| | Update README.md | agentionai | Sep 11, 2026 |
| | Update README.md | agentionai |
| | Update README.md | agentionai |
| | Card: AP-IQ4_M renamed to AP-Q4_K_M | agentionai |
| | Rename AP-IQ4_M -> AP-Q4_K_M (IQ4_M is not a GGUF ftype name; HF's quant parser could not label it) | agentionai |
| | Add AP-IQ4_M: IQ4_XS experts on blk 24-47, IQ3_S on 0-23, 61.22 GiB VRAM | agentionai |
| | Update card: add AP-IQ4_M, held-out KLD for all tiers, n-gram table placement options | agentionai |
| | Add quality-vs-size chart | agentionai |
| | Upload mmproj-F16.gguf with huggingface_hub | agentionai |
| | Upload AP-IQ2_S/Qwen3.8-Flash-Next-AP-IQ2_S.gguf with huggingface_hub | agentionai |
| | card: trim the fork section | agentionai |
| | card: note the fork tracks upstream master (merged 0190529ec) | agentionai |
| | card: add optional fork section with long-context throughput figures | agentionai |
| | Add Agention link | agentionai |
| | Add AP-Q5_K_M: 103.57 GiB, PPL 4.0468 (+1.00%), KLD 0.0204 | agentionai |
| | Add AP-IQ3_XXS: 80.72 GiB, PPL 4.1566 (+3.74%), KLD 0.0787 | agentionai |
| | Card: all five tiers with KLD + VRAM-when-offloaded, offload recipe, sampling params, which-one guidance | agentionai |
| | Add AP-Q5_K_XL: Q5_K experts + Q5_1 PLE, 112.51 GiB, PPL 4.0339 (+0.68%) | agentionai |
| | Add AP-Q4_K_XL: 94.20 GiB, PPL 4.0536 +/- 0.02292 (+1.17% vs reference) | agentionai |
| | Add AP-Q4_K_XL results: 94.20 GiB, PPL 4.0536 (+1.17%) | agentionai |