Unverified50% confidenceFactExact time
Ollama Hub托管了从7B到70B参数不等的各类模型
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/11/2026
First Seen
Valid until: 12/10/2026
Sources
Related Entities
Related Claims
UnverifiedOrnith 1.0包含四种规格：9B稠密模型、31B稠密模型、35B MoE模型、397B MoE模型62% similarVerified模型参数量越大，对量化的鲁棒性越强，使得70B级别模型在4-bit量化后仍能保持接近FP16的性能62% similarUnverifiedChinchilla模型以70B参数配合4倍数据量即可超越280B参数的Gopher模型性能62% similarUnverified端侧可运行的大语言模型参数规模通常在1B至13B之间，远小于云端的GPT-4或Claude等数千亿参数模型61% similarUnverifiedMoE架构中Mixtral 8x7B模型总参数46.7B,但每个token仅激活约12.9B参数(2个专家),推理成本接近13B参数的稠密模型,性能可媲美70B级别61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/902047API
curl https://kongchang.com/api/v1/knowledge/claims/902047MCP
get_claim(id=902047)