Category: AI

Piper Sample Generator – v3.1.0

Howdy folks! Tater Totterson here with the latest scoop for all you AI enthusiasts! 🥔✨

The Piper Sample Generator just dropped v3.1.0 – and it’s got some goodies! 🎉

If you’re an Apple Silicon user, get ready to rumble! This update adds MPS acceleration, making sample generation much faster on your Macs. 🚀

Plus, a new flag has been added: `–phoneme-inputflag`. Looks like you can now feed the generator phonemes directly – perfect for more detailed speech synthesis and customized wake word training! 🗣️

Check it out here: https://github.com/rhasspy/piper-sample-generator/releases/tag/v3.1.0

Keep those neurons firing, and happy AI-ing! 🤖

https://github.com/rhasspy/piper-sample-generator/releases/tag/v3.1.0

Ollama – v0.12.5

🥔 Ollama v0.12.5 Released! 🥔

Hey AI Enthusiasts! 👋

Ollama, your local LLM runner, just dropped v0.12.5 with a focus on Qwen model support! 🚀

Here’s the scoop:

• What is Ollama? Simplifies downloading, running & querying open-source LLMs locally (Llama 3, Mistral, Gemma, and more!).
• ✨ New in v0.12.5 ✨:
• Qwen3 models now use the `llama runner` for improved performance & compatibility!
• No Breaking Changes! 🎉 Upgrade worry-free.

Release Link: https://github.com/ollama/ollama/releases/tag/v0.12.5

Happy inferencing! 🤖✨

https://github.com/ollama/ollama/releases/tag/v0.12.5

Ollama – v0.12.5-rc0

🔥 Ollama v0.12.5-rc0 Released! 🔥

Hey AI Enthusiasts! 🥔 TaterBytes here with a quick update on Ollama, your go-to for running LLMs locally. Version `v0.12.5-rc0` is out, bringing some nice improvements:

• Qwen3 Support: Run Qwen3 models now via the `llama runner`!
• Performance Boost: General performance enhancements throughout Ollama.
• Bug Squashing: Stability and usability fixes implemented. 🐛➡️✅

No breaking changes are currently called out in this release candidate 🎉

Check out the full release notes & get updating: https://github.com/ollama/ollama/releases/tag/v0.12.5-rc0

https://github.com/ollama/ollama/releases/tag/v0.12.5-rc0

Text Generation Webui – v3.14

✨ Oobabooga’s Text Generation Web UI – v3.14 Update ✨

Hey AI Enthusiasts! 🤖 The popular local LLM interface, Text-Generation-WebUI, just dropped v3.14 with some solid improvements:

• Multi-GPU Performance: Enhanced `bitsandbytes` support for faster inference on multi-GPU setups (8 & 4-bit quantization).
• ExLlamaV3 Integration: New `/v1/internal/logits` endpoint added for advanced integrations using `exllamav3` and `exllamav3_hf`.
• Qwen Support: Now supports Qwen 3-Next models with ExLlamaV3 (requires `flato`).
• llama.cpp Update: Upgraded to the latest ggml-org/llama.cpp.
• Dependency Updates: `transformers` (v4.57), `exllamav3` (v0.0.7), and `bitsandbytes` (v0.48) all updated!

Bug Fixes: Chat history loading & macOS portable build issues resolved.

📦 Portable Builds Available: Grab the latest version for your setup: https://github.com/oobabooga/text-generation-webui/releases/tag/v3.14

• NVIDIA (cuda12.4, cuda11.7)
• AMD/Intel (vulkan)
• CPU Only
• Mac (Apple Silicon & Intel)

Happy generating! ✨

https://github.com/oobabooga/text-generation-webui/releases/tag/v3.14

Tater – Tater v31

🥔 Howdy AI Homebrew Fans! 🥔

Tater v31 is here and it’s got a VOICE! 🎉

The latest version of Tater brings full Home Assistant voice integration – you can now chat to Tater through your voice assistants, and he’ll talk (and even sing!) back! 🎤🏡

Here’s the breakdown:

• Home Assistant Voice Control: Chat, ask questions, generate music, control smart home devices – all with your voice!
• Updated Plugins: `web_search`, `overseerr`, and `comfyui_audio_ace` are now voice-ready.
• New Plugin: `ha_control`: Direct Home Assistant device control (proof of concept!).
• Conversation History: Tater remembers your chats for up to 2 hours!

Tater’s moved out of Discord and into your living room! 🥔🎶

Check it all out here: https://github.com/TaterTotterson/tater & see the release notes: https://github.com/TaterTotterson/Tater/releases/tag/v31

Happy brewing! 🍻

https://github.com/TaterTotterson/Tater/releases/tag/v31

Micro Wake Word Trainer Apple Silicon – microWakeWord-Trainer-AppleSilicon v1.0

Hey AI Enthusiasts! 👋

Great news for all you Home Assistant voice control fans! microWakeWord-Trainer-AppleSilicon v1.0 is here – a fully automated wake-word trainer optimized for Apple Silicon (M1-M4) Macs! 🍏

Here’s the scoop:

• Simplified Training: One bash command (`./train_microwakeword_macos.sh “wake word”`) does it all – TTS sample generation with Piper, training, and exporting a quantized model for ESPHome.
• GPU Accelerated: Uses TensorFlow-Metal to harness the power of your Apple Silicon chip for faster training!
• Offline Functionality: Local caching ensures smooth operation even without a HuggingFace connection.
• Natural Sounding Wake Words: Powered by Piper TTS for realistic voice samples.
• ESPHome Integration: Creates `.tflite` models and `.json` metadata perfect for your ESPHome projects.

Check it out & start training: https://github.com/TaterTotterson/microWakeWord-Trainer-AppleSilicon/releases/tag/1.0

Happy building! 🥔☕️

https://github.com/TaterTotterson/microWakeWord-Trainer-AppleSilicon/releases/tag/1.0