Author: Tater Totterson

Ollama – v0.14.0-rc1

Ollama v0.14.0-rc1 just dropped — and it’s generating magic 🖼️🚀

Meet z-image: Ollama’s first foray into local AI image generation. Now you can type `ollama generate z-image “a cat in a spacesuit”` and watch your terminal turn text into visuals — all offline, all on your machine. No cloud. No waiting. Just pure local AI vibes.

This is experimental (yes, bugs ahead!), but it’s huge: Ollama’s going multimodal. Text + images — all from your CLI or API, just like LLMs.

GGUF? Still supported. Custom models? Yep. Now with pixels.

Docs coming soon — but if you’re brave, go ahead and test it. Train your own z-image models. Make a robot squirrel in a trench coat. The future’s local now. 🐱💻

🔗 View Release

Lemonade – v9.1.3

🚀 Lemonade v9.1.3 just dropped — your local LLM rig just got a serious upgrade!

• 🌐 Remote Access: Run Lemonade from anywhere with `lemonade-server` — control your AI locally, even from your phone.
• 💾 Save Custom Loads: Stop retyping params! Use CLI or `/load` to save and reload model configs instantly.
• 🚀 LFM2.5 Support: Powered by FastFlow LM v0.9.25 — faster, smoother inference with zero setup hassle.
• 🐳 Docker Ready: Official image + GitHub Actions pipeline — deploy in seconds, not hours.
• 🐧 Fedora Love: Native RPM packages now available — no more compiling from source.
• 🧹 Cleaner Backend: Removed state from llamacpp — fewer leaks, more stability.
• ✅ Server Detection Fixed: Linux users — auto-detection finally works as intended.

Big props to @SidShetye and the crew!

Grab it, containerize it, remote-control your LLMs — and go play. 🛠️

Full changelog: [v9.1.2…v9.1.3](link)

🔗 View Release

Ollama – v0.14.0-rc0: Add experimental MLX backend and engine with imagegen support (#13648)

Ollama v0.14.0-rc0 just landed — and Apple Silicon fans, this one’s for you 🍏💥

Say hello to experimental MLX backend support — run LLMs natively on M-series chips without CUDA or PyTorch overhead. Faster, leaner, and totally Apple-native.

✨ What’s new?

• 🖼️ Image generation — yes, you can now generate images directly via Ollama (early but wildly cool)
• 🛠️ Built-in build toggles: `cmake –preset MLX` and `go build -tags mlx .` for easy custom compiles
• 🍎 Full macOS support — x86 & ARM builds ready, CPU-only for now (GPU accel coming soon!)
• 📚 Cleaner docs + improved tokenizer guides — because nobody likes cryptic configs

This is still a release candidate, so expect bugs… but if you’re on Mac and want to skip the bloat? Now’s your chance. Break it, tweak it, report it — we’re all in this together 🚀

#MLX #AppleSilicon #ImageGen #Ollama #AIOnMac

🔗 View Release

Text Generation Webui – v3.23

✨ Chat UI got a glow-up! Tables and dividers now look clean, crisp, and way easier to read—perfect for scrolling through long model outputs without eye strain.

🔧 Bug fixes that actually matter:

• Models with `eos_token` disabled? No more crashes! Huge props to @jin-eld 🙌
• Symbolic link issues in `llama-cpp-binaries` fixed—non-portable installs breathe easier now.

🚀 Backend power-up:

• `llama.cpp` updated to latest commit (`55abc39`) → faster, smoother inference
• `bitsandbytes` bumped to 0.49 → better quantization, fewer OOMs, more stable loads

📦 PORTABLE BUILDS ARE LIVE!

Download. Unzip. Run. No install needed.

• NVIDIA? → `cuda12.4`
• AMD/Intel GPU? → `vulkan`
• CPU-only? → `cpu`
• Mac Apple Silicon? → `macos-arm64`

💾 Updating? Just grab the new zip, unzip, and drop your old `user_data` folder in. All your models, settings, themes—still there. Zero reconfiguring.

Go play. No setup. Just pure LLM magic. 🚀

🔗 View Release

ComfyUI – v0.8.2

ComfyUI v0.8.2 is live — quiet updates, big improvements! 🛠️🎨

• Fixed edge-case crashes in heavy workflows (custom nodes + memory-heavy chains, we see you).
• Smoother node connections & improved drag-and-drop feel — tiny tweaks, huge UX win.
• Custom node icons now load reliably after restarts (no more ghost placeholders!).
• Security deps updated — clean, safe, no breaking changes.

Perfect for 24/7 creators who just want their pipelines to work. No flashy new nodes… but if you’ve been battling glitches, this is your upgrade. 💻🧠

Keep crafting — stay comfy!

🔗 View Release

ComfyUI – v0.8.1

🚨 ComfyUI v0.8.1 is live — and it’s a quiet hero update!

• 🛠️ Fixed critical node crashes — KSampler and ImageScaleToTotalPixels now play nice, no more mid-generate meltdowns.
• 🧠 Better memory management — Large models + batch processing? Smoother than ever, even on low-RAM rigs.
• 🔧 Updated torch & numpy — Under-the-hood upgrades for rock-solid stability on Windows, Mac, and Linux.
• ✨ UI polish — Node labels finally stay readable after zooming. No more text soup!

💡 Pro tip: If you’ve been battling “Node crashed” errors — this is your sign to update. No breaking changes, just pure, stable, render-ready vibes.

📦 Grab it: https://github.com/Comfy-Org/ComfyUI/releases/tag/v0.8.1

Your workflows just got a lot more reliable. 🎨✨

🔗 View Release

Lemonade – v9.1.2

Lemonade v9.1.2 just dropped — and your local LLM game just leveled up 🚀

🔥 Custom Model Recipes: Build & share your own configs with `–extra-models-dir` — tweak prompts, quantization, or hardware settings on the fly.

🎮 Native NPU/ROCm Support: Run leaner, faster on ROG Ally X and Ryzen AI Z2 Extreme — no workarounds needed.

📦 LM Studio GGUF Ready: Drop your GGUF files in `–extra-models-dir` and they auto-detect with an `extra.` prefix.

⚡ FastFlowLM 0.9.24: Smoother inference, smarter memory use, fewer crashes — your CPU/GPU will thank you.

🐳 Dockerized Build Guide: New step-by-step docs to containerize Lemonade CPP — perfect for dev environments.

🛠️ UX Upgrades: Cleaner UI, fixed mobile layout, centered subtitles, and smarter model reloads after FLM updates.

🔑 API Key Support: Lock down your local endpoint via `.env` — keep your models private, secure.

🧠 New Model: Meet Nemotron 3 Nano — tiny footprint, big brain power for edge devices.

Plus: sleeker site, better docs, fewer headaches. Your local AI rig just got a serious upgrade. 🛠️💻

🔗 View Release

ComfyUI – v0.8.0

ComfyUI v0.8.0 just dropped—and it’s a game-changer 🎨⚡

• Native WebSockets → Real-time node updates & smoother previews (even on mobile!). No more sluggish HTTP polling.
• Built-in Node Registry → Install custom nodes with a click. No more manual folder diving—just search, install, and go.
• Memory Boost → 50-node workflows? Still running smooth. Better GC = fewer crashes during late-night renders.
• Smart Node Search → Fuzzy matching now live—type “upscale” and find every relevant node in seconds. 🕵️‍♂️
• Python 3.10+ Only → Dropped 3.9 for speed & stability. Time to upgrade!
• Dark Mode Pro → Smoother contrast, crisper icons. Your eyes will thank you at 3 AM.

Windows users: New installer auto-detects CUDA + installs the right torch version. No more “why won’t it run?!” 😅

Plus: 12+ community nodes already live in the registry. Update now, tweak your flows, and let the AI art flow 🚀

🔗 View Release

MLX-LM – v0.30.2

mlx-lm v0.30.2 is out — quiet update, big win for Apple Silicon devs! 🍏

This patch fixes a sneaky build/deploy issue in the release pipeline. No flashy features, but if you’ve been wrestling with install errors or import fails on M-series chips, this is the fix you’ve been waiting for.

Upgrade now and get back to running LLMs smoothly — Hugging Face models, quantization, long-context gen… all working as intended.

🔗 Full details: [changelog](v0.30.1…v0.30.2)

Stay sharp, stay Apple-silicon-powered!

🔗 View Release

MLX-LM – v0.30.1

🔥 MLX LM v0.30.1 is LIVE — Apple Silicon LLMs just got a massive upgrade!

🚀 New Models: RWKV7, Solar Open, K-EXAONE MoE, IQuest Coder V1, YoutuLLM + Minimax M2 (perfect for long-context chats)!

💬 Chat Fixes: Custom DSV32 templates work, non-standard tokenizers behave, and `generation_config` errors? Ignored — no more crashes.

⚡ Performance: GIL starvation fixed in `_generate`, Phi3 (LongRoPE) batched prompts now stable, and `load_config` checks files first — smarter loading.

✨ New Features: `logits_processors` in `batch_generate` (fine-tune outputs like a pro), `model-path` flag for cleaner conversions, and support for mxfp8 & nvfp4 quantization — squeeze more power from your M-series chip!

📋 Bug fixes: `/v1/models` now shows local models correctly.

Big thanks to new contributors: @cubist38, @vyaivanov, @sjugin, @jaycoolslm, @MollySophia, @cxl-git-hub, @lazarust!

Upgrade. Tinker. Crush your next LLM project. 🛠️💻

🔗 View Release