Dan Fu
commited on
Commit
•
afc7050
1
Parent(s):
6d98b24
32K partial checkpoint
Browse files- README.md +13 -0
- config.json +4 -0
- config.yaml +38 -0
- model.bin +3 -0
- model.pt +3 -0
- version.txt +1 -0
README.md
CHANGED
@@ -1,3 +1,16 @@
|
|
1 |
---
|
2 |
license: apache-2.0
|
|
|
|
|
|
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
license: apache-2.0
|
3 |
+
language:
|
4 |
+
- en
|
5 |
+
pipeline_tag: text-classification
|
6 |
---
|
7 |
+
|
8 |
+
# Monarch Mixer-BERT
|
9 |
+
|
10 |
+
The 80M checkpoint for M2-BERT-base from the paper [Monarch Mixer: A Simple Sub-Quadratic GEMM-Based Architecture](https://arxiv.org/abs/2310.12109).
|
11 |
+
This model has been pretrained with sequence length 32K.
|
12 |
+
Note (11/3 evening): this is a partial checkpoint, this one had not finished training before upload.
|
13 |
+
|
14 |
+
This model was trained by Dan Fu, Jon Saad-Falcon, and Simran Arora.
|
15 |
+
|
16 |
+
Check out our [GitHub](https://github.com/HazyResearch/m2/tree/main) for instructions on how to download and fine-tune it!
|
config.json
ADDED
@@ -0,0 +1,4 @@
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"model_type": "m2_bert"
|
3 |
+
}
|
4 |
+
|
config.yaml
ADDED
@@ -0,0 +1,38 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Note that some of the fields in this template haven't been filled in yet.
|
2 |
+
# Please resolve any `null` fields before launching!
|
3 |
+
|
4 |
+
precision: amp_bf16
|
5 |
+
max_seq_len: 32768
|
6 |
+
|
7 |
+
# Tokenizer for dataset creation
|
8 |
+
tokenizer_name: bert-base-uncased
|
9 |
+
|
10 |
+
# Base model config
|
11 |
+
model:
|
12 |
+
name: bert
|
13 |
+
pretrained_model_name: ${tokenizer_name}
|
14 |
+
tokenizer_name: ${tokenizer_name}
|
15 |
+
model_config:
|
16 |
+
num_attention_heads: 12
|
17 |
+
num_hidden_layers: 12
|
18 |
+
attention_probs_dropout_prob: 0.0
|
19 |
+
max_position_embeddings: 32768
|
20 |
+
|
21 |
+
monarch_mixer_sequence_mixing: True
|
22 |
+
long_conv_l_max: 32768
|
23 |
+
long_conv_kernel_learning_rate: 1e-3
|
24 |
+
hyena_lr_pos_emb: 1e-5
|
25 |
+
hyena_w: 10
|
26 |
+
hyena_wd: 0.1
|
27 |
+
hyena_emb_dim: 5
|
28 |
+
hyena_filter_order: 128
|
29 |
+
hyena_training_additions: False
|
30 |
+
|
31 |
+
bidirectional: true
|
32 |
+
residual_long_conv: true
|
33 |
+
|
34 |
+
use_glu_mlp: True
|
35 |
+
use_monarch_mlp: True
|
36 |
+
monarch_mlp_nblocks: 4
|
37 |
+
use_positional_encodings: True
|
38 |
+
|
model.bin
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:f5f6da4ea57ab1b407363530efba622552c121dc29b439cc9b202f042108d7d2
|
3 |
+
size 440736801
|
model.pt
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:069a59ecf30222fa1e67f68b76f7155966a875ac2ab060f1cb2d1213015e3596
|
3 |
+
size 1315397236
|
version.txt
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
1
|