Model release

.gitattributes

@@ -25,3 +25,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zstandard filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+*.json filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,27 @@
+# oBERT-3-downstream-dense-QAT-squadv1
+
+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).
+
+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:
+- 80% unstructured QAT: `neuralmagic/oBERT-3-downstream-pruned-unstructured-80-QAT-squadv1`
+- 80% block-4 QAT: `neuralmagic/oBERT-3-downstream-pruned-block4-80-QAT-squadv1`
+- 90% unstructured QAT: `neuralmagic/oBERT-3-downstream-pruned-unstructured-90-QAT-squadv1`
+- 90% block-4 QAT: `neuralmagic/oBERT-3-downstream-pruned-block4-90-QAT-squadv1`
+
+SQuADv1 dev-set:
+```
+EM = 76.06
+F1 = 84.25
+```
+
+Code: _coming soon_
+
+## BibTeX entry and citation info
+```bibtex
+@article{kurtic2022optimal,
+  title={The Optimal BERT Surgeon: Scalable and Accurate Second-Order Pruning for Large Language Models},
+  author={Kurtic, Eldar and Campos, Daniel and Nguyen, Tuan and Frantar, Elias and Kurtz, Mark and Fineran, Benjamin and Goin, Michael and Alistarh, Dan},
+  journal={arXiv preprint arXiv:2203.07259},
+  year={2022}
+}
+```

all_results.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c7a3a3bb264477e37f982e9638032e3774b516be2bfa380ad191bd049008a496
+size 251

config.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4641c6f957378584d77188cc7e8a6d5d786b0a2f79e3ece5d21661f089938f35
+size 669

eval_nbest_predictions.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6d6303c3b69eebb74dc3091dd82a2cb22279d61304063586452e31dfa9d827d4
+size 49414307

eval_predictions.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f07329d8cea295a6ab2854449b500d58c7373ba3b46063317c753da38287e262
+size 599075

eval_results.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bc85fe97da7484e9e386f97810305c37091c2d88d4e7a7d4304958bbf4333e65
+size 113

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7d374dcc9f2f288c9356474054a0d8c8fd3245413e622598a149851081bef76b
+size 180695361

recipe.yaml

@@ -0,0 +1,14 @@
+!LayerPruningModifier
+    end_epoch: -1.0
+    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']
+    start_epoch: -1.0
+    update_frequency: -1.0
+
+!QuantizationModifier
+    disable_quantization_observer_epoch: 5.0
+    end_epoch: -1.0
+    freeze_bn_stats_epoch: 5.0
+    quantize_embeddings: 1
+    start_epoch: 0.0
+    submodules: ['bert.encoder', 'bert.embeddings', 'qa_outputs']
+

special_tokens_map.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:303df45a03609e4ead04bc3dc1536d0ab19b5358db685b6f3da123d05ec200e3
+size 112

tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5fd1c882abbd30517dced455a2c9768945ec726b96727927e4959348d9de550b
+size 466081

tokenizer_config.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:297a2bab9748edd4a81b858ffc4503ad3b3429e2bfec19c235d7575ab3595d5b
+size 383

train_results.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f007c0787a072d33a21396b2a75ae787829cb57588868399f9e762f2df95e637
+size 159

trainer_state.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b1e381b5e498469ade25ae3ed6ae4962ca2c0f935cb3eae09d7526c448a5419b
+size 7329

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d1dfba59f0b58c44adaf6e6f43c7e7009479949cec4e394dc6ae7f376f8e6fba
+size 2607