TensorFold 1.3.0 boosts GLM-5.3-Flash to 104 tok/s, 24% faster on Blackwells
AIFlow_ML · x · 2026-11-20
TensorFold 1.3.0 is out: GLM-5.3-Flash jumps from 83.8 to 104.1 tok/s (about 24% faster) on two RTX PRO 6000 Blackwells at TP=2 with NVFP4 experts. On Metal, DeepSeek-V4.1's prompt-attention kernel is 1.5× faster. Flash Next frees 3 GiB of page cache on CUDA, Linux gains opt-in on-demand KV page mapping, and tensorfold models/info now estimate which checkpoints fit your machine accounting for context and parallel streams.
More from Infra
- Thermal camera footage proves AI agents can program custom AI chips directly — yaroslavvb · 2026-11-20
- Evidence for scaling EUV lithography power beyond ASML's 1 kW milestone — bronzeagepapi · 2026-11-20
- Reddit asks: What do you actually use local LLMs for, and at what tk/s? — Ok-Option-82 · 2026-11-20
- Liquid d1-2B decision model answers 19-30x more questions/min than same-size LLMs — mervenoyann · 2026-11-20
- Cactus Compute publishes 12 papers in 2026, 5 at NeurIPS, focused on on-device efficiency — joemeno · 2026-11-20
- Modular rebuilds MAX's Module API with PyTorch-like syntax and eager mode — clattner_llvm · 2026-11-20