Training on 时影https://huggingaha.github.io/tags/training/Recent content in Training on 时影Hugo -- gohugo.iozh-cnhuggingaha@gmail.com (时影)huggingaha@gmail.com (时影)© 2026 时影Wed, 01 Oct 2025 00:00:00 +0000LoRA：无憾之选https://huggingaha.github.io/posts/lora-without-regret/Wed, 01 Oct 2025 00:00:00 +0000huggingaha@gmail.com (时影)https://huggingaha.github.io/posts/lora-without-regret/Small Leak Can Sink a Great Ship—Boost RL Training on MoE with 𝑰𝒄𝒆𝑷𝒐𝒑!https://huggingaha.github.io/posts/small-leak-can-sink-a-great-ship/Tue, 23 Sep 2025 00:00:00 +0000huggingaha@gmail.com (时影)https://huggingaha.github.io/posts/small-leak-can-sink-a-great-ship/GSPO：组序列策略优化https://huggingaha.github.io/posts/gspo/Mon, 28 Jul 2025 00:00:00 +0000huggingaha@gmail.com (时影)https://huggingaha.github.io/posts/gspo/