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": 75.55345316934721,
+    "eval_f1": 84.22648496225351,
+    "eval_samples": 10784,
+    "train_loss": 6.046503622577567,
+    "train_runtime": 62400.7352,
+    "train_samples": 88524,
+    "train_samples_per_second": 14.186,
+    "train_steps_per_second": 0.887
+}

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": 75.55345316934721,
+    "eval_f1": 84.22648496225351,
+    "eval_samples": 10784
+}

linear_layer_sparse_stats_total_12M_85.4_relative_sparsity.csv

@@ -0,0 +1,73 @@
+,linear_id,shape,param_count,nnz_count
+0,bert.encoder.layer.0.attention.self.query,"[768, 768]",589824,1757
+1,bert.encoder.layer.0.attention.self.key,"[768, 768]",589824,1800
+2,bert.encoder.layer.0.attention.self.value,"[768, 768]",589824,5416
+3,bert.encoder.layer.0.attention.output.dense,"[768, 768]",589824,7845
+4,bert.encoder.layer.0.intermediate.dense,"[3072, 768]",2359296,701532
+5,bert.encoder.layer.0.output.dense,"[768, 3072]",2359296,591365
+6,bert.encoder.layer.1.attention.self.query,"[768, 768]",589824,27422
+7,bert.encoder.layer.1.attention.self.key,"[768, 768]",589824,27291
+8,bert.encoder.layer.1.attention.self.value,"[768, 768]",589824,22259
+9,bert.encoder.layer.1.attention.output.dense,"[768, 768]",589824,27236
+10,bert.encoder.layer.1.intermediate.dense,"[3072, 768]",2359296,775297
+11,bert.encoder.layer.1.output.dense,"[768, 3072]",2359296,599687
+12,bert.encoder.layer.2.attention.self.query,"[768, 768]",589824,31292
+13,bert.encoder.layer.2.attention.self.key,"[768, 768]",589824,30911
+14,bert.encoder.layer.2.attention.self.value,"[768, 768]",589824,24690
+15,bert.encoder.layer.2.attention.output.dense,"[768, 768]",589824,24902
+16,bert.encoder.layer.2.intermediate.dense,"[3072, 768]",2359296,804492
+17,bert.encoder.layer.2.output.dense,"[768, 3072]",2359296,606736
+18,bert.encoder.layer.3.attention.self.query,"[768, 768]",589824,35285
+19,bert.encoder.layer.3.attention.self.key,"[768, 768]",589824,38990
+20,bert.encoder.layer.3.attention.self.value,"[768, 768]",589824,38379
+21,bert.encoder.layer.3.attention.output.dense,"[768, 768]",589824,39416
+22,bert.encoder.layer.3.intermediate.dense,"[3072, 768]",2359296,803280
+23,bert.encoder.layer.3.output.dense,"[768, 3072]",2359296,592789
+24,bert.encoder.layer.4.attention.self.query,"[768, 768]",589824,28239
+25,bert.encoder.layer.4.attention.self.key,"[768, 768]",589824,30768
+26,bert.encoder.layer.4.attention.self.value,"[768, 768]",589824,39479
+27,bert.encoder.layer.4.attention.output.dense,"[768, 768]",589824,39703
+28,bert.encoder.layer.4.intermediate.dense,"[3072, 768]",2359296,784797
+29,bert.encoder.layer.4.output.dense,"[768, 3072]",2359296,550626
+30,bert.encoder.layer.5.attention.self.query,"[768, 768]",589824,16702
+31,bert.encoder.layer.5.attention.self.key,"[768, 768]",589824,21640
+32,bert.encoder.layer.5.attention.self.value,"[768, 768]",589824,35016
+33,bert.encoder.layer.5.attention.output.dense,"[768, 768]",589824,33592
+34,bert.encoder.layer.5.intermediate.dense,"[3072, 768]",2359296,785249
+35,bert.encoder.layer.5.output.dense,"[768, 3072]",2359296,530621
+36,bert.encoder.layer.6.attention.self.query,"[768, 768]",589824,16901
+37,bert.encoder.layer.6.attention.self.key,"[768, 768]",589824,20927
+38,bert.encoder.layer.6.attention.self.value,"[768, 768]",589824,29547
+39,bert.encoder.layer.6.attention.output.dense,"[768, 768]",589824,26602
+40,bert.encoder.layer.6.intermediate.dense,"[3072, 768]",2359296,706133
+41,bert.encoder.layer.6.output.dense,"[768, 3072]",2359296,458484
+42,bert.encoder.layer.7.attention.self.query,"[768, 768]",589824,15264
+43,bert.encoder.layer.7.attention.self.key,"[768, 768]",589824,21032
+44,bert.encoder.layer.7.attention.self.value,"[768, 768]",589824,34524
+45,bert.encoder.layer.7.attention.output.dense,"[768, 768]",589824,26950
+46,bert.encoder.layer.7.intermediate.dense,"[3072, 768]",2359296,558548
+47,bert.encoder.layer.7.output.dense,"[768, 3072]",2359296,366404
+48,bert.encoder.layer.8.attention.self.query,"[768, 768]",589824,14420
+49,bert.encoder.layer.8.attention.self.key,"[768, 768]",589824,18006
+50,bert.encoder.layer.8.attention.self.value,"[768, 768]",589824,27492
+51,bert.encoder.layer.8.attention.output.dense,"[768, 768]",589824,21056
+52,bert.encoder.layer.8.intermediate.dense,"[3072, 768]",2359296,388041
+53,bert.encoder.layer.8.output.dense,"[768, 3072]",2359296,246678
+54,bert.encoder.layer.9.attention.self.query,"[768, 768]",589824,32843
+55,bert.encoder.layer.9.attention.self.key,"[768, 768]",589824,31522
+56,bert.encoder.layer.9.attention.self.value,"[768, 768]",589824,8125
+57,bert.encoder.layer.9.attention.output.dense,"[768, 768]",589824,6000
+58,bert.encoder.layer.9.intermediate.dense,"[3072, 768]",2359296,148663
+59,bert.encoder.layer.9.output.dense,"[768, 3072]",2359296,82095
+60,bert.encoder.layer.10.attention.self.query,"[768, 768]",589824,19588
+61,bert.encoder.layer.10.attention.self.key,"[768, 768]",589824,19232
+62,bert.encoder.layer.10.attention.self.value,"[768, 768]",589824,5844
+63,bert.encoder.layer.10.attention.output.dense,"[768, 768]",589824,4472
+64,bert.encoder.layer.10.intermediate.dense,"[3072, 768]",2359296,95918
+65,bert.encoder.layer.10.output.dense,"[768, 3072]",2359296,40352
+66,bert.encoder.layer.11.attention.self.query,"[768, 768]",589824,4707
+67,bert.encoder.layer.11.attention.self.key,"[768, 768]",589824,5807
+68,bert.encoder.layer.11.attention.self.value,"[768, 768]",589824,2160
+69,bert.encoder.layer.11.attention.output.dense,"[768, 768]",589824,1310
+70,bert.encoder.layer.11.intermediate.dense,"[3072, 768]",2359296,71477
+71,bert.encoder.layer.11.output.dense,"[768, 3072]",2359296,26152

nncf_bert_config_squad_mvnt_pruning-distill-run2.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": 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 sha256:0a9066033642ddc29a673c4ddbe010d6d14f73dced3460c401ef2bfe9b7eb8f2
+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": 14.257332071901608,
+    "f1": 23.49169853506678
+}

tf_model.h5

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:99005555f720aa0616d0a0b2933db5848c3111fd67dee51ad708f58390392801
+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": 6.046503622577567,
+    "train_runtime": 62400.7352,
+    "train_samples": 88524,
+    "train_samples_per_second": 14.186,
+    "train_steps_per_second": 0.887
+}

trainer_state.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1f5481d7590696c650fd9792b1b96544e6616942da18a79fddd9e5444ddc25e4
+size 31965164

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:86104797dbc9b497fc75531ba856e382a4d7799efe6475a2ab72782259e2ceb2
+size 3055