Unverified50% confidenceTradeoffExact time
The figure "pretraining throughput reaching 170% of a single H100" is jointly affected by batch size, sequence length, model scale, precision format, and interconnect bandwidth, and is not a general conclusion; the ratio may change with a different model or hyperparameters.
1
Sources
50%
Confidence
Long-term
Relevance
10/13/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedScale AI在2024年的分析显示，部分模型在MMLU上的分数膨胀可能高达5-10个百分点，因测试集已泄露到训练数据中70% similarUnverifiedAs the model scales from 8B to 70B, the predictive power of fixed output vector similarity for behavior actually worsens, with a paired mean correlation change of -0.080 (95% confidence interval [-0.127, -0.035]).69% similarUnverified即使只有5-10%的训练样本存在错位，也会导致VLM模型在下游任务上出现显著的性能退化69% similarUnverifiedExperiments conducted by the research team on models such as GPT-2 showed that after only 5–9 generations of recursive training, the quality of model outputs degrades significantly.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1017236API
curl https://kongchang.com/api/v1/knowledge/claims/1017236MCP
get_claim(id=1017236)