The Dataset Viewer has been disabled on this dataset.
my-eval
Fixed evaluation-set composition built with BenchPress. This repo stores
only the recipe (manifest.json) — source data is streamed at load time.
Abilities: external_knowledge_retrieval, contextual_retrieval, temporal_reasoning, commonsense_causal_reasoning
Sources
| benchmark | source repo | config | split | n_samples | license | gated |
|---|---|---|---|---|---|---|
| HLE | cais/hle | - | test | 100 | mit | yes |
| SuperGPQA | m-a-p/SuperGPQA | - | train | 100 | odc-by | no |
Reference scores (expected on this composition)
| model | expected score |
|---|---|
| Claude-3.5-Sonnet | 0.2615 |
| Claude-Sonnet-4 | 0.3075 |
| DeepSeek-v3 | 0.2945 |
| GPT-4o | 0.2215 |
| GPT-oss-120b | 0.341 |
| GPT-oss-20b | 0.2715 |
| Gemma-3-4B | 0.1515 |
| KIMI-K2 | 0.3095 |
| Llama-3.1-8B | 0.129 |
| Phi-4-Mini | 0.119 |
| Qwen2.5-72B | 0.2265 |
| Qwen3-235B | 0.249 |
Usage
from benchpress_hub import load_composition
ds = load_composition("jyjang13/demo-my-eval-075520")
Sources pinned to commit SHAs; loading is deterministic (fixed items, seeded order).
- Downloads last month
- 36