Model release
Browse files- .gitattributes +1 -0
- README.md +27 -0
- all_results.json +3 -0
- config.json +3 -0
- eval_nbest_predictions.json +3 -0
- eval_predictions.json +3 -0
- eval_results.json +3 -0
- pytorch_model.bin +3 -0
- recipe.yaml +14 -0
- special_tokens_map.json +3 -0
- tokenizer.json +3 -0
- tokenizer_config.json +3 -0
- train_results.json +3 -0
- trainer_state.json +3 -0
- training_args.bin +3 -0
- vocab.txt +0 -0
.gitattributes
CHANGED
@@ -25,3 +25,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
|
25 |
*.zip filter=lfs diff=lfs merge=lfs -text
|
26 |
*.zstandard filter=lfs diff=lfs merge=lfs -text
|
27 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
|
|
|
25 |
*.zip filter=lfs diff=lfs merge=lfs -text
|
26 |
*.zstandard filter=lfs diff=lfs merge=lfs -text
|
27 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
28 |
+
*.json filter=lfs diff=lfs merge=lfs -text
|
README.md
ADDED
@@ -0,0 +1,27 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# oBERT-3-downstream-dense-QAT-squadv1
|
2 |
+
|
3 |
+
This model is obtained with [The Optimal BERT Surgeon: Scalable and Accurate Second-Order Pruning for Large Language Models](https://arxiv.org/abs/2203.07259).
|
4 |
+
|
5 |
+
It corresponds to the model presented in the `Table 3 - 3 Layers - 0% Sparsity - QAT`, and it represents an upper bound for performance of the corresponding pruned and quantized models:
|
6 |
+
- 80% unstructured QAT: `neuralmagic/oBERT-3-downstream-pruned-unstructured-80-QAT-squadv1`
|
7 |
+
- 80% block-4 QAT: `neuralmagic/oBERT-3-downstream-pruned-block4-80-QAT-squadv1`
|
8 |
+
- 90% unstructured QAT: `neuralmagic/oBERT-3-downstream-pruned-unstructured-90-QAT-squadv1`
|
9 |
+
- 90% block-4 QAT: `neuralmagic/oBERT-3-downstream-pruned-block4-90-QAT-squadv1`
|
10 |
+
|
11 |
+
SQuADv1 dev-set:
|
12 |
+
```
|
13 |
+
EM = 76.06
|
14 |
+
F1 = 84.25
|
15 |
+
```
|
16 |
+
|
17 |
+
Code: _coming soon_
|
18 |
+
|
19 |
+
## BibTeX entry and citation info
|
20 |
+
```bibtex
|
21 |
+
@article{kurtic2022optimal,
|
22 |
+
title={The Optimal BERT Surgeon: Scalable and Accurate Second-Order Pruning for Large Language Models},
|
23 |
+
author={Kurtic, Eldar and Campos, Daniel and Nguyen, Tuan and Frantar, Elias and Kurtz, Mark and Fineran, Benjamin and Goin, Michael and Alistarh, Dan},
|
24 |
+
journal={arXiv preprint arXiv:2203.07259},
|
25 |
+
year={2022}
|
26 |
+
}
|
27 |
+
```
|
all_results.json
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:c7a3a3bb264477e37f982e9638032e3774b516be2bfa380ad191bd049008a496
|
3 |
+
size 251
|
config.json
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:4641c6f957378584d77188cc7e8a6d5d786b0a2f79e3ece5d21661f089938f35
|
3 |
+
size 669
|
eval_nbest_predictions.json
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:6d6303c3b69eebb74dc3091dd82a2cb22279d61304063586452e31dfa9d827d4
|
3 |
+
size 49414307
|
eval_predictions.json
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:f07329d8cea295a6ab2854449b500d58c7373ba3b46063317c753da38287e262
|
3 |
+
size 599075
|
eval_results.json
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:bc85fe97da7484e9e386f97810305c37091c2d88d4e7a7d4304958bbf4333e65
|
3 |
+
size 113
|
pytorch_model.bin
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:7d374dcc9f2f288c9356474054a0d8c8fd3245413e622598a149851081bef76b
|
3 |
+
size 180695361
|
recipe.yaml
ADDED
@@ -0,0 +1,14 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
!LayerPruningModifier
|
2 |
+
end_epoch: -1.0
|
3 |
+
layers: ['bert.encoder.layer.3', 'bert.encoder.layer.4', 'bert.encoder.layer.5', 'bert.encoder.layer.6', 'bert.encoder.layer.7', 'bert.encoder.layer.8', 'bert.encoder.layer.9', 'bert.encoder.layer.10', 'bert.encoder.layer.11']
|
4 |
+
start_epoch: -1.0
|
5 |
+
update_frequency: -1.0
|
6 |
+
|
7 |
+
!QuantizationModifier
|
8 |
+
disable_quantization_observer_epoch: 5.0
|
9 |
+
end_epoch: -1.0
|
10 |
+
freeze_bn_stats_epoch: 5.0
|
11 |
+
quantize_embeddings: 1
|
12 |
+
start_epoch: 0.0
|
13 |
+
submodules: ['bert.encoder', 'bert.embeddings', 'qa_outputs']
|
14 |
+
|
special_tokens_map.json
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:303df45a03609e4ead04bc3dc1536d0ab19b5358db685b6f3da123d05ec200e3
|
3 |
+
size 112
|
tokenizer.json
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:5fd1c882abbd30517dced455a2c9768945ec726b96727927e4959348d9de550b
|
3 |
+
size 466081
|
tokenizer_config.json
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:297a2bab9748edd4a81b858ffc4503ad3b3429e2bfec19c235d7575ab3595d5b
|
3 |
+
size 383
|
train_results.json
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:f007c0787a072d33a21396b2a75ae787829cb57588868399f9e762f2df95e637
|
3 |
+
size 159
|
trainer_state.json
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:b1e381b5e498469ade25ae3ed6ae4962ca2c0f935cb3eae09d7526c448a5419b
|
3 |
+
size 7329
|
training_args.bin
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:d1dfba59f0b58c44adaf6e6f43c7e7009479949cec4e394dc6ae7f376f8e6fba
|
3 |
+
size 2607
|
vocab.txt
ADDED
The diff for this file is too large to render.
See raw diff
|
|