Fabio
c36510c21e
Some checks failed
Linux Build / Linux cpu (push) Has been cancelled
Linux Build / Linux cpu-clang (push) Has been cancelled
Linux Build / Linux vulkan (push) Has been cancelled
mac-build / macOS CPU (push) Has been cancelled
Nix Build / Nix (cpu) (push) Has been cancelled
Nix Build / Nix (vulkan) (push) Has been cancelled
windows-build / Windows CPU (push) Has been cancelled
340 lines
9.3 KiB
JSON
340 lines
9.3 KiB
JSON
{
|
|
"schema_version": 1,
|
|
"family": "gradium_phonon",
|
|
"display_name": "Gradium Phonon",
|
|
"description": "Gradium's Phonon text-to-speech: a small flow-matching language model with a Mimi decoder, a descendant of Kyutai's pocket-tts, for real-time local synthesis on CPU and GPU with prepared voices.",
|
|
"category": "tts",
|
|
"status": "experimental",
|
|
"tasks": [
|
|
"tts"
|
|
],
|
|
"modes": [
|
|
"offline",
|
|
"streaming"
|
|
],
|
|
"languages": [
|
|
"en"
|
|
],
|
|
"runtime": {
|
|
"tags": [
|
|
"gguf"
|
|
]
|
|
},
|
|
"capabilities": {
|
|
"tts": [
|
|
"built_in_voices",
|
|
"long_form"
|
|
]
|
|
},
|
|
"options": {
|
|
"request": [
|
|
{
|
|
"name": "frames_after_eos",
|
|
"type": "int",
|
|
"description": "Frames kept after EOS, counting the EOS frame; -1 keeps 4 for texts of up to four words and 2 otherwise, as the reference does.",
|
|
"required": false,
|
|
"min": -1,
|
|
"default": -1
|
|
},
|
|
{
|
|
"name": "max_frames",
|
|
"type": "int",
|
|
"description": "Maximum generated audio frames; zero estimates the cap from text length. Captured during session preparation.",
|
|
"required": false,
|
|
"min": 0,
|
|
"default": 0
|
|
},
|
|
{
|
|
"name": "text_chunk_tokens",
|
|
"type": "int",
|
|
"description": "Tokenizer tokens per text chunk; must be positive, default 50. Set during preparation and overridable per request.",
|
|
"required": false,
|
|
"min": 1,
|
|
"default": 50
|
|
},
|
|
{
|
|
"name": "temperature",
|
|
"type": "float",
|
|
"description": "Positive flow noise temperature. Captured during session preparation.",
|
|
"required": false,
|
|
"default": 0.3
|
|
},
|
|
{
|
|
"name": "seed",
|
|
"type": "int",
|
|
"description": "Noise seed; omitted chooses a random seed. Captured during session preparation.",
|
|
"required": false,
|
|
"min": 0,
|
|
"max": 4294967295
|
|
},
|
|
{
|
|
"name": "noise_clamp",
|
|
"type": "float",
|
|
"description": "Noise clamp; a negative value disables clamping. Captured during session preparation.",
|
|
"required": false,
|
|
"default": -1
|
|
},
|
|
{
|
|
"name": "eos_threshold",
|
|
"type": "float",
|
|
"description": "EOS detection threshold on the EOS logit. Captured during session preparation.",
|
|
"required": false,
|
|
"default": 4
|
|
},
|
|
{
|
|
"name": "noise_file",
|
|
"type": "path",
|
|
"description": "Precomputed noise schedule, raw float32 [frames, 32], used as given. Captured during session preparation.",
|
|
"required": false
|
|
},
|
|
{
|
|
"name": "voice_embedding_file",
|
|
"type": "path",
|
|
"description": "Voice file: a speaker_wavs latent file such as those of gradium/phonon-voices, or an exported voice state. Captured during session preparation.",
|
|
"required": false
|
|
},
|
|
{
|
|
"name": "text_chunk_size",
|
|
"type": "int",
|
|
"description": "Optional per-request character chunk size; otherwise use sentence/token budgeting.",
|
|
"required": false
|
|
},
|
|
{
|
|
"name": "export_voice_state_file",
|
|
"type": "path",
|
|
"description": "Export the prepared voice state during session preparation.",
|
|
"required": false
|
|
}
|
|
],
|
|
"session": [
|
|
{
|
|
"name": "voice_state_cache_slots",
|
|
"type": "int",
|
|
"description": "Cached prepared voice states; zero disables reuse.",
|
|
"required": false,
|
|
"min": 0,
|
|
"default": 4
|
|
},
|
|
{
|
|
"name": "offline_graph_capacity_mode",
|
|
"type": "enum",
|
|
"description": "Offline graph capacity policy; CPU/Metal default tiered, other backends double.",
|
|
"required": false,
|
|
"values": [
|
|
"fixed",
|
|
"double",
|
|
"tiered",
|
|
"grow"
|
|
]
|
|
},
|
|
{
|
|
"name": "prompt_graph_capacity",
|
|
"type": "int",
|
|
"description": "Optional reserved prompt capacity.",
|
|
"required": false,
|
|
"min": 1
|
|
},
|
|
{
|
|
"name": "generation_graph_capacity",
|
|
"type": "int",
|
|
"description": "Optional reserved generation capacity.",
|
|
"required": false,
|
|
"min": 1
|
|
},
|
|
{
|
|
"name": "weight_type",
|
|
"type": "enum",
|
|
"description": "Shared matmul weight storage; component matmul settings inherit it. Convolution storage remains independently native unless overridden.",
|
|
"required": false,
|
|
"preset": "weight_type_full"
|
|
},
|
|
{
|
|
"name": "matmul_weight_type",
|
|
"type": "enum",
|
|
"description": "Weight storage type; component-specific settings inherit weight_type.",
|
|
"required": false,
|
|
"preset": "weight_type_full"
|
|
},
|
|
{
|
|
"name": "conv_weight_type",
|
|
"type": "enum",
|
|
"description": "Convolution weight storage.",
|
|
"required": false,
|
|
"preset": "weight_type_conv",
|
|
"default": "native"
|
|
},
|
|
{
|
|
"name": "weight_context_mb",
|
|
"type": "int",
|
|
"description": "Context or graph arena size in MiB.",
|
|
"required": false,
|
|
"min": 1,
|
|
"default": 64
|
|
},
|
|
{
|
|
"name": "flow_weight_context_mb",
|
|
"type": "int",
|
|
"description": "Flow weight context in MiB; inherits gradium_phonon.weight_context_mb when omitted.",
|
|
"required": false,
|
|
"min": 1
|
|
},
|
|
{
|
|
"name": "mimi_decoder_weight_context_mb",
|
|
"type": "int",
|
|
"description": "Mimi decoder weight context in MiB; inherits gradium_phonon.weight_context_mb when omitted.",
|
|
"required": false,
|
|
"min": 1
|
|
},
|
|
{
|
|
"name": "flow_weights_view_context_mb",
|
|
"type": "int",
|
|
"description": "Context or graph arena size in MiB.",
|
|
"required": false,
|
|
"min": 1,
|
|
"default": 256
|
|
},
|
|
{
|
|
"name": "flow_step_graph_arena_mb",
|
|
"type": "int",
|
|
"description": "Flow-step host graph-metadata arena size in MiB, not device workspace.",
|
|
"required": false,
|
|
"min": 1,
|
|
"default": 256
|
|
},
|
|
{
|
|
"name": "mimi_conv_graph_arena_mb",
|
|
"type": "int",
|
|
"description": "Mimi convolution host graph-metadata arena size in MiB, not device workspace.",
|
|
"required": false,
|
|
"min": 1,
|
|
"default": 32
|
|
},
|
|
{
|
|
"name": "mimi_transformer_graph_arena_mb",
|
|
"type": "int",
|
|
"description": "Mimi transformer host graph-metadata arena size in MiB, not device workspace.",
|
|
"required": false,
|
|
"min": 1,
|
|
"default": 96
|
|
},
|
|
{
|
|
"name": "mimi_tail_graph_arena_mb",
|
|
"type": "int",
|
|
"description": "Mimi tail host graph-metadata arena size in MiB, not device workspace.",
|
|
"required": false,
|
|
"min": 1,
|
|
"default": 512
|
|
},
|
|
{
|
|
"name": "mimi_full_chunk_frames",
|
|
"type": "int",
|
|
"description": "Full-sequence decoder chunk frames.",
|
|
"required": false,
|
|
"min": 1,
|
|
"default": 90
|
|
},
|
|
{
|
|
"name": "mimi_stage2_chunk_frames",
|
|
"type": "int",
|
|
"description": "Second-stage decoder chunk frames.",
|
|
"required": false,
|
|
"min": 1,
|
|
"default": 900
|
|
},
|
|
{
|
|
"name": "use_full_mimi",
|
|
"type": "bool",
|
|
"description": "Use full-sequence Mimi decoding.",
|
|
"required": false,
|
|
"default": true
|
|
}
|
|
],
|
|
"load": []
|
|
},
|
|
"package_defaults": {
|
|
"download": {
|
|
"kind": "huggingface_snapshot",
|
|
"repo": "gradium/phonon-31m-en-audiocpp",
|
|
"revision": "main",
|
|
"gated": false
|
|
}
|
|
},
|
|
"packages": [
|
|
{
|
|
"id": "gradium_phonon_31m_en_q8_0",
|
|
"display_name": "Phonon 31m English Q8_0 GGUF",
|
|
"default": true,
|
|
"format": "gguf",
|
|
"precision": "q8_0",
|
|
"target_directory": "phonon-31m-en",
|
|
"files": [
|
|
"phonon-31m-en-q8_0.gguf",
|
|
"default-voice.safetensors"
|
|
]
|
|
},
|
|
{
|
|
"id": "gradium_phonon_31m_en_safetensors",
|
|
"display_name": "Phonon 31m English Safetensors",
|
|
"format": "safetensors",
|
|
"precision": "native",
|
|
"target_directory": "phonon-31m-en-safetensors",
|
|
"files": [
|
|
"config.json",
|
|
"tokenizer.model",
|
|
"model.safetensors",
|
|
"default-voice.safetensors"
|
|
],
|
|
"download": {
|
|
"kind": "huggingface_snapshot",
|
|
"repo": "gradium/phonon-31m-en",
|
|
"revision": "52433c46975bfa13144c56204c9f4ee1c5b148aa",
|
|
"gated": false
|
|
}
|
|
}
|
|
],
|
|
"dependencies": [],
|
|
"ui": {
|
|
"recommended_package": "gradium_phonon_31m_en_q8_0",
|
|
"default_voice": "default",
|
|
"builtin_voices": [
|
|
"default"
|
|
],
|
|
"tags": [
|
|
"TTS",
|
|
"GGUF"
|
|
],
|
|
"docs": [
|
|
"docs/models/gradium_phonon.md",
|
|
"docs/gguf.md"
|
|
]
|
|
},
|
|
"sources": [
|
|
{
|
|
"format": "gguf",
|
|
"roots": {
|
|
"model": ".",
|
|
"weights": "$gguf"
|
|
},
|
|
"files": {
|
|
"config": "model:config.json",
|
|
"tokenizer": "model:tokenizer.model"
|
|
},
|
|
"tensors": {
|
|
"weights": "weights:"
|
|
}
|
|
},
|
|
{
|
|
"format": "safetensors",
|
|
"roots": {
|
|
"model": "."
|
|
},
|
|
"files": {
|
|
"config": "model:config.json",
|
|
"tokenizer": "model:tokenizer.model"
|
|
},
|
|
"tensors": {
|
|
"weights": "model:model.safetensors"
|
|
}
|
|
}
|
|
]
|
|
}
|