Upload AuriStream checkpoint
Browse files- README.md +51 -0
- config.json +11 -9
- model.safetensors +2 -2
README.md
ADDED
|
@@ -0,0 +1,51 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
tags:
|
| 4 |
+
- audio
|
| 5 |
+
- speech
|
| 6 |
+
- language-model
|
| 7 |
+
- auristream
|
| 8 |
+
library_name: transformers
|
| 9 |
+
---
|
| 10 |
+
|
| 11 |
+
# AuriStream100M_40Pred_BigAudioDataset_500k
|
| 12 |
+
|
| 13 |
+
**AuriStream** is a speech language model by **Greta Tuckute** and **Klemen Kotar**.
|
| 14 |
+
|
| 15 |
+
This model predicts cochlear tokens from a tokenizer such as [WavCochCausalV8192](https://huggingface.co/TuKoResearch/WavCochCausalV8192).
|
| 16 |
+
|
| 17 |
+
|
| 18 |
+
|
| 19 |
+
## Model Details
|
| 20 |
+
|
| 21 |
+
| Parameter | Value |
|
| 22 |
+
|-----------|-------|
|
| 23 |
+
| Parameters | ~0.34B |
|
| 24 |
+
| Layers | 12 |
|
| 25 |
+
| Hidden Size | 768 |
|
| 26 |
+
| Attention Heads | 12 |
|
| 27 |
+
| Vocab Size | 8192 |
|
| 28 |
+
| Prediction Steps | 40 |
|
| 29 |
+
|
| 30 |
+
## Usage
|
| 31 |
+
|
| 32 |
+
```python
|
| 33 |
+
from transformers import AutoModel, AutoConfig
|
| 34 |
+
|
| 35 |
+
# Load with trust_remote_code for custom model
|
| 36 |
+
model = AutoModel.from_pretrained(
|
| 37 |
+
"TuKoResearch/AuriStream100M_40Pred_BigAudioDataset_500k",
|
| 38 |
+
trust_remote_code=True,
|
| 39 |
+
)
|
| 40 |
+
|
| 41 |
+
# Or load config first
|
| 42 |
+
config = AutoConfig.from_pretrained("TuKoResearch/AuriStream100M_40Pred_BigAudioDataset_500k", trust_remote_code=True)
|
| 43 |
+
```
|
| 44 |
+
|
| 45 |
+
## Base Model Code
|
| 46 |
+
|
| 47 |
+
This checkpoint uses shared model code from [TuKoResearch/AuriStream-base](https://huggingface.co/TuKoResearch/AuriStream-base).
|
| 48 |
+
|
| 49 |
+
## Tokenizer
|
| 50 |
+
|
| 51 |
+
This model uses cochlear tokens from [WavCochCausalV8192](https://huggingface.co/TuKoResearch/WavCochCausalV8192).
|
config.json
CHANGED
|
@@ -1,19 +1,21 @@
|
|
| 1 |
{
|
| 2 |
-
"
|
| 3 |
-
"architectures": [
|
| 4 |
-
|
|
|
|
| 5 |
"auto_map": {
|
| 6 |
-
"AutoConfig": "configuration_auristream.AuriStreamConfig",
|
| 7 |
-
"AutoModel": "modeling_auristream.
|
| 8 |
-
"AutoModelForCausalLM": "modeling_auristream.
|
| 9 |
},
|
|
|
|
|
|
|
| 10 |
"vocab_size": 8192,
|
| 11 |
"n_embd": 768,
|
| 12 |
"n_layer": 12,
|
| 13 |
"n_head": 12,
|
|
|
|
| 14 |
"dropout": 0.0,
|
| 15 |
"bias": false,
|
| 16 |
-
"
|
| 17 |
-
"n_pred_steps": 40,
|
| 18 |
-
"seq_len": 4096
|
| 19 |
}
|
|
|
|
| 1 |
{
|
| 2 |
+
"model_type": "AuriStream",
|
| 3 |
+
"architectures": [
|
| 4 |
+
"AuriStreamModel"
|
| 5 |
+
],
|
| 6 |
"auto_map": {
|
| 7 |
+
"AutoConfig": "TuKoResearch/AuriStream-base--configuration_auristream.AuriStreamConfig",
|
| 8 |
+
"AutoModel": "TuKoResearch/AuriStream-base--modeling_auristream.AuriStreamModel",
|
| 9 |
+
"AutoModelForCausalLM": "TuKoResearch/AuriStream-base--modeling_auristream.AuriStreamModel"
|
| 10 |
},
|
| 11 |
+
"torch_dtype": "bfloat16",
|
| 12 |
+
"transformers_version": "4.40.0",
|
| 13 |
"vocab_size": 8192,
|
| 14 |
"n_embd": 768,
|
| 15 |
"n_layer": 12,
|
| 16 |
"n_head": 12,
|
| 17 |
+
"n_pred_steps": 40,
|
| 18 |
"dropout": 0.0,
|
| 19 |
"bias": false,
|
| 20 |
+
"input_conv_kernel_size": 0
|
|
|
|
|
|
|
| 21 |
}
|
model.safetensors
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:cfc41a55e57d31f73ec7008aae9edc9ac3d9c460b63cdd1cce511e0d80617632
|
| 3 |
+
size 1371627728
|