Dataset Viewer
Auto-converted to Parquet Duplicate
schema_version
int64
workloads
dict
groups
dict
1
{ "appworld": { "repo_id": "harry1332/hpca2027-agentic-validated-traces-20260711", "revision": "422ddf6ce6c0db80605553f1331fc8f7d1e7b69c", "path": "equal_simpoints_20260715/appworld", "category": "agentic" }, "bc": { "repo_id": "harry1332/ifuse-final-datacenter-traces-20260624", "revision"...
{ "all": [ "appworld", "bc", "bfs", "core_bench", "dfs", "duckdb", "leveldb", "mlgym_fmnist", "pagerank", "rocksdb", "sssp_ego_fb", "terminal_bench" ], "agentic": [ "appworld", "core_bench", "mlgym_fmnist", "terminal_bench" ], "database": [ "...

HPCA 2027 I-Fuse Workloads

This dataset repository is the single installation entry point for the final HPCA 2027 I-Fuse workload set. Its manifest pins validated trace bundles at immutable commits across the existing Hugging Face datasets, avoiding duplicate storage while preserving reproducibility.

Install

python3 -m pip install -U huggingface_hub
hf download harry1332/hpca2027-ifuse-workloads \
  --type dataset --local-dir hpca2027-ifuse-workloads
cd hpca2027-ifuse-workloads
./install.sh

Install selected workloads or groups:

./install.sh appworld leveldb rocksdb
./install.sh agentic
./install.sh database
./install.sh datacenter

Canonical concrete workloads are AppWorld, BC, BFS, CORE-Bench, DFS, LevelDB readrandom, MLGym Fashion-MNIST, PageRank, RocksDB readrandom, SSSP ego-Facebook, and Terminal-Bench FastText.

datacenter is a Scarab suite/group alias rather than a standalone trace. It expands to BC, BFS, DFS, PageRank, RocksDB, and SSSP ego-Facebook.

TaoBench and DuckDB are intentionally excluded from the final evaluation set. TaoBench was used only as an engineering smoke test for runtime I-Fuse.

Each installed bundle is checked for clustering metadata, selected SimPoint files, weights, and replay-ready trace ZIPs. See workloads.json for exact dataset revisions and source paths.

Downloads last month
11