Unverified50% confidenceFactExact time
According to the Chinchilla scaling law, the optimal training data volume for a 30B model is approximately 600B–1T tokens.
1
Sources
50%
Confidence
Long-term
Relevance
8/18/2026
First Seen
Sources
Related Claims
UnverifiedLoRA使可训练参数量通常仅为全参微调的0.1%-1%，可在单张RTX 3090/4090（显存24GB）上微调7B乃至13B级别的模型64% similarUnverifiedIn token generation tests for the Gemma 4 12B model, Spark reached 26.4 tokens/s, Halo 24.6 tokens/s, and M4 Pro was the fastest at 33.8 tokens/s.63% similarUnverified根据Artificial Analysis数据，Fable完成单个任务约用33k token，Opus 5约用37k token62% similarUnverifiedThe Chinchilla model, with 70B parameters and 1.4T tokens, significantly outperforms the Gopher model, which has 280B parameters and 300B tokens, under the same compute budget.62% similarUnverifiedMiniMax M3 每个 token 推理时约激活 23B 参数62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/771365API
curl https://kongchang.com/api/v1/knowledge/claims/771365MCP
get_claim(id=771365)