huggingface/candle
1
0
Fork 0
mirror of https://github.com/huggingface/candle.git synced 2025-06-21 20:22:49 +00:00
Code Issues Packages Projects Releases Wiki Activity

Whisper quantized wasm (#1028)

Browse Source 
* [Whisper] Update to use quantized model

* [whisper] add language detection

* [whisper] change assets location

* [whisper] adapt js example with quantized models

* [whisper] better task parsing

* [whisper] minor fixes
This commit is contained in:
Radamés Ajna
2023-10-04 12:22:57 -07:00
committed by GitHub
parent c18a856e76
commit 27e70a5093
13 changed files with 540 additions and 596 deletions
Download Patch File Download Diff File Expand all files Collapse all files

45
candle-wasm-examples/whisper/src/app.rs
View File

@ -7,7 +7,12 @@ use yew::{html, Component, Context, Html};
use yew_agent::{Bridge, Bridged};
const SAMPLE_NAMES: [&str; 6] = [
"jfk.wav", "a13.wav", "gb0.wav", "gb1.wav", "hp0.wav", "mm0.wav",
"audios/samples_jfk.wav",
"audios/samples_a13.wav",
"audios/samples_gb0.wav",
"audios/samples_gb1.wav",
"audios/samples_hp0.wav",
"audios/samples_mm0.wav",
];
async fn fetch_url(url: &str) -> Result<Vec<u8>, JsValue> {
@ -54,14 +59,46 @@ pub struct App {
}
async fn model_data_load() -> Result<ModelData, JsValue> {
let tokenizer = fetch_url("tokenizer.en.json").await?;
let mel_filters = fetch_url("mel_filters.safetensors").await?;
let weights = fetch_url("tiny.en.safetensors").await?;
let quantized = false;
let is_multilingual = false;
let (tokenizer, mel_filters, weights, config) = if quantized {
console_log!("loading quantized weights");
let tokenizer = fetch_url("quantized/tokenizer-tiny-en.json").await?;
let mel_filters = fetch_url("mel_filters.safetensors").await?;
let weights = fetch_url("quantized/model-tiny-en-q80.gguf").await?;
let config = fetch_url("quantized/config-tiny-en.json").await?;
(tokenizer, mel_filters, weights, config)
} else {
console_log!("loading float weights");
if is_multilingual {
let mel_filters = fetch_url("mel_filters.safetensors").await?;
let tokenizer = fetch_url("whisper-tiny/tokenizer.json").await?;
let weights = fetch_url("whisper-tiny/model.safetensors").await?;
let config = fetch_url("whisper-tiny/config.json").await?;
(tokenizer, mel_filters, weights, config)
} else {
let mel_filters = fetch_url("mel_filters.safetensors").await?;
let tokenizer = fetch_url("whisper-tiny.en/tokenizer.json").await?;
let weights = fetch_url("whisper-tiny.en/model.safetensors").await?;
let config = fetch_url("whisper-tiny.en/config.json").await?;
(tokenizer, mel_filters, weights, config)
}
};
let timestamps = true;
let _task = Some("transcribe".to_string());
console_log!("{}", weights.len());
Ok(ModelData {
tokenizer,
mel_filters,
weights,
config,
quantized,
timestamps,
task: None,
is_multilingual,
language: None,
})
}