Add kerasformers modernbert_large (ModernBERT)
Browse files- README.md +29 -0
- kf_config.json +26 -0
- model.weights.h5 +3 -0
- tokenizer.json +0 -0
README.md
ADDED
|
@@ -0,0 +1,29 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
library_name: kerasformers
|
| 3 |
+
license: apache-2.0
|
| 4 |
+
pipeline_tag: fill-mask
|
| 5 |
+
tags:
|
| 6 |
+
- keras
|
| 7 |
+
- modernbert
|
| 8 |
+
base_model: answerdotai/ModernBERT-large
|
| 9 |
+
---
|
| 10 |
+
|
| 11 |
+
# modernbert_large
|
| 12 |
+
|
| 13 |
+
ModernBERT converted to [kerasformers](https://github.com/IMvision12/KerasFormers)
|
| 14 |
+
(pure Keras 3, runnable on JAX / PyTorch / TensorFlow). Converted from [`answerdotai/ModernBERT-large`](https://huggingface.co/answerdotai/ModernBERT-large).
|
| 15 |
+
|
| 16 |
+
```python
|
| 17 |
+
from kerasformers.models.modernbert import ModernBertModel, ModernBertTokenizer
|
| 18 |
+
|
| 19 |
+
model = ModernBertModel.from_weights("kerasformers/modernbert_large")
|
| 20 |
+
tokenizer = ModernBertTokenizer.from_weights("kerasformers/modernbert_large")
|
| 21 |
+
out = model(tokenizer("Hello, world."))["last_hidden_state"]
|
| 22 |
+
```
|
| 23 |
+
|
| 24 |
+
| layers | embed_dim | heads | mlp_dim |
|
| 25 |
+
|---|---|---|---|
|
| 26 |
+
| 28 | 1024 | 16 | 2624 |
|
| 27 |
+
|
| 28 |
+
The single `model.weights.h5` is the full masked-LM checkpoint; the encoder,
|
| 29 |
+
masked-LM, and task-head classes each load their own subset.
|
kf_config.json
ADDED
|
@@ -0,0 +1,26 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"library_name": "kerasformers",
|
| 3 |
+
"kerasformers_version": "1.2.3",
|
| 4 |
+
"model_module": "kerasformers.models.modernbert",
|
| 5 |
+
"model_class": "ModernBertModel",
|
| 6 |
+
"variant": "modernbert_large",
|
| 7 |
+
"weights": "model.weights.h5",
|
| 8 |
+
"schema_version": 2,
|
| 9 |
+
"weight_dtype": "float32",
|
| 10 |
+
"model_type": "modernbert",
|
| 11 |
+
"text_config": {
|
| 12 |
+
"vocab_size": 50368,
|
| 13 |
+
"embed_dim": 1024,
|
| 14 |
+
"num_layers": 28,
|
| 15 |
+
"num_heads": 16,
|
| 16 |
+
"mlp_dim": 2624,
|
| 17 |
+
"max_position_embeddings": 8192,
|
| 18 |
+
"hidden_act": "gelu",
|
| 19 |
+
"norm_eps": 1e-05,
|
| 20 |
+
"local_attention": 128,
|
| 21 |
+
"global_attn_every_n_layers": 3,
|
| 22 |
+
"global_rope_theta": 160000.0,
|
| 23 |
+
"local_rope_theta": 10000.0,
|
| 24 |
+
"pad_token_id": 50283
|
| 25 |
+
}
|
| 26 |
+
}
|
model.weights.h5
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:a8218e4a8719c83a5be0b1be2ceaf19da96b72bed76423c19203548f63b5fb80
|
| 3 |
+
size 1790662552
|
tokenizer.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|