Hugging Face's logo

Datasets:
ncgc
/
influence_results

Dataset card Files
xet
 Community
ncgc commited on
Commit
ca29b24
·
verified ·
1 Parent(s): d3a9392

Add files using upload-large-folder tool

Browse files
Files changed (50) hide show
  1. bias_statichh-Qwen-1.5B-sft-bf16/query_grads_pure_ekfac_bottom100/aggregated_query_grads.safetensors +3 -0
  2. bias_statichh-Qwen-1.5B-sft-bf16/query_grads_pure_ekfac_bottom100/timing.json +8 -0
  3. bias_statichh-Qwen-3B-sft-bf16/query_grads_pure_ekfac_bottom100/timing.json +8 -0
  4. bias_statichh-Qwen-7B-sft-bf16/factors_ekfac_half/activation_covariance.safetensors +3 -0
  5. bias_statichh-Qwen-7B-sft-bf16/factors_ekfac_half/activation_eigenvalues.safetensors +3 -0
  6. bias_statichh-Qwen-7B-sft-bf16/factors_ekfac_half/activation_eigenvectors.safetensors +3 -0
  7. bias_statichh-Qwen-7B-sft-bf16/factors_ekfac_half/covariance_dataset_metadata.json +5 -0
  8. bias_statichh-Qwen-7B-sft-bf16/factors_ekfac_half/factor_arguments.json +20 -0
  9. bias_statichh-Qwen-7B-sft-bf16/factors_ekfac_half/gradient_eigenvalues.safetensors +3 -0
  10. bias_statichh-Qwen-7B-sft-bf16/factors_ekfac_half/lambda_dataset_metadata.json +5 -0
  11. bias_statichh-Qwen-7B-sft-bf16/factors_ekfac_half/lambda_matrix.safetensors +3 -0
  12. bias_statichh-Qwen-7B-sft-bf16/factors_ekfac_half/num_activation_covariance_processed.safetensors +3 -0
  13. bias_statichh-Qwen-7B-sft-bf16/factors_ekfac_half/num_gradient_covariance_processed.safetensors +3 -0
  14. bias_statichh-Qwen-7B-sft-bf16/factors_ekfac_half/num_lambda_processed.safetensors +3 -0
  15. bias_statichh-Qwen-7B-sft-bf16/query_grads_pure_ekfac_bottom100/aggregated_query_grads.safetensors +3 -0
  16. bias_statichh-Qwen-7B-sft-bf16/query_grads_pure_ekfac_bottom100/timing.json +8 -0
  17. bias_statichh-Qwen-7B-sft-bf16/scores_ekfac_half/factor_arguments.json +20 -0
  18. bias_statichh-Qwen-7B-sft-bf16/scores_ekfac_half/pairwise_scores.safetensors +3 -0
  19. bias_statichh-Qwen-7B-sft-bf16/scores_ekfac_half/query_dataset_metadata.json +5 -0
  20. bias_statichh-Qwen-7B-sft-bf16/scores_ekfac_half/score_arguments.json +19 -0
  21. bias_statichh-Qwen-7B-sft-bf16/scores_ekfac_half/train_dataset_metadata.json +5 -0
  22. truth_statichh-Qwen-1.5B-sft-bf16/scores_ekfac_half/pairwise_scores.safetensors +3 -0
  23. truth_statichh-Qwen-3B-sft-bf16/factors_ekfac_half/activation_covariance.safetensors +3 -0
  24. truth_statichh-Qwen-3B-sft-bf16/factors_ekfac_half/activation_eigenvalues.safetensors +3 -0
  25. truth_statichh-Qwen-3B-sft-bf16/factors_ekfac_half/activation_eigenvectors.safetensors +3 -0
  26. truth_statichh-Qwen-3B-sft-bf16/factors_ekfac_half/covariance_dataset_metadata.json +5 -0
  27. truth_statichh-Qwen-3B-sft-bf16/factors_ekfac_half/factor_arguments.json +20 -0
  28. truth_statichh-Qwen-3B-sft-bf16/factors_ekfac_half/gradient_covariance.safetensors +3 -0
  29. truth_statichh-Qwen-3B-sft-bf16/factors_ekfac_half/gradient_eigenvalues.safetensors +3 -0
  30. truth_statichh-Qwen-3B-sft-bf16/factors_ekfac_half/gradient_eigenvectors.safetensors +3 -0
  31. truth_statichh-Qwen-3B-sft-bf16/factors_ekfac_half/lambda_dataset_metadata.json +5 -0
  32. truth_statichh-Qwen-3B-sft-bf16/factors_ekfac_half/lambda_matrix.safetensors +3 -0
  33. truth_statichh-Qwen-3B-sft-bf16/factors_ekfac_half/num_activation_covariance_processed.safetensors +3 -0
  34. truth_statichh-Qwen-3B-sft-bf16/factors_ekfac_half/num_gradient_covariance_processed.safetensors +3 -0
  35. truth_statichh-Qwen-3B-sft-bf16/factors_ekfac_half/num_lambda_processed.safetensors +3 -0
  36. truth_statichh-Qwen-3B-sft-bf16/scores_ekfac_half/factor_arguments.json +20 -0
  37. truth_statichh-Qwen-3B-sft-bf16/scores_ekfac_half/pairwise_scores.safetensors +3 -0
  38. truth_statichh-Qwen-3B-sft-bf16/scores_ekfac_half/query_dataset_metadata.json +5 -0
  39. truth_statichh-Qwen-3B-sft-bf16/scores_ekfac_half/score_arguments.json +19 -0
  40. truth_statichh-Qwen-3B-sft-bf16/scores_ekfac_half/train_dataset_metadata.json +5 -0
  41. truth_statichh-Qwen-7B-sft-bf16/factors_ekfac_half/covariance_dataset_metadata.json +5 -0
  42. truth_statichh-Qwen-7B-sft-bf16/factors_ekfac_half/factor_arguments.json +20 -0
  43. truth_statichh-Qwen-7B-sft-bf16/factors_ekfac_half/lambda_dataset_metadata.json +5 -0
  44. truth_statichh-Qwen-7B-sft-bf16/factors_ekfac_half/num_gradient_covariance_processed.safetensors +3 -0
  45. truth_statichh-Qwen-7B-sft-bf16/factors_ekfac_half/num_lambda_processed.safetensors +3 -0
  46. truth_statichh-Qwen-7B-sft-bf16/scores_ekfac_half/factor_arguments.json +20 -0
  47. truth_statichh-Qwen-7B-sft-bf16/scores_ekfac_half/pairwise_scores.safetensors +3 -0
  48. truth_statichh-Qwen-7B-sft-bf16/scores_ekfac_half/query_dataset_metadata.json +5 -0
  49. truth_statichh-Qwen-7B-sft-bf16/scores_ekfac_half/score_arguments.json +19 -0
  50. truth_statichh-Qwen-7B-sft-bf16/scores_ekfac_half/train_dataset_metadata.json +5 -0
bias_statichh-Qwen-1.5B-sft-bf16/query_grads_pure_ekfac_bottom100/aggregated_query_grads.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:92a64ca41af7bac256c6a47766676ed45d8d300486bd80b6f7b986bdcb84ab3c
3
+ size 187179768
bias_statichh-Qwen-1.5B-sft-bf16/query_grads_pure_ekfac_bottom100/timing.json ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model_name": "ncgc/statichh-Qwen-1.5B-sft-bf16",
3
+ "analysis_name": "bias_statichh-Qwen-1.5B-sft-bf16",
4
+ "scores_type": "pure_ekfac",
5
+ "bottom_k": 100,
6
+ "num_indices": 100,
7
+ "elapsed_seconds": 9.501603841781616
8
+ }
bias_statichh-Qwen-3B-sft-bf16/query_grads_pure_ekfac_bottom100/timing.json ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model_name": "ncgc/statichh-Qwen-3B-sft-bf16",
3
+ "analysis_name": "bias_statichh-Qwen-3B-sft-bf16",
4
+ "scores_type": "pure_ekfac",
5
+ "bottom_k": 100,
6
+ "num_indices": 100,
7
+ "elapsed_seconds": 15.598024845123291
8
+ }
bias_statichh-Qwen-7B-sft-bf16/factors_ekfac_half/activation_covariance.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:aea8746053de39e61c391cd20dfe0a3cd1ff69d82013b6ee9cf5bbc001e0f800
3
+ size 871935350
bias_statichh-Qwen-7B-sft-bf16/factors_ekfac_half/activation_eigenvalues.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e95f1dd5e5c9409400d91011d3d9bc1b0c78a5d1471bd2b9d7aa758e25066d82
3
+ size 82206
bias_statichh-Qwen-7B-sft-bf16/factors_ekfac_half/activation_eigenvectors.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9334e5da4d592b84f6df92359ee0057d7c1430da1456f1f327b2831032c71f0f
3
+ size 871935350
bias_statichh-Qwen-7B-sft-bf16/factors_ekfac_half/covariance_dataset_metadata.json ADDED
@@ -0,0 +1,5 @@
 
 
 
 
 
 
1
+ {
2
+ "type": "Dataset",
3
+ "dataset_size": 10000,
4
+ "indices": null
5
+ }
bias_statichh-Qwen-7B-sft-bf16/factors_ekfac_half/factor_arguments.json ADDED
@@ -0,0 +1,20 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "strategy": "ekfac",
3
+ "use_empirical_fisher": false,
4
+ "amp_dtype": "torch.bfloat16",
5
+ "amp_scale": 65536.0,
6
+ "has_shared_parameters": false,
7
+ "covariance_max_examples": 100000,
8
+ "covariance_data_partitions": 1,
9
+ "covariance_module_partitions": 1,
10
+ "activation_covariance_dtype": "torch.bfloat16",
11
+ "gradient_covariance_dtype": "torch.bfloat16",
12
+ "eigendecomposition_dtype": "torch.float64",
13
+ "lambda_max_examples": 100000,
14
+ "lambda_data_partitions": 1,
15
+ "lambda_module_partitions": 1,
16
+ "use_iterative_lambda_aggregation": false,
17
+ "offload_activations_to_cpu": false,
18
+ "per_sample_gradient_dtype": "torch.bfloat16",
19
+ "lambda_dtype": "torch.bfloat16"
20
+ }
bias_statichh-Qwen-7B-sft-bf16/factors_ekfac_half/gradient_eigenvalues.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:caf036be2184c264dc1a4aebfc59064a0cf145704eb64742c933d5b2b90e9401
3
+ size 100632
bias_statichh-Qwen-7B-sft-bf16/factors_ekfac_half/lambda_dataset_metadata.json ADDED
@@ -0,0 +1,5 @@
 
 
 
 
 
 
1
+ {
2
+ "type": "Dataset",
3
+ "dataset_size": 10000,
4
+ "indices": null
5
+ }
bias_statichh-Qwen-7B-sft-bf16/factors_ekfac_half/lambda_matrix.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:69c38d9ab3b00facc702183dae108a13140bfcac269649ee7ea7c8eb60de6f1e
3
+ size 466102640
bias_statichh-Qwen-7B-sft-bf16/factors_ekfac_half/num_activation_covariance_processed.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:53c565bf5e98453c684c097ca5098f693b98d0f5557cf78d74ec7220309b749e
3
+ size 1296
bias_statichh-Qwen-7B-sft-bf16/factors_ekfac_half/num_gradient_covariance_processed.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2333befb827abcf272cefbfc25f10b54d4f784a3f9153b9343621322893217c1
3
+ size 1296
bias_statichh-Qwen-7B-sft-bf16/factors_ekfac_half/num_lambda_processed.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:06e2b0ac1580a94cd1409abb85e1545b81ad0e25dbaa7cf33e023e19d8c21053
3
+ size 1296
bias_statichh-Qwen-7B-sft-bf16/query_grads_pure_ekfac_bottom100/aggregated_query_grads.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c9034407d5128f9caecad09473032e880271419d9eb948e94fd2e87b050ab25e
3
+ size 932203264
bias_statichh-Qwen-7B-sft-bf16/query_grads_pure_ekfac_bottom100/timing.json ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model_name": "ncgc/statichh-Qwen-7B-sft-bf16",
3
+ "analysis_name": "bias_statichh-Qwen-7B-sft-bf16",
4
+ "scores_type": "pure_ekfac",
5
+ "bottom_k": 100,
6
+ "num_indices": 100,
7
+ "elapsed_seconds": 30.472189903259277
8
+ }
bias_statichh-Qwen-7B-sft-bf16/scores_ekfac_half/factor_arguments.json ADDED
@@ -0,0 +1,20 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "strategy": "ekfac",
3
+ "use_empirical_fisher": false,
4
+ "amp_dtype": "torch.bfloat16",
5
+ "amp_scale": 65536.0,
6
+ "has_shared_parameters": false,
7
+ "covariance_max_examples": 100000,
8
+ "covariance_data_partitions": 1,
9
+ "covariance_module_partitions": 1,
10
+ "activation_covariance_dtype": "torch.bfloat16",
11
+ "gradient_covariance_dtype": "torch.bfloat16",
12
+ "eigendecomposition_dtype": "torch.float64",
13
+ "lambda_max_examples": 100000,
14
+ "lambda_data_partitions": 1,
15
+ "lambda_module_partitions": 1,
16
+ "use_iterative_lambda_aggregation": false,
17
+ "offload_activations_to_cpu": false,
18
+ "per_sample_gradient_dtype": "torch.bfloat16",
19
+ "lambda_dtype": "torch.bfloat16"
20
+ }
bias_statichh-Qwen-7B-sft-bf16/scores_ekfac_half/pairwise_scores.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b24db75f8c27fbd9c4c125fb81027ddb0da2fafb6fb7adbbfc22317780da371b
3
+ size 193192
bias_statichh-Qwen-7B-sft-bf16/scores_ekfac_half/query_dataset_metadata.json ADDED
@@ -0,0 +1,5 @@
 
 
 
 
 
 
1
+ {
2
+ "type": "Dataset",
3
+ "dataset_size": 936,
4
+ "indices": null
5
+ }
bias_statichh-Qwen-7B-sft-bf16/scores_ekfac_half/score_arguments.json ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "damping_factor": 1e-08,
3
+ "amp_dtype": "torch.bfloat16",
4
+ "offload_activations_to_cpu": false,
5
+ "data_partitions": 1,
6
+ "module_partitions": 1,
7
+ "compute_per_module_scores": false,
8
+ "compute_per_token_scores": false,
9
+ "query_gradient_accumulation_steps": 1,
10
+ "query_gradient_low_rank": null,
11
+ "use_full_svd": false,
12
+ "aggregate_query_gradients": true,
13
+ "aggregate_train_gradients": false,
14
+ "use_measurement_for_self_influence": false,
15
+ "query_gradient_svd_dtype": "torch.float32",
16
+ "per_sample_gradient_dtype": "torch.bfloat16",
17
+ "precondition_dtype": "torch.bfloat16",
18
+ "score_dtype": "torch.bfloat16"
19
+ }
bias_statichh-Qwen-7B-sft-bf16/scores_ekfac_half/train_dataset_metadata.json ADDED
@@ -0,0 +1,5 @@
 
 
 
 
 
 
1
+ {
2
+ "type": "Dataset",
3
+ "dataset_size": 96256,
4
+ "indices": null
5
+ }
truth_statichh-Qwen-1.5B-sft-bf16/scores_ekfac_half/pairwise_scores.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1811b9037b8d6b7e4413ac3bb2a7311695191e54fb09bc63b888def6789e48e9
3
+ size 193192
truth_statichh-Qwen-3B-sft-bf16/factors_ekfac_half/activation_covariance.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:42f0553c9e08f6043a4d7309ec33c39440d86d0edb0558b27dd622ae3899d697
3
+ size 292709750
truth_statichh-Qwen-3B-sft-bf16/factors_ekfac_half/activation_eigenvalues.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a0b2bc9140e7eb86853f8b7ed4facc0b6bf1cc6442340e5740da6cc5a49f9458
3
+ size 47902
truth_statichh-Qwen-3B-sft-bf16/factors_ekfac_half/activation_eigenvectors.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:01c23869870b91dd147c7fcdfcdc94024c8d881591f5baa9f4cb0600b6c1707f
3
+ size 292709750
truth_statichh-Qwen-3B-sft-bf16/factors_ekfac_half/covariance_dataset_metadata.json ADDED
@@ -0,0 +1,5 @@
 
 
 
 
 
 
1
+ {
2
+ "type": "Dataset",
3
+ "dataset_size": 10000,
4
+ "indices": null
5
+ }
truth_statichh-Qwen-3B-sft-bf16/factors_ekfac_half/factor_arguments.json ADDED
@@ -0,0 +1,20 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "strategy": "ekfac",
3
+ "use_empirical_fisher": false,
4
+ "amp_dtype": "torch.bfloat16",
5
+ "amp_scale": 65536.0,
6
+ "has_shared_parameters": false,
7
+ "covariance_max_examples": 100000,
8
+ "covariance_data_partitions": 1,
9
+ "covariance_module_partitions": 1,
10
+ "activation_covariance_dtype": "torch.bfloat16",
11
+ "gradient_covariance_dtype": "torch.bfloat16",
12
+ "eigendecomposition_dtype": "torch.float64",
13
+ "lambda_max_examples": 100000,
14
+ "lambda_data_partitions": 1,
15
+ "lambda_module_partitions": 1,
16
+ "use_iterative_lambda_aggregation": false,
17
+ "offload_activations_to_cpu": false,
18
+ "per_sample_gradient_dtype": "torch.bfloat16",
19
+ "lambda_dtype": "torch.bfloat16"
20
+ }
truth_statichh-Qwen-3B-sft-bf16/factors_ekfac_half/gradient_covariance.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9632b5c6c417ba0a17ebf52845bd8517846ade34e9e5d5adebdf649977d633a3
3
+ size 510133616
truth_statichh-Qwen-3B-sft-bf16/factors_ekfac_half/gradient_eigenvalues.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9bc15d4db788f9c7bd3f1b03ad4fcfdef582eb4a32d4e86ad22961f19cbb3467
3
+ size 58648
truth_statichh-Qwen-3B-sft-bf16/factors_ekfac_half/gradient_eigenvectors.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3c2130c9d7e76a278cd84a038072f608d63acc6f0683ee1ddb54ff3f9c96e03e
3
+ size 510133616
truth_statichh-Qwen-3B-sft-bf16/factors_ekfac_half/lambda_dataset_metadata.json ADDED
@@ -0,0 +1,5 @@
 
 
 
 
 
 
1
+ {
2
+ "type": "Dataset",
3
+ "dataset_size": 10000,
4
+ "indices": null
5
+ }
truth_statichh-Qwen-3B-sft-bf16/factors_ekfac_half/lambda_matrix.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e151a60a3ec70fd550b6652bc5dca8f61a7e88582e1040a494b23b75bfe65204
3
+ size 154147184
truth_statichh-Qwen-3B-sft-bf16/factors_ekfac_half/num_activation_covariance_processed.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a499a0128431c6645b1cacf249a269d583b6ac5996c4912a753a82816e034d1e
3
+ size 1296
truth_statichh-Qwen-3B-sft-bf16/factors_ekfac_half/num_gradient_covariance_processed.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7958f5f400e2bae7ad6a58e9ce18c6798ab8badd1fa2ffae0f7203499fbd7fa4
3
+ size 1296
truth_statichh-Qwen-3B-sft-bf16/factors_ekfac_half/num_lambda_processed.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:cff10ecae1729f5911044a9de269d527ee62cb2985ca4624c3fc4989eafb990f
3
+ size 1296
truth_statichh-Qwen-3B-sft-bf16/scores_ekfac_half/factor_arguments.json ADDED
@@ -0,0 +1,20 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "strategy": "ekfac",
3
+ "use_empirical_fisher": false,
4
+ "amp_dtype": "torch.bfloat16",
5
+ "amp_scale": 65536.0,
6
+ "has_shared_parameters": false,
7
+ "covariance_max_examples": 100000,
8
+ "covariance_data_partitions": 1,
9
+ "covariance_module_partitions": 1,
10
+ "activation_covariance_dtype": "torch.bfloat16",
11
+ "gradient_covariance_dtype": "torch.bfloat16",
12
+ "eigendecomposition_dtype": "torch.float64",
13
+ "lambda_max_examples": 100000,
14
+ "lambda_data_partitions": 1,
15
+ "lambda_module_partitions": 1,
16
+ "use_iterative_lambda_aggregation": false,
17
+ "offload_activations_to_cpu": false,
18
+ "per_sample_gradient_dtype": "torch.bfloat16",
19
+ "lambda_dtype": "torch.bfloat16"
20
+ }
truth_statichh-Qwen-3B-sft-bf16/scores_ekfac_half/pairwise_scores.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f7ad1612cb8827ae499610387021291201663fba6d70d0f20ec61305424e2d49
3
+ size 193192
truth_statichh-Qwen-3B-sft-bf16/scores_ekfac_half/query_dataset_metadata.json ADDED
@@ -0,0 +1,5 @@
 
 
 
 
 
 
1
+ {
2
+ "type": "Dataset",
3
+ "dataset_size": 653,
4
+ "indices": null
5
+ }
truth_statichh-Qwen-3B-sft-bf16/scores_ekfac_half/score_arguments.json ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "damping_factor": 1e-08,
3
+ "amp_dtype": "torch.bfloat16",
4
+ "offload_activations_to_cpu": false,
5
+ "data_partitions": 1,
6
+ "module_partitions": 1,
7
+ "compute_per_module_scores": false,
8
+ "compute_per_token_scores": false,
9
+ "query_gradient_accumulation_steps": 1,
10
+ "query_gradient_low_rank": null,
11
+ "use_full_svd": false,
12
+ "aggregate_query_gradients": true,
13
+ "aggregate_train_gradients": false,
14
+ "use_measurement_for_self_influence": false,
15
+ "query_gradient_svd_dtype": "torch.float32",
16
+ "per_sample_gradient_dtype": "torch.bfloat16",
17
+ "precondition_dtype": "torch.bfloat16",
18
+ "score_dtype": "torch.bfloat16"
19
+ }
truth_statichh-Qwen-3B-sft-bf16/scores_ekfac_half/train_dataset_metadata.json ADDED
@@ -0,0 +1,5 @@
 
 
 
 
 
 
1
+ {
2
+ "type": "Dataset",
3
+ "dataset_size": 96256,
4
+ "indices": null
5
+ }
truth_statichh-Qwen-7B-sft-bf16/factors_ekfac_half/covariance_dataset_metadata.json ADDED
@@ -0,0 +1,5 @@
 
 
 
 
 
 
1
+ {
2
+ "type": "Dataset",
3
+ "dataset_size": 10000,
4
+ "indices": null
5
+ }
truth_statichh-Qwen-7B-sft-bf16/factors_ekfac_half/factor_arguments.json ADDED
@@ -0,0 +1,20 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "strategy": "ekfac",
3
+ "use_empirical_fisher": false,
4
+ "amp_dtype": "torch.bfloat16",
5
+ "amp_scale": 65536.0,
6
+ "has_shared_parameters": false,
7
+ "covariance_max_examples": 100000,
8
+ "covariance_data_partitions": 1,
9
+ "covariance_module_partitions": 1,
10
+ "activation_covariance_dtype": "torch.bfloat16",
11
+ "gradient_covariance_dtype": "torch.bfloat16",
12
+ "eigendecomposition_dtype": "torch.float64",
13
+ "lambda_max_examples": 100000,
14
+ "lambda_data_partitions": 1,
15
+ "lambda_module_partitions": 1,
16
+ "use_iterative_lambda_aggregation": false,
17
+ "offload_activations_to_cpu": false,
18
+ "per_sample_gradient_dtype": "torch.bfloat16",
19
+ "lambda_dtype": "torch.bfloat16"
20
+ }
truth_statichh-Qwen-7B-sft-bf16/factors_ekfac_half/lambda_dataset_metadata.json ADDED
@@ -0,0 +1,5 @@
 
 
 
 
 
 
1
+ {
2
+ "type": "Dataset",
3
+ "dataset_size": 10000,
4
+ "indices": null
5
+ }
truth_statichh-Qwen-7B-sft-bf16/factors_ekfac_half/num_gradient_covariance_processed.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6134a7cf393a840f8649379506886da043c8ee089001a8c29dc4d537e7e9041b
3
+ size 1296
truth_statichh-Qwen-7B-sft-bf16/factors_ekfac_half/num_lambda_processed.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:039f64e1827743bc7907194136ec4e370d860c2907cfb56fda02add920c8fe34
3
+ size 1296
truth_statichh-Qwen-7B-sft-bf16/scores_ekfac_half/factor_arguments.json ADDED
@@ -0,0 +1,20 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "strategy": "ekfac",
3
+ "use_empirical_fisher": false,
4
+ "amp_dtype": "torch.bfloat16",
5
+ "amp_scale": 65536.0,
6
+ "has_shared_parameters": false,
7
+ "covariance_max_examples": 100000,
8
+ "covariance_data_partitions": 1,
9
+ "covariance_module_partitions": 1,
10
+ "activation_covariance_dtype": "torch.bfloat16",
11
+ "gradient_covariance_dtype": "torch.bfloat16",
12
+ "eigendecomposition_dtype": "torch.float64",
13
+ "lambda_max_examples": 100000,
14
+ "lambda_data_partitions": 1,
15
+ "lambda_module_partitions": 1,
16
+ "use_iterative_lambda_aggregation": false,
17
+ "offload_activations_to_cpu": false,
18
+ "per_sample_gradient_dtype": "torch.bfloat16",
19
+ "lambda_dtype": "torch.bfloat16"
20
+ }
truth_statichh-Qwen-7B-sft-bf16/scores_ekfac_half/pairwise_scores.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6e5fa696ffb67ef3712784ed075194ebc254fc469268b2f76702d0c84356fc43
3
+ size 193192
truth_statichh-Qwen-7B-sft-bf16/scores_ekfac_half/query_dataset_metadata.json ADDED
@@ -0,0 +1,5 @@
 
 
 
 
 
 
1
+ {
2
+ "type": "Dataset",
3
+ "dataset_size": 653,
4
+ "indices": null
5
+ }
truth_statichh-Qwen-7B-sft-bf16/scores_ekfac_half/score_arguments.json ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "damping_factor": 1e-08,
3
+ "amp_dtype": "torch.bfloat16",
4
+ "offload_activations_to_cpu": false,
5
+ "data_partitions": 1,
6
+ "module_partitions": 1,
7
+ "compute_per_module_scores": false,
8
+ "compute_per_token_scores": false,
9
+ "query_gradient_accumulation_steps": 1,
10
+ "query_gradient_low_rank": null,
11
+ "use_full_svd": false,
12
+ "aggregate_query_gradients": true,
13
+ "aggregate_train_gradients": false,
14
+ "use_measurement_for_self_influence": false,
15
+ "query_gradient_svd_dtype": "torch.float32",
16
+ "per_sample_gradient_dtype": "torch.bfloat16",
17
+ "precondition_dtype": "torch.bfloat16",
18
+ "score_dtype": "torch.bfloat16"
19
+ }
truth_statichh-Qwen-7B-sft-bf16/scores_ekfac_half/train_dataset_metadata.json ADDED
@@ -0,0 +1,5 @@
 
 
 
 
 
 
1
+ {
2
+ "type": "Dataset",
3
+ "dataset_size": 96256,
4
+ "indices": null
5
+ }