{ "experiment": "EXP-G4", "formula": "gamma = 1 - T_eval*sqrt(2)/theta", "date": "2026-04-18", "results": [ { "prefix": "EleutherAI--pythia-70m", "name": "pythia-70m", "pe_type": "RoPE", "notes": "", "theta": 10000, "d_head": 64, "T_train": 2048, "T_eval_max": 2000.0, "n_points": 9, "gamma_obs": 0.7476017873166874, "gamma_new_pred": 0.717157287525381, "gamma_old_pred": 0.99999561666838, "err_new_pct": 4.245163553501351, "err_old_pct": -25.239493568239496, "R2": 0.9893049417040555 }, { "prefix": "meta-llama--Meta-Llama-3-8B", "name": "Meta-Llama-3-8B", "pe_type": "RoPE", "notes": "", "theta": 500000, "d_head": 128, "T_train": 8192, "T_eval_max": 2000.0, "n_points": 9, "gamma_obs": 1.0454762537473639, "gamma_new_pred": 0.9943431457505076, "gamma_old_pred": 0.7018781400200674, "err_new_pct": 5.1424006104313325, "err_old_pct": 48.95409817400391, "R2": 0.996718622313285 }, { "prefix": "Qwen--Qwen2.5-7B", "name": "Qwen2.5-7B", "pe_type": "RoPE", "notes": "", "theta": 1000000, "d_head": 128, "T_train": 8192, "T_eval_max": 2000.0, "n_points": 9, "gamma_obs": 0.9966953735480816, "gamma_new_pred": 0.9971715728752538, "gamma_old_pred": 0.6666637444455867, "err_new_pct": -0.047755004266634545, "err_old_pct": 49.5049613620367, "R2": 0.9963935840252253 }, { "prefix": "meta-llama--Llama-2-7b-hf", "name": "Llama-2-7b-hf", "pe_type": "RoPE", "notes": "artifact", "theta": 10000, "d_head": 128, "T_train": 4096, "T_eval_max": 2000.0, "n_points": 9, "gamma_obs": 0.2870574377368437, "gamma_new_pred": 0.717157287525381, "gamma_old_pred": 0.99999561666838, "err_new_pct": -59.97287586278841, "err_old_pct": -71.29413039896974, "R2": 0.881768027724978 }, { "prefix": "mistralai--Mistral-7B-v0.1", "name": "Mistral-7B-v0.1", "pe_type": "RoPE", "notes": "only_2pts", "theta": 10000, "d_head": 128, "T_train": 8192, "T_eval_max": 50.0, "n_points": 2, "gamma_obs": 1.213076772373504, "gamma_new_pred": 0.9929289321881345, "gamma_old_pred": 0.99999561666838, "err_new_pct": 22.17156062722697, "err_old_pct": 21.30820897145856, "R2": 1.0 }, { "prefix": "mistralai--Mistral-Nemo-Instruct-2407", "name": "Mistral-Nemo-Instruct-2407", "pe_type": "RoPE", "notes": "partial_7pts", "theta": 1000000, "d_head": 128, "T_train": 131072, "T_eval_max": 500.0, "n_points": 7, "gamma_obs": 0.5407084190220748, "gamma_new_pred": 0.9992928932188134, "gamma_old_pred": 0.6666637444455867, "err_new_pct": -45.8908971842676, "err_old_pct": -18.893381629483898, "R2": 0.9671649502959694 }, { "prefix": "google--gemma-2-9b-it", "name": "gemma-2-9b-it", "pe_type": "RoPE", "notes": "partial", "theta": 10000, "d_head": 256, "T_train": 8192, "T_eval_max": 1000.0, "n_points": 8, "gamma_obs": 0.6586407289285032, "gamma_new_pred": 0.8585786437626906, "gamma_old_pred": 0.99999561666838, "err_new_pct": -23.287082236050793, "err_old_pct": -34.13563840181085, "R2": 0.97730793078849 }, { "prefix": "tiiuae--falcon-7b", "name": "falcon-7b", "pe_type": "ALiBi", "notes": "no_rope", "theta": 10000, "d_head": 64, "T_train": 2048, "T_eval_max": 1000.0, "n_points": 8, "gamma_obs": 0.8928207115404576, "gamma_new_pred": 0.8585786437626906, "gamma_old_pred": 0.99999561666838, "err_new_pct": 3.9882272901294638, "err_old_pct": -10.717537491313212, "R2": 0.9927613215692025 } ] }