ICML Research: Tokenization Impacts and Training Loss Prediction
VectorInst · x · 2026-07-08
During ICML, the Vector Institute shared several research presentations: Colin Raffel discussed the impact of tokenization on language model behavior (TokSuite) and model merging; Vardan Papyan spoke on inter-layer gradient optimization; Chris Maddison presented research on predicting large model training loss; and Rahul G. Krishnan covered sparse training and causal estimation.
Related event: Vector Institute Presents 73 Papers at ICML 2026(4 posts)→
More from Research
- EdgeRock open-sourced: 72 MiB KV cache at 256K context, 99.1% smaller than Qwen3.5-4B — alifcoder · 2026-10-27
- EdgeRock-4B holds KV cache at 72 MiB across 256K context, 99.1% smaller than Qwen3.5-4B — alifcoder · 2026-10-27
- TRL adds Score Centering, a simple fix for RLHF training instability — lvwerra · 2026-10-27
- Latent-Foresight: world models that learn to predict their own future — abursuc · 2026-10-27
- UK AISI papers: no good safeguards yet against misuse of biological AI tools — davidmanheim · 2026-10-27
- CMU's XYBench: LLMs answer queries literally and reinforce user misconceptions — valentina__py · 2026-10-27