memes.cjtrowbridge.com

llama-3.2-vision

The meme is an image of a graph showing the performance of a language model (LLaMA 2) on various tasks compared to a previous model (LLaMA). The tasks are listed on the y-axis, with the x-axis representing the model's size in billions of parameters. The graph shows that LLaMA 2 outperforms the previous model in all tasks, with the largest improvements in "Reasoning" and "Comprehension".

first-seen

2023-12-18T06:47:43+00:00

tesseract

70 72 —e— LLaMA 2 —e— LLaMA 2 65) m= Mistral 70, | Mistral ~ oN m----------------------- S© 604) Berm m rrr rm mmm s5 — : = : 2 68 2 = 3 < 1 = 2° 2 3 : = 1 © 66 : 50 ! oc : AS ; Effective LLaMA 64 Effective LLaMA | | size 23B (3.3x) size 38B (5.4x) | 7 13 34 70 7 13 34 70 Model size (billion parameters) 68 Model size (billion parameters) 70) —e— LLaMA 2 —e LLaMA 2 65) = Mistral 661 m= Mistral = c me O g--------------- 5 60 a 64 . © v , ~ I = 55 OY 62 I S 2. . 250} — c : I UL 60 : | ;Effective LLaMA \Effective LLaMA 45 isize 13B (1.9x) size 21B (3x) 7 13 34 70 7 13 34 70 Model size (billion parameters) Model size (billion parameters)