266 lines
6.5 KiB
JSON
266 lines
6.5 KiB
JSON
{
|
|
"family": "hviske_asr",
|
|
"schema_version": 1,
|
|
"display_name": "Hviske ASR",
|
|
"description": "Danish-optimized Conformer encoder-decoder ASR model fine-tuned from the Hviske v5 family, with selectable Cohere ASR language prompts.",
|
|
"category": "asr",
|
|
"status": "supported",
|
|
"tasks": [
|
|
"asr"
|
|
],
|
|
"modes": [
|
|
"offline"
|
|
],
|
|
"languages": [
|
|
"ar",
|
|
"da",
|
|
"de",
|
|
"el",
|
|
"en",
|
|
"es",
|
|
"fr",
|
|
"it",
|
|
"ja",
|
|
"ko",
|
|
"nl",
|
|
"pl",
|
|
"pt",
|
|
"vi",
|
|
"zh"
|
|
],
|
|
"capabilities": {},
|
|
"dependencies": [],
|
|
"options": {
|
|
"request": [
|
|
{
|
|
"name": "language",
|
|
"type": "string",
|
|
"description": "ASR language code; default da.",
|
|
"required": false,
|
|
"default": "da"
|
|
},
|
|
{
|
|
"name": "punctuation",
|
|
"type": "bool",
|
|
"description": "Enable or disable punctuation tokens in the decoder prompt; default true.",
|
|
"required": false,
|
|
"default": true
|
|
},
|
|
{
|
|
"name": "max_tokens",
|
|
"type": "int",
|
|
"description": "Maximum generated transcript tokens; defaults to the model config.",
|
|
"required": false,
|
|
"min": 1
|
|
},
|
|
{
|
|
"name": "num_beams",
|
|
"type": "int",
|
|
"description": "Beam-search beam count; default 1 uses greedy or sampling decode.",
|
|
"required": false,
|
|
"min": 1,
|
|
"default": 1
|
|
},
|
|
{
|
|
"name": "length_penalty",
|
|
"type": "float",
|
|
"description": "Beam-search length penalty; must be positive when set; default 1.0.",
|
|
"required": false,
|
|
"min": 0.000001,
|
|
"default": 1.0
|
|
},
|
|
{
|
|
"name": "do_sample",
|
|
"type": "bool",
|
|
"description": "Enable sampling instead of greedy decode when num_beams is 1; default false.",
|
|
"required": false,
|
|
"default": false
|
|
},
|
|
{
|
|
"name": "temperature",
|
|
"type": "float",
|
|
"description": "Decoder sampling temperature; must be positive when set; default 1.0.",
|
|
"required": false,
|
|
"min": 0.000001,
|
|
"default": 1.0
|
|
},
|
|
{
|
|
"name": "top_k",
|
|
"type": "int",
|
|
"description": "Top-k sampling limit; default 50, 0 disables top-k.",
|
|
"required": false,
|
|
"min": 0,
|
|
"default": 50
|
|
},
|
|
{
|
|
"name": "top_p",
|
|
"type": "float",
|
|
"description": "Nucleus sampling limit in (0, 1]; default 1.0.",
|
|
"required": false,
|
|
"min": 0.000001,
|
|
"max": 1.0,
|
|
"default": 1.0
|
|
},
|
|
{
|
|
"name": "seed",
|
|
"type": "int",
|
|
"description": "Decoder sampling seed; random if omitted.",
|
|
"required": false,
|
|
"min": 0
|
|
},
|
|
{
|
|
"name": "audio_chunk_mode",
|
|
"type": "enum",
|
|
"description": "Audio chunking mode; default auto uses quiet-energy splitting only when audio exceeds the model clip window.",
|
|
"values": [
|
|
"auto",
|
|
"fixed",
|
|
"quiet_energy",
|
|
"none"
|
|
],
|
|
"required": false,
|
|
"default": "auto"
|
|
},
|
|
{
|
|
"name": "audio_chunk_duration_sec",
|
|
"type": "float",
|
|
"description": "Maximum audio chunk duration in seconds; defaults to the model clip window.",
|
|
"required": false,
|
|
"min": 0.000001
|
|
}
|
|
],
|
|
"session": [
|
|
{
|
|
"name": "weight_type",
|
|
"type": "enum",
|
|
"description": "Matmul weight storage type; default native.",
|
|
"preset": "weight_type_full",
|
|
"required": false,
|
|
"default": "native"
|
|
},
|
|
{
|
|
"name": "conv_weight_type",
|
|
"type": "enum",
|
|
"description": "Convolution weight storage type; defaults to weight_type when set, otherwise native.",
|
|
"preset": "weight_type_conv",
|
|
"required": false
|
|
},
|
|
{
|
|
"name": "weight_context_mb",
|
|
"type": "int",
|
|
"description": "Weight descriptor context size in MiB; default 32.",
|
|
"required": false,
|
|
"min": 1,
|
|
"default": 32
|
|
},
|
|
{
|
|
"name": "encoder_graph_arena_mb",
|
|
"type": "int",
|
|
"description": "Encoder graph arena size in MiB; default 512.",
|
|
"required": false,
|
|
"min": 1,
|
|
"default": 512
|
|
},
|
|
{
|
|
"name": "decoder_prefill_graph_arena_mb",
|
|
"type": "int",
|
|
"description": "Decoder prefill graph arena size in MiB; default 512.",
|
|
"required": false,
|
|
"min": 1,
|
|
"default": 512
|
|
},
|
|
{
|
|
"name": "decoder_decode_graph_arena_mb",
|
|
"type": "int",
|
|
"description": "Decoder cached-step graph arena size in MiB; default 512.",
|
|
"required": false,
|
|
"min": 1,
|
|
"default": 512
|
|
}
|
|
],
|
|
"load": []
|
|
},
|
|
"runtime": {
|
|
"tags": [
|
|
"gguf"
|
|
]
|
|
},
|
|
"ui": {
|
|
"recommended_package": "hviske_asr_q8_0",
|
|
"tags": [
|
|
"ASR",
|
|
"GGUF"
|
|
],
|
|
"docs": [
|
|
"docs/asr.md",
|
|
"docs/gguf.md"
|
|
]
|
|
},
|
|
"packages": [
|
|
{
|
|
"id": "hviske_asr_q8_0",
|
|
"display_name": "Hviske v5.3 Q8_0 GGUF",
|
|
"default": true,
|
|
"format": "gguf",
|
|
"precision": "q8_0",
|
|
"target_directory": "Hviske-v5.3-GGUF",
|
|
"files": [
|
|
"Audio to text (ASR)/hviske-v5.3_Q8.gguf"
|
|
],
|
|
"strip_prefix": "Audio to text (ASR)",
|
|
"download": {
|
|
"kind": "huggingface_snapshot",
|
|
"repo": "mirek190/audio.cpp"
|
|
}
|
|
},
|
|
{
|
|
"id": "hviske_asr_safetensors",
|
|
"display_name": "Hviske v5.3 Safetensors",
|
|
"format": "safetensors",
|
|
"precision": "native",
|
|
"target_directory": "hviske-v5.3",
|
|
"files": [
|
|
"config.json",
|
|
"generation_config.json",
|
|
"model.safetensors",
|
|
"tokenizer.model"
|
|
],
|
|
"download": {
|
|
"kind": "huggingface_snapshot",
|
|
"repo": "syvai/hviske-v5.3"
|
|
}
|
|
}
|
|
],
|
|
"sources": [
|
|
{
|
|
"format": "gguf",
|
|
"roots": {
|
|
"model": ".",
|
|
"weights": "$gguf"
|
|
},
|
|
"files": {
|
|
"config": "model:config.json",
|
|
"generation_config": "model:generation_config.json",
|
|
"tokenizer": "model:tokenizer.model"
|
|
},
|
|
"tensors": {
|
|
"weights": "weights:"
|
|
}
|
|
},
|
|
{
|
|
"format": "safetensors",
|
|
"roots": {
|
|
"model": "."
|
|
},
|
|
"files": {
|
|
"config": "model:config.json",
|
|
"generation_config": "model:generation_config.json",
|
|
"tokenizer": "model:tokenizer.model"
|
|
},
|
|
"tensors": {
|
|
"weights": "model:model.safetensors"
|
|
}
|
|
}
|
|
]
|
|
}
|