Monday, October 26, 2026i10X News · Section desk7 stories
Studies
Efficiency & Training
Training methods, scaling, distillation, inference efficiency
FilterCompanies & models
-
Chunked KV Compression Creates Periodic Blind Spots Up to 40 Points Wide
DeepSeek-V4 family retrieval accuracy swings by as much as 40 pp depending on a token’s phase inside the compression window — gaps average scores hide.
By
Editor's picks
- Chunked KV Compression Creates Periodic Blind Spots Up to 40 Points Wide
- Diffusion-Augmented LLMs Deliver Up to 3× Lossless Speedups Over Autoregressive Decoding
- A Small RL Teacher Can Lift a Larger Student Past Itself via On-Policy Distillation
- On-Policy Distillation Isn’t the Magic — Learning Rate and KL Direction Are
- MiniMax Sparse Attention: 14× Prefill Speedup at 1M Context Without Quality Loss
More stories
-
Kimi K3: A 2.8T Open MoE With 2.5× Better Scaling Than K2
Moonshot’s Kimi K3 activates 104B of 2.8T params, adds native vision and 1M context. KDA, Attention Residuals and Stable LatentMoE deliver ~2.5× scaling efficiency over Kimi…
-
DeepSeek-V4.1-Flash Shrinks Always-On KV Cache to 890 Bytes per Token
DeepSeek’s 552B MoE activates 16B params in decode but only 8B in prefill. CSA2 + FP4 KV cuts global HBM KV to ~1/4 of V4-Flash; SWA…
· RSS