Unverified50% confidenceBenchmarkExact time
On the LSTM time-series prediction task, parameters were reduced from 12,051 to 2,048, with a validation set MSE as low as 0.00006.
1
Sources
50%
Confidence
Long-term
Relevance
10/9/2026
First Seen
Sources
Related Entities
Related Claims
Unverified针对有限意图集合训练的轻量级专用模型参数量通常在10M以下，其推理延迟可比通用LLM调用低2-3个数量级66% similarUnverified领域微调后的7B模型在任务准确率上往往超越未经微调的70B通用模型，同时推理成本降低一个数量级65% similarUnverifiedReducing model parameters from 70B to 7B lowers both the VRAM required for inference and the per-request computation by approximately one order of magnitude.64% similarUnverified任务的输出熵越低（答案越可预测），推测性解码的加速效益越显著；GSM8K接受率从位置0的0.92衰减至0.53，MTBench从0.78急跌至0.1364% similarUnverifiedModels' standard accuracy on the passage-level ranking task was only 0% to 13%.64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/991517API
curl https://kongchang.com/api/v1/knowledge/claims/991517MCP
get_claim(id=991517)