Using this for my personal assistant harness now - ONNX conversions available on my hf :)
Today we’re releasing Underdog Bark 3.0, the #1 frontier speech model
It's a tiny 163 MB download as accurate as NVIDIA’s Parakeet (15x bigger). It beats OpenAI's Whisper large-v3 (57x bigger)
It's fast- transcribes 1 min of audio in <0.25s on a Macbook. On recorded meetings (AMI), Bark 3.0 gets 7.82% of words wrong. Parakeet gets 9.42%. OpenAI's Whisper large-v3 gets 13.63% error rate
Bark runs on both your Laptop & Phone. It’s free and open source under CC-BY 4.0
Try private on-device meeting notes & dictation, powered by Bark, in Underdog today: underdog.ai/bark
Ok I can finally see how linux nerds get into the whole "yeah lets just customize everything" vibe. This is awesome
What happened to macOS man. I've been traveling for a week and omg I've encountered soooo many bugs.
funny weekend projects: watch animals duke it out in chess and bet coins on them. pawnimals.tv/#/tv
Pick left because you can't spend money that doubles, someone else will also get infinitely doubling bills and then your money is useless.
I may have cooked more than I should've.
Running the decision index benchmarks today and the extrapolation is looking promising!
I've been working on a series of decision models ranging from a tiny jev alternative. I didn't realize how big they blew up while I was playing around teaching this model. After I saw that openAI was going to launch theirs I kinda figured it's time to do something else 🤣 I don't have the compute nor data to play in the big leagues
No clue if it's good, but the benchmarks did well for a 0.6B model. It performs quite well on jevbench, with a 66% completion rate, it outperforms gpt6 luna, and generally punches in the 27-35B category with a P95 of 36ms on a DGX spark.
The model is a culmination of 1000s of individual training runs. It sucks at computer use and coding, but, it seems quite well versed on logic, arithmetic, and policy forming. If you want - give it a try and let me know what you think. I figured it'd be better to throw it out in public than to leave it in the back of my projects folder.
Aside from that there's also some experimental new API surfaces such as sorting large lists (useful for e.g. search re-rankers), and image/video support. Although I didn't want to fit all that into the 0.6B model🙂
huggingface.co/velddev/nest-…
My friend and I have been collaborating together on making our own llm tools. Ofcourse we also gotta have bots now.
I'm going to lose my mind why do I need to log in to google cloud every day and why does cloudflare need to 500 every time on 2FA.
Just when work became so much smoother these companies are ensloppening their auth stack so much that I cannot do anything without needing twelve different auth steps.
Concerning that nearly all of OpenAI dev day releases were reactive with no product launches that were done to push their own vision forward.
I've been working on a series of decision models ranging from a tiny jev alternative. I didn't realize how big they blew up while I was playing around teaching this model. After I saw that openAI was going to launch theirs I kinda figured it's time to do something else 🤣 I don't have the compute nor data to play in the big leagues
No clue if it's good, but the benchmarks did well for a 0.6B model. It performs quite well on jevbench, with a 66% completion rate, it outperforms gpt6 luna, and generally punches in the 27-35B category with a P95 of 36ms on a DGX spark.
The model is a culmination of 1000s of individual training runs. It sucks at computer use and coding, but, it seems quite well versed on logic, arithmetic, and policy forming. If you want - give it a try and let me know what you think. I figured it'd be better to throw it out in public than to leave it in the back of my projects folder.
Aside from that there's also some experimental new API surfaces such as sorting large lists (useful for e.g. search re-rankers), and image/video support. Although I didn't want to fit all that into the 0.6B model🙂
huggingface.co/velddev/nest-…