microsoft/VibeVoice-ASR-BitNet
Automatic Speech Recognition • 0.3B • Updated • 1.75k • 94
ActiveEnergy
islameissa
AI & ML interests
None yet
Recent Activity
liked a model 5 days ago
microsoft/VibeVoice-ASR-BitNet liked a model about 1 month ago
deepreinforce-ai/Ornith-1.0-35B-GGUF new activity about 1 month ago
deepreinforce-ai/Ornith-1.0-35B-GGUF:Good JobOrganizations
None yet
liked a
model 5 days ago
liked a
model about 1 month ago
Good Job
👍🔥 6
#8 opened about 1 month ago
by islameissa
reacted to constannnt's post with 🔥🚀❤️ about 1 month ago
Post
10397
We are excited to announce Sipp.sh: a high-performance library for running AI inference locally and in the cloud through a unified API.
We began to realize that an LLM isn't just a chat interface for information retrieval. It can be integrated directly into web, games, or productivity apps to handle continuous monitoring and decision-making. It can act as a sort of "second brain,” the silent hand that guides and helps a user without them even realizing it. We see this as the next frontier of UX design, but this is only possible if developers have access to low-cost, zero-latency compute and absolute data privacy.
That's why we created Sipp. It’s an opinionated library that lets developers integrate local AI into any application, giving them the superpowers to completely rethink user experiences across the web, games, and desktop.
To achieve this, we built an entirely new stack in Rust and C++, working alongside the llama.cpp project. Through our work, we were able to contribute back to that community to help upgrade the GGML WebGPU backend. This deep optimization is what enables our fast, responsive decode speeds directly in the browser. Sipp ships as a zero-dependency library for desktop and web, achieving 3x to 5x speedup in token decode compared to popular alternatives.
We are already seeing some incredible use cases emerge from this, from continuous monitoring using local vision to the dynamic generation of game elements in a real-time wizard vs. wizard game.
The best part? It's fully open-source!
We see this as the start of a dialogue about what the future of user interaction is going to look like, and we built Sipp to lay the foundation for that exciting future. Check out the live demos on our site, run your own benchmarks, or come hang out with us in our Discord.
Website: https://www.sipp.sh/
Github: https://github.com/noumena-labs/Sipp
We began to realize that an LLM isn't just a chat interface for information retrieval. It can be integrated directly into web, games, or productivity apps to handle continuous monitoring and decision-making. It can act as a sort of "second brain,” the silent hand that guides and helps a user without them even realizing it. We see this as the next frontier of UX design, but this is only possible if developers have access to low-cost, zero-latency compute and absolute data privacy.
That's why we created Sipp. It’s an opinionated library that lets developers integrate local AI into any application, giving them the superpowers to completely rethink user experiences across the web, games, and desktop.
To achieve this, we built an entirely new stack in Rust and C++, working alongside the llama.cpp project. Through our work, we were able to contribute back to that community to help upgrade the GGML WebGPU backend. This deep optimization is what enables our fast, responsive decode speeds directly in the browser. Sipp ships as a zero-dependency library for desktop and web, achieving 3x to 5x speedup in token decode compared to popular alternatives.
We are already seeing some incredible use cases emerge from this, from continuous monitoring using local vision to the dynamic generation of game elements in a real-time wizard vs. wizard game.
The best part? It's fully open-source!
We see this as the start of a dialogue about what the future of user interaction is going to look like, and we built Sipp to lay the foundation for that exciting future. Check out the live demos on our site, run your own benchmarks, or come hang out with us in our Discord.
Website: https://www.sipp.sh/
Github: https://github.com/noumena-labs/Sipp
- 1 reply
I really like this approach. Could you share more details on how this is architecturally implemented? Specifically, is the Neural Reasoning System (NRS) baked into the model via fine-tuning/parameter adapters (like LoRA), or does it function as an external inference-time execution layer (e.g., a custom logits processor, verifier agent, or decoding wrapper) that intercepts token generation?
liked a
model 3 months ago
updated a
model 3 months ago
published a
model 3 months ago
This is very interesting. Can we create multiple shadow layers and attach the appropriate layer when needed and detach it after?
liked a
dataset 3 months ago
Qwen3.6 GGUF Benchmarks
🔥👍 23
26
#10 opened 3 months ago
by 
danielhanchen
2-bit Qwen3.6 Tool-calling is amazing!!
🔥🚀 15
13
#2 opened 3 months ago
by danielhanchen
Nice Quants
👍🤝 3
#1 opened 4 months ago
by islameissa
liked 2
models 4 months ago
Verbose looping
2
#28 opened 5 months ago
by islameissa
upvoted an article 5 months ago
Article
GGML and llama.cpp join HF to ensure the long-term progress of Local AI
- +4
ggerganov, ngxson, allozaur, lysandre, victor, julien-c
• • 508 Large or medium versions
#1 opened 5 months ago
by islameissa