Unverified85% confidenceFactTime unknown
Mistral AI的7B模型在发布时性能超越了两倍参数量的Llama 2 13B
1
Sources
85%
Confidence
Long-term
Relevance
6/1/2026
First Seen
Sources
AnythingLLM：6万Star本地AI知识库，隐私优先的RAG平台
githubMintplex-Labs
Related Entities
Related Claims
UnverifiedMistral的7B模型在发布时超越了Llama 2 13B的表现85% similarUnverifiedMistral AI的Mixtral 8x7B采用MoE架构，总参数量约47B，但每次推理实际激活的参数仅约13B75% similarUnverifiedMistral 7B证明精心训练的7B模型可以匹敌早期的13B甚至30B模型75% similarUnverified在HumanEval、MBPP、GSM8K等基准测试上，Gemma 7B的表现显著优于同参数量级的LLaMA 2 7B和Mistral 7B71% similarVerifiedMistral 7B于2023年9月发布，拥有73亿参数，在多项基准测试中击败了参数量达130亿甚至340亿的竞品模型66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/33059API
curl https://kongchang.com/api/v1/knowledge/claims/33059MCP
get_claim(id=33059)