Hacker News Reader

emadmon Stable Cascade

Tbh I think these models will largely be trained on synthetic datasets in the future. They are mostly trained on garbage now. We have been doing opt outs on these, has been interesting to see quality differential (or lack thereof), eg removing books3 from stableLM 3b zephyr https://stability.wandb.io/stability-llm/stable-lm/reports/S...

emadmon Stable Cascade

We should, we haven't yet...

Instead we've given 10m+ supercomputer hours in grants to all sorts of projects, now we have our grant team in place & there is a huge increase in available funding for folk that can actually build stuff we can tap into.

emadmon Stable Cascade

Code is MIT, weights are under the NC license for now.

emadmon Stable Cascade

https://x.com/EMostaque/status/1649152422634221593?s=20

We now have top models of every type, sites like www.stableaudio.com, memberships, custom model deals etc so lots of demand

We're the only AI company that can make a model of any type for anyone from scratch & are the most liked / one of the most downloaded on HuggingFace (https://x.com/Jarvis_Data/status/1730394474285572148?s=20, https://x.com/EMostaque/status/1727055672057962634?s=20)

Its going ok, team working hard and shipping good models, the team are accelerating their work on building ComfyUI to bring it all together.

My favourite recent model was CheXagent, I think medical models should be open & will really save lives: https://x.com/Kseniase_/status/1754575702824038717?s=20

emadmon Stable Cascade

Dom and Pablo both work for Stability AI (Dom finishing his degree).

All the original Stable Diffusion researchers (Robin Rombach, Patrick Esser, Dominik Lorenz, Andreas Blattman) also work for Stability AI.

emadmon Stable-Audio-Demo

stableaudio.com is fully licensed, music is an interesting area

https://www.musicbusinessworldwide.com/stability-ai-launches...

emadmon Stable-Audio-Demo

Yeah, we build ComfyUI so you can imagine what is coming soon around that.

Need to add more stuff to my Soundcloud https://on.soundcloud.com/XrqNb

emadmon Stable-Audio-Demo

Particularly interesting with the scaled up version of https://www.text-description-to-speech.com

Do try https://www.stableaudio.com for rights licensed model you can use commercially.

emadmon Stable-Audio-Demo

This is part of a paper on the prior version of the model: https://x.com/stableaudio/status/1755558334797685089?s=20

https://arxiv.org/abs/2402.04825

Which outperforms similar music models.

The pace is accelerating and even better ones are coming with far greater cohesion and... stuff. Will be quite the year for music.

emadmon Stable-Audio-Demo

There is a CC licensed version soon plus API.

Models are advancing very fast, will be quite the year for music.

emadmon Stable-Audio-Demo

It's a complex topic and perceptions change.

Ed still likes Stability, especially as we fully trained stable audio on rights licensed data (bit different in audio to other media types), offer opt out of datasets etc.

emadmon Google TPU v5p beats Nvidia H100

Training on 1k+ v5p and h100 this isn’t true. We get 400 tflops per chip on v5p int8 and 650 fp8 for h100s

H100s are less stable training though.

V5ps are about the same as intel gaudi2s and both are cheaper than h100s price/perf

Gaudi3s are 3-4x faster…

emadmon Stable Code 3B: Coding on the Edge

There are not enough cats on the internet. We are working to fix this.

emadmon Stable Code 3B: Coding on the Edge

Yeah, Phi-2 is weird on chat, StableLM beats it on some metrics, Phi-2 does on others but also doesn't really have system integration yet.

The base model of StableLM 3b zephyr is actually under an even more permissive license (we didn't change in retrospect) and is the best base to train on for MacBooks with 8gb RAM, edge devices etc.

With LLM Farm quantised you can run it faster than you can read on a iPhone or whatever.

https://huggingface.co/stabilityai/stablelm-3b-4e1t

It's also one of the only models with fully dataset, training and other transparency: https://stability.wandb.io/stability-llm/stable-lm/reports/S...

emadmon Stable Code 3B: Coding on the Edge

Yeah there is going to be a big push into Comfy and some very interesting new models coming ^_^

emadmon If you can't reproduce the model then it's not open-source

We made our last language model fully reproducible including all datasets, training details, hyper parameters etc: https://stability.wandb.io/stability-llm/stable-lm/reports/S...

emadmon Stable Code 3B: Coding on the Edge

Membership with upsell to support, custom models and more

Plus licensed variant models like stable audio and on chip installation like arm for specialist models eg Japanese law or Indonesian accounting

emadmon Stable Code 3B: Coding on the Edge

We released a competitor to runway recently that beat it on blind tests, plus way faster image in sdxl turbo

We have been working on ComfyUI for the next step and new image models

Midjourney and others are pipelines versus models so we have a higher bar to jump but the og stable diffusion team are working hard!

emadmon Stable Code 3B: Coding on the Edge

3b is good for 8gb MacBook Air etc. 7b is slightly too big.

Sure these will continue to improve, phi2 is a good base as well

emadmon Stable Code 3B: Coding on the Edge

Business model is bundling so you have a one stop shop for good quality models of every modality and cultural variants of them.

These go on bedrock, on chip, on prem etc and our consulting partners take them to the end user.

On the innovation side stable diffusion turbo does like 100 cats with hats per second and the video model outperforms runway, pika etc on blind tests.

Stable audio was one of the time innovation of the year winners on music and we released a sota 3d model.

Stable LM zephyr is the best 3b chat model works great on a MacBook Air.

Most of the pixels in the world will be generated so fast high quality image/video are the core and these other models are to support them.

It’s really hard to build good solid models and we are the only company that can build a model of any type for anyone.

emadmon Run Mistral 7B on M1 Mac

Just run stable lm zephyr https://huggingface.co/TheBloke/stablelm-zephyr-3b-GGUF

emadmon GigaGPT: GPT-3 sized models in 565 lines of code

StableLM-Zephyr outperforms original GPT3, Falcon 40b instruct, Guacano 65b etc and runs on cpu

On windows you can get it via lmstudio.ai for example

Here is ur running on a MacBook M2 Air, we have smaller, more performant models coming

https://x.com/emostaque/status/1732912442282312099?s=46

emadmon Mistral AI Valued at $2B

With all the new national supercomputers scale isn’t really going to be an issue, they all want large language models on 10k GH200s or whatever and the libraries are getting easier to use

emadmon Mistral AI Valued at $2B

I think its also tough at the early stage of the diffusion (aha) of innovation curve, we are at the point of early adopters and high churn before mass adoption of these technologies over the coming years as they are good enough, fast enough and cheap enough.

AGI is a bit of a canard imo, its not really actionable on a business sense.

emadmon Mistral AI Valued at $2B

With convertible debt and many of these rounds investors get the first money out, so the first 450m would go to the investors.

emadmon Mistral AI Valued at $2B

It’s kinda weird thinking deep tech companies should be profitable a year in.

Like it takes time to make lots of money and it’s really hard to build state of the art models.

Reality is this market is huge and growing massively as it is so much more efficient to use these models than many (but not all) tasks.

At stability I told team to focus on shipping models as next year is the year for generative media where we are the leader as language models go to the edge.

emadmon Mistral AI Valued at $2B

Main barrier right now is access to supercompute and how to run it, everything is standardising quickly in the space

emadmon Mistral AI Valued at $2B

Aleph Alpha raised even more ^_^

https://sifted.eu/articles/ai-startup-aleph-alpha-raises-500...

emadmon Mistral AI Valued at $2B

They already do, we just released a model equivalent to most 40-60b base models that runs on a MacBook Air no problem.

It's like 1.6gb, ones coming are better and smaller https://x.com/EMostaque/status/1732912442282312099?s=20

I think the large language model paradigm is pretty much done as we move to satisficing tbh

emadmon Mistral AI Valued at $2B

Yeah they shouldn't worry, they'll get a big French government deal at worst