Datasets:

9Tempest
/

lp-gemm-qwen3-5-9b

Tags:

License:

Dataset card Data Studio Files Files and versions

xet

Community

Dataset Preview

Duplicate

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

Error code:   DatasetGenerationError
Exception:    CastError
Message:      Couldn't cast
format: string
lhs_dtype: string
lhs_shape: list<item: int64>
  child 0, item: int64
meta_path: string
model_id: string
module: string
path: string
rhs_dtype: string
rhs_shape: list<item: int64>
  child 0, item: int64
sample_idx: int64
out_features: int64
in_features: int64
to
{'format': Value('string'), 'in_features': Value('int64'), 'lhs_dtype': Value('string'), 'lhs_shape': List(Value('int64')), 'model_id': Value('string'), 'module': Value('string'), 'out_features': Value('int64'), 'rhs_dtype': Value('string'), 'rhs_shape': List(Value('int64')), 'sample_idx': Value('int64')}
because column names don't match
Traceback:    Traceback (most recent call last):
                File "/usr/local/lib/python3.12/site-packages/datasets/builder.py", line 1779, in _prepare_split_single
                  for key, table in generator:
                                    ^^^^^^^^^
                File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 609, in wrapped
                  for item in generator(*args, **kwargs):
                              ^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/packaged_modules/json/json.py", line 299, in _generate_tables
                  self._cast_table(pa_table, json_field_paths=json_field_paths),
                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/packaged_modules/json/json.py", line 128, in _cast_table
                  pa_table = table_cast(pa_table, self.info.features.arrow_schema)
                             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/table.py", line 2321, in table_cast
                  return cast_table_to_schema(table, schema)
                         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/table.py", line 2249, in cast_table_to_schema
                  raise CastError(
              datasets.table.CastError: Couldn't cast
              format: string
              lhs_dtype: string
              lhs_shape: list<item: int64>
                child 0, item: int64
              meta_path: string
              model_id: string
              module: string
              path: string
              rhs_dtype: string
              rhs_shape: list<item: int64>
                child 0, item: int64
              sample_idx: int64
              out_features: int64
              in_features: int64
              to
              {'format': Value('string'), 'in_features': Value('int64'), 'lhs_dtype': Value('string'), 'lhs_shape': List(Value('int64')), 'model_id': Value('string'), 'module': Value('string'), 'out_features': Value('int64'), 'rhs_dtype': Value('string'), 'rhs_shape': List(Value('int64')), 'sample_idx': Value('int64')}
              because column names don't match
              
              The above exception was the direct cause of the following exception:
              
              Traceback (most recent call last):
                File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 1342, in compute_config_parquet_and_info_response
                  parquet_operations, partial, estimated_dataset_info = stream_convert_to_parquet(
                                                                        ^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 907, in stream_convert_to_parquet
                  builder._prepare_split(split_generator=splits_generators[split], file_format="parquet")
                File "/usr/local/lib/python3.12/site-packages/datasets/builder.py", line 1646, in _prepare_split
                  for job_id, done, content in self._prepare_split_single(
                                               ^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/builder.py", line 1832, in _prepare_split_single
                  raise DatasetGenerationError("An error occurred while generating the dataset") from e
              datasets.exceptions.DatasetGenerationError: An error occurred while generating the dataset

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.

format string	in_features int64	lhs_dtype string	lhs_shape list	model_id string	module string	out_features int64	rhs_dtype string	rhs_shape list
linear_gemm_operand_pair_v1	12,288	torch.float32	[ 12, 12288 ]	Qwen/Qwen3.5-9B	model.layers.0.mlp.down_proj	4,096	torch.float32	[ 12288, 4096 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.0.mlp.gate_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.0.mlp.up_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	12,288	torch.float32	[ 12, 12288 ]	Qwen/Qwen3.5-9B	model.layers.10.mlp.down_proj	4,096	torch.float32	[ 12288, 4096 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.10.mlp.gate_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.10.mlp.up_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	12,288	torch.float32	[ 12, 12288 ]	Qwen/Qwen3.5-9B	model.layers.11.mlp.down_proj	4,096	torch.float32	[ 12288, 4096 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.11.mlp.gate_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.11.mlp.up_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.11.self_attn.k_proj	1,024	torch.float32	[ 4096, 1024 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.11.self_attn.o_proj	4,096	torch.float32	[ 4096, 4096 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.11.self_attn.q_proj	8,192	torch.float32	[ 4096, 8192 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.11.self_attn.v_proj	1,024	torch.float32	[ 4096, 1024 ]
linear_gemm_operand_pair_v1	12,288	torch.float32	[ 12, 12288 ]	Qwen/Qwen3.5-9B	model.layers.12.mlp.down_proj	4,096	torch.float32	[ 12288, 4096 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.12.mlp.gate_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.12.mlp.up_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	12,288	torch.float32	[ 12, 12288 ]	Qwen/Qwen3.5-9B	model.layers.13.mlp.down_proj	4,096	torch.float32	[ 12288, 4096 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.13.mlp.gate_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.13.mlp.up_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	12,288	torch.float32	[ 12, 12288 ]	Qwen/Qwen3.5-9B	model.layers.14.mlp.down_proj	4,096	torch.float32	[ 12288, 4096 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.14.mlp.gate_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.14.mlp.up_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	12,288	torch.float32	[ 12, 12288 ]	Qwen/Qwen3.5-9B	model.layers.15.mlp.down_proj	4,096	torch.float32	[ 12288, 4096 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.15.mlp.gate_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.15.mlp.up_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.15.self_attn.k_proj	1,024	torch.float32	[ 4096, 1024 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.15.self_attn.o_proj	4,096	torch.float32	[ 4096, 4096 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.15.self_attn.q_proj	8,192	torch.float32	[ 4096, 8192 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.15.self_attn.v_proj	1,024	torch.float32	[ 4096, 1024 ]
linear_gemm_operand_pair_v1	12,288	torch.float32	[ 12, 12288 ]	Qwen/Qwen3.5-9B	model.layers.16.mlp.down_proj	4,096	torch.float32	[ 12288, 4096 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.16.mlp.gate_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.16.mlp.up_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	12,288	torch.float32	[ 12, 12288 ]	Qwen/Qwen3.5-9B	model.layers.17.mlp.down_proj	4,096	torch.float32	[ 12288, 4096 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.17.mlp.gate_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.17.mlp.up_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	12,288	torch.float32	[ 12, 12288 ]	Qwen/Qwen3.5-9B	model.layers.18.mlp.down_proj	4,096	torch.float32	[ 12288, 4096 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.18.mlp.gate_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.18.mlp.up_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	12,288	torch.float32	[ 12, 12288 ]	Qwen/Qwen3.5-9B	model.layers.19.mlp.down_proj	4,096	torch.float32	[ 12288, 4096 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.19.mlp.gate_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.19.mlp.up_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.19.self_attn.k_proj	1,024	torch.float32	[ 4096, 1024 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.19.self_attn.o_proj	4,096	torch.float32	[ 4096, 4096 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.19.self_attn.q_proj	8,192	torch.float32	[ 4096, 8192 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.19.self_attn.v_proj	1,024	torch.float32	[ 4096, 1024 ]
linear_gemm_operand_pair_v1	12,288	torch.float32	[ 12, 12288 ]	Qwen/Qwen3.5-9B	model.layers.1.mlp.down_proj	4,096	torch.float32	[ 12288, 4096 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.1.mlp.gate_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.1.mlp.up_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	12,288	torch.float32	[ 12, 12288 ]	Qwen/Qwen3.5-9B	model.layers.20.mlp.down_proj	4,096	torch.float32	[ 12288, 4096 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.20.mlp.gate_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.20.mlp.up_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	12,288	torch.float32	[ 12, 12288 ]	Qwen/Qwen3.5-9B	model.layers.21.mlp.down_proj	4,096	torch.float32	[ 12288, 4096 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.21.mlp.gate_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.21.mlp.up_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	12,288	torch.float32	[ 12, 12288 ]	Qwen/Qwen3.5-9B	model.layers.22.mlp.down_proj	4,096	torch.float32	[ 12288, 4096 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.22.mlp.gate_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.22.mlp.up_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	12,288	torch.float32	[ 12, 12288 ]	Qwen/Qwen3.5-9B	model.layers.23.mlp.down_proj	4,096	torch.float32	[ 12288, 4096 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.23.mlp.gate_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.23.mlp.up_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.23.self_attn.k_proj	1,024	torch.float32	[ 4096, 1024 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.23.self_attn.o_proj	4,096	torch.float32	[ 4096, 4096 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.23.self_attn.q_proj	8,192	torch.float32	[ 4096, 8192 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.23.self_attn.v_proj	1,024	torch.float32	[ 4096, 1024 ]
linear_gemm_operand_pair_v1	12,288	torch.float32	[ 12, 12288 ]	Qwen/Qwen3.5-9B	model.layers.24.mlp.down_proj	4,096	torch.float32	[ 12288, 4096 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.24.mlp.gate_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.24.mlp.up_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	12,288	torch.float32	[ 12, 12288 ]	Qwen/Qwen3.5-9B	model.layers.25.mlp.down_proj	4,096	torch.float32	[ 12288, 4096 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.25.mlp.gate_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.25.mlp.up_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	12,288	torch.float32	[ 12, 12288 ]	Qwen/Qwen3.5-9B	model.layers.26.mlp.down_proj	4,096	torch.float32	[ 12288, 4096 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.26.mlp.gate_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.26.mlp.up_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	12,288	torch.float32	[ 12, 12288 ]	Qwen/Qwen3.5-9B	model.layers.27.mlp.down_proj	4,096	torch.float32	[ 12288, 4096 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.27.mlp.gate_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.27.mlp.up_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.27.self_attn.k_proj	1,024	torch.float32	[ 4096, 1024 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.27.self_attn.o_proj	4,096	torch.float32	[ 4096, 4096 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.27.self_attn.q_proj	8,192	torch.float32	[ 4096, 8192 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.27.self_attn.v_proj	1,024	torch.float32	[ 4096, 1024 ]
linear_gemm_operand_pair_v1	12,288	torch.float32	[ 12, 12288 ]	Qwen/Qwen3.5-9B	model.layers.28.mlp.down_proj	4,096	torch.float32	[ 12288, 4096 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.28.mlp.gate_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.28.mlp.up_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	12,288	torch.float32	[ 12, 12288 ]	Qwen/Qwen3.5-9B	model.layers.29.mlp.down_proj	4,096	torch.float32	[ 12288, 4096 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.29.mlp.gate_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.29.mlp.up_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	12,288	torch.float32	[ 12, 12288 ]	Qwen/Qwen3.5-9B	model.layers.2.mlp.down_proj	4,096	torch.float32	[ 12288, 4096 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.2.mlp.gate_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.2.mlp.up_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	12,288	torch.float32	[ 12, 12288 ]	Qwen/Qwen3.5-9B	model.layers.30.mlp.down_proj	4,096	torch.float32	[ 12288, 4096 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.30.mlp.gate_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.30.mlp.up_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	12,288	torch.float32	[ 12, 12288 ]	Qwen/Qwen3.5-9B	model.layers.31.mlp.down_proj	4,096	torch.float32	[ 12288, 4096 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.31.mlp.gate_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.31.mlp.up_proj	12,288	torch.float32	[ 4096, 12288 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.31.self_attn.k_proj	1,024	torch.float32	[ 4096, 1024 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.31.self_attn.o_proj	4,096	torch.float32	[ 4096, 4096 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.31.self_attn.q_proj	8,192	torch.float32	[ 4096, 8192 ]
linear_gemm_operand_pair_v1	4,096	torch.float32	[ 12, 4096 ]	Qwen/Qwen3.5-9B	model.layers.31.self_attn.v_proj	1,024	torch.float32	[ 4096, 1024 ]
linear_gemm_operand_pair_v1	12,288	torch.float32	[ 12, 12288 ]	Qwen/Qwen3.5-9B	model.layers.3.mlp.down_proj	4,096	torch.float32	[ 12288, 4096 ]

End of preview.

Low Precision GEMM Operand Samples

This dataset contains real Linear-layer GEMM operand pairs extracted from public language-model checkpoints for low-precision GEMM numerics studies.

Each .pt sample is a dictionary:

{
    "lhs": Tensor[tokens, in_features],
    "rhs": Tensor[in_features, out_features],
    "meta": {...},
}

For a PyTorch nn.Linear, the captured GEMM is:

output = lhs @ rhs

where lhs is the flattened input activation to the layer and rhs is linear.weight.T.

Source Models

Qwen/Qwen3.5-9B

Samples: 128
Source folder name: qwen3_5_9b
Manifest: manifest.jsonl

Module type counts:

down_proj: 32
gate_proj: 32
k_proj: 8
o_proj: 8
q_proj: 8
up_proj: 32
v_proj: 8

Intended Use

Use this dataset to evaluate low-precision GEMM numerics and quantization schemes on real model operand distributions. It is not a text dataset and is not intended for model training.

Produced With

This dataset was produced with the NumericsBenchmark extractor: https://github.com/9Tempest/NumericsBenchmark

git clone https://github.com/9Tempest/NumericsBenchmark.git
cd NumericsBenchmark
python -m venv .venv
source .venv/bin/activate
pip install -e '.[extract]'

numerics-bench-extract \
    --model Qwen/Qwen3.5-9B \
    --out-dir extracted_gemms/qwen3_5_9b \
    --prompt "Explain why low precision GEMM numerics matter." \
    --max-samples-per-module 1 \
    --max-rows 4096

The extractor registers forward hooks on torch.nn.Linear modules, captures the flattened layer input as lhs, stores linear.weight.T as rhs, and writes one .pt operand-pair sample plus JSON metadata per captured module.

Run The Benchmark

The recommended way to evaluate these samples is to use the same NumericsBenchmark repo:

git clone https://github.com/9Tempest/NumericsBenchmark.git
cd NumericsBenchmark
python -m venv .venv
source .venv/bin/activate
pip install -e '.[extract]'

python - <<'PY'
from huggingface_hub import snapshot_download
snapshot_download(
    repo_id="9Tempest/lp-gemm-qwen3-5-9b",
    repo_type="dataset",
    local_dir="extracted_gemms/qwen3_5_9b",
)
PY

numerics-bench-real-eval \
    --data-dir extracted_gemms/qwen3_5_9b \
    --scheme nvfp4 \
    --rht off,on \
    --gemm-mode qdq_fp32 \
    --csv outputs/qwen3_5_9b_nvfp4_rht_real_eval.csv

Format

The companion manifest.jsonl lists each sample path, model id, module name, tensor shapes, dtypes, and format version.

Downloads last month: 4

Total file size:

21.2 GB