| { |
| "schema_version": 1, |
| "repo_type": "model", |
| "name": "Inflect-Nano GGUF Quantized Weights", |
| "description": "English-only GGUF quantized acoustic and vocoder weights for the InflectNanoTTS C++/GGML runtime. This repository contains quantized derivatives of the base model, not the base training checkpoint.", |
| "format": "GGUF", |
| "runtime": { |
| "name": "InflectNanoTTS", |
| "library_name": "ggml", |
| "task": "text-to-speech", |
| "sample_rate_hz": 24000 |
| }, |
| "language": [ |
| "en" |
| ], |
| "base_model": { |
| "id": "owensong/Inflect-Nano-v1", |
| "relation": "quantized" |
| }, |
| "license": "apache-2.0", |
| "components": { |
| "acoustic": { |
| "description": "Text/prosody-to-mel acoustic model weights.", |
| "files": [ |
| { |
| "quantization": "f16", |
| "path": "acoustic/inflect_acoustic_f16.gguf", |
| "bytes": 6995552, |
| "sha256": "3ac767c2ea538b4b7832cee5f02d711b0b8f22fb7ac4e775f82f0b8aa9d25434" |
| }, |
| { |
| "quantization": "q2_k", |
| "path": "acoustic/inflect_acoustic_q2_k.gguf", |
| "bytes": 2164704, |
| "sha256": "cc0ce00e6ec39c29335a8f3bfee53e26728e65bfff4f71bef71951648b5015ac" |
| }, |
| { |
| "quantization": "q3_k", |
| "path": "acoustic/inflect_acoustic_q3_k.gguf", |
| "bytes": 2611616, |
| "sha256": "2823f42f4f5869c7d7a9a810036175bba1ecda4670cfb29c4d8493d90fd3ad11" |
| }, |
| { |
| "quantization": "q3_k_dynamic", |
| "path": "acoustic/inflect_acoustic_q3_k_dynamic.gguf", |
| "bytes": 2399488, |
| "sha256": "6b700bb9063d3f7792378c5d075d8a26eb00ba2b97ec2ddf31e94bf3ce786bf2", |
| "notes": "Mixed/dynamic quantization recipe; see model card for compatibility caveats." |
| }, |
| { |
| "quantization": "q4_0", |
| "path": "acoustic/inflect_acoustic_q4_0.gguf", |
| "bytes": 2646240, |
| "sha256": "d2cdbb4d7cd7eca5aa266571ea5777c5a2fda3144b02f30e750d2fd6c42202b6" |
| }, |
| { |
| "quantization": "q4_0_E", |
| "path": "acoustic/inflect_acoustic_q4_0_E.gguf", |
| "bytes": 2213824, |
| "sha256": "0201c26fe154520d6f535c81d255c505ee3fd1bdd3c6a7a9295fa790e830df2b", |
| "notes": "Experimental optimized quant with fp16 biases and the last layer quantized to q4_0 for ESP32 8MB PSRAM targets." |
| }, |
| { |
| "quantization": "q4_k", |
| "path": "acoustic/inflect_acoustic_q4_k.gguf", |
| "bytes": 2807712, |
| "sha256": "2c34152a397a59e07fa76ffab352f9970bea4b0279f42022a440ae66203d12fa" |
| }, |
| { |
| "quantization": "q5_0", |
| "path": "acoustic/inflect_acoustic_q5_0.gguf", |
| "bytes": 3073760, |
| "sha256": "dd1ff88f03d15ba50301b7a9eaccf219c15a6fdb269daa5e0822125d827b4456" |
| }, |
| { |
| "quantization": "q5_k", |
| "path": "acoustic/inflect_acoustic_q5_k.gguf", |
| "bytes": 3744288, |
| "sha256": "06143d20d675898a41189284db63b6a28efefb92d79a4dc475a9b12d16b40e75" |
| }, |
| { |
| "quantization": "q6_k", |
| "path": "acoustic/inflect_acoustic_q6_k.gguf", |
| "bytes": 4328736, |
| "sha256": "fc99259b14eed08da76948127685ae51c489dfd26fd938e681edefac7bacff4f" |
| }, |
| { |
| "quantization": "q8_0", |
| "path": "acoustic/inflect_acoustic_q8_0.gguf", |
| "bytes": 4356288, |
| "sha256": "0d12d38c3f768b48b2e848c6861c0a8d4752e1b2f3cf88a92660be7d429a3ed8" |
| } |
| ] |
| }, |
| "vocoder": { |
| "description": "Mel-to-waveform vocoder weights.", |
| "files": [ |
| { |
| "quantization": "f16", |
| "path": "vocoder/inflect_vocoder_f16.gguf", |
| "bytes": 2357696, |
| "sha256": "d0d7a9880f557f1550cb4aab9c90dcd7bbae52cc534e33f2c3a2a5a286483177" |
| }, |
| { |
| "quantization": "q4_0", |
| "path": "vocoder/inflect_vocoder_q4_0.gguf", |
| "bytes": 1010592, |
| "sha256": "dd58127273a6a1262f97079f87e11eed93e06b36ea9141eea815c434c0f7cada" |
| }, |
| { |
| "quantization": "q4_0_E", |
| "path": "vocoder/inflect_vocoder_q4_0_E.gguf", |
| "bytes": 702976, |
| "sha256": "8adbb4fbd1bd7fffdd305248ed4a7dcd50cc281b1be026f82618eb793a48868e", |
| "notes": "Experimental optimized quant with fp16 biases and the last layer quantized to q4_0 for ESP32 8MB PSRAM targets." |
| }, |
| { |
| "quantization": "q5_k", |
| "path": "vocoder/inflect_vocoder_q5_k.gguf", |
| "bytes": 1306592, |
| "sha256": "0f9026dcd60f7f0fb4e4caba6198a9c57332fe69dd1d53199c6c52547717d7e4" |
| }, |
| { |
| "quantization": "q8_0", |
| "path": "vocoder/inflect_vocoder_q8_0.gguf", |
| "bytes": 1498688, |
| "sha256": "add2489c11b896a0056bed889157898f0d53a997430ccfb65ad1c40fee000c76" |
| } |
| ] |
| } |
| }, |
| "support_files": [ |
| { |
| "name": "cmudict", |
| "path": "cmudict.bin", |
| "bytes": 2460494, |
| "sha256": "67019e21cd01cb600709f11d75c03d16fa219169187d116cb98c51ba5f4112e0", |
| "notes": "Compact binary pronunciation dictionary used by the text frontend." |
| }, |
| { |
| "name": "cmudict_index", |
| "path": "cmudict.idx", |
| "bytes": 8942, |
| "sha256": "49e9cf6d176312c9c7a4920ce73f9acf7a248c357fb049f986041ee8867969c1", |
| "notes": "Index file for cmudict.bin that makes dictionary initialization faster." |
| } |
| ], |
| "recommended_pairings": [ |
| { |
| "name": "esp32_8mb_psram_experimental", |
| "acoustic": "acoustic/inflect_acoustic_q4_0_E.gguf", |
| "vocoder": "vocoder/inflect_vocoder_q4_0_E.gguf", |
| "notes": "Experimental optimized q4_0 pair with fp16 biases and last-layer q4_0 quantization, sized for ESP32 8MB PSRAM deployments." |
| }, |
| { |
| "name": "smallest_experimental", |
| "acoustic": "acoustic/inflect_acoustic_q2_k.gguf", |
| "vocoder": "vocoder/inflect_vocoder_q4_0.gguf", |
| "notes": "Smallest listed pairing; quality should be verified per release." |
| }, |
| { |
| "name": "balanced", |
| "acoustic": "acoustic/inflect_acoustic_q4_0.gguf", |
| "vocoder": "vocoder/inflect_vocoder_q4_0.gguf", |
| "notes": "Compact pairing for size-sensitive deployments." |
| }, |
| { |
| "name": "higher_quality_quantized", |
| "acoustic": "acoustic/inflect_acoustic_q6_k.gguf", |
| "vocoder": "vocoder/inflect_vocoder_q5_k.gguf", |
| "notes": "Larger quantized pairing with less aggressive compression." |
| }, |
| { |
| "name": "reference_quantized_runtime", |
| "acoustic": "acoustic/inflect_acoustic_f16.gguf", |
| "vocoder": "vocoder/inflect_vocoder_f16.gguf", |
| "notes": "F16 GGUF runtime reference; still a converted GGUF derivative, not the original checkpoint." |
| } |
| ], |
| "known_limitations": [ |
| "English text only.", |
| "These are quantized/converter outputs and are not suitable as base training checkpoints.", |
| "Quality varies by acoustic/vocoder quantization pairing; run listening tests before production use.", |
| "Some experimental low-bit variants may trade intelligibility or prosody quality for file size." |
| ] |
| } |
|
|
