π“Ÿ host1.bitchan.one

𓀐 @+wSqT=92

⏱ 2d seenβ›“ 0πŸ‘ 0β§– 1

filter [-]

languages:onother

Recent Highlights from the --GLM-5.5 training pipeline and MoE upcycling vs distillation techniques: >110253536 >110253579 >110253639 >110253655 >110253685 >110253723 >110253692 >110254255 >110254309 >110253599 >110253642 >110256841 --Comparing GLM and Inkling benchmarks and local hardware requirements: >110254699 >110254738 >110255038 >110255445 >110255465 >110255473 >110255484 >110255395 >110255417 >110254754 --Anon shares air-gapped setup and AGENTS.md for DeepSeek tool calling: >110256400 >110256415 >110256547 >110256560 >110257347 --Irony over Inkling 2 Small's 594B parameter count: >110254596 >110254624 >110254767 >110254820 >110254857 --MTP slowing down GPU inference and consuming extra VRAM: >110257211 >110257230 >110257298 --Seeking small models with reliable JSON and tool-calling for games: >110253337 >110253344 >110253367 >110253376 >110253470 >110253500 --PLaMo 3 Translate open weights release for Japanese translation: >110256815 --Local abliteration methods and toolings for removing Gemma's safeguards: >110257290 >110257300 >110257319 >110257345 --Anon reports high tokens per second using Basalt engine and NVFP4: >110255496 >110255500 >110255786 >110257165 --Depth Hybrid Attention architectural design claiming superior performance over Full Attention: >110256911 --Comparing Ripple-65B and Qwen-27B performance benchmarks: >110255529 >110255533 >110256915 >110257011 --Gemini Flash leaks internal monologue and system prompt guidelines: >110254009 >110254045 --Logs: >110253937 >110254009 >110254490 >110256952 >110256983 --Miku, Gemma, Inkling, Rin, Qwen (free space): >110253337 >110253879 >110253937 >110254136 >110254830 >110255538 >110255863 >110256034 >110256275 >110256456 >110256615 >110256694 >110257136 >110257201 Recent Highlight Posts from the Why?: Enable Links: https://rentry.org/lmg-recap-script⏱ 2d↩