Dataset Viewer
Auto-converted to Parquet Duplicate
task_id
string
operator
string
batch
string
active
bool
backend
string
task_spec
string
readme
string
source_dir
string
test_script
string
case_manifests
list
golden_tensor_files
int64
editable_prefixes
list
component_license
string
component_license_file
string
source_commit
string
add_rms_norm
add_rms_norm
batch1
true
nn
datasets/tasks/batch1/add_rms_norm/task.json
datasets/tasks/batch1/add_rms_norm/README.md
datasets/ops_src/ops_nn/test_kernels/kernels/add_rms_norm
datasets/ops_src/ops_nn/test_kernels/kernels/add_rms_norm/add_rms_norm_test.py
[ "datasets/ops_src/ops_nn/test_kernels/kernels/add_rms_norm/test_data/cases_manifest.json" ]
20
[ "op_host/", "op_kernel/" ]
CANN Open Software License Agreement Version 2.0
datasets/LICENSES/ops-nn/LICENSE
51b0b55bc313d7cdeb4a4bbd53b071258dce07f4
clipped_swiglu
clipped_swiglu
batch1
true
nn
datasets/tasks/batch1/clipped_swiglu/task.json
datasets/tasks/batch1/clipped_swiglu/README.md
datasets/ops_src/ops_nn/test_kernels/kernels/clipped_swiglu
datasets/ops_src/ops_nn/test_kernels/kernels/clipped_swiglu/clipped_swiglu_test.py
[ "datasets/ops_src/ops_nn/test_kernels/kernels/clipped_swiglu/test_data/cases_manifest.json" ]
20
[ "op_host/", "op_kernel/" ]
CANN Open Software License Agreement Version 2.0
datasets/LICENSES/ops-nn/LICENSE
51b0b55bc313d7cdeb4a4bbd53b071258dce07f4
rms_norm
rms_norm
batch1
true
nn
datasets/tasks/batch1/rms_norm/task.json
datasets/tasks/batch1/rms_norm/README.md
datasets/ops_src/ops_nn/test_kernels/kernels/rms_norm
datasets/ops_src/ops_nn/test_kernels/kernels/rms_norm/rms_norm_test.py
[ "datasets/ops_src/ops_nn/test_kernels/kernels/rms_norm/test_data/cases_manifest.json" ]
20
[ "op_host/", "op_kernel/" ]
CANN Open Software License Agreement Version 2.0
datasets/LICENSES/ops-nn/LICENSE
51b0b55bc313d7cdeb4a4bbd53b071258dce07f4
swi_glu
swi_glu
batch1
true
nn
datasets/tasks/batch1/swi_glu/task.json
datasets/tasks/batch1/swi_glu/README.md
datasets/ops_src/ops_nn/test_kernels/kernels/swi_glu
datasets/ops_src/ops_nn/test_kernels/kernels/swi_glu/swi_glu_test.py
[ "datasets/ops_src/ops_nn/test_kernels/kernels/swi_glu/test_data/cases_manifest.json" ]
20
[ "op_host/", "op_kernel/" ]
CANN Open Software License Agreement Version 2.0
datasets/LICENSES/ops-nn/LICENSE
51b0b55bc313d7cdeb4a4bbd53b071258dce07f4
interleave_rope
interleave_rope
batch1
true
transformer
datasets/tasks/batch1/interleave_rope/task.json
datasets/tasks/batch1/interleave_rope/README.md
datasets/ops_src/ops_transformer/test_kernels/kernels/interleave_rope
datasets/ops_src/ops_transformer/test_kernels/kernels/interleave_rope/interleave_rope_test.py
[ "datasets/ops_src/ops_transformer/test_kernels/kernels/interleave_rope/test_data/cases_manifest.json" ]
20
[ "op_host/", "op_kernel/" ]
CANN Open Software License Agreement Version 2.0
datasets/LICENSES/ops-transformer/LICENSE
51b0b55bc313d7cdeb4a4bbd53b071258dce07f4
lightning_indexer
lightning_indexer
batch1
true
transformer
datasets/tasks/batch1/lightning_indexer/task.json
datasets/tasks/batch1/lightning_indexer/README.md
datasets/ops_src/ops_transformer/test_kernels/kernels/lightning_indexer
datasets/ops_src/ops_transformer/test_kernels/kernels/lightning_indexer/lightning_indexer_test.py
[ "datasets/ops_src/ops_transformer/test_kernels/kernels/lightning_indexer/test_data/cases_manifest.json" ]
20
[ "op_host/", "op_kernel/" ]
CANN Open Software License Agreement Version 2.0
datasets/LICENSES/ops-transformer/LICENSE
51b0b55bc313d7cdeb4a4bbd53b071258dce07f4
mla_prolog
mla_prolog
batch1
true
transformer
datasets/tasks/batch1/mla_prolog/task.json
datasets/tasks/batch1/mla_prolog/README.md
datasets/ops_src/ops_transformer/test_kernels/kernels/mla_prolog
datasets/ops_src/ops_transformer/test_kernels/kernels/mla_prolog/mla_prolog_test.py
[ "datasets/ops_src/ops_transformer/test_kernels/kernels/mla_prolog/test_data/cases_manifest.json" ]
20
[ "op_host/", "op_kernel/" ]
CANN Open Software License Agreement Version 2.0
datasets/LICENSES/ops-transformer/LICENSE
51b0b55bc313d7cdeb4a4bbd53b071258dce07f4
moe_fused_topk
moe_fused_topk
batch1
true
transformer
datasets/tasks/batch1/moe_fused_topk/task.json
datasets/tasks/batch1/moe_fused_topk/README.md
datasets/ops_src/ops_transformer/test_kernels/kernels/moe_fused_topk
datasets/ops_src/ops_transformer/test_kernels/kernels/moe_fused_topk/moe_fused_topk_test.py
[ "datasets/ops_src/ops_transformer/test_kernels/kernels/moe_fused_topk/test_data/cases_manifest.json" ]
20
[ "op_host/", "op_kernel/" ]
CANN Open Software License Agreement Version 2.0
datasets/LICENSES/ops-transformer/LICENSE
51b0b55bc313d7cdeb4a4bbd53b071258dce07f4
moe_gating_top_k
moe_gating_top_k
batch1
true
transformer
datasets/tasks/batch1/moe_gating_top_k/task.json
datasets/tasks/batch1/moe_gating_top_k/README.md
datasets/ops_src/ops_transformer/test_kernels/kernels/moe_gating_top_k
datasets/ops_src/ops_transformer/test_kernels/kernels/moe_gating_top_k/moe_gating_top_k_test.py
[ "datasets/ops_src/ops_transformer/test_kernels/kernels/moe_gating_top_k/test_data/cases_manifest.json" ]
20
[ "op_host/", "op_kernel/" ]
CANN Open Software License Agreement Version 2.0
datasets/LICENSES/ops-transformer/LICENSE
51b0b55bc313d7cdeb4a4bbd53b071258dce07f4
moe_token_unpermute
moe_token_unpermute
batch1
true
transformer
datasets/tasks/batch1/moe_token_unpermute/task.json
datasets/tasks/batch1/moe_token_unpermute/README.md
datasets/ops_src/ops_transformer/test_kernels/kernels/moe_token_unpermute
datasets/ops_src/ops_transformer/test_kernels/kernels/moe_token_unpermute/moe_token_unpermute_test.py
[ "datasets/ops_src/ops_transformer/test_kernels/kernels/moe_token_unpermute/test_data/cases_manifest.json" ]
20
[ "op_host/", "op_kernel/" ]
CANN Open Software License Agreement Version 2.0
datasets/LICENSES/ops-transformer/LICENSE
51b0b55bc313d7cdeb4a4bbd53b071258dce07f4
qkv_rms_norm_rope_cache
qkv_rms_norm_rope_cache
batch1
true
transformer
datasets/tasks/batch1/qkv_rms_norm_rope_cache/task.json
datasets/tasks/batch1/qkv_rms_norm_rope_cache/README.md
datasets/ops_src/ops_transformer/test_kernels/kernels/qkv_rms_norm_rope_cache
datasets/ops_src/ops_transformer/test_kernels/kernels/qkv_rms_norm_rope_cache/qkv_rms_norm_rope_cache_test.py
[ "datasets/ops_src/ops_transformer/test_kernels/kernels/qkv_rms_norm_rope_cache/test_data/cases_manifest.json" ]
20
[ "op_host/", "op_kernel/" ]
CANN Open Software License Agreement Version 2.0
datasets/LICENSES/ops-transformer/LICENSE
51b0b55bc313d7cdeb4a4bbd53b071258dce07f4
recurrent_gated_delta_rule
recurrent_gated_delta_rule
batch1
true
transformer
datasets/tasks/batch1/recurrent_gated_delta_rule/task.json
datasets/tasks/batch1/recurrent_gated_delta_rule/README.md
datasets/ops_src/ops_transformer/test_kernels/kernels/recurrent_gated_delta_rule
datasets/ops_src/ops_transformer/test_kernels/kernels/recurrent_gated_delta_rule/recurrent_gated_delta_rule_test.py
[ "datasets/ops_src/ops_transformer/test_kernels/kernels/recurrent_gated_delta_rule/test_data/cases_manifest.json" ]
20
[ "op_host/", "op_kernel/" ]
CANN Open Software License Agreement Version 2.0
datasets/LICENSES/ops-transformer/LICENSE
51b0b55bc313d7cdeb4a4bbd53b071258dce07f4
rope_with_sin_cos_cache
rope_with_sin_cos_cache
batch1
true
transformer
datasets/tasks/batch1/rope_with_sin_cos_cache/task.json
datasets/tasks/batch1/rope_with_sin_cos_cache/README.md
datasets/ops_src/ops_transformer/test_kernels/kernels/rope_with_sin_cos_cache
datasets/ops_src/ops_transformer/test_kernels/kernels/rope_with_sin_cos_cache/rope_with_sin_cos_cache_test.py
[ "datasets/ops_src/ops_transformer/test_kernels/kernels/rope_with_sin_cos_cache/test_data/cases_manifest.json" ]
20
[ "op_host/", "op_kernel/" ]
CANN Open Software License Agreement Version 2.0
datasets/LICENSES/ops-transformer/LICENSE
51b0b55bc313d7cdeb4a4bbd53b071258dce07f4
sparse_flash_attention
sparse_flash_attention
batch1
true
transformer
datasets/tasks/batch1/sparse_flash_attention/task.json
datasets/tasks/batch1/sparse_flash_attention/README.md
datasets/ops_src/ops_transformer/test_kernels/kernels/sparse_flash_attention
datasets/ops_src/ops_transformer/test_kernels/kernels/sparse_flash_attention/sparse_flash_attention_test.py
[ "datasets/ops_src/ops_transformer/test_kernels/kernels/sparse_flash_attention/test_data/cases_manifest.json" ]
20
[ "op_host/", "op_kernel/" ]
CANN Open Software License Agreement Version 2.0
datasets/LICENSES/ops-transformer/LICENSE
51b0b55bc313d7cdeb4a4bbd53b071258dce07f4
fused_gdn_gating
fused_gdn_gating
batch1
true
vllm_ascend
datasets/tasks/batch1/fused_gdn_gating/task.json
datasets/tasks/batch1/fused_gdn_gating/README.md
datasets/ops_src/vllm_ascend/kernels/fused_gdn_gating
datasets/ops_src/vllm_ascend/kernels/fused_gdn_gating/fused_gdn_gating_test.py
[]
0
[ "op_host/", "op_kernel/" ]
Apache-2.0
datasets/LICENSES/vllm-ascend/LICENSE
51b0b55bc313d7cdeb4a4bbd53b071258dce07f4
hc_post
hc_post
batch1
true
vllm_ascend
datasets/tasks/batch1/hc_post/task.json
datasets/tasks/batch1/hc_post/README.md
datasets/ops_src/vllm_ascend/kernels/hc_post
datasets/ops_src/vllm_ascend/kernels/hc_post/hc_post_test.py
[]
0
[ "op_host/", "op_kernel/" ]
Apache-2.0
datasets/LICENSES/vllm-ascend/LICENSE
51b0b55bc313d7cdeb4a4bbd53b071258dce07f4
hc_pre_sinkhorn
hc_pre_sinkhorn
batch1
true
vllm_ascend
datasets/tasks/batch1/hc_pre_sinkhorn/task.json
datasets/tasks/batch1/hc_pre_sinkhorn/README.md
datasets/ops_src/vllm_ascend/kernels/hc_pre_sinkhorn
datasets/ops_src/vllm_ascend/kernels/hc_pre_sinkhorn/hc_pre_sinkhorn_test.py
[]
0
[ "op_host/", "op_kernel/" ]
Apache-2.0
datasets/LICENSES/vllm-ascend/LICENSE
51b0b55bc313d7cdeb4a4bbd53b071258dce07f4

AscendGYM — Kernel Optimization Benchmark on Ascend

AscendGYM evaluates the ability of models and coding-agent harnesses to optimize operators for Ascend hardware. This release provides task definitions, operator baseline snapshots, frozen correctness tests, golden tensors, and the evaluation client code, preserving the upstream directory layout.

The benchmark targets kernel optimization in real-world training and inference workloads. The benchmark package includes the files needed for local task preparation. Compilation, correctness testing, and performance profiling require access to an Ascend execution environment. For the complete workflow, see the original project instructions.

Benchmark tasks

Official evaluation: Batch 1

The default batch1 subset contains 17 evaluation tasks. They are grouped below by their source component; each operator links to its task guide. All tasks include frozen correctness tests.

Additional reference materials: 3 tasks outside the official benchmark

The preserved_golden subset retains the following ops-transformer sources and golden references. These tasks are separate from the official Batch 1 evaluation.

Task indexes and files

The Dataset Viewer and load_dataset() expose task metadata, one row per task. They do not load tensors, execute tests, or run the benchmark. Both subsets use the test split; no training split is provided. The underlying indexes are Batch 1 and reference materials.

Each index row includes the task ID, operator, batch, active status, backend, paths to the task specification and tests, case manifests, editable source prefixes, tensor count, baseline component license, and source revision. Paths are relative to the full downloaded repository root. License fields describe the baseline component; file-level notices continue to apply.

The golden_tensor_files field counts packaged .pt files, not test cases. Tasks without packaged .pt files still include frozen tests.

README.md                    # This Dataset Card
README_PROJECT.md            # Original upstream project instructions
metadata/                    # Viewer indexes and release provenance
datasets/
  registry.json
  tasks/                     # Task specifications and task documentation
  ops_src/                   # Operator baselines, tests and frozen tensors
  LICENSES/                  # Third-party licenses and notices
  SOURCE_MANIFEST.json        # Upstream file sizes and SHA-256 hashes
bridge/                      # Task preparation and evaluation entry points
eval/                        # Native and bare harness materials
environment/                 # Reference stack and pinned upstream dependencies
docs/                        # Technical report, guides, and documentation assets
examples/
LICENSE                      # Component license summary and full license texts
LICENSE-APACHE-2.0            # Unmodified original project license text
LICENSES.md                  # Component scope and third-party notice links
CHECKSUMS.sha256              # Release file integrity list

Load the task index

Set HF_REPO_ID to the actual Dataset repository ID hosting this release:

export HF_REPO_ID='SLAI-AITP/AscendGYM'
python -m pip install datasets huggingface_hub
import os
from datasets import load_dataset

tasks = load_dataset(os.environ["HF_REPO_ID"], "batch1", split="test")
preserved = load_dataset(os.environ["HF_REPO_ID"], "preserved_golden", split="test")
print(tasks[0]["task_id"], tasks[0]["source_dir"])

For reproducible use, pass revision= with the Hugging Face commit of this release. The GitHub source commit above is provenance and is not a Hugging Face revision identifier.

Download the complete benchmark

import os
from huggingface_hub import snapshot_download

snapshot_download(
    repo_id=os.environ["HF_REPO_ID"],
    repo_type="dataset",
    local_dir="./AscendGYM",
)

Check the downloaded file hashes before generating local task packages, then prepare the tasks:

cd AscendGYM
sha256sum --check CHECKSUMS.sha256
source datasets/activate.sh
python3 bridge/op_manager.py scan
python3 bridge/op_manager.py build --all

The scan and build commands prepare local files; they do not compile kernels, contact a model, or establish NPU correctness or performance. Evaluation client dependencies and the complete run procedure are in README_PROJECT.md.

Execution requirements

The client requires Python 3.10+ and can run separately from the execution host. Actual compilation and profiling require a compatible Ascend NPU host, driver, CANN stack, operator dependencies, and a reachable execution service. Uploading or loading this dataset does not provision those resources.

See the container guide, self-hosted environment guide, execution protocol, reference stack, and pinned build dependencies. The reference stack is a record of selected checks, not a guarantee that every operator works on every Ascend installation.

Provenance, integrity and limitations

metadata/release.json records the GitHub source commit and release overlays. The original datasets/SOURCE_MANIFEST.json is preserved unchanged. Export verification checked 1282 source files against that manifest. CHECKSUMS.sha256 additionally covers the complete release payload, including the task indexes and this card.

Golden tensors and frozen tests are correctness references. They are not model weights, model training data, or historical benchmark scores. Keep them unchanged when optimizing candidate operator code; allowed edits are listed in each task. Results depend on the hardware, software stack, model endpoint, harness and evaluation budget. Local release validation does not claim new NPU evaluation results. No train/test independence beyond the supplied benchmark design is claimed.

Licenses

Project-authored code, task definitions, tests, configurations, and documentation without another license declaration are licensed under Apache License 2.0, as stated by the upstream project. This file preserves the original project license text without modification.

Included third-party materials retain their own licenses and copyright notices:

  • ops-nn and ops-transformer: the CANN Open Software License Agreement Version 2.0, including its stated Huawei AI Processor purpose restrictions, and the applicable third-party notices.
  • vLLM-Ascend: Apache License 2.0 and the existing third-party notices.
  • Frozen inputs and golden tensors: the scope described in the original dataset provenance and licensing statement. Third-party or derived materials remain subject to their original terms.

The project Apache license does not replace these component-specific terms. See the component license overview and the full license texts for details.

Citation and updates

When referring to this snapshot, cite the upstream source revision and record the Hugging Face revision used for download. Refer to the upstream technical report for its author-provided citation details. Report benchmark issues through the upstream issue tracker.

Downloads last month
74

Paper for SLAI-AITP/AscendGYM