English(EN) I think I should just store every single token I put into a LLM and get out of them. I ponder about training very tiny models (like 1B to 2B) for highly specifi

用户考虑存储LLM token用于专门的模型训练

作者 PulseAugur 编辑部 · [1 个来源] · 2026-08-06 07:20

用户正在考虑存储大语言模型（LLM）处理过的所有token，以帮助训练更小的、专门化的模型。这种方法可以通过检索和编辑机制，显著提高特定任务的性能。用户指出，尽管LLM的成本已变得非常低廉，但这种数据策略仍可能带来好处。 AI

影响该策略可以实现更高效、更具成本效益的任务特定小型AI模型的微调。

排序理由用户对潜在AI策略的个人思考，并非正式公告或研究。

在 Mastodon — fosstodon.org 阅读 →

其他

AI 生成摘要 · Google Gemini · 来自 1 个来源。我们如何撰写摘要 →

报道来源 [1]

Mastodon — fosstodon.org TIER_1 English(EN) · [email protected] · 2026-08-06 07:20

I think I should just store every single token I put into a LLM and get out of them. I ponder about training very tiny models (like 1B to 2B) for highly specifi

I think I should just store every single token I put into a LLM and get out of them. I ponder about training very tiny models (like 1B to 2B) for highly specific tasks and that data would help me big time. Something something like retrieval and edit models. Could make a lot of ta…

报道来源 [1]

I think I should just store every single token I put into a LLM and get out of them. I ponder about training very tiny models (like 1B to 2B) for highly specifi

相关实体

相关话题