osanseviero

6 ms·

1.
by osanseviero 3mo ago
Hi! What implementation are you using? Right now VLLM is the one recommended. llama.cpp is in an early draft
2.
by osanseviero 1y ago
Hi! The model is 8B if you also load the vision and audio components. We just used the text model in LMArena.
3.
by osanseviero 1y ago
Hi! Omar from the Gemma team here. Last time we only released the quantized GGUFs. Only llama.cpp users could use it (+ Ollama, but without vision). Now, we released the unquantized checkpoints, so anyone can quantize themselves and use in
4.
by osanseviero 2y ago
Please make sure to update to the latest llama.cpp version
5.
by osanseviero 2y ago
Nat Friedman leads the project. He was GitHub's CEO, among many other things. He funds many interesting ambitious projects, such as the Vesuvius Challenge ( https://scrollprize.org/ )
6.
by osanseviero 2y ago
Yes, they are still used - Encoder based models have much faster inference (are auto-regressive) and are smaller. They are great for applications where speed and efficiency are key. - Most embedding models are BERT-based (see MTEB leaderbo
7.
IBM and NASA Release OS Model for Weather and Climate Applications(newsroom.ibm.com)
18 points by osanseviero 2y ago|2 comments
8.
by osanseviero 2y ago
Yes, there are a few dozen full open source models (license, code, data, models)
9.
by osanseviero 2y ago
Hi all! I'm Omar from Hugging Face. Happy to answer any questions you might have about Hugging Face in general, llamas, and open ML!
10.
State of Open AI – July Edition(docs.google.com)
2 points by osanseviero 2y ago|0 comments
11.
Brew Install Llama.cpp(twitter.com)
1 points by osanseviero 2y ago|0 comments
12.
Zephyr 141B, a Mixtral 8x22B fine-tune, is now available in Hugging Chat(huggingface.co)
30 points by osanseviero 2y ago|12 comments
13.
by osanseviero 2y ago
Zephyr 141B is a Mixtral 8x22B fine-tune. Here are some interesting details - Base model: Mixtral 8x22B, 8 experts, 141B total params, 35B activated params - Fine-tuned with ORPO, a new alignment algorithm with no SFT step (hence much faste
14.
by osanseviero 3y ago
The model is also at https://huggingface.co/xai-org
15.
European Space Agency open-sources largest Sentinel-2 dataset(twitter.com)
4 points by osanseviero 3y ago|1 comments
16.
by osanseviero 3y ago
The dataset has - 2 million patches - 1068x1068 pixel patches - 2.5 trillion pixels Read more in https://huggingface.co/posts/aliFrancis/293058125194160
17.
The Stack v2 – dataset with 900B tokens of code(huggingface.co)
3 points by osanseviero 3y ago|0 comments
18.
Introduction to Matryoshka Embeddings(huggingface.co)
1 points by osanseviero 3y ago|0 comments
19.
InternLM – SOTA OS 7B and 20B model with 200K context length(twitter.com)
3 points by osanseviero 3y ago|1 comments
20.
The Llama Hitchiking Guide to Local LLMs(osanseviero.github.io)
3 points by osanseviero 3y ago|0 comments
21.
Everything you wanted to know about sentence embeddings (and maybe a bit more)(osanseviero.github.io)
2 points by osanseviero 3y ago|0 comments
22.
by osanseviero 3y ago
Hey @godelski! Author of the blog post here. I really appreciate you taking the time to provide all this feedback. This feedback + additional resources are extremely useful. I agree that the subtitle is not as accurate as it could be. I&#x2
23.
by osanseviero 3y ago
Yes, you're correct. I tried to connect a common training problem (gradient explosion and vanishing gradient) with the issue of softmax being sensitive to large values. I agree it's misleading/inaccurate, so will rewrite that
24.
by osanseviero 3y ago
Just math, and not even that fancy. Let's say you want to predict if you'll pass an exam based on how many hours you studied (x1) and how many exercises you did (x2). A neuron will learn a weight for each variable (w1 and w2). If
25.
Random Transformer – Understand how transformers work by demystifying their math(osanseviero.github.io)
3 points by osanseviero 3y ago|0 comments
26.
StarVector: Generating Scalable Vector Graphics Code from Images(huggingface.co)
3 points by osanseviero 3y ago|0 comments
27.
by osanseviero 3y ago
Base model, not fine-tuned
28.
by osanseviero 3y ago
This blog post might be interesting - https://huggingface.co/blog/moe MoEs are especially useful for much faster pre-training. During inference, the model will be fast but still require a very high amount of VRAM. MoEs
29.
Mixture of Experts Explained(huggingface.co)
29 points by osanseviero 3y ago|2 comments
30.
by osanseviero 3y ago
Model at https://huggingface.co/meta-llama/LlamaGuard-7b Run in free Google Colab https://colab.research.google.com/drive/16s0tlCSEDtczjPzdIK3...

More ›