oMLX 0.3.9.dev2 release notes from the angle of Codex/Copilot on Mac local LLMs: Gemma 4 VLM MTP, DFlash, omlx launch copilot, SSD KV cache — what each changes for agent workflows.
VoxCPM2 sits in the tokenizer-free corner. Mapped vs F5-TTS, CosyVoice2, Irodori-TTS, Style-Bert-VITS2; plus why Japanese TTS still leans on OpenJTalk.
Gemini API File Search now indexes images alongside text in the same store. Metadata filters can isolate NPC memories by chapter and character, and a single-character prototype costs under $1/month on Flash-Lite. Notes on tier limits, pricing breakdown, and what to test first.
A DEV Community article proposes cross-modal distillation for wildfire evacuation routing that encodes road closures and AQI thresholds directly into the loss function. I look at the teacher-student gap when the student drops satellite imagery, why 23ms edge inference is irrelevant if sensor data is 5 minutes old, and what's missing for production.
Checked Fortress Token Optimizer's DEV article and npm/PyPI packages. Polite filler words shrink 11-22%, but running it blindly on system prompts or RAG context can strip constraints that control model output.
158K lines of AI-generated C# for a Cities: Skylines II total conversion mod. CivicRAG for codebase indexing, 300+ custom Roslyn analyzers as compile-time design rules, and manual visual debugging for render bugs AI couldn't see.
Tested Klein 9B + 9B NSFW LoRA on M1 Max 64GB via mflux 0.17.5: 1m51s/512, 5m37s/1024 q4, 224/224 LoRA keys match, NSFW prompts uncensored, Japanese subjects work with helper tokens.
Vektor Memory v1.5.4 supersession chains positioned against YourMemory decay, Cloudflare key-overwrite, and CTX, with a BM25 vs cosine threshold trap and a 5-field minimum schema for agent memory.
The paper argues that RAG, vector stores, and scratchpads are retrieval, not learning. Read alongside CTX and OCR-Memory, the gap between 'better search' and 'weight-level learning' becomes concrete.
Tested Gemma 4 MTP drafter on M1 Max 64GB with mlx-vlm 0.5.0. Only the 26B A4B MoE got +13%; 31B Dense and E4B got slower. Code gen vs short haiku prompts flip the result.