Initial model commit

.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
+*trainer_state.json filter=lfs diff=lfs merge=lfs -text

all_results.json

@@ -0,0 +1,11 @@
+{
+    "epoch": 10.0,
+    "eval_exact_match": 79.00662251655629,
+    "eval_f1": 86.94488067162003,
+    "eval_samples": 10784,
+    "train_loss": 12.042027303979902,
+    "train_runtime": 55561.5375,
+    "train_samples": 88524,
+    "train_samples_per_second": 15.933,
+    "train_steps_per_second": 0.996
+}

config.json

@@ -0,0 +1,25 @@
+{
+  "_name_or_path": "bert-base-uncased",
+  "architectures": [
+    "NNCFNetwork"
+  ],
+  "attention_probs_dropout_prob": 0.1,
+  "gradient_checkpointing": false,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 768,
+  "initializer_range": 0.02,
+  "intermediate_size": 3072,
+  "layer_norm_eps": 1e-12,
+  "max_position_embeddings": 512,
+  "model_type": "bert",
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "pad_token_id": 0,
+  "position_embedding_type": "absolute",
+  "torch_dtype": "float32",
+  "transformers_version": "4.9.1",
+  "type_vocab_size": 2,
+  "use_cache": true,
+  "vocab_size": 30522
+}

eval_results.json

@@ -0,0 +1,6 @@
+{
+    "epoch": 10.0,
+    "eval_exact_match": 79.00662251655629,
+    "eval_f1": 86.94488067162003,
+    "eval_samples": 10784
+}

linear_layer_sparse_stats_total_23M_72.9_relative_sparsity.csv

@@ -0,0 +1,73 @@
+,linear_id,shape,param_count,nnz_count
+0,bert.encoder.layer.0.attention.self.query,"[768, 768]",589824,16252
+1,bert.encoder.layer.0.attention.self.key,"[768, 768]",589824,13054
+2,bert.encoder.layer.0.attention.self.value,"[768, 768]",589824,31394
+3,bert.encoder.layer.0.attention.output.dense,"[768, 768]",589824,44074
+4,bert.encoder.layer.0.intermediate.dense,"[3072, 768]",2359296,1159424
+5,bert.encoder.layer.0.output.dense,"[768, 3072]",2359296,1061306
+6,bert.encoder.layer.1.attention.self.query,"[768, 768]",589824,58240
+7,bert.encoder.layer.1.attention.self.key,"[768, 768]",589824,54729
+8,bert.encoder.layer.1.attention.self.value,"[768, 768]",589824,57604
+9,bert.encoder.layer.1.attention.output.dense,"[768, 768]",589824,65732
+10,bert.encoder.layer.1.intermediate.dense,"[3072, 768]",2359296,1230849
+11,bert.encoder.layer.1.output.dense,"[768, 3072]",2359296,1063370
+12,bert.encoder.layer.2.attention.self.query,"[768, 768]",589824,76074
+13,bert.encoder.layer.2.attention.self.key,"[768, 768]",589824,72798
+14,bert.encoder.layer.2.attention.self.value,"[768, 768]",589824,69367
+15,bert.encoder.layer.2.attention.output.dense,"[768, 768]",589824,68078
+16,bert.encoder.layer.2.intermediate.dense,"[3072, 768]",2359296,1260119
+17,bert.encoder.layer.2.output.dense,"[768, 3072]",2359296,1081396
+18,bert.encoder.layer.3.attention.self.query,"[768, 768]",589824,79954
+19,bert.encoder.layer.3.attention.self.key,"[768, 768]",589824,84045
+20,bert.encoder.layer.3.attention.self.value,"[768, 768]",589824,97323
+21,bert.encoder.layer.3.attention.output.dense,"[768, 768]",589824,95918
+22,bert.encoder.layer.3.intermediate.dense,"[3072, 768]",2359296,1263350
+23,bert.encoder.layer.3.output.dense,"[768, 3072]",2359296,1069440
+24,bert.encoder.layer.4.attention.self.query,"[768, 768]",589824,77306
+25,bert.encoder.layer.4.attention.self.key,"[768, 768]",589824,77738
+26,bert.encoder.layer.4.attention.self.value,"[768, 768]",589824,103253
+27,bert.encoder.layer.4.attention.output.dense,"[768, 768]",589824,103673
+28,bert.encoder.layer.4.intermediate.dense,"[3072, 768]",2359296,1253106
+29,bert.encoder.layer.4.output.dense,"[768, 3072]",2359296,1031342
+30,bert.encoder.layer.5.attention.self.query,"[768, 768]",589824,59269
+31,bert.encoder.layer.5.attention.self.key,"[768, 768]",589824,66896
+32,bert.encoder.layer.5.attention.self.value,"[768, 768]",589824,97790
+33,bert.encoder.layer.5.attention.output.dense,"[768, 768]",589824,93833
+34,bert.encoder.layer.5.intermediate.dense,"[3072, 768]",2359296,1263747
+35,bert.encoder.layer.5.output.dense,"[768, 3072]",2359296,1014643
+36,bert.encoder.layer.6.attention.self.query,"[768, 768]",589824,59641
+37,bert.encoder.layer.6.attention.self.key,"[768, 768]",589824,67416
+38,bert.encoder.layer.6.attention.self.value,"[768, 768]",589824,95731
+39,bert.encoder.layer.6.attention.output.dense,"[768, 768]",589824,85998
+40,bert.encoder.layer.6.intermediate.dense,"[3072, 768]",2359296,1181526
+41,bert.encoder.layer.6.output.dense,"[768, 3072]",2359296,920703
+42,bert.encoder.layer.7.attention.self.query,"[768, 768]",589824,44935
+43,bert.encoder.layer.7.attention.self.key,"[768, 768]",589824,56868
+44,bert.encoder.layer.7.attention.self.value,"[768, 768]",589824,89913
+45,bert.encoder.layer.7.attention.output.dense,"[768, 768]",589824,74029
+46,bert.encoder.layer.7.intermediate.dense,"[3072, 768]",2359296,995977
+47,bert.encoder.layer.7.output.dense,"[768, 3072]",2359296,777939
+48,bert.encoder.layer.8.attention.self.query,"[768, 768]",589824,49884
+49,bert.encoder.layer.8.attention.self.key,"[768, 768]",589824,57649
+50,bert.encoder.layer.8.attention.self.value,"[768, 768]",589824,92159
+51,bert.encoder.layer.8.attention.output.dense,"[768, 768]",589824,71883
+52,bert.encoder.layer.8.intermediate.dense,"[3072, 768]",2359296,757196
+53,bert.encoder.layer.8.output.dense,"[768, 3072]",2359296,568698
+54,bert.encoder.layer.9.attention.self.query,"[768, 768]",589824,68469
+55,bert.encoder.layer.9.attention.self.key,"[768, 768]",589824,69557
+56,bert.encoder.layer.9.attention.self.value,"[768, 768]",589824,32458
+57,bert.encoder.layer.9.attention.output.dense,"[768, 768]",589824,23895
+58,bert.encoder.layer.9.intermediate.dense,"[3072, 768]",2359296,359909
+59,bert.encoder.layer.9.output.dense,"[768, 3072]",2359296,245729
+60,bert.encoder.layer.10.attention.self.query,"[768, 768]",589824,42730
+61,bert.encoder.layer.10.attention.self.key,"[768, 768]",589824,44139
+62,bert.encoder.layer.10.attention.self.value,"[768, 768]",589824,17533
+63,bert.encoder.layer.10.attention.output.dense,"[768, 768]",589824,12605
+64,bert.encoder.layer.10.intermediate.dense,"[3072, 768]",2359296,233883
+65,bert.encoder.layer.10.output.dense,"[768, 3072]",2359296,130882
+66,bert.encoder.layer.11.attention.self.query,"[768, 768]",589824,11427
+67,bert.encoder.layer.11.attention.self.key,"[768, 768]",589824,14775
+68,bert.encoder.layer.11.attention.self.value,"[768, 768]",589824,6865
+69,bert.encoder.layer.11.attention.output.dense,"[768, 768]",589824,3223
+70,bert.encoder.layer.11.intermediate.dense,"[3072, 768]",2359296,190784
+71,bert.encoder.layer.11.output.dense,"[768, 3072]",2359296,71760

nncf_bert_config_squad_mvnt_pruning-distill-run8.json

@@ -0,0 +1,31 @@
+{
+    "input_info": [
+        {
+            "sample_size": [1, 384],
+            "type": "long"
+        },
+        {
+            "sample_size": [1, 384],
+            "type": "long"
+        },
+        {
+            "sample_size": [1, 384],
+            "type": "long"
+        }
+    ],
+    "compression": {
+        "algorithm": "movement_sparsity",
+        "params": {
+            "schedule": "poly_threshold",
+            "power": 3,
+            "init_importance_threshold": 0.0,
+            "final_importance_threshold": 0.1,
+            "warmup_start_epoch": 1,
+            "warmup_end_epoch": 2.5,
+            "steps_per_epoch": 5533,
+            "regu_final_lambda": 400,
+            "update_per_optimizer_step": true,
+        },
+        "ignored_scopes": ["{re}.*NNCFEmbedding", "{re}.*qa_outputs*"]
+    },
+}

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:600721be352243b4f8c1751c5c5b00d348401f728ee94076ed7b8baeb9730696
+size 435643185

special_tokens_map.json

@@ -0,0 +1 @@
+{"unk_token": "[UNK]", "sep_token": "[SEP]", "pad_token": "[PAD]", "cls_token": "[CLS]", "mask_token": "[MASK]"}

tf_eval_results.json

@@ -0,0 +1,4 @@
+{
+    "exact_match": 31.05960264900662,
+    "f1": 39.84460263383556
+}

tf_model.h5

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c6882972800d1a921103ecfe0ae42607d1f675a092d58878441bc071b85b3e0a
+size 435842064

tokenizer_config.json

@@ -0,0 +1 @@
+{"do_lower_case": true, "unk_token": "[UNK]", "sep_token": "[SEP]", "pad_token": "[PAD]", "cls_token": "[CLS]", "mask_token": "[MASK]", "tokenize_chinese_chars": true, "strip_accents": null, "model_max_length": 512, "special_tokens_map_file": null, "name_or_path": "bert-base-uncased", "tokenizer_class": "BertTokenizer"}

train_results.json

@@ -0,0 +1,8 @@
+{
+    "epoch": 10.0,
+    "train_loss": 12.042027303979902,
+    "train_runtime": 55561.5375,
+    "train_samples": 88524,
+    "train_samples_per_second": 15.933,
+    "train_steps_per_second": 0.996
+}

trainer_state.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f9771dceb7b8cca0a6d0861451e815fd0dbaf5518a4ab1f6246522437ab67719
+size 31647642

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:19466a89c86a5728654b838639bee6be304cd6da6c20fa07ca9bde40b079d3ce
+size 3055