Hugging Face's logo

vuiseng9
/
r0.010-squad-bert-b-mvmt-8bit

ONNX
Model card Files Community
vchua commited on
Commit
4088e0c
•
1 Parent(s): 5ceb4bf

Add collaterals

Browse files
Files changed (14) hide show
  1. all_results.json +5 -0
  2. compressed_graph.dot +0 -0
  3. eval_predictions.json +0 -0
  4. eval_results.json +5 -0
  5. ir/log.bapp.tput.10k-iter +63 -0
  6. ir/sparsity_structures.csv +73 -0
  7. ir/sparsity_structures.md +74 -0
  8. ir/sparsity_structures.pkl +0 -0
  9. ir/squad-BertForQuestionAnswering.cropped.8bit.bin +3 -0
  10. ir/squad-BertForQuestionAnswering.cropped.8bit.mapping +0 -0
  11. ir/squad-BertForQuestionAnswering.cropped.8bit.onnx +3 -0
  12. ir/squad-BertForQuestionAnswering.cropped.8bit.xml +0 -0
  13. nncf-mvmt-p3.json +67 -0
  14. original_graph.dot +0 -0
all_results.json ADDED
@@ -0,0 +1,5 @@
 
 
 
 
 
 
1
+ {
2
+ "eval_exact_match": 83.53831598864711,
3
+ "eval_f1": 89.99254134200535,
4
+ "eval_samples": 10784
5
+ }
compressed_graph.dot ADDED
The diff for this file is too large to render. See raw diff
 
eval_predictions.json ADDED
The diff for this file is too large to render. See raw diff
 
eval_results.json ADDED
@@ -0,0 +1,5 @@
 
 
 
 
 
 
1
+ {
2
+ "eval_exact_match": 83.53831598864711,
3
+ "eval_f1": 89.99254134200535,
4
+ "eval_samples": 10784
5
+ }
ir/log.bapp.tput.10k-iter ADDED
@@ -0,0 +1,63 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [Step 1/11] Parsing and validating input arguments
2
+ [ WARNING ] -nstreams default value is determined automatically for a device. Although the automatic selection usually provides a reasonable performance, but it still may be non-optimal for some cases, for more information look at README.
3
+ [Step 2/11] Loading OpenVINO
4
+ [ WARNING ] PerformanceMode was not explicitly specified in command line. Device CPU performance hint will be set to THROUGHPUT.
5
+ [ INFO ] OpenVINO:
6
+ API version............. 2022.1.0-6935-7cd3c8e86e9
7
+ [ INFO ] Device info
8
+ CPU
9
+ openvino_intel_cpu_plugin version 2022.1
10
+ Build................... 2022.1.0-6935-7cd3c8e86e9
11
+
12
+ [Step 3/11] Setting device configuration
13
+ [ WARNING ] -nstreams default value is determined automatically for CPU device. Although the automatic selection usually provides a reasonable performance, but it still may be non-optimal for some cases, for more information look at README.
14
+ [Step 4/11] Reading network files
15
+ [ INFO ] Read model took 64.90 ms
16
+ [Step 5/11] Resizing network to match image sizes and given batch
17
+ [ WARNING ] Batch dimension is not specified for input 'input.0'. The first dimension will be interpreted as batch size.
18
+ [ WARNING ] Batch dimension is not specified for input 'input.1'. The first dimension will be interpreted as batch size.
19
+ [ WARNING ] Batch dimension is not specified for input 'input.2'. The first dimension will be interpreted as batch size.
20
+ [ INFO ] Reshaping model: 'input.0': {1,384}, 'input.1': {1,384}, 'input.2': {1,384}
21
+ [ INFO ] Reshape model took 0.05 ms
22
+ [ INFO ] Network batch size: 1
23
+ [Step 6/11] Configuring input of the model
24
+ [ INFO ] Model input 'input.0' precision i64, dimensions ([N,...]): 1 384
25
+ [ INFO ] Model input 'input.1' precision i64, dimensions ([N,...]): 1 384
26
+ [ INFO ] Model input 'input.2' precision i64, dimensions ([N,...]): 1 384
27
+ [ INFO ] Model output 'output.0' precision f32, dimensions ([...]): 1 384
28
+ [ INFO ] Model output 'output.1' precision f32, dimensions ([...]): 1 384
29
+ [Step 7/11] Loading the model to the device
30
+ [ INFO ] Compile model took 788.57 ms
31
+ [Step 8/11] Querying optimal runtime parameters
32
+ [ INFO ] DEVICE: CPU
33
+ [ INFO ] AVAILABLE_DEVICES , ['']
34
+ [ INFO ] RANGE_FOR_ASYNC_INFER_REQUESTS , (1, 1, 1)
35
+ [ INFO ] RANGE_FOR_STREAMS , (1, 152)
36
+ [ INFO ] FULL_DEVICE_NAME , Intel(R) Xeon(R) Platinum 8368 CPU @ 2.40GHz
37
+ [ INFO ] OPTIMIZATION_CAPABILITIES , ['WINOGRAD', 'FP32', 'FP16', 'INT8', 'BIN', 'EXPORT_IMPORT']
38
+ [ INFO ] CACHE_DIR ,
39
+ [ INFO ] NUM_STREAMS , 19
40
+ [ INFO ] INFERENCE_NUM_THREADS , 0
41
+ [ INFO ] PERF_COUNT , False
42
+ [ INFO ] PERFORMANCE_HINT_NUM_REQUESTS , 0
43
+ [Step 9/11] Creating infer requests and preparing input data
44
+ [ INFO ] Create 19 infer requests took 1.50 ms
45
+ [ WARNING ] No input files were given for input 'input.0'!. This input will be filled with random values!
46
+ [ WARNING ] No input files were given for input 'input.1'!. This input will be filled with random values!
47
+ [ WARNING ] No input files were given for input 'input.2'!. This input will be filled with random values!
48
+ [ INFO ] Fill input 'input.0' with random values
49
+ [ INFO ] Fill input 'input.1' with random values
50
+ [ INFO ] Fill input 'input.2' with random values
51
+ [ WARNING ] Number of iterations was aligned by request number from 10000 to 10013 using number of requests 19
52
+ [Step 10/11] Measuring performance (Start inference asynchronously, 19 inference requests using 19 streams for CPU, inference only: True, limits: 10013 iterations)
53
+ [ INFO ] Benchmarking in inference only mode (inputs filling are not included in measurement loop).
54
+ [ INFO ] First inference took 37.43 ms
55
+ [Step 11/11] Dumping statistics report
56
+ Count: 10013 iterations
57
+ Duration: 24302.14 ms
58
+ Latency:
59
+ Median: 43.95 ms
60
+ AVG: 45.97 ms
61
+ MIN: 39.02 ms
62
+ MAX: 126.30 ms
63
+ Throughput: 412.02 FPS
ir/sparsity_structures.csv ADDED
@@ -0,0 +1,73 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ pt_module_name,block_id,orig_w_shape,final_w_shape,orig_b_shape,final_b_shape,prune_by,id_to_keep,head_id_to_keep,nncf_graph_node
2
+ nncf_module.bert.encoder.layer.0.attention.output.dense,0,"(768, 768)","(768, 192)","(768,)","(768,)",group of 64 cols,See pkl,"[3, 8, 10]",BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[0]/BertAttention[attention]/BertSelfOutput[output]/NNCFLinear[dense]/linear_0
3
+ nncf_module.bert.encoder.layer.0.attention.self.query,0,"(768, 768)","(192, 768)","(768,)","(192,)",group of 64 rows,See pkl,"[3, 8, 10]",BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[0]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[query]/linear_0
4
+ nncf_module.bert.encoder.layer.0.attention.self.value,0,"(768, 768)","(192, 768)","(768,)","(192,)",group of 64 rows,See pkl,"[3, 8, 10]",BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[0]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[value]/linear_0
5
+ nncf_module.bert.encoder.layer.0.attention.self.key,0,"(768, 768)","(192, 768)","(768,)","(192,)",group of 64 rows,See pkl,"[3, 8, 10]",BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[0]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[key]/linear_0
6
+ nncf_module.bert.encoder.layer.0.intermediate.dense,1,"(3072, 768)","(2094, 768)","(3072,)","(2094,)",row,See pkl,,BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[0]/BertIntermediate[intermediate]/NNCFLinear[dense]/linear_0
7
+ nncf_module.bert.encoder.layer.0.output.dense,1,"(768, 3072)","(768, 2094)","(768,)","(768,)",col,See pkl,,BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[0]/BertOutput[output]/NNCFLinear[dense]/linear_0
8
+ nncf_module.bert.encoder.layer.1.attention.self.key,2,"(768, 768)","(320, 768)","(768,)","(320,)",group of 64 rows,See pkl,"[1, 4, 7, 10, 11]",BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[1]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[key]/linear_0
9
+ nncf_module.bert.encoder.layer.1.attention.self.value,2,"(768, 768)","(320, 768)","(768,)","(320,)",group of 64 rows,See pkl,"[1, 4, 7, 10, 11]",BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[1]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[value]/linear_0
10
+ nncf_module.bert.encoder.layer.1.attention.output.dense,2,"(768, 768)","(768, 320)","(768,)","(768,)",group of 64 cols,See pkl,"[1, 4, 7, 10, 11]",BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[1]/BertAttention[attention]/BertSelfOutput[output]/NNCFLinear[dense]/linear_0
11
+ nncf_module.bert.encoder.layer.1.attention.self.query,2,"(768, 768)","(320, 768)","(768,)","(320,)",group of 64 rows,See pkl,"[1, 4, 7, 10, 11]",BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[1]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[query]/linear_0
12
+ nncf_module.bert.encoder.layer.1.output.dense,3,"(768, 3072)","(768, 2062)","(768,)","(768,)",col,See pkl,,BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[1]/BertOutput[output]/NNCFLinear[dense]/linear_0
13
+ nncf_module.bert.encoder.layer.1.intermediate.dense,3,"(3072, 768)","(2062, 768)","(3072,)","(2062,)",row,See pkl,,BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[1]/BertIntermediate[intermediate]/NNCFLinear[dense]/linear_0
14
+ nncf_module.bert.encoder.layer.2.attention.self.key,4,"(768, 768)","(768, 768)","(768,)","(768,)",group of 64 rows,See pkl,"[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11]",BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[2]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[key]/linear_0
15
+ nncf_module.bert.encoder.layer.2.attention.output.dense,4,"(768, 768)","(768, 768)","(768,)","(768,)",group of 64 cols,See pkl,"[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11]",BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[2]/BertAttention[attention]/BertSelfOutput[output]/NNCFLinear[dense]/linear_0
16
+ nncf_module.bert.encoder.layer.2.attention.self.value,4,"(768, 768)","(768, 768)","(768,)","(768,)",group of 64 rows,See pkl,"[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11]",BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[2]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[value]/linear_0
17
+ nncf_module.bert.encoder.layer.2.attention.self.query,4,"(768, 768)","(768, 768)","(768,)","(768,)",group of 64 rows,See pkl,"[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11]",BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[2]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[query]/linear_0
18
+ nncf_module.bert.encoder.layer.2.output.dense,5,"(768, 3072)","(768, 2229)","(768,)","(768,)",col,See pkl,,BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[2]/BertOutput[output]/NNCFLinear[dense]/linear_0
19
+ nncf_module.bert.encoder.layer.2.intermediate.dense,5,"(3072, 768)","(2229, 768)","(3072,)","(2229,)",row,See pkl,,BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[2]/BertIntermediate[intermediate]/NNCFLinear[dense]/linear_0
20
+ nncf_module.bert.encoder.layer.3.attention.self.query,6,"(768, 768)","(448, 768)","(768,)","(448,)",group of 64 rows,See pkl,"[0, 1, 5, 8, 9, 10, 11]",BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[3]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[query]/linear_0
21
+ nncf_module.bert.encoder.layer.3.attention.output.dense,6,"(768, 768)","(768, 448)","(768,)","(768,)",group of 64 cols,See pkl,"[0, 1, 5, 8, 9, 10, 11]",BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[3]/BertAttention[attention]/BertSelfOutput[output]/NNCFLinear[dense]/linear_0
22
+ nncf_module.bert.encoder.layer.3.attention.self.value,6,"(768, 768)","(448, 768)","(768,)","(448,)",group of 64 rows,See pkl,"[0, 1, 5, 8, 9, 10, 11]",BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[3]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[value]/linear_0
23
+ nncf_module.bert.encoder.layer.3.attention.self.key,6,"(768, 768)","(448, 768)","(768,)","(448,)",group of 64 rows,See pkl,"[0, 1, 5, 8, 9, 10, 11]",BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[3]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[key]/linear_0
24
+ nncf_module.bert.encoder.layer.3.intermediate.dense,7,"(3072, 768)","(2155, 768)","(3072,)","(2155,)",row,See pkl,,BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[3]/BertIntermediate[intermediate]/NNCFLinear[dense]/linear_0
25
+ nncf_module.bert.encoder.layer.3.output.dense,7,"(768, 3072)","(768, 2155)","(768,)","(768,)",col,See pkl,,BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[3]/BertOutput[output]/NNCFLinear[dense]/linear_0
26
+ nncf_module.bert.encoder.layer.4.attention.output.dense,8,"(768, 768)","(768, 768)","(768,)","(768,)",group of 64 cols,See pkl,"[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11]",BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[4]/BertAttention[attention]/BertSelfOutput[output]/NNCFLinear[dense]/linear_0
27
+ nncf_module.bert.encoder.layer.4.attention.self.query,8,"(768, 768)","(768, 768)","(768,)","(768,)",group of 64 rows,See pkl,"[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11]",BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[4]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[query]/linear_0
28
+ nncf_module.bert.encoder.layer.4.attention.self.key,8,"(768, 768)","(768, 768)","(768,)","(768,)",group of 64 rows,See pkl,"[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11]",BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[4]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[key]/linear_0
29
+ nncf_module.bert.encoder.layer.4.attention.self.value,8,"(768, 768)","(768, 768)","(768,)","(768,)",group of 64 rows,See pkl,"[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11]",BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[4]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[value]/linear_0
30
+ nncf_module.bert.encoder.layer.4.output.dense,9,"(768, 3072)","(768, 1973)","(768,)","(768,)",col,See pkl,,BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[4]/BertOutput[output]/NNCFLinear[dense]/linear_0
31
+ nncf_module.bert.encoder.layer.4.intermediate.dense,9,"(3072, 768)","(1973, 768)","(3072,)","(1973,)",row,See pkl,,BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[4]/BertIntermediate[intermediate]/NNCFLinear[dense]/linear_0
32
+ nncf_module.bert.encoder.layer.5.attention.self.key,10,"(768, 768)","(768, 768)","(768,)","(768,)",group of 64 rows,See pkl,"[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11]",BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[5]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[key]/linear_0
33
+ nncf_module.bert.encoder.layer.5.attention.output.dense,10,"(768, 768)","(768, 768)","(768,)","(768,)",group of 64 cols,See pkl,"[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11]",BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[5]/BertAttention[attention]/BertSelfOutput[output]/NNCFLinear[dense]/linear_0
34
+ nncf_module.bert.encoder.layer.5.attention.self.value,10,"(768, 768)","(768, 768)","(768,)","(768,)",group of 64 rows,See pkl,"[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11]",BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[5]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[value]/linear_0
35
+ nncf_module.bert.encoder.layer.5.attention.self.query,10,"(768, 768)","(768, 768)","(768,)","(768,)",group of 64 rows,See pkl,"[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11]",BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[5]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[query]/linear_0
36
+ nncf_module.bert.encoder.layer.5.intermediate.dense,11,"(3072, 768)","(1947, 768)","(3072,)","(1947,)",row,See pkl,,BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[5]/BertIntermediate[intermediate]/NNCFLinear[dense]/linear_0
37
+ nncf_module.bert.encoder.layer.5.output.dense,11,"(768, 3072)","(768, 1947)","(768,)","(768,)",col,See pkl,,BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[5]/BertOutput[output]/NNCFLinear[dense]/linear_0
38
+ nncf_module.bert.encoder.layer.6.attention.self.value,12,"(768, 768)","(768, 768)","(768,)","(768,)",group of 64 rows,See pkl,"[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11]",BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[6]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[value]/linear_0
39
+ nncf_module.bert.encoder.layer.6.attention.output.dense,12,"(768, 768)","(768, 768)","(768,)","(768,)",group of 64 cols,See pkl,"[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11]",BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[6]/BertAttention[attention]/BertSelfOutput[output]/NNCFLinear[dense]/linear_0
40
+ nncf_module.bert.encoder.layer.6.attention.self.query,12,"(768, 768)","(768, 768)","(768,)","(768,)",group of 64 rows,See pkl,"[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11]",BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[6]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[query]/linear_0
41
+ nncf_module.bert.encoder.layer.6.attention.self.key,12,"(768, 768)","(768, 768)","(768,)","(768,)",group of 64 rows,See pkl,"[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11]",BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[6]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[key]/linear_0
42
+ nncf_module.bert.encoder.layer.6.output.dense,13,"(768, 3072)","(768, 1539)","(768,)","(768,)",col,See pkl,,BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[6]/BertOutput[output]/NNCFLinear[dense]/linear_0
43
+ nncf_module.bert.encoder.layer.6.intermediate.dense,13,"(3072, 768)","(1539, 768)","(3072,)","(1539,)",row,See pkl,,BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[6]/BertIntermediate[intermediate]/NNCFLinear[dense]/linear_0
44
+ nncf_module.bert.encoder.layer.7.attention.self.value,14,"(768, 768)","(768, 768)","(768,)","(768,)",group of 64 rows,See pkl,"[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11]",BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[7]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[value]/linear_0
45
+ nncf_module.bert.encoder.layer.7.attention.self.key,14,"(768, 768)","(768, 768)","(768,)","(768,)",group of 64 rows,See pkl,"[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11]",BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[7]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[key]/linear_0
46
+ nncf_module.bert.encoder.layer.7.attention.self.query,14,"(768, 768)","(768, 768)","(768,)","(768,)",group of 64 rows,See pkl,"[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11]",BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[7]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[query]/linear_0
47
+ nncf_module.bert.encoder.layer.7.attention.output.dense,14,"(768, 768)","(768, 768)","(768,)","(768,)",group of 64 cols,See pkl,"[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11]",BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[7]/BertAttention[attention]/BertSelfOutput[output]/NNCFLinear[dense]/linear_0
48
+ nncf_module.bert.encoder.layer.7.output.dense,15,"(768, 3072)","(768, 1151)","(768,)","(768,)",col,See pkl,,BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[7]/BertOutput[output]/NNCFLinear[dense]/linear_0
49
+ nncf_module.bert.encoder.layer.7.intermediate.dense,15,"(3072, 768)","(1151, 768)","(3072,)","(1151,)",row,See pkl,,BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[7]/BertIntermediate[intermediate]/NNCFLinear[dense]/linear_0
50
+ nncf_module.bert.encoder.layer.8.attention.output.dense,16,"(768, 768)","(768, 384)","(768,)","(768,)",group of 64 cols,See pkl,"[1, 2, 6, 9, 10, 11]",BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[8]/BertAttention[attention]/BertSelfOutput[output]/NNCFLinear[dense]/linear_0
51
+ nncf_module.bert.encoder.layer.8.attention.self.value,16,"(768, 768)","(384, 768)","(768,)","(384,)",group of 64 rows,See pkl,"[1, 2, 6, 9, 10, 11]",BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[8]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[value]/linear_0
52
+ nncf_module.bert.encoder.layer.8.attention.self.query,16,"(768, 768)","(384, 768)","(768,)","(384,)",group of 64 rows,See pkl,"[1, 2, 6, 9, 10, 11]",BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[8]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[query]/linear_0
53
+ nncf_module.bert.encoder.layer.8.attention.self.key,16,"(768, 768)","(384, 768)","(768,)","(384,)",group of 64 rows,See pkl,"[1, 2, 6, 9, 10, 11]",BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[8]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[key]/linear_0
54
+ nncf_module.bert.encoder.layer.8.output.dense,17,"(768, 3072)","(768, 714)","(768,)","(768,)",col,See pkl,,BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[8]/BertOutput[output]/NNCFLinear[dense]/linear_0
55
+ nncf_module.bert.encoder.layer.8.intermediate.dense,17,"(3072, 768)","(714, 768)","(3072,)","(714,)",row,See pkl,,BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[8]/BertIntermediate[intermediate]/NNCFLinear[dense]/linear_0
56
+ nncf_module.bert.encoder.layer.9.attention.self.value,18,"(768, 768)","(320, 768)","(768,)","(320,)",group of 64 rows,See pkl,"[0, 2, 6, 8, 11]",BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[9]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[value]/linear_0
57
+ nncf_module.bert.encoder.layer.9.attention.self.query,18,"(768, 768)","(320, 768)","(768,)","(320,)",group of 64 rows,See pkl,"[0, 2, 6, 8, 11]",BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[9]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[query]/linear_0
58
+ nncf_module.bert.encoder.layer.9.attention.self.key,18,"(768, 768)","(320, 768)","(768,)","(320,)",group of 64 rows,See pkl,"[0, 2, 6, 8, 11]",BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[9]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[key]/linear_0
59
+ nncf_module.bert.encoder.layer.9.attention.output.dense,18,"(768, 768)","(768, 320)","(768,)","(768,)",group of 64 cols,See pkl,"[0, 2, 6, 8, 11]",BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[9]/BertAttention[attention]/BertSelfOutput[output]/NNCFLinear[dense]/linear_0
60
+ nncf_module.bert.encoder.layer.9.output.dense,19,"(768, 3072)","(768, 266)","(768,)","(768,)",col,See pkl,,BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[9]/BertOutput[output]/NNCFLinear[dense]/linear_0
61
+ nncf_module.bert.encoder.layer.9.intermediate.dense,19,"(3072, 768)","(266, 768)","(3072,)","(266,)",row,See pkl,,BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[9]/BertIntermediate[intermediate]/NNCFLinear[dense]/linear_0
62
+ nncf_module.bert.encoder.layer.10.attention.self.query,20,"(768, 768)","(256, 768)","(768,)","(256,)",group of 64 rows,See pkl,"[3, 7, 10, 11]",BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[10]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[query]/linear_0
63
+ nncf_module.bert.encoder.layer.10.attention.self.value,20,"(768, 768)","(256, 768)","(768,)","(256,)",group of 64 rows,See pkl,"[3, 7, 10, 11]",BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[10]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[value]/linear_0
64
+ nncf_module.bert.encoder.layer.10.attention.output.dense,20,"(768, 768)","(768, 256)","(768,)","(768,)",group of 64 cols,See pkl,"[3, 7, 10, 11]",BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[10]/BertAttention[attention]/BertSelfOutput[output]/NNCFLinear[dense]/linear_0
65
+ nncf_module.bert.encoder.layer.10.attention.self.key,20,"(768, 768)","(256, 768)","(768,)","(256,)",group of 64 rows,See pkl,"[3, 7, 10, 11]",BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[10]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[key]/linear_0
66
+ nncf_module.bert.encoder.layer.10.output.dense,21,"(768, 3072)","(768, 297)","(768,)","(768,)",col,See pkl,,BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[10]/BertOutput[output]/NNCFLinear[dense]/linear_0
67
+ nncf_module.bert.encoder.layer.10.intermediate.dense,21,"(3072, 768)","(297, 768)","(3072,)","(297,)",row,See pkl,,BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[10]/BertIntermediate[intermediate]/NNCFLinear[dense]/linear_0
68
+ nncf_module.bert.encoder.layer.11.attention.output.dense,22,"(768, 768)","(768, 256)","(768,)","(768,)",group of 64 cols,See pkl,"[1, 2, 3, 4]",BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[11]/BertAttention[attention]/BertSelfOutput[output]/NNCFLinear[dense]/linear_0
69
+ nncf_module.bert.encoder.layer.11.attention.self.value,22,"(768, 768)","(256, 768)","(768,)","(256,)",group of 64 rows,See pkl,"[1, 2, 3, 4]",BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[11]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[value]/linear_0
70
+ nncf_module.bert.encoder.layer.11.attention.self.key,22,"(768, 768)","(256, 768)","(768,)","(256,)",group of 64 rows,See pkl,"[1, 2, 3, 4]",BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[11]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[key]/linear_0
71
+ nncf_module.bert.encoder.layer.11.attention.self.query,22,"(768, 768)","(256, 768)","(768,)","(256,)",group of 64 rows,See pkl,"[1, 2, 3, 4]",BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[11]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[query]/linear_0
72
+ nncf_module.bert.encoder.layer.11.intermediate.dense,23,"(3072, 768)","(322, 768)","(3072,)","(322,)",row,See pkl,,BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[11]/BertIntermediate[intermediate]/NNCFLinear[dense]/linear_0
73
+ nncf_module.bert.encoder.layer.11.output.dense,23,"(768, 3072)","(768, 322)","(768,)","(768,)",col,See pkl,,BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[11]/BertOutput[output]/NNCFLinear[dense]/linear_0
ir/sparsity_structures.md ADDED
@@ -0,0 +1,74 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ | | pt_module_name | block_id | orig_w_shape | final_w_shape | orig_b_shape | final_b_shape | prune_by | id_to_keep | head_id_to_keep | nncf_graph_node |
2
+ |---:|:---------------------------------------------------------|-----------:|:---------------|:----------------|:---------------|:----------------|:-----------------|:-------------|:---------------------------------------|:--------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
3
+ | 0 | nncf_module.bert.encoder.layer.0.attention.output.dense | 0 | (768, 768) | (768, 192) | (768,) | (768,) | group of 64 cols | See pkl | [3, 8, 10] | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[0]/BertAttention[attention]/BertSelfOutput[output]/NNCFLinear[dense]/linear_0 |
4
+ | 1 | nncf_module.bert.encoder.layer.0.attention.self.query | 0 | (768, 768) | (192, 768) | (768,) | (192,) | group of 64 rows | See pkl | [3, 8, 10] | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[0]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[query]/linear_0 |
5
+ | 2 | nncf_module.bert.encoder.layer.0.attention.self.value | 0 | (768, 768) | (192, 768) | (768,) | (192,) | group of 64 rows | See pkl | [3, 8, 10] | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[0]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[value]/linear_0 |
6
+ | 3 | nncf_module.bert.encoder.layer.0.attention.self.key | 0 | (768, 768) | (192, 768) | (768,) | (192,) | group of 64 rows | See pkl | [3, 8, 10] | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[0]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[key]/linear_0 |
7
+ | 4 | nncf_module.bert.encoder.layer.0.intermediate.dense | 1 | (3072, 768) | (2094, 768) | (3072,) | (2094,) | row | See pkl | | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[0]/BertIntermediate[intermediate]/NNCFLinear[dense]/linear_0 |
8
+ | 5 | nncf_module.bert.encoder.layer.0.output.dense | 1 | (768, 3072) | (768, 2094) | (768,) | (768,) | col | See pkl | | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[0]/BertOutput[output]/NNCFLinear[dense]/linear_0 |
9
+ | 6 | nncf_module.bert.encoder.layer.1.attention.self.key | 2 | (768, 768) | (320, 768) | (768,) | (320,) | group of 64 rows | See pkl | [1, 4, 7, 10, 11] | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[1]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[key]/linear_0 |
10
+ | 7 | nncf_module.bert.encoder.layer.1.attention.self.value | 2 | (768, 768) | (320, 768) | (768,) | (320,) | group of 64 rows | See pkl | [1, 4, 7, 10, 11] | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[1]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[value]/linear_0 |
11
+ | 8 | nncf_module.bert.encoder.layer.1.attention.output.dense | 2 | (768, 768) | (768, 320) | (768,) | (768,) | group of 64 cols | See pkl | [1, 4, 7, 10, 11] | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[1]/BertAttention[attention]/BertSelfOutput[output]/NNCFLinear[dense]/linear_0 |
12
+ | 9 | nncf_module.bert.encoder.layer.1.attention.self.query | 2 | (768, 768) | (320, 768) | (768,) | (320,) | group of 64 rows | See pkl | [1, 4, 7, 10, 11] | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[1]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[query]/linear_0 |
13
+ | 10 | nncf_module.bert.encoder.layer.1.output.dense | 3 | (768, 3072) | (768, 2062) | (768,) | (768,) | col | See pkl | | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[1]/BertOutput[output]/NNCFLinear[dense]/linear_0 |
14
+ | 11 | nncf_module.bert.encoder.layer.1.intermediate.dense | 3 | (3072, 768) | (2062, 768) | (3072,) | (2062,) | row | See pkl | | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[1]/BertIntermediate[intermediate]/NNCFLinear[dense]/linear_0 |
15
+ | 12 | nncf_module.bert.encoder.layer.2.attention.self.key | 4 | (768, 768) | (768, 768) | (768,) | (768,) | group of 64 rows | See pkl | [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11] | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[2]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[key]/linear_0 |
16
+ | 13 | nncf_module.bert.encoder.layer.2.attention.output.dense | 4 | (768, 768) | (768, 768) | (768,) | (768,) | group of 64 cols | See pkl | [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11] | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[2]/BertAttention[attention]/BertSelfOutput[output]/NNCFLinear[dense]/linear_0 |
17
+ | 14 | nncf_module.bert.encoder.layer.2.attention.self.value | 4 | (768, 768) | (768, 768) | (768,) | (768,) | group of 64 rows | See pkl | [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11] | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[2]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[value]/linear_0 |
18
+ | 15 | nncf_module.bert.encoder.layer.2.attention.self.query | 4 | (768, 768) | (768, 768) | (768,) | (768,) | group of 64 rows | See pkl | [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11] | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[2]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[query]/linear_0 |
19
+ | 16 | nncf_module.bert.encoder.layer.2.output.dense | 5 | (768, 3072) | (768, 2229) | (768,) | (768,) | col | See pkl | | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[2]/BertOutput[output]/NNCFLinear[dense]/linear_0 |
20
+ | 17 | nncf_module.bert.encoder.layer.2.intermediate.dense | 5 | (3072, 768) | (2229, 768) | (3072,) | (2229,) | row | See pkl | | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[2]/BertIntermediate[intermediate]/NNCFLinear[dense]/linear_0 |
21
+ | 18 | nncf_module.bert.encoder.layer.3.attention.self.query | 6 | (768, 768) | (448, 768) | (768,) | (448,) | group of 64 rows | See pkl | [0, 1, 5, 8, 9, 10, 11] | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[3]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[query]/linear_0 |
22
+ | 19 | nncf_module.bert.encoder.layer.3.attention.output.dense | 6 | (768, 768) | (768, 448) | (768,) | (768,) | group of 64 cols | See pkl | [0, 1, 5, 8, 9, 10, 11] | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[3]/BertAttention[attention]/BertSelfOutput[output]/NNCFLinear[dense]/linear_0 |
23
+ | 20 | nncf_module.bert.encoder.layer.3.attention.self.value | 6 | (768, 768) | (448, 768) | (768,) | (448,) | group of 64 rows | See pkl | [0, 1, 5, 8, 9, 10, 11] | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[3]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[value]/linear_0 |
24
+ | 21 | nncf_module.bert.encoder.layer.3.attention.self.key | 6 | (768, 768) | (448, 768) | (768,) | (448,) | group of 64 rows | See pkl | [0, 1, 5, 8, 9, 10, 11] | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[3]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[key]/linear_0 |
25
+ | 22 | nncf_module.bert.encoder.layer.3.intermediate.dense | 7 | (3072, 768) | (2155, 768) | (3072,) | (2155,) | row | See pkl | | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[3]/BertIntermediate[intermediate]/NNCFLinear[dense]/linear_0 |
26
+ | 23 | nncf_module.bert.encoder.layer.3.output.dense | 7 | (768, 3072) | (768, 2155) | (768,) | (768,) | col | See pkl | | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[3]/BertOutput[output]/NNCFLinear[dense]/linear_0 |
27
+ | 24 | nncf_module.bert.encoder.layer.4.attention.output.dense | 8 | (768, 768) | (768, 768) | (768,) | (768,) | group of 64 cols | See pkl | [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11] | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[4]/BertAttention[attention]/BertSelfOutput[output]/NNCFLinear[dense]/linear_0 |
28
+ | 25 | nncf_module.bert.encoder.layer.4.attention.self.query | 8 | (768, 768) | (768, 768) | (768,) | (768,) | group of 64 rows | See pkl | [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11] | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[4]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[query]/linear_0 |
29
+ | 26 | nncf_module.bert.encoder.layer.4.attention.self.key | 8 | (768, 768) | (768, 768) | (768,) | (768,) | group of 64 rows | See pkl | [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11] | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[4]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[key]/linear_0 |
30
+ | 27 | nncf_module.bert.encoder.layer.4.attention.self.value | 8 | (768, 768) | (768, 768) | (768,) | (768,) | group of 64 rows | See pkl | [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11] | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[4]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[value]/linear_0 |
31
+ | 28 | nncf_module.bert.encoder.layer.4.output.dense | 9 | (768, 3072) | (768, 1973) | (768,) | (768,) | col | See pkl | | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[4]/BertOutput[output]/NNCFLinear[dense]/linear_0 |
32
+ | 29 | nncf_module.bert.encoder.layer.4.intermediate.dense | 9 | (3072, 768) | (1973, 768) | (3072,) | (1973,) | row | See pkl | | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[4]/BertIntermediate[intermediate]/NNCFLinear[dense]/linear_0 |
33
+ | 30 | nncf_module.bert.encoder.layer.5.attention.self.key | 10 | (768, 768) | (768, 768) | (768,) | (768,) | group of 64 rows | See pkl | [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11] | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[5]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[key]/linear_0 |
34
+ | 31 | nncf_module.bert.encoder.layer.5.attention.output.dense | 10 | (768, 768) | (768, 768) | (768,) | (768,) | group of 64 cols | See pkl | [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11] | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[5]/BertAttention[attention]/BertSelfOutput[output]/NNCFLinear[dense]/linear_0 |
35
+ | 32 | nncf_module.bert.encoder.layer.5.attention.self.value | 10 | (768, 768) | (768, 768) | (768,) | (768,) | group of 64 rows | See pkl | [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11] | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[5]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[value]/linear_0 |
36
+ | 33 | nncf_module.bert.encoder.layer.5.attention.self.query | 10 | (768, 768) | (768, 768) | (768,) | (768,) | group of 64 rows | See pkl | [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11] | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[5]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[query]/linear_0 |
37
+ | 34 | nncf_module.bert.encoder.layer.5.intermediate.dense | 11 | (3072, 768) | (1947, 768) | (3072,) | (1947,) | row | See pkl | | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[5]/BertIntermediate[intermediate]/NNCFLinear[dense]/linear_0 |
38
+ | 35 | nncf_module.bert.encoder.layer.5.output.dense | 11 | (768, 3072) | (768, 1947) | (768,) | (768,) | col | See pkl | | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[5]/BertOutput[output]/NNCFLinear[dense]/linear_0 |
39
+ | 36 | nncf_module.bert.encoder.layer.6.attention.self.value | 12 | (768, 768) | (768, 768) | (768,) | (768,) | group of 64 rows | See pkl | [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11] | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[6]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[value]/linear_0 |
40
+ | 37 | nncf_module.bert.encoder.layer.6.attention.output.dense | 12 | (768, 768) | (768, 768) | (768,) | (768,) | group of 64 cols | See pkl | [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11] | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[6]/BertAttention[attention]/BertSelfOutput[output]/NNCFLinear[dense]/linear_0 |
41
+ | 38 | nncf_module.bert.encoder.layer.6.attention.self.query | 12 | (768, 768) | (768, 768) | (768,) | (768,) | group of 64 rows | See pkl | [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11] | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[6]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[query]/linear_0 |
42
+ | 39 | nncf_module.bert.encoder.layer.6.attention.self.key | 12 | (768, 768) | (768, 768) | (768,) | (768,) | group of 64 rows | See pkl | [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11] | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[6]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[key]/linear_0 |
43
+ | 40 | nncf_module.bert.encoder.layer.6.output.dense | 13 | (768, 3072) | (768, 1539) | (768,) | (768,) | col | See pkl | | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[6]/BertOutput[output]/NNCFLinear[dense]/linear_0 |
44
+ | 41 | nncf_module.bert.encoder.layer.6.intermediate.dense | 13 | (3072, 768) | (1539, 768) | (3072,) | (1539,) | row | See pkl | | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[6]/BertIntermediate[intermediate]/NNCFLinear[dense]/linear_0 |
45
+ | 42 | nncf_module.bert.encoder.layer.7.attention.self.value | 14 | (768, 768) | (768, 768) | (768,) | (768,) | group of 64 rows | See pkl | [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11] | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[7]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[value]/linear_0 |
46
+ | 43 | nncf_module.bert.encoder.layer.7.attention.self.key | 14 | (768, 768) | (768, 768) | (768,) | (768,) | group of 64 rows | See pkl | [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11] | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[7]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[key]/linear_0 |
47
+ | 44 | nncf_module.bert.encoder.layer.7.attention.self.query | 14 | (768, 768) | (768, 768) | (768,) | (768,) | group of 64 rows | See pkl | [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11] | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[7]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[query]/linear_0 |
48
+ | 45 | nncf_module.bert.encoder.layer.7.attention.output.dense | 14 | (768, 768) | (768, 768) | (768,) | (768,) | group of 64 cols | See pkl | [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11] | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[7]/BertAttention[attention]/BertSelfOutput[output]/NNCFLinear[dense]/linear_0 |
49
+ | 46 | nncf_module.bert.encoder.layer.7.output.dense | 15 | (768, 3072) | (768, 1151) | (768,) | (768,) | col | See pkl | | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[7]/BertOutput[output]/NNCFLinear[dense]/linear_0 |
50
+ | 47 | nncf_module.bert.encoder.layer.7.intermediate.dense | 15 | (3072, 768) | (1151, 768) | (3072,) | (1151,) | row | See pkl | | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[7]/BertIntermediate[intermediate]/NNCFLinear[dense]/linear_0 |
51
+ | 48 | nncf_module.bert.encoder.layer.8.attention.output.dense | 16 | (768, 768) | (768, 384) | (768,) | (768,) | group of 64 cols | See pkl | [1, 2, 6, 9, 10, 11] | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[8]/BertAttention[attention]/BertSelfOutput[output]/NNCFLinear[dense]/linear_0 |
52
+ | 49 | nncf_module.bert.encoder.layer.8.attention.self.value | 16 | (768, 768) | (384, 768) | (768,) | (384,) | group of 64 rows | See pkl | [1, 2, 6, 9, 10, 11] | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[8]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[value]/linear_0 |
53
+ | 50 | nncf_module.bert.encoder.layer.8.attention.self.query | 16 | (768, 768) | (384, 768) | (768,) | (384,) | group of 64 rows | See pkl | [1, 2, 6, 9, 10, 11] | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[8]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[query]/linear_0 |
54
+ | 51 | nncf_module.bert.encoder.layer.8.attention.self.key | 16 | (768, 768) | (384, 768) | (768,) | (384,) | group of 64 rows | See pkl | [1, 2, 6, 9, 10, 11] | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[8]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[key]/linear_0 |
55
+ | 52 | nncf_module.bert.encoder.layer.8.output.dense | 17 | (768, 3072) | (768, 714) | (768,) | (768,) | col | See pkl | | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[8]/BertOutput[output]/NNCFLinear[dense]/linear_0 |
56
+ | 53 | nncf_module.bert.encoder.layer.8.intermediate.dense | 17 | (3072, 768) | (714, 768) | (3072,) | (714,) | row | See pkl | | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[8]/BertIntermediate[intermediate]/NNCFLinear[dense]/linear_0 |
57
+ | 54 | nncf_module.bert.encoder.layer.9.attention.self.value | 18 | (768, 768) | (320, 768) | (768,) | (320,) | group of 64 rows | See pkl | [0, 2, 6, 8, 11] | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[9]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[value]/linear_0 |
58
+ | 55 | nncf_module.bert.encoder.layer.9.attention.self.query | 18 | (768, 768) | (320, 768) | (768,) | (320,) | group of 64 rows | See pkl | [0, 2, 6, 8, 11] | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[9]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[query]/linear_0 |
59
+ | 56 | nncf_module.bert.encoder.layer.9.attention.self.key | 18 | (768, 768) | (320, 768) | (768,) | (320,) | group of 64 rows | See pkl | [0, 2, 6, 8, 11] | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[9]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[key]/linear_0 |
60
+ | 57 | nncf_module.bert.encoder.layer.9.attention.output.dense | 18 | (768, 768) | (768, 320) | (768,) | (768,) | group of 64 cols | See pkl | [0, 2, 6, 8, 11] | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[9]/BertAttention[attention]/BertSelfOutput[output]/NNCFLinear[dense]/linear_0 |
61
+ | 58 | nncf_module.bert.encoder.layer.9.output.dense | 19 | (768, 3072) | (768, 266) | (768,) | (768,) | col | See pkl | | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[9]/BertOutput[output]/NNCFLinear[dense]/linear_0 |
62
+ | 59 | nncf_module.bert.encoder.layer.9.intermediate.dense | 19 | (3072, 768) | (266, 768) | (3072,) | (266,) | row | See pkl | | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[9]/BertIntermediate[intermediate]/NNCFLinear[dense]/linear_0 |
63
+ | 60 | nncf_module.bert.encoder.layer.10.attention.self.query | 20 | (768, 768) | (256, 768) | (768,) | (256,) | group of 64 rows | See pkl | [3, 7, 10, 11] | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[10]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[query]/linear_0 |
64
+ | 61 | nncf_module.bert.encoder.layer.10.attention.self.value | 20 | (768, 768) | (256, 768) | (768,) | (256,) | group of 64 rows | See pkl | [3, 7, 10, 11] | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[10]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[value]/linear_0 |
65
+ | 62 | nncf_module.bert.encoder.layer.10.attention.output.dense | 20 | (768, 768) | (768, 256) | (768,) | (768,) | group of 64 cols | See pkl | [3, 7, 10, 11] | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[10]/BertAttention[attention]/BertSelfOutput[output]/NNCFLinear[dense]/linear_0 |
66
+ | 63 | nncf_module.bert.encoder.layer.10.attention.self.key | 20 | (768, 768) | (256, 768) | (768,) | (256,) | group of 64 rows | See pkl | [3, 7, 10, 11] | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[10]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[key]/linear_0 |
67
+ | 64 | nncf_module.bert.encoder.layer.10.output.dense | 21 | (768, 3072) | (768, 297) | (768,) | (768,) | col | See pkl | | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[10]/BertOutput[output]/NNCFLinear[dense]/linear_0 |
68
+ | 65 | nncf_module.bert.encoder.layer.10.intermediate.dense | 21 | (3072, 768) | (297, 768) | (3072,) | (297,) | row | See pkl | | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[10]/BertIntermediate[intermediate]/NNCFLinear[dense]/linear_0 |
69
+ | 66 | nncf_module.bert.encoder.layer.11.attention.output.dense | 22 | (768, 768) | (768, 256) | (768,) | (768,) | group of 64 cols | See pkl | [1, 2, 3, 4] | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[11]/BertAttention[attention]/BertSelfOutput[output]/NNCFLinear[dense]/linear_0 |
70
+ | 67 | nncf_module.bert.encoder.layer.11.attention.self.value | 22 | (768, 768) | (256, 768) | (768,) | (256,) | group of 64 rows | See pkl | [1, 2, 3, 4] | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[11]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[value]/linear_0 |
71
+ | 68 | nncf_module.bert.encoder.layer.11.attention.self.key | 22 | (768, 768) | (256, 768) | (768,) | (256,) | group of 64 rows | See pkl | [1, 2, 3, 4] | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[11]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[key]/linear_0 |
72
+ | 69 | nncf_module.bert.encoder.layer.11.attention.self.query | 22 | (768, 768) | (256, 768) | (768,) | (256,) | group of 64 rows | See pkl | [1, 2, 3, 4] | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[11]/BertAttention[attention]/BertSelfAttention[self]/NNCFLinear[query]/linear_0 |
73
+ | 70 | nncf_module.bert.encoder.layer.11.intermediate.dense | 23 | (3072, 768) | (322, 768) | (3072,) | (322,) | row | See pkl | | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[11]/BertIntermediate[intermediate]/NNCFLinear[dense]/linear_0 |
74
+ | 71 | nncf_module.bert.encoder.layer.11.output.dense | 23 | (768, 3072) | (768, 322) | (768,) | (768,) | col | See pkl | | BertForQuestionAnswering/BertModel[bert]/BertEncoder[encoder]/ModuleList[layer]/BertLayer[11]/BertOutput[output]/NNCFLinear[dense]/linear_0 |
ir/sparsity_structures.pkl ADDED
Binary file (185 kB). View file
 
ir/squad-BertForQuestionAnswering.cropped.8bit.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6865b9646ebc6413b581926390f088cf95b4fe03584cf834fdd452f6676e7255
3
+ size 68417348
ir/squad-BertForQuestionAnswering.cropped.8bit.mapping ADDED
The diff for this file is too large to render. See raw diff
 
ir/squad-BertForQuestionAnswering.cropped.8bit.onnx ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:dae06d09997062857b9d9e381d0e8304cc8ecb2f1263cec5b2b53a1181d54d4d
3
+ size 272689860
ir/squad-BertForQuestionAnswering.cropped.8bit.xml ADDED
The diff for this file is too large to render. See raw diff
 
nncf-mvmt-p3.json ADDED
@@ -0,0 +1,67 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "input_info": [
3
+ {
4
+ "sample_size": [1, 384],
5
+ "type": "long"
6
+ },
7
+ {
8
+ "sample_size": [1, 384],
9
+ "type": "long"
10
+ },
11
+ {
12
+ "sample_size": [1, 384],
13
+ "type": "long"
14
+ }
15
+ ],
16
+ "compression":
17
+ [
18
+ {
19
+ "algorithm": "movement_sparsity",
20
+ "params": {
21
+ "schedule": "threshold_polynomial_decay",
22
+ "power": 3,
23
+ "init_importance_threshold": 0.0,
24
+ "final_importance_threshold": 0.1,
25
+ "warmup_start_epoch": 1,
26
+ "warmup_end_epoch": 10,
27
+ "steps_per_epoch": 5533,
28
+ "importance_regularization_factor": 0.010,
29
+ "update_per_optimizer_step": true,
30
+ },
31
+ "sparse_structure_by_scopes": [
32
+ ["block", [32, 32], "{re}.*BertAttention*"],
33
+ ["per_dim", [0], "{re}.*BertIntermediate*"],
34
+ ["per_dim", [1], "{re}.*BertOutput*"]
35
+ ],
36
+ "ignored_scopes": ["{re}.*NNCFEmbedding", "{re}.*qa_outputs*"]
37
+ },
38
+ {
39
+ "algorithm": "quantization",
40
+ "initializer": {
41
+ "range": {
42
+ "num_init_samples": 32,
43
+ "type": "percentile",
44
+ "params":
45
+ {
46
+ "min_percentile": 0.01,
47
+ "max_percentile": 99.99
48
+ }
49
+ },
50
+
51
+ "batchnorm_adaptation": {
52
+ "num_bn_adaptation_samples": 200
53
+ }
54
+ },
55
+ "activations":
56
+ {
57
+ "mode": "symmetric"
58
+ },
59
+ "weights":
60
+ {
61
+ "mode": "symmetric",
62
+ "signed": true,
63
+ "per_channel": false
64
+ }
65
+ }
66
+ ]
67
+ }
original_graph.dot ADDED
The diff for this file is too large to render. See raw diff