Dataset Viewer
The dataset viewer is not available for this split.
Cannot load the dataset split (in streaming mode) to extract the first rows.
Error code: StreamingRowsError
Exception: CastError
Message: Couldn't cast
format: string
synthetic: bool
full_expert_weights: bool
files: list<item: struct<path: string, bytes: int64, sha256: string>>
child 0, item: struct<path: string, bytes: int64, sha256: string>
child 0, path: string
child 1, bytes: int64
child 2, sha256: string
input: struct<text: string, add_bos: bool, chat_template: bool, token_ids: list<item: int64>, token_positio (... 41 chars omitted)
child 0, text: string
child 1, add_bos: bool
child 2, chat_template: bool
child 3, token_ids: list<item: int64>
child 0, item: int64
child 4, token_position: int64
child 5, token_id: int64
child 6, phase: string
prefix: struct<implementation: string, tensor_parallel_world_size: int64, attention_head_chunk: int64, round (... 12 chars omitted)
child 0, implementation: string
child 1, tensor_parallel_world_size: int64
child 2, attention_head_chunk: int64
child 3, rounding: string
source: struct<model: string, url: string, revision: string, config_sha256: string, index_sha256: string, to (... 14339 chars omitted)
child 0, model: string
child 1, url: string
child 2, revision: string
child 3, config_sha256: string
child 4, index_sha256: string
child 5, tokenizer_sha256: string
child 6, tensor_spans: struct<embed.weight.row0: struct<tensor: string, shard: string, dtype: string, source_shape: list<it (... 14200 chars omitted)
child 0, embed.weight.row0: struct<tensor: string, shard: string, dtype: string, source_shape: list<item: in
...
child 0, file: string
child 1, shape: list<item: int64>
child 0, item: int64
child 2, dtype: string
child 3, bytes: int64
child 4, sha256: string
child 7, eh_int8: struct<file: string, shape: list<item: int64>, dtype: string, bytes: int64, sha256: string>
child 0, file: string
child 1, shape: list<item: int64>
child 0, item: int64
child 2, dtype: string
child 3, bytes: int64
child 4, sha256: string
child 8, down_bf16: struct<file: string, shape: list<item: int64>, dtype: string, bytes: int64, sha256: string>
child 0, file: string
child 1, shape: list<item: int64>
child 0, item: int64
child 2, dtype: string
child 3, bytes: int64
child 4, sha256: string
child 9, weighted_bf16: struct<file: string, shape: list<item: int64>, dtype: string, bytes: int64, sha256: string>
child 0, file: string
child 1, shape: list<item: int64>
child 0, item: int64
child 2, dtype: string
child 3, bytes: int64
child 4, sha256: string
child 10, routed_bf16: struct<file: string, shape: list<item: int64>, dtype: string, bytes: int64, sha256: string>
child 0, file: string
child 1, shape: list<item: int64>
child 0, item: int64
child 2, dtype: string
child 3, bytes: int64
child 4, sha256: string
route_weights_fp32: list<item: list<item: double>>
child 0, item: list<item: double>
child 0, item: double
to
{'format': Value('string'), 'synthetic': Value('bool'), 'created_at': Value('string'), 'source': {'model': Value('string'), 'url': Value('string'), 'revision': Value('string'), 'config_sha256': Value('string'), 'index_sha256': Value('string'), 'tokenizer_sha256': Value('string'), 'tensor_spans': {'embed.weight.row0': {'tensor': Value('string'), 'shard': Value('string'), 'dtype': Value('string'), 'source_shape': List(Value('int64')), 'tensor_byte_offset': Value('int64'), 'nbytes': Value('int64'), 'sha256': Value('string'), 'etag': Value('string')}, 'embed.weight.row2788': {'tensor': Value('string'), 'shard': Value('string'), 'dtype': Value('string'), 'source_shape': List(Value('int64')), 'tensor_byte_offset': Value('int64'), 'nbytes': Value('int64'), 'sha256': Value('string'), 'etag': Value('string')}, 'embed.weight.row9068': {'tensor': Value('string'), 'shard': Value('string'), 'dtype': Value('string'), 'source_shape': List(Value('int64')), 'tensor_byte_offset': Value('int64'), 'nbytes': Value('int64'), 'sha256': Value('string'), 'etag': Value('string')}, 'embed.weight.row24479': {'tensor': Value('string'), 'shard': Value('string'), 'dtype': Value('string'), 'source_shape': List(Value('int64')), 'tensor_byte_offset': Value('int64'), 'nbytes': Value('int64'), 'sha256': Value('string'), 'etag': Value('string')}, 'embed.weight.row46202': {'tensor': Value('string'), 'shard': Value('string'), 'dtype': Value('string'), 'source_shape': List(Value('int64')), 'tensor_byte_offset': Val
...
rence': Value('bool'), 'w13_status_zero': Value('bool'), 'swiglu_repeat': Value('bool'), 'qh_reference': Value('bool'), 'eh_reference': Value('bool'), 'w2_split_k': Value('bool'), 'w2_bf16_reference': Value('bool'), 'w2_accumulator_reference': Value('bool'), 'w2_common_exponent_reference': Value('bool'), 'w2_status_zero': Value('bool'), 'weighted_sum_fused_vs_materialized': Value('bool'), 'activation_status_zero': Value('bool'), 'x_bf16_disk_readback': Value('bool'), 'qx_int8_disk_readback': Value('bool'), 'ex_int8_disk_readback': Value('bool'), 'hidden_bf16_disk_readback': Value('bool'), 'qh_int8_disk_readback': Value('bool'), 'eh_int8_disk_readback': Value('bool'), 'down_bf16_disk_readback': Value('bool'), 'weighted_bf16_disk_readback': Value('bool'), 'routed_bf16_disk_readback': Value('bool'), 'gate_bf16_disk_readback': Value('bool'), 'up_bf16_disk_readback': Value('bool'), 'gate_up_reconstruct_original_bitwise': Value('bool'), 'other_tensor_files_unchanged': Value('bool')}, 'validated': Value('bool'), 'export_transform': {'operation': Value('string'), 'exported_at': Value('string'), 'source_repo': Value('string'), 'source_revision': Value('string'), 'source_manifest_sha256': Value('string'), 'source_tensor': {'file': Value('string'), 'shape': List(Value('int64')), 'dtype': Value('string'), 'bytes': Value('int64'), 'sha256': Value('string')}, 'column_ranges': {'gate_bf16': List(Value('int64')), 'up_bf16': List(Value('int64'))}, 'roundtrip_bitwise_verified': Value('bool')}}
because column names don't match
Traceback: Traceback (most recent call last):
File "/src/services/worker/src/worker/utils.py", line 147, in get_rows_or_raise
return get_rows(
dataset=dataset,
...<4 lines>...
column_names=column_names,
)
File "/src/libs/libcommon/src/libcommon/utils.py", line 272, in decorator
return func(*args, **kwargs)
File "/src/services/worker/src/worker/utils.py", line 127, in get_rows
rows_plus_one = list(itertools.islice(safe_iter(ds, dataset=dataset), rows_max_number + 1))
File "/src/services/worker/src/worker/utils.py", line 483, in safe_iter
yield from ds.decode(False) if ds.features else ds
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2840, in __iter__
for key, example in ex_iterable:
^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2373, in __iter__
for key, pa_table in self._iter_arrow():
~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2398, in _iter_arrow
for key, pa_table in self.ex_iterable._iter_arrow():
~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 536, in _iter_arrow
for key, pa_table in iterator:
^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 419, in _iter_arrow
for key, pa_table in self.generate_tables_fn(**gen_kwags):
~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 343, in _generate_tables
self._cast_table(pa_table, json_field_paths=json_field_paths),
~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 132, in _cast_table
pa_table = table_cast(pa_table, self.info.features.arrow_schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2378, in table_cast
return cast_table_to_schema(table, schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2306, in cast_table_to_schema
raise CastError(
...<3 lines>...
)
datasets.table.CastError: Couldn't cast
format: string
synthetic: bool
full_expert_weights: bool
files: list<item: struct<path: string, bytes: int64, sha256: string>>
child 0, item: struct<path: string, bytes: int64, sha256: string>
child 0, path: string
child 1, bytes: int64
child 2, sha256: string
input: struct<text: string, add_bos: bool, chat_template: bool, token_ids: list<item: int64>, token_positio (... 41 chars omitted)
child 0, text: string
child 1, add_bos: bool
child 2, chat_template: bool
child 3, token_ids: list<item: int64>
child 0, item: int64
child 4, token_position: int64
child 5, token_id: int64
child 6, phase: string
prefix: struct<implementation: string, tensor_parallel_world_size: int64, attention_head_chunk: int64, round (... 12 chars omitted)
child 0, implementation: string
child 1, tensor_parallel_world_size: int64
child 2, attention_head_chunk: int64
child 3, rounding: string
source: struct<model: string, url: string, revision: string, config_sha256: string, index_sha256: string, to (... 14339 chars omitted)
child 0, model: string
child 1, url: string
child 2, revision: string
child 3, config_sha256: string
child 4, index_sha256: string
child 5, tokenizer_sha256: string
child 6, tensor_spans: struct<embed.weight.row0: struct<tensor: string, shard: string, dtype: string, source_shape: list<it (... 14200 chars omitted)
child 0, embed.weight.row0: struct<tensor: string, shard: string, dtype: string, source_shape: list<item: in
...
child 0, file: string
child 1, shape: list<item: int64>
child 0, item: int64
child 2, dtype: string
child 3, bytes: int64
child 4, sha256: string
child 7, eh_int8: struct<file: string, shape: list<item: int64>, dtype: string, bytes: int64, sha256: string>
child 0, file: string
child 1, shape: list<item: int64>
child 0, item: int64
child 2, dtype: string
child 3, bytes: int64
child 4, sha256: string
child 8, down_bf16: struct<file: string, shape: list<item: int64>, dtype: string, bytes: int64, sha256: string>
child 0, file: string
child 1, shape: list<item: int64>
child 0, item: int64
child 2, dtype: string
child 3, bytes: int64
child 4, sha256: string
child 9, weighted_bf16: struct<file: string, shape: list<item: int64>, dtype: string, bytes: int64, sha256: string>
child 0, file: string
child 1, shape: list<item: int64>
child 0, item: int64
child 2, dtype: string
child 3, bytes: int64
child 4, sha256: string
child 10, routed_bf16: struct<file: string, shape: list<item: int64>, dtype: string, bytes: int64, sha256: string>
child 0, file: string
child 1, shape: list<item: int64>
child 0, item: int64
child 2, dtype: string
child 3, bytes: int64
child 4, sha256: string
route_weights_fp32: list<item: list<item: double>>
child 0, item: list<item: double>
child 0, item: double
to
{'format': Value('string'), 'synthetic': Value('bool'), 'created_at': Value('string'), 'source': {'model': Value('string'), 'url': Value('string'), 'revision': Value('string'), 'config_sha256': Value('string'), 'index_sha256': Value('string'), 'tokenizer_sha256': Value('string'), 'tensor_spans': {'embed.weight.row0': {'tensor': Value('string'), 'shard': Value('string'), 'dtype': Value('string'), 'source_shape': List(Value('int64')), 'tensor_byte_offset': Value('int64'), 'nbytes': Value('int64'), 'sha256': Value('string'), 'etag': Value('string')}, 'embed.weight.row2788': {'tensor': Value('string'), 'shard': Value('string'), 'dtype': Value('string'), 'source_shape': List(Value('int64')), 'tensor_byte_offset': Value('int64'), 'nbytes': Value('int64'), 'sha256': Value('string'), 'etag': Value('string')}, 'embed.weight.row9068': {'tensor': Value('string'), 'shard': Value('string'), 'dtype': Value('string'), 'source_shape': List(Value('int64')), 'tensor_byte_offset': Value('int64'), 'nbytes': Value('int64'), 'sha256': Value('string'), 'etag': Value('string')}, 'embed.weight.row24479': {'tensor': Value('string'), 'shard': Value('string'), 'dtype': Value('string'), 'source_shape': List(Value('int64')), 'tensor_byte_offset': Value('int64'), 'nbytes': Value('int64'), 'sha256': Value('string'), 'etag': Value('string')}, 'embed.weight.row46202': {'tensor': Value('string'), 'shard': Value('string'), 'dtype': Value('string'), 'source_shape': List(Value('int64')), 'tensor_byte_offset': Val
...
rence': Value('bool'), 'w13_status_zero': Value('bool'), 'swiglu_repeat': Value('bool'), 'qh_reference': Value('bool'), 'eh_reference': Value('bool'), 'w2_split_k': Value('bool'), 'w2_bf16_reference': Value('bool'), 'w2_accumulator_reference': Value('bool'), 'w2_common_exponent_reference': Value('bool'), 'w2_status_zero': Value('bool'), 'weighted_sum_fused_vs_materialized': Value('bool'), 'activation_status_zero': Value('bool'), 'x_bf16_disk_readback': Value('bool'), 'qx_int8_disk_readback': Value('bool'), 'ex_int8_disk_readback': Value('bool'), 'hidden_bf16_disk_readback': Value('bool'), 'qh_int8_disk_readback': Value('bool'), 'eh_int8_disk_readback': Value('bool'), 'down_bf16_disk_readback': Value('bool'), 'weighted_bf16_disk_readback': Value('bool'), 'routed_bf16_disk_readback': Value('bool'), 'gate_bf16_disk_readback': Value('bool'), 'up_bf16_disk_readback': Value('bool'), 'gate_up_reconstruct_original_bitwise': Value('bool'), 'other_tensor_files_unchanged': Value('bool')}, 'validated': Value('bool'), 'export_transform': {'operation': Value('string'), 'exported_at': Value('string'), 'source_repo': Value('string'), 'source_revision': Value('string'), 'source_manifest_sha256': Value('string'), 'source_tensor': {'file': Value('string'), 'shape': List(Value('int64')), 'dtype': Value('string'), 'bytes': Value('int64'), 'sha256': Value('string')}, 'column_ranges': {'gate_bf16': List(Value('int64')), 'up_bf16': List(Value('int64'))}, 'roundtrip_bitwise_verified': Value('bool')}}
because column names don't matchNeed help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
W4A8 HW 真实推理张量
使用本机 DeepSeek-V4-Pro 真实权重，输入文本 “请解释矩阵乘法。”，取第 0 层、prefill 最后一个 token。真实权重按 HW 规则量化后，在 CUDA 上计算以下张量。未使用随机输入或合成权重。
下载全部张量（压缩包） · 来源、shape 与 SHA256
| 文件名，不含 .bin（点击下载） | Shape | 含义 |
|---|---|---|
| x_bf16 | [1,7168] |
FFN 原始输入 |
| qx_int8、ex_int8 | [1,7168]、[1,56] |
量化输入及指数 |
| gate_bf16 | [6,3072] |
6 个专家的完整 gate（W1） 投影结果 |
| up_bf16 | [6,3072] |
6 个专家的完整 up（W3） 投影结果 |
| hidden_bf16 | [6,3072] |
SwiGLU 输出 |
| qh_int8、eh_int8 | [6,3072]、[6,24] |
W2 的量化输入及指数 |
| down_bf16 | [6,7168] |
完整 W2 输出，尚未乘路由权重 |
| weighted_bf16 | [6,7168] |
各专家路由加权后的结果 |
| routed_bf16 | [1,7168] |
6 个完整专家的路由加权求和结果 |
6 行专家依次为 225、184、230、348、333、158。每个专家使用完整的 3072 维中间层，gate 和 up 分别保存在独立文件中。
前置 embedding、attention、mHC 和归一化按原生 TP=1 计算；路由 FFN 使用完整专家权重（TP=1）。W2 对全部 3072 个通道执行 HW 累加和最终 BF16 舍入。routed_bf16 为完整路由分支输出，尚未加上共享专家输出。
文件为小端、行优先、无文件头。BF16 保存原始 16 位数据；INT8 保存有符号字节。指数 E 表示 scale=2**E，-128 表示全零组。两个 GEMM 已通过独立整数参考的逐位校验；SwiGLU 使用本仓库 HW CUDA 实现。
下载并读取示例（无需登录）：
import torch
from huggingface_hub import snapshot_download
root = snapshot_download(
"chiruan/deepseek-v4-pro-w4a8-hw-golden", repo_type="dataset",
allow_patterns=["tensors/*.bin", "manifest.json"],
)
x = torch.from_file(f"{root}/tensors/x_bf16.bin", dtype=torch.bfloat16,
size=7168).reshape(1, 7168)
- Downloads last month
- 28