blixt commited on
Commit
f4635a6
·
verified ·
1 Parent(s): e258ac1

Phonon 31M weights converted for WebGPU (int8, from gradium/phonon-31m-en@92741c5)

Browse files
Files changed (2) hide show
  1. README.md +36 -0
  2. phonon-31m-en.int8.bin +3 -0
README.md ADDED
@@ -0,0 +1,36 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ language:
4
+ - en
5
+ base_model: gradium/phonon-31m-en
6
+ pipeline_tag: text-to-speech
7
+ tags:
8
+ - text-to-speech
9
+ - webgpu
10
+ - phonon
11
+ ---
12
+
13
+ # Phonon 31M for WebGPU
14
+
15
+ This repository holds one file, `phonon-31m-en.int8.bin`: the weights of [gradium/phonon-31m-en](https://huggingface.co/gradium/phonon-31m-en) at revision `92741c5b07d085ff840e26fea8ce43110b87852f`, converted for [blixt/phonon](https://github.com/blixt/phonon), a WebGPU runtime that runs the model in the browser. A browser downloads this 43.7 MB file in place of the 103 MB checkpoint.
16
+
17
+ The model is by [Gradium](https://gradium.ai). Gradium did not make this conversion. The weights stay under the Apache License 2.0.
18
+
19
+ ## What the conversion does
20
+
21
+ - Tensors with 16,384 values or more become int8, with one f16 scale for each 32 values (scale = max |value| / 127, values rounded to f16 first, quantized values rounded half to even). All other tensors are f16.
22
+ - The rows of q and k move from interleaved rotary pairs to halves in each head.
23
+ - The flow head's time embedding for its one step is added to the bias of `cond_embed`.
24
+ - The codec's latent projection is merged into its upsampling weights, and its layer scales into the weights they scale.
25
+ - Convolution weights are reordered so that each output gathers its inputs.
26
+
27
+ The file starts with a JSON header (each tensor's offset, length and scale offset, and the vectors the speed control needs), followed by the vectors and the weights. `packModel` and `unpackModel` in [src/checkpoint.ts](https://github.com/blixt/phonon/blob/main/src/checkpoint.ts) write and read it.
28
+
29
+ ## Rebuild it
30
+
31
+ ```sh
32
+ git clone https://github.com/blixt/phonon && cd phonon && npm install
33
+ npm run build-weights -- path/to/gradium/phonon-31m-en/model.safetensors phonon-31m-en.int8.bin
34
+ shasum -a 256 phonon-31m-en.int8.bin
35
+ # 94f9899fe623251260f30892db8356c7293da09196dc9df9c3c4bb1ddc4f5a85
36
+ ```
phonon-31m-en.int8.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:94f9899fe623251260f30892db8356c7293da09196dc9df9c3c4bb1ddc4f5a85
3
+ size 43690044