Datasets:

hongpingzhang
/

ecocompute-energy-efficiency

Exception:    SplitsNotFoundError
Message:      The split names could not be parsed from the dataset config.
Traceback:    Traceback (most recent call last):
                File "/usr/local/lib/python3.12/site-packages/datasets/inspect.py", line 286, in get_dataset_config_info
                  for split_generator in builder._split_generators(
                                         ^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/packaged_modules/json/json.py", line 91, in _split_generators
                  pa_table = next(iter(self._generate_tables(**splits[0].gen_kwargs, allow_full_read=False)))[1]
                             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/packaged_modules/json/json.py", line 193, in _generate_tables
                  examples = [ujson_loads(line) for line in batch.splitlines()]
                              ^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/utils/json.py", line 20, in ujson_loads
                  return pd.io.json.ujson_loads(*args, **kwargs)
                         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
              ValueError: Expected object or value
              
              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/split_names.py", line 65, in compute_split_names_from_streaming_response
                  for split in get_dataset_split_names(
                               ^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/inspect.py", line 340, in get_dataset_split_names
                  info = get_dataset_config_info(
                         ^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/inspect.py", line 291, in get_dataset_config_info
                  raise SplitsNotFoundError("The split names could not be parsed from the dataset config.") from err
              datasets.inspect.SplitsNotFoundError: The split names could not be parsed from the dataset config.

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.

EcoCompute: LLM Energy Efficiency Measurements

93+ empirical measurements of energy consumption during LLM inference, collected at 10 Hz via NVML across 3 NVIDIA GPU architectures.

Key Findings

INT8 Energy Paradox: Default bitsandbytes INT8 (load_in_8bit=True) increases energy by 17–147% vs FP16
NF4 Small-Model Penalty: 4-bit quantization wastes 11–29% more energy on models ≤3B parameters
Batch Size Impact: BS=1 → BS=64 reduces energy per request by 95.7%

Data Coverage

GPU	Architecture	Measurements	Config Types
RTX 5090	Blackwell	8	FP16, NF4
RTX 4090D	Ada Lovelace	15	FP16, NF4, INT8, Pure INT8
A800	Ampere	70	Pure INT8 × 7 batch sizes
Total	3 architectures	93+	5 config types

Models Tested

Qwen/Qwen2-1.5B (1.5B params)
microsoft/Phi-3-mini-4k-instruct (3.8B params)
01-ai/Yi-1.5-6B (6B params)
mistralai/Mistral-7B-Instruct-v0.2 (7B params)
Qwen/Qwen2.5-7B-Instruct (7B params)

Dataset Structure

metadata/
├── README.md                     — Dataset documentation
├── COMPLETE_DATASET_MEMO.md      — Full 93+ measurement documentation
├── rtx5090_metadata.json         — RTX 5090 Blackwell results
├── rtx4090d_metadata.json        — RTX 4090D Ada Lovelace results
├── pure_int8_metadata.json       — Pure INT8 ablation study
├── a800_metadata.json            — A800 Ampere results
└── batch_size_experiment/        — A800 batch size sweep (BS 1–64)
    ├── *_raw_*.csv               — Raw per-run measurements (70 rows)
    ├── *_summary_*.csv           — Aggregated statistics
    └── *_results_*.png           — Visualization

Measurement Methodology

Tool: NVIDIA Management Library (NVML) via pynvml
Frequency: 10 Hz (100ms polling)
Metric: GPU board power (watts)
Protocol: 3 warmup + 10 measured runs + 30s thermal stabilization
Quality: CV < 2% (throughput), CV < 5% (power)
Generation: Greedy decoding, max_new_tokens=256, fixed prompt

Software Environment

GPU	PyTorch	CUDA	Driver	transformers	bitsandbytes
RTX 5090	2.6.0	12.6	570.86.15	4.48.0	0.45.3
RTX 4090D	2.4.1	12.1	560.35.03	4.47.0	0.45.0
A800	2.4.1	12.1	535.183.01	4.47.0	0.45.0

Related Resources

Interactive Dashboard: hongping-zh.github.io/ecocompute-dynamic-eval
GitHub Repository: github.com/hongping-zh/ecocompute-dynamic-eval
Zenodo Archive: zenodo.org/records/18900289
HuggingFace Optimum PR: huggingface/optimum#2410

Citation

@dataset{zhang2026ecocompute,
  title={EcoCompute: LLM Energy Efficiency Measurements},
  author={Zhang, Hongping},
  year={2026},
  publisher={Hugging Face},
  url={https://huggingface.co/datasets/hongpingzhang/ecocompute-energy-efficiency}
}

License

MIT

Downloads last month: 7