iAccount based inUnited States
About this account
- Account based in
- United States
- Connected via
- Web
Account-level information from X, not a live location or the device used for a specific post.
On-demand GPUs, managed clusters to develop, train and deploy AI in one place. : https://nitter.cf/t.co/a69wnEBpKH
- Tweets5.2K
- Following89
- Followers47.6K
- Likes3.8K
ALT This image is a presentation slide for "Meta Muse Glimmer (30B)," a vision perception encoder model. The slide features a detailed diagram of the model architecture, which includes multiple "Post MLP layers," "Linear layers," "ROPE," and embeddings. It highlights a vocabulary size of 32K and provides specifications like a hidden dimension of 6144. Supported context length is 8192 tokens. On the right, there are performance graphs comparing "DGX Spark + Ollama" in prompt processing, token generation, and memory usage, with lower values indicating better performance.