Initial Chatterbox-TTS GGUF release

.gitattributes

@@ -33,3 +33,14 @@ 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
+codec-f16.gguf filter=lfs diff=lfs merge=lfs -text
+codec-f32.gguf filter=lfs diff=lfs merge=lfs -text
+codec-q4_k_m.gguf filter=lfs diff=lfs merge=lfs -text
+codec-q5_k_m.gguf filter=lfs diff=lfs merge=lfs -text
+codec-q8_0.gguf filter=lfs diff=lfs merge=lfs -text
+s3t.gguf filter=lfs diff=lfs merge=lfs -text
+t3-f16.gguf filter=lfs diff=lfs merge=lfs -text
+t3-f32.gguf filter=lfs diff=lfs merge=lfs -text
+t3-q4_k_m.gguf filter=lfs diff=lfs merge=lfs -text
+t3-q5_k_m.gguf filter=lfs diff=lfs merge=lfs -text
+t3-q8_0.gguf filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,84 @@
+---
+license: mit
+language:
+- en
+tags:
+- gguf
+- tts
+- chatterbox
+- text-to-speech
+base_model: ResembleAI/chatterbox
+---
+
+# Chatterbox-TTS GGUF
+
+GGUF conversions of [Resemble AI's Chatterbox-TTS](https://huggingface.co/ResembleAI/chatterbox).
+
+The Chatterbox stack is split into two halves so each can run inside the
+runtime that owns it:
+
+- **LLM-part (T3)** — text + conditioning to discrete speech token IDs,
+  run by a `chatterbox_t3`-aware llama.cpp build.
+- **Codec-part (S3G / S3T)** — discrete speech tokens to/from 24 kHz
+  PCM, run by [codec.cpp](https://github.com/mybigday/codec.cpp).
+
+Only payload speech token IDs (`0..6560`) cross the boundary between the
+two runtimes.
+
+## Files
+
+### LLM-part (T3, `t3_cfg.safetensors`)
+
+`t3-<quant>.gguf` — 30-layer, hidden 1024, MLP 4096 Llama-style backbone
+plus T3-specific embeddings, heads, positional embeddings, and the
+conditioning encoder (perceiver / speaker / emotion). Architecture:
+`chatterbox_t3`.
+
+| File              | Size    |
+| ----------------- | ------- |
+| `t3-f32.gguf`     | 2031 MB |
+| `t3-f16.gguf`     | 1016 MB |
+| `t3-q8_0.gguf`    |  540 MB |
+| `t3-q5_k_m.gguf`  |  349 MB |
+| `t3-q4_k_m.gguf`  |  286 MB |
+
+### Codec-part (S3G, `s3gen.safetensors`)
+
+`codec[-<quant>].gguf` — Chatterbox S3Gen flow-matching decoder + HiFi-GAN
+vocoder. Architecture: `chatterbox_s3g`. Built-in conditioning (`conds.pt`)
+is baked in.
+
+| File                | Size    |
+| ------------------- | ------- |
+| `codec-f32.gguf`    |  535 MB |
+| `codec-f16.gguf`    |  268 MB |
+| `codec-q8_0.gguf`   |  196 MB |
+| `codec-q5_k_m.gguf` |  167 MB |
+| `codec-q4_k_m.gguf` |  158 MB |
+
+### Codec-part (S3T, `s3gen.safetensors[tokenizer.*]`)
+
+`s3t.gguf` — S3 audio tokenizer (24 kHz mel front-end, 16 kHz token rate
+25 Hz, codebook 6561). Architecture: `chatterbox_s3t`. Only needed for
+voice-cloning / reference-token paths; F16 only because most weights are
+already small.
+
+## Boundary contract
+
+- token rate: 25 Hz
+- codebook size: 6561 (payload IDs `0..6560`)
+- `n_q = 1`
+- encode sample rate: 16 kHz, decode sample rate: 24 kHz
+- `start_speech_token = 6561`, `stop_speech_token = 6562`
+
+## Notes
+
+- Quantization (`Q8_0`, `Q5_K_M`, `Q4_K_M`) is applied to 2-D `.weight`
+  tensors whose last dimension is divisible by the quant block size (32
+  for Q8, 256 for K-quants); norms, biases, and small tensors stay F16.
+- Source weights: `t3_cfg.safetensors` (T3 base config) and
+  `s3gen.safetensors` from `ResembleAI/chatterbox`. Multilingual T3
+  variants (`t3_23lang`, `t3_mtl23ls_v2/v3`) are not converted here.
+- llama.cpp does not yet support the `chatterbox_t3` architecture
+  upstream; the T3 GGUFs are forward-compatible with that future runtime
+  and otherwise consumable by tooling that reads GGUF metadata/weights.

codec-f16.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8382d0cd5eddcdf454415ff6a804aa624dd480c80102d3693db9bd9e91e09050
+size 280878048

codec-f32.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:069c9ad0e42dea31090753bf54f50cae9dd38337f526da20ff8294ab265b9f94
+size 561480800

codec-q4_k_m.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0ada4835a7fda716307fedb23fba6fe795a55693abc8951f19a754377b07a454
+size 165200928

codec-q5_k_m.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9e0ab68a6c8ac9278c1c0b9b86ff4edacb7126330767fd6368f72ada07f08a2b
+size 175259808

codec-q8_0.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3992234e8d3071e8f79911900bc53a1a2a96aa1ac41496eb3d91131911896ac6
+size 205114848

s3t.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:26592ce171dd40bb54468a32dd9a3b697e15bfc23ebc8f8d218e34c3962e69c4
+size 247487280

t3-f16.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ba91576bc71065b8a9af656b03f7754dd847387f6e792080c9eb5215a151c929
+size 1064828992

t3-f32.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e551adae11ec294c2f75be6cf2b2450e633e0b73fe838c0ab0b56c3cd7dd8cd2
+size 2129639488

t3-q4_k_m.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:93180910778985f557e11c7da12e93dc3509fd3f414e80b23de6415866c215a8
+size 299645120

t3-q5_k_m.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3b164d7639bc1f016728a0bac3d0be68524e46a4b3e72c6efd688b3bc22b4653
+size 366182848

t3-q8_0.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4f898d5b403f29cb7c9008ef9f828106c156039228763eaa12edfe6a12df5f2c
+size 565796032