҈̿ ̶̢̢̧̡̼̜̝̬͍̜̘̙͉̘͎͓͍̣̰͖̹͖͚̭̘̖̟͕̬̠̬͇̹̮͎̣̱̎̾͌̊́̇͛̅͂̀̀͑̃̈̀̓̏͌͌͋̐̾̒͋͋̏͋̽̈́͐͑̐̂͆̊̈́̾͌̓͌̕̚͝͝͠͠͝ͅ Repos: https://xitter.cf/t.co/3x0DctKuld
Joined March 2018
- Tweets3.7K
- Following137
- Followers1.8K
- Likes541
̷̨̨̛̛̲͎̣̞̘͈̯͚͂̈́̂̄̽̎́̽̔͑̄̏̽̏͒̾́̅̐̈́̾̎̆͆̽́͌̽̀̚̕̚̕͠͝͝ retweeted
I love the #ChatGPT Cheat Sheet by Ricky Costa (@Quantum_Stat)
which includes
🔹NLP Tasks
🔹Code
🔹Structured Output Styles
🔹Unstructured Output Styles
🔹Media Types
🔹Meta ChatGPT
🔹Expert Prompting
Get your hands on this amazing resource at:i.mtr.cool/ehyhxpfexx
̷̨̨̛̛̲͎̣̞̘͈̯͚͂̈́̂̄̽̎́̽̔͑̄̏̽̏͒̾́̅̐̈́̾̎̆͆̽́͌̽̀̚̕̚̕͠͝͝ retweeted
Exciting news from our latest LLM compression research! 🚀 Together with @ISTAustria and @neuralmagic, we’ve been exploring sparse finetuning for LLMs and achieved 7.7 tokens/second on a single core and at 26.7 tokens/second on 4 cores of an AMD Ryzen CPU! (1/n)
̷̨̨̛̛̲͎̣̞̘͈̯͚͂̈́̂̄̽̎́̽̔͑̄̏̽̏͒̾́̅̐̈́̾̎̆͆̽́͌̽̀̚̕̚̕͠͝͝ retweeted
Want to prune your #ML models at higher levels without impacting accuracy? ✂️
Join us for a virtual session 📺 on April 6 where we'll discuss second-order pruning methods that enable higher sparsity by removing weights that directly affect the loss function the least.
1/3
̷̨̨̛̛̲͎̣̞̘͈̯͚͂̈́̂̄̽̎́̽̔͑̄̏̽̏͒̾́̅̐̈́̾̎̆͆̽́͌̽̀̚̕̚̕͠͝͝ retweeted
Deploying a GPT-175B requires 5 A100 80GB GPUs, each costing $15,000.
That's $75,000 for inference 💰
You can reduce the model’s size by removing 50% of the weights without losing accuracy 🤯
Let's explore how to do that with the #SparseGPT algorithm.
-A quick thread-
🧵👇
̷̨̨̛̛̲͎̣̞̘͈̯͚͂̈́̂̄̽̎́̽̔͑̄̏̽̏͒̾́̅̐̈́̾̎̆͆̽́͌̽̀̚̕̚̕͠͝͝ retweeted
My recent talk given during our external research seminars at @neuralmagic is online!
youtu.be/npcT5VO6YOA
̷̨̨̛̛̲͎̣̞̘͈̯͚͂̈́̂̄̽̎́̽̔͑̄̏̽̏͒̾́̅̐̈́̾̎̆͆̽́͌̽̀̚̕̚̕͠͝͝ retweeted
🚨 Introducing Sparse BioBERT, our new #NLP model that is specifically fine-tuned on four unique #biology related datasets.
Benchmarked against its dense variant on the ArXiv-biology dataset, the performance improvement was over 9x faster for BioBERT⚡️
neuralmagic.com/blog/revolut…
̷̨̨̛̛̲͎̣̞̘͈̯͚͂̈́̂̄̽̎́̽̔͑̄̏̽̏͒̾́̅̐̈́̾̎̆͆̽́͌̽̀̚̕̚̕͠͝͝ retweeted
#YOLOv5 has been awesome for building impressive object detection demos that scale from laptops to servers!
Check out the 20+ sparsified versions @neuralmagic has made, from nano🤏 to xlarge🦣
sparsezoo.neuralmagic.com/?r…
Announcing our new partnership 📣
Deploy #YOLOv5 with GPU-class performance on CPUs with @neuralmagic 🌟
Neural Magic is leading a new software-delivered AI movement by bringing hyper-performant and scalable ML inference to commodity CPU infrastructure
bit.ly/3IvPNs3
̷̨̨̛̛̲͎̣̞̘͈̯͚͂̈́̂̄̽̎́̽̔͑̄̏̽̏͒̾́̅̐̈́̾̎̆͆̽́͌̽̀̚̕̚̕͠͝͝ retweeted
A #StableDiffusion v2.1 embedding fine-tuned for generating images that look like Web-UI designs.
Example Prompt: “Landing page for space rocket, trendy online store, modern futuristic online store, clean and modern design, web_design by web-ui-v2-700”
huggingface.co/pmejna/web-ui…
̷̨̨̛̛̲͎̣̞̘͈̯͚͂̈́̂̄̽̎́̽̔͑̄̏̽̏͒̾́̅̐̈́̾̎̆͆̽́͌̽̀̚̕̚̕͠͝͝ retweeted
Accelerate your @huggingface 🤗 Inference Endpoints with DeepSparse to achieve 43x CPU speedup and 97% cost reduction over @PyTorch.
Side note: DeepSparse is even faster than a T4 GPU 🤯
Learn more in our blog: neuralmagic.com/blog/acceler… – at Massachusetts, USA
ALT DeepSparse cost efficiency on inference endpoint compared to PyTorch CPU and PyTorch T4 GPU
̷̨̨̛̛̲͎̣̞̘͈̯͚͂̈́̂̄̽̎́̽̔͑̄̏̽̏͒̾́̅̐̈́̾̎̆͆̽́͌̽̀̚̕̚̕͠͝͝ retweeted
Introducing SetFit v0.5 - now with support to optimize few-shot models for inference 🏎!
The new features include:
🧑🏫 A knowledge distillation trainer
🕸 Export to @onnxai and quantize w/ @onnxruntime or @IntelAI 's Neural Compressor
Code & examples: github.com/huggingface/setfi…
̷̨̨̛̛̲͎̣̞̘͈̯͚͂̈́̂̄̽̎́̽̔͑̄̏̽̏͒̾́̅̐̈́̾̎̆͆̽́͌̽̀̚̕̚̕͠͝͝ retweeted
How does FSDP make large-scale AI training more efficient?
FSDP parallelizes data, model parameters, optimizer states and gradients associated with the model.
Learn how to boost your distributed training speed and model sizes: bit.ly/3W6ogl4
1/4
̷̨̨̛̛̲͎̣̞̘͈̯͚͂̈́̂̄̽̎́̽̔͑̄̏̽̏͒̾́̅̐̈́̾̎̆͆̽́͌̽̀̚̕̚̕͠͝͝ retweeted
We’re celebrating the first 6 months of Cohere For AI and the moments that made this year special! Here’s a look back at our 2022 highlights! 🧵
txt.cohere.ai/c4ai-2022/
̷̨̨̛̛̲͎̣̞̘͈̯͚͂̈́̂̄̽̎́̽̔͑̄̏̽̏͒̾́̅̐̈́̾̎̆͆̽́͌̽̀̚̕̚̕͠͝͝ retweeted
To better understand a complex topic, it sometimes helps to have it explained in multiple ways. @soniakmurthy & coauthors' ACCoRD system generates sets of descriptions of scientific concepts. CC @Marktechpost: marktechpost.com/2022/12/13/…
̷̨̨̛̛̲͎̣̞̘͈̯͚͂̈́̂̄̽̎́̽̔͑̄̏̽̏͒̾́̅̐̈́̾̎̆͆̽́͌̽̀̚̕̚̕͠͝͝ retweeted
Microsoft is pleased to announce that seven of its research papers have been selected for oral presentations at #NeurIPS2022, and two others have received Outstanding Paper Awards. msft.it/6011eEqqW
̷̨̨̛̛̲͎̣̞̘͈̯͚͂̈́̂̄̽̎́̽̔͑̄̏̽̏͒̾́̅̐̈́̾̎̆͆̽́͌̽̀̚̕̚̕͠͝͝ retweeted
It gets even better when you combine int8-Quantization and pruning! And quantization aware training is really easy to do with @neuralmagic
̷̨̨̛̛̲͎̣̞̘͈̯͚͂̈́̂̄̽̎́̽̔͑̄̏̽̏͒̾́̅̐̈́̾̎̆͆̽́͌̽̀̚̕̚̕͠͝͝ retweeted
🚨 New blog post alert 🚨
I had a blast 💥 playing around with @neuralmagic's (NM) software to train and run a face detector at GPU speed on CPU!
👉 Post: francescopochetti.com/neural…
📒 Notebook: github.com/FraPochetti/Kaggl…
In this one, we'll learn how to 👇
̷̨̨̛̛̲͎̣̞̘͈̯͚͂̈́̂̄̽̎́̽̔͑̄̏̽̏͒̾́̅̐̈́̾̎̆͆̽́͌̽̀̚̕̚̕͠͝͝ retweeted
1) Instead of spending large amounts of time, engineering, and money on labeling and perfecting a dataset, utilize the power of large language models to classify text sequences without training for annotation or deployment on your data.
̷̨̨̛̛̲͎̣̞̘͈̯͚͂̈́̂̄̽̎́̽̔͑̄̏̽̏͒̾́̅̐̈́̾̎̆͆̽́͌̽̀̚̕̚̕͠͝͝ retweeted
Stoked to receive @neuralmagic swag 🚀 for a recent viral tweet of mine about all the awesome work they are doing to make neural networks more accessible!
So kind of them 🙏
Btw, I am working on a Neural Magic + SageMaker post 😉. Stay tuned for updates!
̷̨̨̛̛̲͎̣̞̘͈̯͚͂̈́̂̄̽̎́̽̔͑̄̏̽̏͒̾́̅̐̈́̾̎̆͆̽́͌̽̀̚̕̚̕͠͝͝ retweeted
📣📣 You can now deploy the DeepSparse Engine on AWS SageMaker! 📣📣
In this example, we deployed sparse DistilBERT to achieve a 7x increase in model performance. 📈
You can do the same with a single CLI command!
neuralmagic.com/blog/deepspa…
#sagemaker #machinelearning #sparsity
̷̨̨̛̛̲͎̣̞̘͈̯͚͂̈́̂̄̽̎́̽̔͑̄̏̽̏͒̾́̅̐̈́̾̎̆͆̽́͌̽̀̚̕̚̕͠͝͝ retweeted
CVPR Main conference is starting today! Check out the Stanford AI Lab papers and talks in our new blog post #cvpr2022 ai.stanford.edu/blog/cvpr-20…
