{ "kernel_benchmark": [ { "model_id": "openai-community/gpt2", "model_name": "GPT-2 (124M)", "n_params_M": 124.4, "seq_len": 56, "device": "cpu", "baseline_ms": 39.176, "s20_injected_ms": 41.556, "s20_overhead": 1.061, "avg_perplexity": 175.7, "perplexity_per_prompt": [ 89.63, 202.44, 235.02 ] }, { "model_id": "distilbert/distilgpt2", "model_name": "DistilGPT-2 (82M)", "n_params_M": 81.9, "seq_len": 56, "device": "cpu", "baseline_ms": 21.623, "s20_injected_ms": 21.331, "s20_overhead": 0.987, "avg_perplexity": 302.02, "perplexity_per_prompt": [ 223.34, 339.5, 343.22 ] }, { "model_id": "facebook/opt-125m", "model_name": "OPT-125M", "n_params_M": 125.2, "seq_len": 57, "device": "cpu", "baseline_ms": 29.008, "s20_injected_ms": 29.36, "s20_overhead": 1.012, "avg_perplexity": 199.73, "perplexity_per_prompt": [ 148.67, 265.95, 184.57 ] }, { "model_id": "bigscience/bloom-560m", "model_name": "BLOOM-560M", "n_params_M": 559.2, "seq_len": 48, "device": "cpu", "baseline_ms": 122.361, "s20_injected_ms": 117.968, "s20_overhead": 0.964, "avg_perplexity": 138.99, "perplexity_per_prompt": [ 177.74, 209.88, 29.35 ] } ] }