Unverified85% confidenceFactTime unknown
Mistral的7B模型在发布时超越了Llama 2 13B的表现
1
Sources
85%
Confidence
Long-term
Relevance
6/1/2026
First Seen
Sources
Unsloth：本地微调大模型速度提升5倍的开源利器
githubunslothai
Related Entities
Related Claims
UnverifiedMistral AI的7B模型在发布时性能超越了两倍参数量的Llama 2 13B85% 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79% 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70% 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 7B69% 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13B66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/33075API
curl https://kongchang.com/api/v1/knowledge/claims/33075MCP
get_claim(id=33075)