You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
We introduce Mistral 7B v0.1, a 7-billion-parameter language model engineeredfor superior performance and efficiency. Mistral 7B outperforms Llama 2 13Bacross all evaluated benchmarks, and Llama 1 34B in reasoning, mathematics, andcode generation. Our model leverages grouped-query attention (GQA) for fasterinference, coupled with sliding window attention (SWA) to effectively handlesequences of arbitrary length with a reduced inference cost. We also provide amodel fine-tuned to follow instructions, Mistral 7B -- Instruct, that surpassesthe Llama 2 13B -- Chat model both on human and automated benchmarks. Ourmodels are released under the Apache 2.0 license.
URL
Affiliations
Abstract
Translation (by gpt-3.5-turbo)
Summary (by gpt-3.5-turbo)
The text was updated successfully, but these errors were encountered: