Datasets:

Jeesup
/

glue-lora-bitwidth-results

Tags:

Dataset Preview

Split (1)

train

The full dataset viewer is not available (click to read why). Only showing a preview of the rows.

The dataset generation failed because of a cast error

Error code:   DatasetGenerationCastError
Exception:    DatasetGenerationCastError
Message:      An error occurred while generating the dataset

All the data files must have the same columns, but at some point there are 19 new columns ({'test_majority_baseline', 'test_accuracy', 'n_classes', 'run_name', 'train_runtime_sec', 'lora_init_source', 'train_steps', 'val_macro_f1', 'val_loss', 'train_loss', 'trainable_params', 'gpu', 'test_loss', 'throughput_tokens_per_sec', 'val_accuracy', 'peak_gpu_mem_reserved_gib', 'throughput_samples_per_sec', 'test_macro_f1', 'peak_gpu_mem_gib'}) and 29 missing columns ({'sv1', 'sv2', 'sv8', 'energy_top2', 'fro_norm', 'sv5', 'energy_top1', 'stable_rank', 'energy_top16', 'layer', 'sv15', 'sv16', 'sv9', 'module', 'sv7', 'in_features', 'sv10', 'out_features', 'energy_top4', 'sv12', 'energy_top8', 'sv13', 'sv3', 'spectral_norm', 'sv4', 'sv14', 'effective_rank', 'sv11', 'sv6'}).

This happened while the csv dataset builder was generating data using

hf://datasets/Jeesup/glue-lora-bitwidth-results/aggregate_results.csv (at revision 66d10c988fbfb6a2d94131cfc19c5ca381ad6c0d), ['hf://datasets/Jeesup/glue-lora-bitwidth-results@66d10c988fbfb6a2d94131cfc19c5ca381ad6c0d/adapter_layer_metrics.csv', 'hf://datasets/Jeesup/glue-lora-bitwidth-results@66d10c988fbfb6a2d94131cfc19c5ca381ad6c0d/aggregate_results.csv', 'hf://datasets/Jeesup/glue-lora-bitwidth-results@66d10c988fbfb6a2d94131cfc19c5ca381ad6c0d/aggregate_results_by_condition.csv', 'hf://datasets/Jeesup/glue-lora-bitwidth-results@66d10c988fbfb6a2d94131cfc19c5ca381ad6c0d/aggregate_results_by_task.csv', 'hf://datasets/Jeesup/glue-lora-bitwidth-results@66d10c988fbfb6a2d94131cfc19c5ca381ad6c0d/gamma_table.csv', 'hf://datasets/Jeesup/glue-lora-bitwidth-results@66d10c988fbfb6a2d94131cfc19c5ca381ad6c0d/pairwise_geometry_metrics.csv', 'hf://datasets/Jeesup/glue-lora-bitwidth-results@66d10c988fbfb6a2d94131cfc19c5ca381ad6c0d/swap_penalty.csv', 'hf://datasets/Jeesup/glue-lora-bitwidth-results@66d10c988fbfb6a2d94131cfc19c5ca381ad6c0d/swap_penalty_by_task.csv', 'hf://datasets/Jeesup/glue-lora-bitwidth-results@66d10c988fbfb6a2d94131cfc19c5ca381ad6c0d/swap_raw.csv']

Please either edit the data files to have matching columns, or separate them into different configurations (see docs at https://hf.co/docs/hub/datasets-manual-configuration#multiple-configurations)
Traceback:    Traceback (most recent call last):
                File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1837, in _prepare_split_single
                  writer.write_table(table)
                  ~~~~~~~~~~~~~~~~~~^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/arrow_writer.py", line 765, in write_table
                  self._write_table(pa_table, writer_batch_size=writer_batch_size)
                  ~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/arrow_writer.py", line 773, in _write_table
                  pa_table = table_cast(pa_table, self._schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2369, in table_cast
                  return cast_table_to_schema(table, schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2297, in cast_table_to_schema
                  raise CastError(
                  ...<3 lines>...
                  )
              datasets.table.CastError: Couldn't cast
              run_name: string
              model_size: string
              task: string
              bitwidth: string
              seed: int64
              n_classes: int64
              val_accuracy: double
              val_macro_f1: double
              val_loss: double
              test_accuracy: double
              test_macro_f1: double
              test_loss: double
              test_majority_baseline: double
              train_loss: double
              train_steps: int64
              train_runtime_sec: double
              throughput_samples_per_sec: double
              throughput_tokens_per_sec: double
              peak_gpu_mem_gib: double
              peak_gpu_mem_reserved_gib: double
              trainable_params: int64
              gpu: string
              lora_init_source: string
              -- schema metadata --
              pandas: '{"index_columns": [{"kind": "range", "name": null, "start": 0, "' + 3173
              to
              {'model_size': Value('string'), 'task': Value('string'), 'bitwidth': Value('string'), 'seed': Value('int64'), 'layer': Value('int64'), 'module': Value('string'), 'out_features': Value('int64'), 'in_features': Value('int64'), 'fro_norm': Value('float64'), 'spectral_norm': Value('float64'), 'effective_rank': Value('float64'), 'stable_rank': Value('float64'), 'energy_top1': Value('float64'), 'energy_top2': Value('float64'), 'energy_top4': Value('float64'), 'energy_top8': Value('float64'), 'energy_top16': Value('float64'), 'sv1': Value('float64'), 'sv2': Value('float64'), 'sv3': Value('float64'), 'sv4': Value('float64'), 'sv5': Value('float64'), 'sv6': Value('float64'), 'sv7': Value('float64'), 'sv8': Value('float64'), 'sv9': Value('float64'), 'sv10': Value('float64'), 'sv11': Value('float64'), 'sv12': Value('float64'), 'sv13': Value('float64'), 'sv14': Value('float64'), 'sv15': Value('float64'), 'sv16': Value('float64')}
              because column names don't match
              
              During handling of the above exception, another exception occurred:
              
              Traceback (most recent call last):
                File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 1369, in compute_config_parquet_and_info_response
                  parquet_operations, partial, estimated_dataset_info = stream_convert_to_parquet(
                                                                        ~~~~~~~~~~~~~~~~~~~~~~~~~^
                      builder, max_dataset_size_bytes=max_dataset_size_bytes
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                  )
                  ^
                File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 948, in stream_convert_to_parquet
                  builder._prepare_split(split_generator=splits_generators[split], file_format="parquet")
                  ~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1683, in _prepare_split
                  for job_id, done, content in self._prepare_split_single(
                                               ~~~~~~~~~~~~~~~~~~~~~~~~~~^
                      gen_kwargs=gen_kwargs, job_id=job_id, **_prepare_split_args
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                  ):
                  ^
                File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1839, in _prepare_split_single
                  raise DatasetGenerationCastError.from_cast_error(
                  ...<4 lines>...
                  )
              datasets.exceptions.DatasetGenerationCastError: An error occurred while generating the dataset
              
              All the data files must have the same columns, but at some point there are 19 new columns ({'test_majority_baseline', 'test_accuracy', 'n_classes', 'run_name', 'train_runtime_sec', 'lora_init_source', 'train_steps', 'val_macro_f1', 'val_loss', 'train_loss', 'trainable_params', 'gpu', 'test_loss', 'throughput_tokens_per_sec', 'val_accuracy', 'peak_gpu_mem_reserved_gib', 'throughput_samples_per_sec', 'test_macro_f1', 'peak_gpu_mem_gib'}) and 29 missing columns ({'sv1', 'sv2', 'sv8', 'energy_top2', 'fro_norm', 'sv5', 'energy_top1', 'stable_rank', 'energy_top16', 'layer', 'sv15', 'sv16', 'sv9', 'module', 'sv7', 'in_features', 'sv10', 'out_features', 'energy_top4', 'sv12', 'energy_top8', 'sv13', 'sv3', 'spectral_norm', 'sv4', 'sv14', 'effective_rank', 'sv11', 'sv6'}).
              
              This happened while the csv dataset builder was generating data using
              
              hf://datasets/Jeesup/glue-lora-bitwidth-results/aggregate_results.csv (at revision 66d10c988fbfb6a2d94131cfc19c5ca381ad6c0d), ['hf://datasets/Jeesup/glue-lora-bitwidth-results@66d10c988fbfb6a2d94131cfc19c5ca381ad6c0d/adapter_layer_metrics.csv', 'hf://datasets/Jeesup/glue-lora-bitwidth-results@66d10c988fbfb6a2d94131cfc19c5ca381ad6c0d/aggregate_results.csv', 'hf://datasets/Jeesup/glue-lora-bitwidth-results@66d10c988fbfb6a2d94131cfc19c5ca381ad6c0d/aggregate_results_by_condition.csv', 'hf://datasets/Jeesup/glue-lora-bitwidth-results@66d10c988fbfb6a2d94131cfc19c5ca381ad6c0d/aggregate_results_by_task.csv', 'hf://datasets/Jeesup/glue-lora-bitwidth-results@66d10c988fbfb6a2d94131cfc19c5ca381ad6c0d/gamma_table.csv', 'hf://datasets/Jeesup/glue-lora-bitwidth-results@66d10c988fbfb6a2d94131cfc19c5ca381ad6c0d/pairwise_geometry_metrics.csv', 'hf://datasets/Jeesup/glue-lora-bitwidth-results@66d10c988fbfb6a2d94131cfc19c5ca381ad6c0d/swap_penalty.csv', 'hf://datasets/Jeesup/glue-lora-bitwidth-results@66d10c988fbfb6a2d94131cfc19c5ca381ad6c0d/swap_penalty_by_task.csv', 'hf://datasets/Jeesup/glue-lora-bitwidth-results@66d10c988fbfb6a2d94131cfc19c5ca381ad6c0d/swap_raw.csv']
              
              Please either edit the data files to have matching columns, or separate them into different configurations (see docs at https://hf.co/docs/hub/datasets-manual-configuration#multiple-configurations)

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

model_size string	task string	bitwidth string	seed int64	layer int64	module string	out_features int64	in_features int64	fro_norm float64	spectral_norm float64	effective_rank float64	stable_rank float64	energy_top1 float64	energy_top2 float64	energy_top4 float64	energy_top8 float64	energy_top16 float64	sv1 float64	sv2 float64	sv3 float64	sv4 float64	sv5 float64	sv6 float64	sv7 float64	sv8 float64	sv9 float64	sv10 float64	sv11 float64	sv12 float64	sv13 float64	sv14 float64	sv15 float64	sv16 float64
1B	mnli	bf16	42	0	k_proj	512	2,048	0.478136	0.310471	11.830018	2.371702	0.421638	0.687144	0.843664	0.94075	1	0.310471	0.24637	0.154217	0.109544	0.09273	0.076511	0.063621	0.060785	0.054018	0.048658	0.047053	0.041414	0.037193	0.03441	0.032083	0.027093
1B	mnli	bf16	42	0	o_proj	2,048	2,048	0.888789	0.518331	13.327409	2.940231	0.340109	0.531859	0.73751	0.895371	1	0.518331	0.389194	0.340692	0.215363	0.204104	0.193881	0.165473	0.134431	0.120828	0.117049	0.109882	0.103095	0.09631	0.091316	0.088292	0.078989
1B	mnli	bf16	42	0	q_proj	2,048	2,048	1.04181	0.650642	12.65524	2.563854	0.390038	0.576144	0.81379	0.912971	1	0.650642	0.449438	0.423397	0.280477	0.188861	0.164903	0.15257	0.146661	0.141586	0.127115	0.112317	0.108451	0.099496	0.097417	0.090933	0.078859
1B	mnli	bf16	42	0	v_proj	512	2,048	0.438832	0.306725	12.108855	2.04692	0.488539	0.720204	0.825744	0.926386	1	0.306725	0.211217	0.112638	0.08739	0.083111	0.071471	0.061523	0.059836	0.05333	0.051704	0.045922	0.043571	0.039134	0.034888	0.033676	0.027728
1B	mnli	bf16	42	1	k_proj	512	2,048	0.462451	0.247758	13.453991	3.483966	0.287029	0.487598	0.732822	0.886742	1	0.247758	0.207108	0.162713	0.161146	0.11345	0.087383	0.08135	0.076111	0.074626	0.064725	0.059508	0.053684	0.050735	0.046952	0.044337	0.035996
1B	mnli	bf16	42	1	o_proj	2,048	2,048	1.010072	0.648706	12.500945	2.42443	0.412468	0.679533	0.814992	0.916371	1	0.648706	0.521989	0.28684	0.236483	0.190455	0.16604	0.145722	0.135479	0.125932	0.118509	0.108962	0.105182	0.095903	0.095057	0.089582	0.078894
1B	mnli	bf16	42	1	q_proj	2,048	2,048	1.029521	0.708345	13.281681	2.112425	0.47339	0.616031	0.741713	0.885521	1	0.708345	0.388829	0.273353	0.241846	0.227463	0.193491	0.181431	0.174151	0.149895	0.139819	0.136334	0.122884	0.117365	0.111992	0.099302	0.09724
1B	mnli	bf16	42	1	v_proj	512	2,048	0.466233	0.347362	9.977397	1.801536	0.555082	0.863449	0.918804	0.966002	1	0.347362	0.258903	0.083517	0.071116	0.064737	0.053657	0.040424	0.039442	0.037771	0.036893	0.03258	0.030578	0.029263	0.026577	0.02315	0.022525
1B	mnli	bf16	42	2	k_proj	512	2,048	0.47884	0.285264	12.966253	2.817647	0.354906	0.61452	0.763637	0.906944	1	0.285264	0.24398	0.139071	0.12186	0.110738	0.09287	0.079348	0.075331	0.066144	0.061342	0.055051	0.051231	0.049489	0.044557	0.040397	0.038431
1B	mnli	bf16	42	2	o_proj	2,048	2,048	1.018573	0.685337	11.917879	2.208903	0.452713	0.668478	0.854815	0.933667	1	0.685337	0.473132	0.374083	0.231054	0.170264	0.155564	0.125565	0.113362	0.112094	0.109873	0.10276	0.089824	0.088354	0.080834	0.079272	0.07022
1B	mnli	bf16	42	2	q_proj	2,048	2,048	1.112087	0.722506	13.550971	2.369163	0.42209	0.558394	0.731102	0.872195	1	0.722506	0.410575	0.36308	0.285951	0.236701	0.215557	0.196167	0.18309	0.173842	0.160484	0.151545	0.14106	0.132219	0.128075	0.114974	0.11008
1B	mnli	bf16	42	2	v_proj	512	2,048	0.519771	0.434275	10.186154	1.4325	0.69808	0.826088	0.911999	0.9659	1	0.434275	0.185965	0.117337	0.097169	0.072047	0.058525	0.056262	0.052733	0.045698	0.042496	0.039042	0.034444	0.02998	0.026439	0.023138	0.021779
1B	mnli	bf16	42	3	k_proj	512	2,048	0.548404	0.367223	12.154858	2.230191	0.448392	0.638691	0.818666	0.935618	1	0.367223	0.239232	0.191229	0.132509	0.111578	0.102803	0.082871	0.072714	0.066835	0.057742	0.053771	0.048162	0.045075	0.041303	0.036964	0.035307
1B	mnli	bf16	42	3	o_proj	2,048	2,048	1.056567	0.693695	13.195705	2.319839	0.431064	0.583719	0.729755	0.898639	1	0.693695	0.412812	0.30579	0.263661	0.253719	0.229345	0.203983	0.173059	0.150596	0.144825	0.129461	0.118486	0.113235	0.09997	0.095801	0.081894
1B	mnli	bf16	42	3	q_proj	2,048	2,048	1.097508	0.808894	11.898837	1.840907	0.54321	0.72444	0.846423	0.931755	1	0.808894	0.467221	0.312484	0.222003	0.178707	0.161143	0.154225	0.145242	0.137276	0.120794	0.106711	0.094917	0.092722	0.08676	0.080605	0.075819
1B	mnli	bf16	42	3	v_proj	512	2,048	0.553195	0.42971	11.33683	1.657314	0.603386	0.752205	0.866037	0.949342	1	0.42971	0.213406	0.152123	0.108137	0.097278	0.084642	0.067877	0.06526	0.05558	0.051722	0.047775	0.044584	0.04046	0.039826	0.034693	0.032273
1B	mnli	bf16	42	4	k_proj	512	2,048	0.558635	0.429351	11.548218	1.692899	0.590703	0.707329	0.852861	0.945331	1	0.429351	0.190777	0.175888	0.120333	0.10627	0.082325	0.075543	0.071275	0.064973	0.053387	0.051012	0.043525	0.041732	0.039123	0.036411	0.029906
1B	mnli	bf16	42	4	o_proj	2,048	2,048	0.949071	0.627225	12.59419	2.289554	0.436766	0.677989	0.798901	0.912196	1	0.627225	0.466131	0.260954	0.202021	0.181238	0.164037	0.147244	0.143573	0.134659	0.113866	0.105951	0.095828	0.090529	0.088441	0.078925	0.073039
1B	mnli	bf16	42	4	q_proj	2,048	2,048	1.011137	0.665778	11.717894	2.306542	0.433549	0.736994	0.854773	0.938482	1	0.665778	0.556993	0.283488	0.200129	0.161624	0.150381	0.142293	0.128839	0.112528	0.107524	0.099919	0.087753	0.083919	0.073465	0.069167	0.061354
1B	mnli	bf16	42	4	v_proj	512	2,048	0.567309	0.467367	10.109594	1.473411	0.678697	0.800741	0.919681	0.970007	1	0.467367	0.198188	0.164549	0.105846	0.07649	0.063936	0.057142	0.054709	0.04364	0.04192	0.034996	0.034515	0.032594	0.031529	0.027956	0.027152
1B	mnli	bf16	42	5	k_proj	512	2,048	0.5713	0.38981	12.526484	2.147938	0.465563	0.648154	0.804953	0.91968	1	0.38981	0.244121	0.181193	0.135447	0.108837	0.1027	0.089803	0.083591	0.07542	0.065637	0.060125	0.056149	0.054407	0.052225	0.047428	0.03891
1B	mnli	bf16	42	5	o_proj	2,048	2,048	0.946153	0.598205	12.364101	2.501632	0.399739	0.66548	0.797841	0.927277	1	0.598205	0.487743	0.269033	0.214737	0.2052	0.170707	0.166178	0.130416	0.118372	0.112606	0.094417	0.08865	0.087619	0.080318	0.062688	0.059823
1B	mnli	bf16	42	5	q_proj	2,048	2,048	1.096806	0.702937	13.71684	2.434594	0.410746	0.519268	0.697227	0.868716	1	0.702937	0.361317	0.340424	0.313357	0.263365	0.234224	0.214065	0.190403	0.182395	0.160055	0.153729	0.140076	0.130481	0.117741	0.116565	0.106382
1B	mnli	bf16	42	5	v_proj	512	2,048	0.531679	0.408677	11.367978	1.69254	0.590828	0.725645	0.86443	0.948808	1	0.408677	0.195219	0.163855	0.111281	0.09966	0.076927	0.069374	0.056477	0.05487	0.050073	0.045026	0.044163	0.042006	0.035343	0.034356	0.027955
1B	mnli	bf16	42	6	k_proj	512	2,048	0.635088	0.409102	11.374869	2.409926	0.414951	0.688003	0.875306	0.953229	1	0.409102	0.331861	0.207797	0.179908	0.116912	0.086477	0.073148	0.070225	0.060282	0.057814	0.052616	0.047489	0.045125	0.042686	0.040198	0.037286
1B	mnli	bf16	42	6	o_proj	2,048	2,048	0.869999	0.712906	11.016452	1.489274	0.671468	0.78134	0.87658	0.953027	1	0.712906	0.288378	0.215157	0.160609	0.147013	0.124875	0.10689	0.096077	0.084262	0.078318	0.077403	0.070755	0.060718	0.055294	0.0507	0.044808
1B	mnli	bf16	42	6	q_proj	2,048	2,048	1.08946	0.820271	11.177371	1.764044	0.566879	0.757427	0.887583	0.947052	1	0.820271	0.475569	0.340172	0.196897	0.153835	0.136584	0.120141	0.117607	0.110924	0.098462	0.097594	0.091962	0.0835	0.07979	0.074589	0.062948
1B	mnli	bf16	42	6	v_proj	512	2,048	0.503888	0.41172	9.940239	1.497835	0.66763	0.833481	0.913383	0.972579	1	0.41172	0.205208	0.115036	0.083987	0.077898	0.058578	0.057284	0.047427	0.041972	0.033793	0.03196	0.02846	0.027237	0.023936	0.022119	0.020572
1B	mnli	bf16	42	7	k_proj	512	2,048	1.04801	0.998692	7.278714	1.101207	0.908094	0.942753	0.969294	0.987982	1	0.998692	0.195105	0.138304	0.100116	0.083892	0.076151	0.065579	0.058205	0.053712	0.048538	0.044486	0.039111	0.036914	0.03488	0.031859	0.029259
1B	mnli	bf16	42	7	o_proj	2,048	2,048	0.852187	0.68204	10.780377	1.561172	0.640544	0.801935	0.889886	0.95851	1	0.68204	0.342353	0.20056	0.153777	0.134188	0.114995	0.098079	0.0948	0.077405	0.073809	0.063729	0.061965	0.057335	0.0544	0.053914	0.040462
1B	mnli	bf16	42	7	q_proj	2,048	2,048	1.085218	0.805492	8.912915	1.815143	0.550921	0.902783	0.94026	0.979879	1	0.805492	0.643729	0.164411	0.130787	0.12626	0.115484	0.100053	0.085852	0.072215	0.069475	0.058852	0.053905	0.050694	0.047403	0.037534	0.032558
1B	mnli	bf16	42	7	v_proj	512	2,048	0.495345	0.333744	10.930369	2.202882	0.453951	0.731645	0.890269	0.958566	1	0.333744	0.261031	0.174655	0.091743	0.072614	0.070226	0.058416	0.056044	0.047601	0.042044	0.040392	0.034476	0.031657	0.0304	0.029792	0.022338
1B	mnli	bf16	42	8	k_proj	512	2,048	0.763863	0.726937	7.296813	1.104176	0.905653	0.941995	0.968388	0.988556	1	0.726937	0.14562	0.095039	0.079795	0.063072	0.061538	0.048026	0.041192	0.037039	0.033054	0.03121	0.030706	0.027717	0.024387	0.022701	0.020433
1B	mnli	bf16	42	8	o_proj	2,048	2,048	1.008201	0.737695	9.429694	1.867847	0.535376	0.854277	0.940388	0.974097	1	0.737695	0.569345	0.259811	0.14152	0.105398	0.09866	0.084549	0.079202	0.074219	0.067826	0.063677	0.06185	0.054465	0.04744	0.044343	0.034019
1B	mnli	bf16	42	8	q_proj	2,048	2,048	0.886662	0.841531	7.279574	1.110134	0.900792	0.931384	0.967342	0.990206	1	0.841531	0.155082	0.128758	0.108122	0.081269	0.075121	0.056794	0.050014	0.042161	0.03978	0.033031	0.028439	0.026718	0.025962	0.024815	0.020893
1B	mnli	bf16	42	8	v_proj	512	2,048	0.623003	0.512711	9.295307	1.476503	0.677276	0.876549	0.93808	0.974893	1	0.512711	0.278108	0.125574	0.090074	0.06853	0.062806	0.055049	0.051156	0.048029	0.040848	0.039949	0.036681	0.033537	0.030116	0.020796	0.019074
1B	mnli	bf16	42	9	k_proj	512	2,048	0.347441	0.202997	11.880726	2.929451	0.341361	0.614666	0.842384	0.944391	1	0.202997	0.181638	0.139131	0.090176	0.065618	0.056712	0.052487	0.045132	0.040689	0.033238	0.032281	0.026855	0.026148	0.024803	0.02264	0.019437
1B	mnli	bf16	42	9	o_proj	2,048	2,048	0.615907	0.428267	11.880395	2.06824	0.483503	0.725299	0.850905	0.9329	1	0.428267	0.302858	0.167972	0.139403	0.099054	0.090527	0.087418	0.073859	0.072308	0.068674	0.060222	0.057168	0.05111	0.049852	0.042588	0.041271
1B	mnli	bf16	42	9	q_proj	2,048	2,048	0.559172	0.401956	11.497271	1.935235	0.516733	0.695361	0.865309	0.94889	1	0.401956	0.236331	0.184375	0.138362	0.097991	0.087762	0.069094	0.063679	0.058073	0.052173	0.048889	0.043848	0.042964	0.040158	0.033521	0.031483
1B	mnli	bf16	42	9	v_proj	512	2,048	0.317267	0.228403	11.542101	1.929498	0.51827	0.720992	0.863264	0.94528	1	0.228403	0.142848	0.095703	0.071846	0.055121	0.046072	0.040578	0.038053	0.033619	0.031923	0.029297	0.026754	0.024059	0.021572	0.020124	0.018316
1B	mnli	bf16	42	10	k_proj	512	2,048	0.251937	0.16936	12.65948	2.212912	0.451893	0.628667	0.778885	0.921223	1	0.16936	0.105925	0.075695	0.061684	0.058045	0.046899	0.044911	0.038062	0.03102	0.02965	0.026602	0.024812	0.024193	0.022969	0.020824	0.017
1B	mnli	bf16	42	10	o_proj	2,048	2,048	0.532705	0.419522	10.745091	1.612369	0.620205	0.792601	0.889257	0.957557	1	0.419522	0.221182	0.129341	0.103438	0.080643	0.071997	0.064029	0.059961	0.057801	0.04697	0.042391	0.039679	0.032309	0.029325	0.025636	0.023761
1B	mnli	bf16	42	10	q_proj	2,048	2,048	0.469166	0.287811	12.707808	2.657294	0.376323	0.571101	0.788389	0.917833	1	0.287811	0.20706	0.170847	0.136529	0.104228	0.086677	0.075659	0.066273	0.063853	0.058056	0.054053	0.044407	0.041249	0.039365	0.037397	0.033094
1B	mnli	bf16	42	10	v_proj	512	2,048	0.234694	0.154238	12.736726	2.315369	0.431897	0.638882	0.775958	0.910376	1	0.154238	0.106776	0.066251	0.056224	0.04925	0.045419	0.040473	0.035741	0.03464	0.029868	0.028373	0.023961	0.022413	0.019332	0.01779	0.016521
1B	mnli	bf16	42	11	k_proj	512	2,048	0.343711	0.252963	11.384301	1.846175	0.541661	0.688068	0.861171	0.953971	1	0.252963	0.131515	0.108864	0.092728	0.068477	0.055898	0.040936	0.038387	0.034049	0.031792	0.026207	0.025619	0.024563	0.023391	0.020236	0.019094
1B	mnli	bf16	42	11	o_proj	2,048	2,048	0.758357	0.660967	9.28608	1.316402	0.759646	0.857487	0.940752	0.976351	1	0.660967	0.23721	0.175814	0.130292	0.087102	0.073237	0.063065	0.059543	0.052908	0.048711	0.046473	0.044365	0.037314	0.037165	0.028107	0.027151
1B	mnli	bf16	42	11	q_proj	2,048	2,048	0.721841	0.564721	10.463219	1.633862	0.612047	0.749189	0.90626	0.966369	1	0.564721	0.267317	0.243025	0.150934	0.102369	0.096573	0.084086	0.066661	0.06576	0.056361	0.049555	0.046317	0.040252	0.037801	0.035526	0.033327
1B	mnli	bf16	42	11	v_proj	512	2,048	0.295228	0.212376	11.31363	1.932425	0.517485	0.756786	0.86479	0.945465	1	0.212376	0.144421	0.079624	0.05544	0.053	0.039992	0.038215	0.034102	0.032414	0.03127	0.027842	0.025072	0.022173	0.020023	0.015417	0.01381
1B	mnli	bf16	42	12	k_proj	512	2,048	0.236703	0.142567	12.759238	2.756586	0.362768	0.552197	0.778223	0.919345	1	0.142567	0.103021	0.089815	0.067801	0.054548	0.045545	0.042244	0.032749	0.030108	0.027765	0.026177	0.025462	0.022475	0.021851	0.016821	0.015572
1B	mnli	bf16	42	12	o_proj	2,048	2,048	0.564006	0.391309	11.207813	2.077443	0.481361	0.693089	0.878428	0.954671	1	0.391309	0.259521	0.201867	0.134933	0.096393	0.076434	0.070824	0.064059	0.057935	0.051568	0.048706	0.042009	0.038634	0.031774	0.030955	0.028392
1B	mnli	bf16	42	12	q_proj	2,048	2,048	0.505664	0.316187	12.648951	2.55762	0.390989	0.608284	0.799578	0.916154	1	0.316187	0.235715	0.164074	0.1483	0.10396	0.08825	0.077449	0.072208	0.069816	0.063352	0.057845	0.047867	0.045617	0.043502	0.040944	0.035559
1B	mnli	bf16	42	12	v_proj	512	2,048	0.251046	0.146815	12.380927	2.923918	0.342007	0.601934	0.812929	0.930041	1	0.146815	0.127991	0.086498	0.076262	0.053574	0.04239	0.039948	0.033436	0.030712	0.026811	0.026379	0.025334	0.022202	0.01843	0.017853	0.016065
1B	mnli	bf16	42	13	k_proj	512	2,048	0.290231	0.221816	11.640681	1.711984	0.584118	0.6997	0.84115	0.94511	1	0.221816	0.098671	0.08858	0.063784	0.052994	0.052513	0.044252	0.035109	0.033347	0.027629	0.026602	0.024831	0.022884	0.019326	0.016571	0.015881
1B	mnli	bf16	42	13	o_proj	2,048	2,048	0.504243	0.372986	10.672233	1.827663	0.547147	0.71599	0.89736	0.964727	1	0.372986	0.207196	0.165898	0.136357	0.080438	0.072818	0.053532	0.049902	0.044434	0.044018	0.037745	0.033509	0.02882	0.026223	0.025804	0.018032
1B	mnli	bf16	42	13	q_proj	2,048	2,048	0.727863	0.635318	9.717136	1.312559	0.761871	0.856533	0.923107	0.97062	1	0.635318	0.223943	0.146701	0.117257	0.100089	0.079368	0.070122	0.062747	0.059363	0.052503	0.0499	0.044848	0.040276	0.035319	0.033248	0.028428
1B	mnli	bf16	42	13	v_proj	512	2,048	0.271593	0.204115	10.99699	1.77047	0.564822	0.748861	0.889345	0.951293	1	0.204115	0.116513	0.089564	0.048383	0.038427	0.033748	0.032756	0.029681	0.028585	0.026846	0.022953	0.021445	0.020514	0.015731	0.014606	0.01366
1B	mnli	bf16	42	14	k_proj	512	2,048	0.310103	0.259472	10.970112	1.428342	0.700112	0.780424	0.875915	0.952726	1	0.259472	0.087881	0.078791	0.05454	0.048988	0.044002	0.041121	0.036871	0.031049	0.029336	0.025608	0.024801	0.023219	0.019981	0.01802	0.013693
1B	mnli	bf16	42	14	o_proj	2,048	2,048	0.454395	0.3248	10.217476	1.957203	0.510933	0.773872	0.903141	0.973245	1	0.3248	0.233003	0.12722	0.102498	0.073805	0.064927	0.052231	0.045651	0.036586	0.034414	0.028683	0.025559	0.023939	0.021599	0.01675	0.014327
1B	mnli	bf16	42	14	q_proj	2,048	2,048	0.789338	0.712999	8.305222	1.225603	0.815925	0.888052	0.949678	0.985385	1	0.712999	0.211989	0.165914	0.104255	0.092332	0.079928	0.061761	0.059326	0.047368	0.045091	0.041869	0.032756	0.030851	0.023574	0.016608	0.014821
1B	mnli	bf16	42	14	v_proj	512	2,048	0.290734	0.233817	9.906421	1.546115	0.646782	0.818126	0.913666	0.973152	1	0.233817	0.120346	0.072204	0.053501	0.045631	0.038684	0.02875	0.024959	0.022746	0.021335	0.019603	0.017799	0.013655	0.013064	0.012033	0.009684
1B	mnli	bf16	42	15	k_proj	512	2,048	0.284576	0.219658	11.308485	1.678431	0.595795	0.730932	0.8733	0.94992	1	0.219658	0.104613	0.086184	0.064045	0.044361	0.043067	0.036185	0.032755	0.030047	0.026861	0.02439	0.022192	0.021245	0.019216	0.01695	0.015365
1B	mnli	bf16	42	15	o_proj	2,048	2,048	0.799081	0.748817	6.738235	1.138757	0.87815	0.939442	0.9742	0.993698	1	0.748817	0.19783	0.113401	0.096615	0.075455	0.055704	0.046589	0.038517	0.03296	0.030184	0.026896	0.020458	0.018591	0.016187	0.012773	0.010662
1B	mnli	bf16	42	15	q_proj	2,048	2,048	0.402176	0.368378	8.810912	1.19191	0.83899	0.886415	0.945224	0.978253	1	0.368378	0.087583	0.076828	0.060079	0.045591	0.036854	0.032329	0.029333	0.027583	0.025924	0.024192	0.022154	0.018923	0.01645	0.014568	0.012948
1B	mnli	bf16	42	15	v_proj	512	2,048	0.219354	0.157181	11.361529	1.947566	0.513461	0.726961	0.867169	0.948218	1	0.157181	0.101355	0.064891	0.050352	0.035813	0.034828	0.027163	0.025814	0.024301	0.020962	0.019853	0.019549	0.016542	0.012957	0.011583	0.010467
1B	mnli	bf16	43	0	k_proj	512	2,048	0.525034	0.370646	11.838099	2.006572	0.498362	0.67992	0.835457	0.943989	1	0.370646	0.223715	0.166136	0.123589	0.108974	0.087715	0.075267	0.068435	0.058817	0.048866	0.047497	0.043904	0.041712	0.039392	0.033876	0.031145
1B	mnli	bf16	43	0	o_proj	2,048	2,048	0.943528	0.658422	12.204793	2.053527	0.486967	0.68301	0.82109	0.928414	1	0.658422	0.417764	0.284822	0.204453	0.186299	0.155885	0.146549	0.122721	0.113369	0.105714	0.09965	0.086728	0.084156	0.08012	0.072155	0.059514
1B	mnli	bf16	43	0	q_proj	2,048	2,048	0.994196	0.632905	13.048338	2.46756	0.405259	0.579734	0.765672	0.89896	1	0.632905	0.415279	0.349407	0.248396	0.212656	0.177659	0.169132	0.162342	0.146966	0.139974	0.120029	0.108944	0.102871	0.092377	0.087608	0.074916
1B	mnli	bf16	43	0	v_proj	512	2,048	0.431672	0.242943	11.87098	3.157168	0.31674	0.623846	0.845368	0.94255	1	0.242943	0.23922	0.169015	0.112749	0.0854	0.071746	0.05798	0.048028	0.047143	0.043307	0.041434	0.036768	0.032262	0.031212	0.029632	0.025407
1B	mnli	bf16	43	1	k_proj	512	2,048	0.449253	0.25412	13.737665	3.12541	0.319958	0.497645	0.692892	0.868256	1	0.25412	0.189374	0.155156	0.123827	0.103013	0.09699	0.09045	0.084815	0.077145	0.075058	0.063566	0.054471	0.049813	0.045085	0.044574	0.038681
1B	mnli	bf16	43	1	o_proj	2,048	2,048	1.008146	0.652155	12.796964	2.389712	0.41846	0.665858	0.794603	0.900546	1	0.652155	0.501443	0.281978	0.226582	0.185709	0.164256	0.158472	0.14524	0.143934	0.129443	0.121673	0.118342	0.100394	0.094924	0.089891	0.087346
1B	mnli	bf16	43	1	q_proj	2,048	2,048	0.977634	0.565689	14.285599	2.986736	0.334814	0.467776	0.63732	0.834866	1	0.565689	0.356484	0.297139	0.271576	0.248358	0.232348	0.193218	0.18923	0.162297	0.157286	0.15436	0.14506	0.140112	0.126702	0.118134	0.110637
1B	mnli	bf16	43	1	v_proj	512	2,048	0.476923	0.384321	10.870882	1.539957	0.649369	0.786611	0.88587	0.956101	1	0.384321	0.176682	0.111868	0.100313	0.078683	0.065892	0.055429	0.048675	0.044899	0.043509	0.039251	0.036966	0.034128	0.028409	0.026257	0.022533
1B	mnli	bf16	43	2	k_proj	512	2,048	0.478025	0.350643	12.682701	1.858539	0.538057	0.655171	0.788146	0.910546	1	0.350643	0.16359	0.132476	0.113295	0.103334	0.082092	0.078133	0.06669	0.062809	0.060583	0.050202	0.049794	0.048247	0.046602	0.042732	0.038734
1B	mnli	bf16	43	2	o_proj	2,048	2,048	0.910749	0.545678	12.884722	2.785638	0.358984	0.609002	0.779764	0.907327	1	0.545678	0.45539	0.30587	0.219284	0.204501	0.16248	0.137749	0.136429	0.122452	0.113439	0.109312	0.09684	0.094086	0.084967	0.077965	0.074356
1B	mnli	bf16	43	2	q_proj	2,048	2,048	1.063551	0.620186	13.151608	2.940853	0.340037	0.635758	0.77097	0.888032	1	0.620186	0.578362	0.304282	0.245676	0.203927	0.180129	0.171998	0.169699	0.151783	0.141086	0.131574	0.130226	0.120156	0.11725	0.111792	0.093568
1B	mnli	bf16	43	2	v_proj	512	2,048	0.48345	0.342049	11.529667	1.997673	0.500582	0.666989	0.875654	0.9442	1	0.342049	0.197213	0.191818	0.109434	0.071897	0.070558	0.055212	0.053148	0.047822	0.046923	0.045622	0.040222	0.038269	0.035323	0.03382	0.031588
1B	mnli	bf16	43	3	k_proj	512	2,048	0.608124	0.421878	11.715813	2.077829	0.481272	0.734368	0.855108	0.939286	1	0.421878	0.305939	0.165206	0.13175	0.110713	0.085326	0.078373	0.073824	0.069074	0.061431	0.054808	0.053339	0.049313	0.047191	0.042016	0.040435
1B	mnli	bf16	43	3	o_proj	2,048	2,048	1.098714	0.845931	11.655677	1.686943	0.592788	0.702811	0.845333	0.945415	1	0.845931	0.364439	0.307449	0.278433	0.222759	0.173372	0.154078	0.131897	0.112924	0.108157	0.099367	0.09755	0.082281	0.078287	0.070054	0.065173
1B	mnli	bf16	43	3	q_proj	2,048	2,048	1.082156	0.705556	12.761952	2.352432	0.425092	0.62019	0.797359	0.911831	1	0.705556	0.477987	0.356695	0.283275	0.222395	0.178708	0.162818	0.1617	0.142246	0.129233	0.117196	0.114576	0.10631	0.105145	0.096177	0.088582
1B	mnli	bf16	43	3	v_proj	512	2,048	0.522757	0.397189	11.18111	1.732228	0.577291	0.736952	0.881299	0.948763	1	0.397189	0.208881	0.180697	0.082434	0.080392	0.071025	0.061439	0.05616	0.052088	0.047109	0.045836	0.042241	0.040387	0.038479	0.033087	0.031268
1B	mnli	bf16	43	4	k_proj	512	2,048	0.622087	0.5203	10.427949	1.429536	0.699528	0.813285	0.895916	0.963723	1	0.5203	0.209817	0.142252	0.10836	0.099189	0.085829	0.069323	0.065039	0.05724	0.050046	0.04721	0.041294	0.038711	0.035811	0.029612	0.025812
1B	mnli	bf16	43	4	o_proj	2,048	2,048	0.955416	0.578482	13.185519	2.727755	0.366602	0.598651	0.745534	0.896828	1	0.578482	0.460238	0.265994	0.251645	0.224422	0.19181	0.165306	0.153694	0.140029	0.12455	0.119644	0.106274	0.10115	0.094367	0.092177	0.076253
1B	mnli	bf16	43	4	q_proj	2,048	2,048	1.017293	0.740741	11.820342	1.886073	0.530202	0.764178	0.841404	0.932601	1	0.740741	0.492075	0.204849	0.194826	0.174228	0.155859	0.144173	0.137641	0.116614	0.112354	0.098619	0.093042	0.08665	0.08457	0.078176	0.066129
1B	mnli	bf16	43	4	v_proj	512	2,048	0.538822	0.420563	10.979108	1.641455	0.609216	0.749738	0.880811	0.956532	1	0.420563	0.201984	0.156316	0.116704	0.082721	0.078753	0.070849	0.062607	0.05583	0.051629	0.044549	0.038914	0.031677	0.030729	0.027907	0.024737
1B	mnli	bf16	43	5	k_proj	512	2,048	0.56062	0.398466	13.024425	1.979496	0.505179	0.627365	0.752831	0.891044	1	0.398466	0.195966	0.155125	0.123975	0.112974	0.111042	0.099428	0.09198	0.08701	0.077111	0.071726	0.070252	0.061826	0.057666	0.045032	0.038363
1B	mnli	bf16	43	5	o_proj	2,048	2,048	0.914655	0.581399	12.151553	2.474941	0.40405	0.672142	0.823298	0.929573	1	0.581399	0.473586	0.283622	0.214511	0.180519	0.148774	0.134902	0.126448	0.112603	0.104994	0.099744	0.089321	0.081146	0.066422	0.062051	0.049412
1B	mnli	bf16	43	5	q_proj	2,048	2,048	1.092491	0.730368	13.265662	2.237444	0.446939	0.559279	0.738447	0.89297	1	0.730368	0.366172	0.354402	0.297056	0.252561	0.218232	0.203426	0.177862	0.154723	0.152961	0.14216	0.125363	0.120152	0.10943	0.098715	0.091248
1B	mnli	bf16	43	5	v_proj	512	2,048	0.49224	0.397978	11.339079	1.529797	0.653681	0.767925	0.860616	0.947654	1	0.397978	0.166376	0.108174	0.103718	0.081643	0.081016	0.067017	0.058043	0.053306	0.046295	0.043028	0.038704	0.036353	0.035219	0.031854	0.027795
1B	mnli	bf16	43	6	k_proj	512	2,048	0.729459	0.484688	10.970994	2.265043	0.441493	0.682403	0.883912	0.959452	1	0.484688	0.358038	0.27889	0.171597	0.12078	0.109916	0.086655	0.077571	0.072831	0.067112	0.057262	0.050587	0.04983	0.039206	0.033078	0.028553
1B	mnli	bf16	43	6	o_proj	2,048	2,048	0.881184	0.642235	12.361602	1.882543	0.531196	0.688941	0.811179	0.920016	1	0.642235	0.349981	0.228664	0.206467	0.169631	0.145621	0.134981	0.127709	0.119617	0.105922	0.092129	0.087607	0.080593	0.074406	0.0665	0.062948
1B	mnli	bf16	43	6	q_proj	2,048	2,048	1.112751	0.857365	11.237574	1.68448	0.593655	0.764686	0.883656	0.947443	1	0.857365	0.460188	0.300681	0.238542	0.153305	0.152016	0.137839	0.115639	0.114335	0.103169	0.099193	0.089848	0.084903	0.080297	0.074617	0.064994
1B	mnli	bf16	43	6	v_proj	512	2,048	0.526819	0.419101	10.412677	1.580106	0.632869	0.813559	0.908433	0.963203	1	0.419101	0.223939	0.12961	0.097634	0.073962	0.064341	0.056992	0.048399	0.047108	0.042297	0.038593	0.036235	0.033458	0.030485	0.028471	0.023296
1B	mnli	bf16	43	7	k_proj	512	2,048	0.821196	0.745361	8.591827	1.213838	0.823833	0.898825	0.950014	0.980809	1	0.745361	0.224882	0.159957	0.094516	0.085838	0.075771	0.066904	0.056409	0.05383	0.048078	0.045836	0.041025	0.037665	0.03161	0.029859	0.025277
1B	mnli	bf16	43	7	o_proj	2,048	2,048	0.758662	0.514481	12.112744	2.174486	0.459879	0.667431	0.826834	0.931628	1	0.514481	0.34563	0.262455	0.151211	0.143416	0.124755	0.118567	0.100626	0.087378	0.082617	0.078366	0.074355	0.066464	0.064451	0.051378	0.044849
1B	mnli	bf16	43	7	q_proj	2,048	2,048	1.309998	1.268944	6.397517	1.065751	0.938306	0.960962	0.978865	0.990928	1	1.268944	0.197183	0.150119	0.090482	0.083293	0.076385	0.067426	0.058156	0.054542	0.053132	0.047865	0.043861	0.042833	0.04096	0.033785	0.030029
1B	mnli	bf16	43	7	v_proj	512	2,048	0.539899	0.385385	10.626718	1.96261	0.509526	0.751293	0.901781	0.964491	1	0.385385	0.265468	0.185023	0.098143	0.077701	0.074044	0.067928	0.046318	0.045439	0.039993	0.039253	0.034723	0.034172	0.033136	0.029532	0.02832
1B	mnli	bf16	43	8	k_proj	512	2,048	0.997032	0.962142	5.945685	1.073844	0.931234	0.958187	0.98553	0.994666	1	0.962142	0.163689	0.13818	0.089928	0.054602	0.049132	0.044124	0.041704	0.034666	0.030715	0.027756	0.024307	0.023304	0.022579	0.01941	0.019149
1B	mnli	bf16	43	8	o_proj	2,048	2,048	1.006823	0.777847	9.948487	1.675399	0.596873	0.840463	0.923957	0.96838	1	0.777847	0.496916	0.257452	0.135485	0.122866	0.112806	0.097342	0.087948	0.081386	0.073042	0.065018	0.063068	0.060906	0.059866	0.049645	0.04616
1B	mnli	bf16	43	8	q_proj	2,048	2,048	0.979087	0.916909	7.438919	1.140224	0.877021	0.918164	0.967768	0.988773	1	0.916909	0.198597	0.186208	0.113477	0.086065	0.076523	0.059666	0.057555	0.052286	0.049258	0.041794	0.03775	0.032006	0.023243	0.021301	0.020306
1B	mnli	bf16	43	8	v_proj	512	2,048	0.616029	0.473115	8.565449	1.69538	0.589838	0.915004	0.952688	0.981988	1	0.473115	0.35128	0.093542	0.074503	0.066404	0.053393	0.046801	0.040848	0.036754	0.034593	0.031192	0.029634	0.029135	0.025307	0.023049	0.020402

End of preview.

GLUE LoRA x Backbone Bit-Width — combined results

Aggregated metrics, LoRA geometry analysis and figures for a controlled study of whether backbone bit-width (bf16 / int8 / nf4) changes what a LoRA adapter learns.

Grid: 2 model sizes (1B, 3B) x 4 GLUE tasks (MNLI, QQP, SST-2, RTE) x 3 bit-widths x 3 seeds = 72 runs (0 present here).

For each (model size, seed) the adapter initialisation is identical across bit-widths, so cross-bit differences are attributable to the backbone. The headline statistic is

Gamma = mean cross-bit-width adapter distance / mean within-bit-width seed distance

Gamma > 1 means bit-width perturbs the adapter more than the random seed does.

Files

results_summary.md — the report, answering the 8 study questions
aggregate_results.csv — per-run metrics
aggregate_results_by_task.csv, aggregate_results_by_condition.csv
adapter_layer_metrics.csv — per-adapter per-layer geometry (norms, effective rank, stable rank, singular values, energy)
pairwise_geometry_metrics.csv — per-pair per-layer comparisons (Frobenius distance, cosine, principal angles, projection overlap, sign agreement)
gamma_table.csv — bit-width vs seed effect sizes
plots/ — 7 figures

Adapters live in Jeesup/llama32-<size>-<task>-<bitwidth>-lora-seed<seed>.

Downloads last month: 40

Total file size:

9.35 MB