Add kerasformers electra_large_generator (ELECTRA)
Browse files- README.md +22 -0
- kf_config.json +24 -0
- model.weights.h5 +3 -0
- tokenizer.json +0 -0
README.md
ADDED
|
@@ -0,0 +1,22 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
library_name: kerasformers
|
| 3 |
+
license: apache-2.0
|
| 4 |
+
pipeline_tag: fill-mask
|
| 5 |
+
tags:
|
| 6 |
+
- keras
|
| 7 |
+
- electra
|
| 8 |
+
base_model: google/electra-large-generator
|
| 9 |
+
---
|
| 10 |
+
|
| 11 |
+
# electra_large_generator
|
| 12 |
+
|
| 13 |
+
ELECTRA masked-LM checkpoint converted to [kerasformers](https://github.com/IMvision12/KerasFormers) (pure Keras 3, runnable on JAX / PyTorch / TensorFlow). Converted from [`google/electra-large-generator`](https://huggingface.co/google/electra-large-generator).
|
| 14 |
+
|
| 15 |
+
```python
|
| 16 |
+
from kerasformers.models.electra import ElectraMaskedLM, ElectraTokenizer
|
| 17 |
+
|
| 18 |
+
model = ElectraMaskedLM.from_weights("kerasformers/electra_large_generator")
|
| 19 |
+
tokenizer = ElectraTokenizer.from_weights("kerasformers/electra_large_generator")
|
| 20 |
+
```
|
| 21 |
+
|
| 22 |
+
embed_dim 256 | layers 24 | heads 4
|
kf_config.json
ADDED
|
@@ -0,0 +1,24 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"library_name": "kerasformers",
|
| 3 |
+
"kerasformers_version": "1.2.4",
|
| 4 |
+
"model_module": "kerasformers.models.electra",
|
| 5 |
+
"model_class": "ElectraMaskedLM",
|
| 6 |
+
"variant": "electra_large_generator",
|
| 7 |
+
"weights": "model.weights.h5",
|
| 8 |
+
"schema_version": 2,
|
| 9 |
+
"weight_dtype": "float32",
|
| 10 |
+
"model_type": "electra",
|
| 11 |
+
"text_config": {
|
| 12 |
+
"vocab_size": 30522,
|
| 13 |
+
"embedding_size": 1024,
|
| 14 |
+
"embed_dim": 256,
|
| 15 |
+
"num_layers": 24,
|
| 16 |
+
"num_heads": 4,
|
| 17 |
+
"mlp_dim": 1024,
|
| 18 |
+
"max_position_embeddings": 512,
|
| 19 |
+
"type_vocab_size": 2,
|
| 20 |
+
"hidden_act": "gelu",
|
| 21 |
+
"layer_norm_eps": 1e-12,
|
| 22 |
+
"pad_token_id": 0
|
| 23 |
+
}
|
| 24 |
+
}
|
model.weights.h5
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:9fd2ccdcdaa11a4c22426dec4a8f13e142a8a5f28914c68eb909f2a2289ce13e
|
| 3 |
+
size 331105840
|
tokenizer.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|