Tater – Tater v56

🚨 Tater v56 — Cerberus Upgrade is LIVE! 🚨

The core of Tater just got a massive intelligence overhaul — meet Cerberus, the new 3-phase reasoning engine:

🔹 Plan → Execute → Validate — ensures smarter, step-by-step execution with fewer surprises.

🛡️ Tool Safety Upgraded

• Tools only fire when absolutely intended
• Malformed or accidental calls? Nope. Not today.

✉️ Smarter Messaging

• `send_message` now ignores casual chat — only triggers on clear intent
• “Send it here” ✅, “Hey, send that” ❌ (nope!)

🧠 Context That Actually Makes Sense

• Clean, scoped memory per conversation
• “Do that again?” → Works now
• Topic shifts reset context intelligently (no more weird carryover!)

📊 Behind-the-Curtain Wins

✅ Smoother error handling & retries

✅ Reduced token bloat = lower costs & faster responses

✅ More deterministic behavior (yes, predictable LLMs are a thing now)

🎯 What This Means for You

✔️ Reliable multi-step workflows (finally!)

✔️ Fewer “why did it do that?!” moments

✔️ Natural, fluid follow-ups

✔️ A rock-solid foundation for future agent smarts

🔮 What’s Next?

Cerberus sets the stage for long-horizon tasks, learning, and advanced agent behavior — but v56 is all about stability, safety, and raw, reliable power.

🐶 Cerberus is awake. 🐶🔥

👉 Check the README to upgrade & explore!

🔗 View Release

Heretic – v1.2.0

🚨 Heretic v1.2.0 is live — and it’s packing serious upgrades! 🚨

The team just dropped a massive update to Heretic, the fully automatic censorship-removal tool for LLMs — and this one’s packed with performance boosts, new features, and rock-solid stability fixes. Here’s what’s new:

🔹 Memory & Stability Wins

• `amax_memory` setting to cap RAM usage 🧠
• Smarter iteration logic to avoid getting stuck in low-divergence traps ⚙️
• Magnitude-Preserving Orthogonal Ablation — finer control over refusal suppression 🎯

🔹 LoRA + Quantization Power-Ups

• Brand-new LoRA-based abliteration engine 🪄
• Full 4-bit quantization support — run decensoring on consumer GPUs 🖥️
• Auto-GPU detection at startup (no more manual config headaches!) 🤖

🔹 Resilience & Flexibility

• Run extra optimization trials after the main run finishes 🔄
• Save/resume support for long jobs — no more lost progress! 💾
• Fixed MXFP4 loading issues 🛠️
• Refactored save/load machinery = way more reliable 📦

🔹 Vision-Language Models? Now Supported!

• Broad support for VL models (e.g., LLaVA-style) 📸➡️🧠
• Plus: full type checking, debug tools, slop-reduction configs, and bug fixes 🐞

🌟 Shoutout to @noctrex, @accemlcc, @anrp, and @salmanmkc — welcome to the crew!

🔗 Full changelog: v1.1.0 → v1.2.0

🚀 Grab it now and keep decensoring smarter, faster, and safer!

🔗 View Release

Ollama – v0.16.2-rc0: mlxrunner fixes (#14247)

🚨 Ollama v0.16.2-rc0 is here — and it’s bringing serious Apple Silicon love! 🍏

This release candidate focuses on fixing critical bugs in the `mlxrunner` backend (hello, M1/M2/M3 users 👋), making it way more reliable for cutting-edge models. Here’s what’s new:

🔹 GLM4-MoE-Lite now loads smoothly via `mlxrunner` — no more crashes on M-series!

🔹 Diffusion models (like Stable Diffusion) — finally load without hiccups 🖼️

🔹 Cleaner logs — less noise, more signal (goodbye, endless debug spam 🧹)

🔹 `–imagegen` flag now actually works for image generation workflows ✅

💡 Pro tip: This is a release candidate, so it’s the perfect time to test stability before v0.16.2 drops — especially if you’re into MoE models or local image gen on Mac.

Ready to give it a spin? 🧪 Let us know how it goes!

🔗 View Release

Home Assistant Voice Pe – 26.2.1

🚀 Home Assistant Voice PE v26.2.1 is live!

The latest update (from v25.12.4 → v26.2.1) is all about polish and reliability—perfect for those relying on voice control in their smart homes, especially offline. Here’s what’s improved:

✅ Media playback is smoother & more stable

No more skips or dropouts—your voice-triggered music, alerts, and announcements now play cleanly.

🛠️ TTS timeouts fixed!

Text-to-speech responses now fully render and play—no more truncated or missing voice replies. 🎙️✨

💡 Bonus: The project is now officially sponsored by the Open Home Foundation—a big vote of confidence in its mission for private, local-first voice control. 🏡🔐

78 releases down, more innovation ahead! 🛠️

Check it out if you’re building or expanding a local, privacy-first voice assistant setup. 🎯

🔗 View Release

Ollama – v0.16.1

🚨 Ollama v0.16.1 is live! 🚨

Hey AI tinkerers & local LLM lovers — fresh update incoming! 🔥

What’s new in v0.16.1?

🔹 New model config added: `minimax-m2.5` 🧠

• Looks like a fresh MiniMax model variant (internal/experimental for now — keep an eye out for docs!).
• You can already pull it via `ollama pull minimax-m2.5` if you’re feeling adventurous 🛠️

🔹 Lightweight patch release — no breaking changes, just lean & mean model support upgrades.

📦 Binaries are rolling out for macOS, Windows, and Linux — grab the latest from GitHub or update via your package manager.

👉 v0.16.1 Release Notes

Let us know if you get `minimax-m2.5` running — curious to hear your benchmarks and use cases! 🧪✨

🔗 View Release

Lemonade – v9.3.2

🚀 Lemonade v9.3.2 is live!

This one’s a quick but important patch—especially if you’re rocking AMD GPUs on Linux.

🔧 What’s new/fixed:

• ✅ Fixed incorrect path for Stable Diffusion ROCm artifacts on Linux

→ Fixes runtime hiccups and ensures proper loading of AMD GPU binaries

→ PR: #1085 | Commit: `5a382c5` (GPG verified!)

🎯 Why it matters:

• ROCm users on Linux can now run SD models reliably—no more path-related crashes or config headaches.
• No breaking changes, no flashy new features… just solid, quiet reliability 🛠️

If you’re using Lemonade with AMD/NPU/GPU acceleration on Linux—update now! 🐧✨

Full details: lemonade-sdk/lemonade

🔗 View Release

MLX-LM – v0.30.7

🚀 MLX-LM v0.30.7 is live — and it’s packed with model love, speed boosts, and polish!

🔥 New Models Added:

• ✅ GLM-5 — a powerful new contender in the LLM space 🧠
• ✅ Qwen3.5 (text-only) — ideal for high-performance, non-vision tasks
• ✅ DeepSeek V3.2 improvements — faster indexer & smoother weight loading 🛠️
• ✅ Kimi Linear bugs squashed — now stable & reliable

🛠️ Tooling Upgrades:

• 🐍 Pythonic tool calling for LFM2 models (huge thanks to @viktike!)
• 🧰 New Mistral tool parser — cleaner, more intuitive function/tool integration

⚡ Performance & Training:

• 📈 Faster DSV3.2 generation — thanks to kernel & op-level optimizations
• 📏 LongCat MLA support — smarter attention for long-context generations
• 🔁 Validation set now optional in training — faster prototyping!

👏 Shoutout to our newest contributors: @viktike & @JJJYmmm — welcome to the crew!

👉 Dive into the details: [v0.30.6 → v0.30.7 Changelog](link-to-changelog)

Let’s push the limits on Apple silicon — together! 🛠️💻✨

🔗 View Release

Home Assistant Voice Pe – 26.2.0

🚨 Home Assistant Voice PE v26.2.0 is live! 🚨

Hey AI tinkerers & smart home wizards — big shoutout to the latest update of Home Assistant Voice PE, now powered by the awesome folks at the Open Home Foundation 🙌

🔥 What’s new in v26.2.0?

✅ Media playback stability improved — fewer stutters, smoother audio/video responses during voice interactions

🎙️ TTS timeout bug squashed — no more cut-off replies! Full text now plays reliably, every time

💡 Bonus context: This release builds on 78+ prior releases — and now with offline-first voice control (no internet needed!), it’s perfect for privacy-focused automations.

Ready to make your smart home talk back? 🛠️✨

Check the changelog (25.12.4 → 26.2.0) and upgrade!

🔗 View Release

Ollama – v0.16.0

🚨 Ollama v0.16.0 is live! 🚨

The latest drop from the Ollama crew just landed — and while the release notes are light on flashy new features, this one’s a quiet but meaningful polish pass. Here’s the lowdown:

🔹 API Docs Fixed!

The OpenAPI schema for `/api/ps` (list running processes) and `/api/tags` (list local models) has been corrected — meaning better Swagger compatibility, smoother SDK generation, and fewer headaches for integrators. 🛠️

🔹 Stability & Under-the-Hood Tweaks

Expect refined model loading, improved streaming behavior, and likely minor bug fixes — especially around context handling and memory usage. No breaking changes, just smoother sailing.

🔹 Still GGUF-Friendly

All your favorite quantized models (Llama 3, DeepSeek-R1, Phi-4, etc.) keep rolling — no format changes here.

💡 Pro Tip: If you’re building tools or dashboards against Ollama’s REST API, this update makes your life easier. Run `ollama pull ollama/ollama:latest` or grab the latest binary from GitHub.

👉 Full details (when they land): v0.16.0 Release

Happy local LLM-ing! 🤖✨

🔗 View Release

Ollama – v0.16.0-rc2

🚨 Ollama v0.16.0-rc2 is out! 🚨

This release candidate is a light but tidy patch focused on API docs & stability—perfect for keeping your integrations humming. Here’s the lowdown:

🔹 Fixed OpenAPI schema for two key endpoints:

• `/api/ps` — now correctly documents the list running processes response
• `/api/tags` — updated to reflect accurate model tag listing behavior

✅ Why it matters: If you’re using SDKs, auto-generated clients, or UI tools that rely on the OpenAPI spec (like Swagger), this ensures they’ll work exactly as expected.

📦 Binaries for macOS, Linux & Windows are already up (2 assets).

📅 Released by `sam18` on Feb 12 @ 01:37 UTC

🔗 Commit: `f8dc7c9`

No flashy new models or breaking changes—just solid polish for the upcoming v0.16.0! 🛠️

Want a sneak peek at what’s actually new in v0.16 (beyond rc2)? Just say the word! 😄

🔗 View Release