| | Add Gyro-L (64 GB tier): TQP pair-trellis experts, 50.2 GiB GPU weights | agentionai | Oct 12, 2026 |
| | Card: Gyro-M update 2026-10-11 | agentionai |
| | Gyro-M update: improved encode (KLD 0.291) | agentionai |
| | Card: point Strata instructions at rc2 | agentionai |
| | Card: Gyro-S update 2026-10-10 | agentionai |
| | Gyro-S update: improved encode (KLD 0.401) | agentionai |
| | Docker Hub image, sponsor link; note on multi-request MTP fix | agentionai |
| | Card: note on the updated MTP draft (upstream names; needs a 2026-10-05+ build) | agentionai |
| | MTP draft: upstream tensor names (blk.48.nextn.hc_head_*) + MTP-layer compress ratio; weights unchanged. Loads in upstream llama.cpp b11386+ and agentionai/llama.cpp main 2026-10-05+ | agentionai |
| | Card v2: headline numbers first, slimmer layout, Strata quick start, LiveCodeBench and MMLU-Pro | agentionai |
| | Card header image | agentionai |
| | Card: Windows note for --ngram-on-disk (fast reader on Windows from main 2026-10-05; lazy-mode fallback) | agentionai |
| | Card: CUDA kernel update and crash fix are in agentionai/llama.cpp main (2026-10-04) | agentionai |
| | Card: corrections (sizes, loop results over 10 seeds, Vulkan or CUDA, 5090 session 2 numbers, Strata quick start, MMLU-Pro) | agentionai |
| | Card: min-p 0.05; CUDA status and RTX 5090/A6000 numbers (prefill ~800, 128k/160k context); MTP draft fits on 32 GB up to 24k/32k; Ampere CUDA crash workaround | agentionai |
| | Card: two-GPU setup (model on one card, draft on the other); keep --ngram-on-disk | agentionai |
| | Card: 16-24 GB cards supported with --n-cpu-moe (new CPU kernels, 2026-10-02) | agentionai |
| | Card: footer typo fixes | agentionai |
| | Card: pipeline image-text-to-text; support footer | agentionai |
| | Card: "Which setup?" table per hardware; drafting only on 48 GB+/unified memory; build and platform notes | agentionai |