Landscape 16:9 log-scaled plot of training loss vs compute, four curves for different model sizes.
X-axis "Training compute (FLOPs)" with log ticks "1e20", "1e21", "1e22", "1e23", "1e24". Y-axis "Validation loss (cross-entropy)" with linear decreasing ticks "3.5", "3.0", "2.5", "2.0", "1.5".
Four descending curves with ±1σ shaded bands, labels near tails:
"70M params" (slate gray), "1B params" (muted navy), "10B params" (dusty teal), "70B params" (soft terracotta).
Warm-copper dashed diagonal line labeled "compute-optimal frontier"; open circles at isoflop crossover points. Legend box top-right.
Title: "Empirical scaling laws: loss vs training compute". Subtitle: "four model sizes on a fixed data mixture; shaded bands = ±1 std over 3 seeds."
中文翻譯
橫向 16:9 的對數座標圖，繪製訓練損失 vs 計算量，四條曲線代表不同的模型規模。
X 軸“訓練計算量 (FLOPs)”，對數刻度爲“1e20”、“1e21”、“1e22”、“1e23”、“1e24”。Y 軸“驗證損失 (交叉熵)”，線性遞減刻度爲“3.5”、“3.0”、“2.5”、“2.0”、“1.5”。
四條下降曲線，帶 ±1σ 陰影帶，曲線尾端附近標註：
“70M 參數”（石板灰）、“1B 參數”（低飽和藏青）、“10B 參數”（灰青色）、“70B 參數”（柔和赤陶色）。
暖銅色虛線對角線，標註爲“計算最優前沿”；在等 FLOP 交叉點處爲空心圓點。右上角圖例框。
標題：“經驗縮放定律：損失 vs 訓練計算量”。副標題：“四組模型規模在同一固定數據配比上；陰影帶 = 3 個隨機種子上的 ±1 標準差。”
