Is Llama 3 Really Better Than Mistral?

With the recent launch of the much anticipated Llama 3, I decided to use both Mistral, which is one of the best small (7B) language models out there, and Llama 3, which according to its benchmark scores claims to outperform Mistral. But is it really better when it comes to using it as the LLM in your RAG applications? To test this, I put the same questions to both Mistral and Llama 3 and the results will surprise you.

Link to Collab

I created a RAG application using Ollama. In case you want to know how you can do it yourself, you can check out this post. I used the Elden Ring Wikipedia article as the document for contextual retrieval. I was using conversation buffer memory, which just passes the entire conversational history as context back to the language model. Furthermore, I asked the same question to both LLMs, and at the end we also asked the same question to the current king of LLMs, GPT-4. The question was –

"How many awards did Elden Ring Win, and did it win Game of the year award ?"

The entire prompt with the context was –

Be precise in your response. Given the context - Elden Ring winning Game of the Year
at the 23rd Game Developers Choice
AwardsSome reviewers criticized a number of the game's menu and accessibility systems.[84][85] Reviewers
complained about the poor performance of the Window s version; framerate issues were commonly
mentioned.[81][86] Reviewers noted the story of Elden Ring lacks Martin's writing style. Kyle Orland of Ars
Technica said the game's storytelling is "characteristically sparse and cryptic", and differs from the
expectations of Martin's fans.[76] Chris Carter of Destructoid called the story "low key" but said it is better-
told than those of previous FromSoftware games.[80] Aoife Wilson of Eurogam er said George R. R.
Martin's heavy inclusion in the marketing was "baffling" when his contributions to the overall narrative
were unclear.[72] Mitchell Saltzman did not mind the lack of Martin's style, saying the side-stories rather
than any gr and, ove rarching pl ot kept him "enthralled".[70]

120. Mejia, Ozzie (January 26, 2023). "Elden Ring & Stray lead Game Developers Choice
Awards 2023 nominees" (https://www.shacknews.com/article/133863/gdc-2023-award-nomi
nees). Shacknews. Archived (https://web.archive.org/web/20230127040625/https://www.sha
cknews.com/article/133863/gdc-2023-award-nominees) from the original on January 27,
2023. Retrieved January 27, 2023.
121. Beth Elderkin (March 22, 2023). "'Elden Ring' Wins Game Of The Year At The 2023 Game
Developers Choice Awards" (https://gdconf.com/news/elden-ring-wins-game-year-2023-gam
e-developers-choice-awards). Game Developers Choice Conference. Archived (https://web.
archive.org/web/20230323091858/https://gdconf.com/news/elden-ring-wins-game-year-2023
-game-developers-choice-awards) from the original on March 23, 2023. Retrieved March 23,
2023.
122. "gamescom award 2021: These were the best games of the year" (https://www.gamescom.gl
obal/en/gamescom/gamescom-award/gamescom-award-review-2021). Gamescom.

Tin Pan Alley Award for
Best Music in a GameNominated
The Steam
AwardsJanuary 3, 2023Game of the Year Won
[132]
Best Game You Suck At Won
The Streamer
AwardsMarch 11, 2023Stream Game of the
YearWon[133]
1. Sawyer, Will; Franey, Joel (April 8, 2022). "Where Elden Ring takes place and the story
explained" (https://www.gamesradar.com/elden-ring-where-does-it-take-place-setting-story-l
ore/). gamesradar. Archived (https://web.archive.org/web/20220402212714/https://www.gam
esradar.com/elden-ring-where-does-it-take-place-setting-story-lore/) from the original on April
2, 2022. Retrieved July 26, 2022.
2. Knapp, Mark (June 16, 2021). "Elden Ring: Release Date, Gameplay, and What We Know
So Far" (https://www.ign.com/articles/elden-ring-release-date-news-gameplay-trailer-story).
IGN. Ziff Davis. Archived (https://web.archive.org/web/20220303124310/https://www.ign.co
m/articles/elden-ring-release-date-news-gameplay-trailer-story) from the original on March 3,
2022. Retrieved July 6, 2021.

Elden Ring sold 13.4 million copies worldwide by the end of March 2022,[87] and 20 million by February
2023.[88] It was the best-selling game in several regions between February and March 2022,[89][90] and is
the fastest-selling Bandai Namco game of all time.[91][92] It was the second-bestselling game of 2022 in the
US after Call of Duty: Modern Warfare II,[93] the third-bestselling in Europe,[94] and the tenth-bestselling in
Japan at retail.[95]
Elden Ring won many awards and accolades. It was listed as the
2022' s Game of the Year by several publications, including Ars
Technica,[96] Destructoid,[97] EGM,[98] Eurogam er,[99] Game
Informer,[100] GamesRadar +,[101] GameSpot,[102] IGN,[103] PC
Gamer,[104] and Polygon.[105] It is among the most awarded games
in history with 324 G ame of the Year awards.[106][107]S a l e s
A w a r d s,  Answer the following question - How many awards did Elden Ring Win, and did it win Game of the year award ?.

Mistral

1. "Elden Ring" (https://www.metacritic.com/game/pc/elden-ring). Metacritic. Archived (https://web.
archive.org/web/20220402212714/https://www.metacritic.com/game/pc/elden-ring) from the
original on April 2, 2022. Retrieved April 2, 2022.
2. "Elden Ring" (https://www.met

Llama 3 – 8B

Elden Ring winning Game of the Year
at the 23rd Game Developers Choice
AwardsSome reviewers criticized a number of the game's menu and accessibility systems.[84][85] Reviewers
complained about the poor performance of the Window s version; framerate issues were commonly
mentioned.[81][86] Reviewers noted the story of Elden Ring lacks Martin's writing style. Kyle Orland of Ars
Technica said the game's storytelling is "characteristically sparse and cryptic", and differs from the
expectations of Martin's fans.[76] Chris Carter

We can clearly see that Llama 3 – 8B got one part right, however, from the context it couldn’t find that it won 324 awards in total.

GPT-4

'Elden Ring won 324 Game of the Year awards. Yes, it did win the Game of the Year award at the 23rd Game Developers Choice Awards.'

GPT-4 is still far ahead of smaller LLMs, but Llama 3 8B has improved compared to Mistral.

Comments

Leave a comment