LLM Fine-Tuning Performance Benchmarks Across 959 Hardware Configurations

Performance benchmarks for Large Language Model fine-tuning across various hardware and software configurations. The dataset includes throughput measurements for 959 valid configurations, collected over 1000 GPU hours on a Kubernetes cluster. It was created by ibm-research and last updated on 2026-05-15.

Use Cases

• Predicting fine-tuning throughput based on hardware and software configurations.
• Benchmarking different GPU setups for LLM training efficiency.
• Researching methods for handling categorical variables in performance modeling.

Strengths

• Contains 959 valid fine-tuning configurations.
• Data collection involved over 1000 GPU hours of compute.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

ibm-research

Collection Method

Benchmarks collected from a Kubernetes cluster.

Freshness

Last updated 2026-05-15 15:29:24.