden [32,896]
+                     frame(codes [1,11,1], pos, noise_slow [2,4097], window [10], noise_fast [9,4096], forced, use_forced)
+                       = the slow step + the semantic draw (top-k 50 / top-p 0.9 / T 0.7, Gumbel-max, RAS) + the fast AR's
+                         ten rows with nine codebook draws — ONE call per frame                 ─▶ semantic · codes [10]
+                     first_frame(logits, hidden, …) for the frame right after the prefill
+  host: the 10-token RAS window, the uniform draws (a seeded stream), stop at eos 151645 or 512 frames
+  2. codec.aimodel   codes [1,10,160] ─▶ wav [1, 327680] at 44.1 kHz                     (every op causal: 160-frame windows, keep the last 32)
+voice registration:  wav 44.1 kHz [1,1,442368] ─▶ encoder.aimodel ─▶ codes [1,10,216]    (0.5–10 s reference; the codes + transcript are the voice)
+```
+
+A frame is 2,048 samples (21.5 frames/s). The prompt is the publisher's: `<|im_start|>system\n` · "convert the provided
+text to speech" (or, with a voice, "… reference to the following:\n\nText:\n" · `<|speaker:0|>` + transcript ·
+"\n\nSpeech:\n" · the reference's codebook-0 codes as semantic ids) · `<|im_end|>\n<|im_start|>user\n` · text ·
+`<|im_end|>\n<|im_start|>assistant\n<|voice|>` — segments encoded separately, asserted id for id against the publisher's
+processor on all 18 fixtures, in Python and in the Swift host.
+
+### Graph contracts
+
+```
+dualar   prefill     in  codes[1,11,32] i32 (row 0 ids, rows 1-10 codebooks; pad 151643 / 0) · pos[1] i32   out logits[32,4097] f16 · hidden[32,896] f16
+         frame       in  codes[1,11,1] i32 (the previous frame) · pos[1] · noise_slow[2,4097] f32 · window[10] i32 (-1 = none)
+                         noise_fast[9,4096] f32 · forced[11] i32 · use_forced[1] f32           out semantic[1] i32 · codes[10] i32 · logits · hidden · fast_logits[9,4096] · sampled_*
+         first_frame in  logits[4097] f16 · hidden[896] f16 · the same noise / window / forced   out as frame
+         state       k_cache / v_cache [24,1,2,2048,64] f16 (prefill and frame)
+         precision   slow AR linears int8 (weight-only, per-block-32, symmetric with clipping); embeddings, the 4,097-row head, norms, the fast AR fp16
+codec    main        in  codes[1,10,160] i32 (right-pad 0; codebook 0 < 4096, codebooks 1-9 < 1024, clamped in-graph)   out wav[1,327680] f16
+encoder  main        in  audio[1,1,442368] f32 mono 44.1 kHz (right-pad 0)                                          out codes[1,10,216] i32 (keep ceil(samples/2048) frames)
+```
+
+The head is the embedding table's 4,097 rows (the semantic ids and eos) instead of 155,776: the publisher's sampler sets
+every other logit to -inf before drawing, so nothing is lost (7 MB instead of 279 MB).
+
+## What made it work
+
+1. **One graph call per frame, the sampler inside.** The publisher's ONNX runtime — and this port's first export —
+   runs a frame as a slow-AR call plus nine fast-AR calls with sampling on the host; on the raw Core AI runtime path
+   that is ~48 ms of engine time per 46 ms frame on an M4 Max, a call costing milliseconds before any arithmetic. The
+   shipped `frame` function does the slow step, the semantic draw and the fast AR's ten rows in one call, with the
+   publisher's top-k / top-p / temperature / Gumbel-max / RAS sampler written without a sort (`topk(50)` +
+   `logsumexp` + a cumulative sum); the uniform draws are inputs, so a recorded draw replays the oracle's choice exactly.
+   Details: [`knowledge/audio8-tts-port.md`](https://github.com/john-rocky/coreai-model-zoo/blob/main/knowledge/audio8-tts-port.md).
+2. **int8 on the slow AR, fp16 on the fast AR.** On 15,093 teacher-forced codebook draws the fp16 fast AR reproduces the
+   oracle's draw 99.5 % of the time; int8 drops to 95.8 % for 53 MB. The slow AR takes int8 well (1,671 / 1,695 semantic
+   draws, 21 of the 24 misses at a Gumbel margin under 0.13; fp16: 1,688).
+3. **The codec encoder needs fp32 arithmetic.** Its vector quantizers pick nearest codebook entries; the whole-fp16
+   export flips those choices and the nine residual books cascade (44–121 of ~200 frames exact). fp16 weights with fp32
+   arithmetic (the Fun-ASR encoder recipe) gives 691 / 712 frames exact and codebook 0 exact on every frame.
+4. **Kept from the publisher's code:** bfloat16-rounded RoPE tables (interleaved pairs), RMSNorm eps 1e-6 in fp32, the
+   RAS window that skips the first frame's token, the clamp of codebooks 1–9 to 1,023. The slow AR's residual peaks at
+   12,827 — a fifth of float16's range — so no rescale was needed.
+
+## Numerics gate (18 fixtures: 6 ja + 6 en + 2 zh sentences of our own, 4 voice-clone fixtures on FLEURS test clips)
+
+Oracle = the publisher's `modeling_arktts.py` in fp32 on the CPU, seeded (`generate` reproduced call for call, every
+uniform draw recorded). Teacher-forced = the oracle's tokens fed to the graphs; a *draw* comparison asks whether the
+in-graph sampler, given the port's logits and the oracle's noise, makes the oracle's choice.
+
+| stage (Mac GPU) | result |
+| --- | --- |
+| tokenizer-only prompt vs the publisher's processor | 18 / 18 prompts identical (Python and Swift) |
+| plain-torch re-author, eager fp32, teacher-forced | every argmax and every replayed draw equal to the oracle; codec wav cos 1.000000 |
+| **dualar int8 (ship)**, teacher-forced, 1,695 slow steps | logits cos min 0.99985, argmax 1,668, **draw 1,671 / 1,695** |
+| same, 15,093 fast-AR codebook draws | logits cos min 0.99805, argmax 14,705, **draw 14,711 / 15,093** |
+| codec decoder on the oracle's codes, 18 utterances | wav cos min 0.99978, log-mel cos min 0.99876 |
+| codec encoder (fp16w32) on the 4 reference clips, 712 frames | 691 frames exact; codebook 0 exact on every frame, every codebook ≥ 97.7 % |
+| free run (the port's own loop, the oracle's draws) | reached eos 18 / 18; Swift host == Python engine on 1,637 / 1,688 frames (10 / 18 utterances identical end to end) |
+| ASR round trip vs the fixture text (Fun-ASR fp32): ja CER / en WER / zh CER | port **2.3 % / 0.9 % / 0.0 %** · the oracle's own audio 6.5 / 0.0 / 0.0 |
+| speaker cosine to the reference clip (WavLM-Base-Plus-SV), 4 clone fixtures | port mean 0.855 / min 0.561 · oracle 0.854 / 0.582 |
+
+A miss is fp16 GPU or int8 arithmetic moving a draw that sat within a hair of the runner-up; under sampling, one flip
+changes every later frame, so end-to-end identity is not the bar — the ASR and speaker rows are. They are eight to
+fourteen sentences per language, our own measurement with one normalizer: the port's speech is as intelligible and as
+speaker-faithful as the publisher's fp32 output on these fixtures, not more.
+
+## Speed
+
+Measured with [`apps/Audio8Gate`](https://github.com/john-rocky/coreai-model-zoo/tree/main/apps/Audio8Gate) (the kit's `Audio8TTS` in a Release build, the 18 fixtures,
+78 s of audio; medians; 2026-09-28) **while another model conversion ran on the same Mac (load average 5–7)**. Both
+assets are JIT: the first load specializes them, later loads read the cache.
+
+| | frame (slow step + sampling + fast AR) | codec (160-frame window) | time to first audio (32-frame chunk) | RTF median / p90 | first load (cold) → later loads | footprint |
+|---|---|---|---|---|---|---|
+| M4 Max (GPU, macOS 27 26A428, GPU lock held) | 36–40 ms | ~0.7 s per utterance | 1.5 s | **1.05** / 1.09 (bench on one 5 s sentence: 0.99) | 5.4 s (cache 0 → 1.27 GB) → 0.8 s | 0.85–0.95 GB |
+| iPhone 18 Pro | not yet measured (the device was held by another lane on 2026-09-28) | | | | | |
+
+A 46 ms frame costs ~36 ms of engine time, so the Mac synthesizes at about real time and streams the first 1.5 s of
+audio after 1.5 s. The per-frame cost is the raw runtime path's per-layer dispatch (a 512-slot cache, AOT compilation
+and Apple's composite RMSNorm/RoPE change it by 0–10 %); Apple's pipelined engine drives a same-sized Qwen3-0.6B at
+2.8 ms per token, and moving the slow step onto it is the lever for a several-times faster frame — a separate round.
+
+## Use it
+
+CoreAIKit `Audio8TTS` (catalog enrollment pending):
+
+```swift
+import CoreAIKit
+
+let tts = try await Audio8TTS(paths: .standard(root: modelDir))      // dualar + codec .aimodel, tokenizer/
+let audio = try await tts.synthesize("明日の午後、駅前の喫茶店で待ち合わせましょう。")   // 44.1 kHz mono [Float]
+// streaming: tts.synthesizeStreaming(text) { chunk in play(chunk) }   ~1.5 s chunks
+// voice cloning: Audio8Voice(referenceText:codes:) from register_voice.py / the encoder graph
+let cloned = try await tts.synthesize("Turn left at the second traffic light.", voice: voice)
+```
+
+## ⬇️ Bundle
+
+**[mlboydaisuke/Audio8-TTS-Preview-0.6b-CoreAI](https://huggingface.co/mlboydaisuke/Audio8-TTS-Preview-0.6b-CoreAI)** —
+`audio8_dualar_int8_cl2048_w32.aimodel/` (876 MB) + `audio8_codec_decoder_fp16_t160.aimodel/` (261 MB) +
+`audio8_codec_encoder_fp16w32_t216.aimodel/` (416 MB, voice registration only) + `tokenizer/`; one subtree for macOS and
+iOS (JIT `.aimodel`s). Apache-2.0: LICENSE + NOTICE from the publisher's repository. Mirror:
+`coreai-community/Audio8-TTS-Preview-0.6b-CoreAI`.
+
+Convert yourself — [`conversion/audio8_tts/`](https://github.com/john-rocky/coreai-model-zoo/tree/main/conversion/audio8_tts): `oracle_audio8.py` → `parity_audio8.py` →
+`export_audio8_frame.py --mode int8`, `export_audio8.py --part codec`, `audio8_encoder.py --frames 216 --dtype fp16w32`,
+gated by `gate_audio8_frame.py` (recipe: [`recipe.toml`](https://github.com/john-rocky/coreai-model-zoo/blob/main/models/audio8-tts/recipe.toml)). Generated speech can be misused for
+impersonation; the publisher asks for consent before cloning a voice and disclosure of synthetic audio, and so does
+this port.

audio8_codec_decoder_fp16_t160.aimodel/main.hash

@@ -0,0 +1 @@
+�#���1U��8�l~wz�ާ:ҼG���zӱi

audio8_codec_decoder_fp16_t160.aimodel/main.mlirb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9d23fad5cf3155adfd38d76c7e777a8012dea73ad2bc47ef069f1dc67ad3b169
+size 261101172

audio8_codec_decoder_fp16_t160.aimodel/metadata.json

@@ -0,0 +1,6 @@
+{
+  "producer" : "coreai-core 1.0.0b2",
+  "license" : "apache-2.0",
+  "creationDate" : "20260927T212440Z",
+  "assetVersion" : "2.0"
+}

audio8_codec_encoder_fp16_t216.aimodel/main.mlirb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9c77708867a6ac00aec8ab5b8ed5999f926f4f081b2a5413d29fce81cda97b48
+size 416030239

audio8_codec_encoder_fp16_t216.aimodel/metadata.json

@@ -0,0 +1,6 @@
+{
+  "creationDate" : "20260927T222131Z",
+  "assetVersion" : "2.0",
+  "license" : "apache-2.0",
+  "producer" : "coreai-core 1.0.0b2"
+}

audio8_dualar_int8_cl2048_w32.aimodel/main.hash

@@ -0,0 +1 @@
+k0+�i��=���a�k�_�oҤO�KY(�6 

audio8_dualar_int8_cl2048_w32.aimodel/main.mlirb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6b302bf769aaee3dbe8ae78f61a36bd95fee9183dd6fd2a44fb04b5928993620
+size 876260029

audio8_dualar_int8_cl2048_w32.aimodel/metadata.json

@@ -0,0 +1,6 @@
+{
+  "creationDate" : "20260927T220903Z",
+  "producer" : "coreai-core 1.0.0b2",
+  "license" : "apache-2.0",
+  "assetVersion" : "2.0"
+}

config.json

@@ -0,0 +1,56 @@
+{
+  "architectures": [
+    "ArkttsModel"
+  ],
+  "attention_o_bias": false,
+  "attention_qk_norm": false,
+  "attention_qkv_bias": true,
+  "auto_map": {
+    "AutoConfig": "configuration_arktts.ArkttsConfig",
+    "AutoModel": "modeling_arktts.ArkttsModel",
+    "AutoModelForCausalLM": "modeling_arktts.ArkttsModel"
+  },
+  "codebook_size": 4096,
+  "codec_filename": "codec.pth",
+  "codec_frame_size": 2048,
+  "codec_post_intermediate_size": 1216,
+  "codec_post_n_head": 16,
+  "codec_post_n_layer": 8,
+  "codec_post_n_local_heads": 8,
+  "codec_sample_rate": 44100,
+  "dim": 896,
+  "dropout": 0.0,
+  "dtype": "bfloat16",
+  "eos_token_id": 151645,
+  "fast_attention_o_bias": false,
+  "fast_attention_qk_norm": false,
+  "fast_attention_qkv_bias": false,
+  "fast_dim": 896,
+  "fast_head_dim": 64,
+  "fast_intermediate_size": 4864,
+  "fast_n_head": 14,
+  "fast_n_local_heads": 2,
+  "head_dim": 64,
+  "initializer_range": 0.02,
+  "intermediate_size": 4864,
+  "max_seq_len": 2048,
+  "model_type": "arktts",
+  "n_fast_layer": 4,
+  "n_head": 14,
+  "n_layer": 24,
+  "n_local_heads": 2,
+  "norm_eps": 1e-06,
+  "norm_fastlayer_input": true,
+  "num_codebooks": 10,
+  "pad_token_id": 151643,
+  "ras_temperature": 1.0,
+  "ras_top_p": 0.9,
+  "ras_window_size": 10,
+  "rope_base": 1000000,
+  "semantic_begin_id": 151678,
+  "semantic_end_id": 155773,
+  "tie_word_embeddings": true,
+  "transformers_version": "4.57.5",
+  "use_gradient_checkpointing": false,
+  "vocab_size": 155776
+}

metadata.json

@@ -0,0 +1,187 @@
+{
+ "metadata_version": "0.2",
+ "kind": "tts",
+ "name": "Audio8-TTS-Preview-0.6b-CoreAI",
+ "source": {
+  "model_definition": "torch",
+  "hf_model_id": "Edge0/Audio8-TTS-Preview-0.6b",
+  "hf_revision": "f07040f3d151f1ba0253bfb92cb2f5dd38b44594",
+  "model_safetensors_sha256": "62dcff0adf6c2535b3260467a7c1d482b556da57266c96a444518b76e140d2c3",
+  "codec_pth_sha256": "c310505aa11fe2f6cc63b8d3130dc7e77e73227774f5c62575769b1f47a8d048",
+  "license": "apache-2.0"
+ },
+ "audio": {
+  "sample_rate": 44100,
+  "frame_samples": 2048,
+  "frames_per_second": 21.533203125
+ },
+ "graphs": {
+  "dualar": {
+   "asset": "audio8_dualar_int8_cl2048_w32.aimodel",
+   "functions": {
+    "prefill": {
+     "inputs": {
+      "codes": "[1, 11, 32] int32 (row 0 ids, rows 1-10 codebooks; pad rows: 151643 / 0)",
+      "pos": "[1] int32 first position"
+     },
+     "outputs": {
+      "logits": "[32, 4097] float16 (semantic 0..4095 then eos)",
+      "hidden": "[32, 896] float16"
+     }
+    },
+    "frame": {
+     "inputs": {
+      "codes": "[1, 11, 1] int32 (the previous frame: semantic id, 10 codebooks)",
+      "pos": "[1] int32 (its position)",
+      "noise_slow": "[2, 4097] float32 uniform draws (normal branch, RAS-high branch)",
+      "window": "[10] int32 RAS window (-1 = none)",
+      "noise_fast": "[9, 4096] float32 uniform draws",
+      "forced": "[11] int32 (teacher forcing)",
+      "use_forced": "[1] float32 0/1"
+     },
+     "outputs": {
+      "semantic": "[1] int32 token id (151645 = eos)",
+      "codes": "[10] int32 codebooks 0..9",
+      "logits": "[4097] float16",
+      "hidden": "[896] float16",
+      "fast_logits": "[9, 4096] float16",
+      "sampled_semantic": "[1] int32",
+      "sampled_codes": "[10] int32"
+     }
+    },
+    "first_frame": {
+     "inputs": {
+      "logits": "[4097] float16 (the prefill's last row)",
+      "hidden": "[896] float16",
+      "noise_slow": "…",
+      "window": "…",
+      "noise_fast": "…",
+      "forced": "…",
+      "use_forced": "…"
+     },
+     "outputs": "as frame"
+    }
+   },
+   "state": {
+    "k_cache": "[24, 1, 2, 2048, 64] float16",
+    "v_cache": "[24, 1, 2, 2048, 64] float16 (prefill and frame)"
+   },
+   "compression": "slow AR: int8 weight-only, symmetric with clipping, per-block-32 (input axis) on the 24 layers' linears; embeddings, codebook embeddings, the 4,097-row head, norms, the fast AR (4 layers) float16"
+  },
+  "codec_decoder": {
+   "asset": "audio8_codec_decoder_fp16_t160.aimodel",
+   "functions": {
+    "main": {
+     "inputs": {
+      "codes": "[1, 10, 160] int32 (right-pad with 0; codebook 0 < 4096, codebooks 1-9 < 1024, clamped in-graph)"
+     },
+     "outputs": {
+      "wav": "[1, 327680] float16, 44.1 kHz"
+     }
+    }
+   },
+   "note": "every op is causal: a frame's samples depend on at most the 127 frames before it (the codec transformer's window) — decode a stream in 160-frame windows and keep the last 32",
+   "compression": null
+  },
+  "codec_encoder": {
+   "asset": "audio8_codec_encoder_fp16_t216.aimodel",
+   "functions": {
+    "main": {
+     "inputs": {
+      "audio": "[1, 1, 442368] float32 mono 44.1 kHz, right-pad with 0 (216 frames = 10.03 s)"
+     },
+     "outputs": {
+      "codes": "[1, 10, 216] int32; keep the first ceil(samples / 2048) frames"
+     }
+    }
+   },
+   "note": "voice registration (zero-shot cloning): codes + the exact transcript make an Audio8Voice",
+   "compression": null
+  }
+ },
+ "sampling": {
+  "top_k": 50,
+  "top_p": 0.9,
+  "temperature": 0.7,
+  "max_new_tokens": 512,
+  "ras": {
+   "window": 10,
+   "top_p": 0.9,
+   "temperature": 1.0
+  },
+  "rule": "argmax(softmax(processed) / -log(u)) with u ~ U(0,1); processed = top-k/top-p mask then / temperature; a semantic id repeated within the 10-frame window is replaced by the (0.9, 1.0) draw",
+  "eos": 151645,
+  "semantic_begin": 151678,
+  "semantic_end": 155773,
+  "pad": 151643
+ },
+ "prompt": {
+  "spec": "conversion/audio8_tts/prompt.py (segments encoded one at a time with the tokenizer, no special tokens added)",
+  "tokenizer": {
+   "files": "tokenizer/",
+   "class_retag": "PreTrainedTokenizerFast -> Qwen2Tokenizer for swift-transformers"
+  }
+ },
+ "languages": [
+  "yue",
+  "zh",
+  "nl",
+  "en",
+  "fr",
+  "de",
+  "it",
+  "ja",
+  "ko",
+  "pl",
+  "es"
+ ],
+ "host": {
+  "swift": "CoreAIKit Audio8TTS (Sources/CoreAIKit/Audio8TTS)",
+  "python_gate": "conversion/audio8_tts/gate_audio8.py"
+ },
+ "files": {
+  "audio8_dualar_int8_cl2048_w32.aimodel/main.hash": {
+   "sha256": "73ba75f6998692b3c361a27939f2a2db36c3735236eed96f9ea955be6c70c42a",
+   "bytes": 32
+  },
+  "audio8_dualar_int8_cl2048_w32.aimodel/main.mlirb": {
+   "sha256": "6b302bf769aaee3dbe8ae78f61a36bd95fee9183dd6fd2a44fb04b5928993620",
+   "bytes": 876260029
+  },
+  "audio8_dualar_int8_cl2048_w32.aimodel/metadata.json": {
+   "sha256": "11dd0938dfd88fea0a7f862f27297a43f31d7886ce1754c3d57b552ee8100387",
+   "bytes": 133
+  },
+  "audio8_codec_decoder_fp16_t160.aimodel/main.hash": {
+   "sha256": "41cd54a438da85d253f6a733fbb806605a38d54558c8d51ada7babe251a6dccd",
+   "bytes": 32
+  },
+  "audio8_codec_decoder_fp16_t160.aimodel/main.mlirb": {
+   "sha256": "9d23fad5cf3155adfd38d76c7e777a8012dea73ad2bc47ef069f1dc67ad3b169",
+   "bytes": 261101172
+  },
+  "audio8_codec_decoder_fp16_t160.aimodel/metadata.json": {
+   "sha256": "0221caad005eed1da346fda823fe97cfbce4cbcd8a262b3dc982279d15505f4a",
+   "bytes": 133
+  },
+  "audio8_codec_encoder_fp16_t216.aimodel/main.hash": {
+   "sha256": "c24c6679b723b2eaccede267c5bbb88a77cf6b53230321339364a0d1db9c3241",
+   "bytes": 32
+  },
+  "audio8_codec_encoder_fp16_t216.aimodel/main.mlirb": {
+   "sha256": "9c77708867a6ac00aec8ab5b8ed5999f926f4f081b2a5413d29fce81cda97b48",
+   "bytes": 416030239
+  },
+  "audio8_codec_encoder_fp16_t216.aimodel/metadata.json": {
+   "sha256": "393db855a49f603faf5e09edd9334b47ddc621a8b1a78464a966d07e1714f681",
+   "bytes": 133
+  }
+ },
+ "compilation": {
+  "date": "2026-09-27T23:09:59.812117+00:00",
+  "targets": [
+   "macos",
+   "ios (jit)"
+  ]
+ }
+}

tokenizer/tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f24e08099d45a8adf3f52f5f0b03276e433bb9d689bb15fcbcc48ce58744588b
+size 12217872