MDMM hardware-aware metrics (FPGA + PACA), tests and docs #46
There are no files selected for viewing
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,72 @@ | ||
| # file: /src/pquant/configs/config_mdmm_fpga.yaml | ||
| # MDMM pruning with the hardware-aware FPGA metric (DSP/BRAM resource grouping). | ||
|
|
||
| pruning_parameters: | ||
| pruning_method: mdmm | ||
| enable_pruning: true | ||
| disable_pruning_for_layers: [] # Disable pruning for these layers, even if enable_pruning is true | ||
| constraint_type: "Equality" | ||
| target_value: 0.0 | ||
| metric_type: "FPGAAwareSparsity" | ||
| target_sparsity: 0.9 | ||
| rf: 4 | ||
| epsilon: 1.0e-03 | ||
| scale: 50.0 | ||
| damping: 1.0 | ||
| use_grad: false | ||
| constraint_lr: 1.0e-3 | ||
| # FPGAAwareSparsityMetric-specific | ||
| precision: 16 # weight bit-width | ||
| target_resource: "DSP" # "DSP" or "BRAM" | ||
| bram_width: 36 # BRAM word width (only used when target_resource == "BRAM") | ||
|
|
||
| quantization_parameters: | ||
| enable_quantization: true | ||
| default_weight_keep_negatives: 1. | ||
| default_weight_integer_bits: 0. | ||
| default_weight_fractional_bits: 7. | ||
| default_data_keep_negatives: 0. | ||
| default_data_integer_bits: 0. | ||
| default_data_fractional_bits: 7. | ||
| dynamic_data_quantization: false | ||
| granularity: "per_tensor" | ||
| quantize_input: true | ||
| quantize_output: false | ||
| hgq_beta: 1e-5 | ||
| hgq_gamma: 0.0003 | ||
| hgq_heterogeneous: True | ||
| layer_specific: {} | ||
| use_high_granularity_quantization: false | ||
| use_real_tanh: false | ||
| use_relu_multiplier: false | ||
| use_symmetric_quantization: false | ||
| overflow_mode_parameters: SAT | ||
| overflow_mode_data: SAT | ||
| round_mode: RND | ||
| training_parameters: | ||
| epochs: 200 | ||
| fine_tuning_epochs: 30 | ||
| pretraining_epochs: 0 | ||
| pruning_first: true | ||
| rewind: never | ||
| rounds: 1 | ||
| save_weights_epoch: -1 | ||
| fitcompress_parameters: | ||
| enable_fitcompress : false | ||
| optimize_quantization : true | ||
| quantization_schedule : [7.,4.,3.,2] | ||
| pruning_schedule : {start : 0, end : -3, steps : 40} | ||
| compression_goal : 0.10 | ||
| optimize_pruning : false | ||
| greedy_astar : true | ||
| approximate : true | ||
| f_lambda : 1 | ||
| hpo_parameters: | ||
| experiment_name: experiment_name | ||
| model_name: jet_tagger | ||
| num_trials: 1 | ||
| sampler: | ||
| type: RandomSampler | ||
| hyperparameter_search: | ||
| numerical: {} | ||
| categorical: {} | ||
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,74 @@ | ||
| # file: /src/pquant/configs/config_mdmm_paca.yaml | ||
| # MDMM pruning with the PACA pattern metric. PACA always pairs with an equality | ||
| # constraint at target 0 (enforced by the config model); the values below are listed | ||
| # for clarity. | ||
|
|
||
| pruning_parameters: | ||
| pruning_method: mdmm | ||
| enable_pruning: true | ||
| disable_pruning_for_layers: [] # Disable pruning for these layers, even if enable_pruning is true | ||
|
Collaborator
There was a problem hiding this comment. same, too much redundant parameters
Contributor
Author
There was a problem hiding this comment. Same as above |
||
| constraint_type: "Equality" # Forced to Equality/0 for PACA by MDMMPruningModel. | ||
| target_value: 0.0 | ||
| metric_type: "PACAPatternSparsity" | ||
| target_sparsity: 0.9 | ||
| rf: 1 | ||
| epsilon: 1.0e-03 | ||
| scale: 1.0 | ||
| damping: 1.0 | ||
| use_grad: false | ||
| constraint_lr: 1.0e-3 | ||
| # PACAPatternMetric-specific | ||
|
Collaborator
There was a problem hiding this comment. Clarification about PACAPatternMetric-specific values; are these values result in the best results?
Contributor
Author
There was a problem hiding this comment. They were the values (hand-tuned) that worked best for the SmartPixels. Do you think it would be better to scan for the best hyperparameters? If Yes, Kindly let me know if you also have any dataset in mind, that you would like this to be tuned to.
Collaborator
There was a problem hiding this comment. I would say, that in this situation it would be nice just to leave the best values based on your point of view. If they are the same, leave them as it's. |
||
| num_patterns_to_keep: 16 | ||
| beta: 0.85 # cumulative pattern-frequency coverage kept (0..1) | ||
| distance_metric: "cosine" # "hamming", "valued_hamming", or "cosine" | ||
|
|
||
| quantization_parameters: | ||
| enable_quantization: true | ||
| default_weight_keep_negatives: 1. | ||
| default_weight_integer_bits: 0. | ||
| default_weight_fractional_bits: 7. | ||
| default_data_keep_negatives: 0. | ||
| default_data_integer_bits: 0. | ||
| default_data_fractional_bits: 7. | ||
| dynamic_data_quantization: false | ||
| granularity: "per_tensor" | ||
| quantize_input: true | ||
| quantize_output: false | ||
| hgq_beta: 1e-5 | ||
| hgq_gamma: 0.0003 | ||
| hgq_heterogeneous: True | ||
| layer_specific: {} | ||
| use_high_granularity_quantization: false | ||
| use_real_tanh: false | ||
| use_relu_multiplier: false | ||
| use_symmetric_quantization: false | ||
| overflow_mode_parameters: SAT | ||
| overflow_mode_data: SAT | ||
| round_mode: RND | ||
| training_parameters: | ||
| epochs: 200 | ||
| fine_tuning_epochs: 30 | ||
| pretraining_epochs: 0 | ||
| pruning_first: true | ||
| rewind: never | ||
| rounds: 1 | ||
| save_weights_epoch: -1 | ||
| fitcompress_parameters: | ||
| enable_fitcompress : false | ||
| optimize_quantization : true | ||
| quantization_schedule : [7.,4.,3.,2] | ||
| pruning_schedule : {start : 0, end : -3, steps : 40} | ||
| compression_goal : 0.10 | ||
| optimize_pruning : false | ||
| greedy_astar : true | ||
| approximate : true | ||
| f_lambda : 1 | ||
| hpo_parameters: | ||
| experiment_name: experiment_name | ||
| model_name: jet_tagger | ||
| num_trials: 1 | ||
| sampler: | ||
| type: RandomSampler | ||
| hyperparameter_search: | ||
| numerical: {} | ||
| categorical: {} | ||
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Oops, something went wrong.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
too much redundant parameters in config files
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Sure Happy to trim them. these two were clones from the
config_mdmm.yamlto stay consistent and trimming them might make it a bit confusing for the user (ommited = default or ommited = forgotten). Do you think we should trim just these two or all of them? the minimal versions should be ~15 linesThere was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Maybe we can make separate variation of files, including different mdmm parameters for different settings? All others blocks (training, hpo, quantization) should stay the same within the files.