.jpeg)
Multi-Instance GPUs on Runpod: Stop Paying for Compute You Don't Need
With MIG, we can partition RTX 6000 Pro cards into isolated 24 GB instances. Here's when it makes sense for your workloads.
All
Welcome to the Runpod Roundup for the week ending August 05, 2023. In this issue, we'll be discussing the newest advancements in AI models over the past week, with a focus on new offerings that you can run in a Runpod instance right this second. In this issue, we'll be looking at audio generation and some new LLM options that have been released.
Meta AI Releases Audiocraft For AI Music and Sound Effects
Believe it or not, there is actually a very litigious nature around the use of stock sound effects, one of the most prominent recent lawsuits being over the infamous Roblox "oof" sound, which was about a fifth of a second in length. DMCA takedowns over sound effect usage on YouTube are also common. It's actually also somewhat understandable, as there can be considerable expense, time, and effort involved in Foley sound creation. Meta has some samples of their model's output on their website, such as whistling, sirens, trains, and so on. A full tutorial will be on the way soon, but for now you can grab the AudioGen and MusicGen models from Meta's HuggingFace.
Alibaba Cloud Open Sources Two Chinese Open Source LLM Models
Alibaba Cloud has released two 7b Chinese-language LLM models (Qwen and Qwen-7b), which are free for commercial use for companies with fewer than 100 monthly active users. These models are related to their their Tongyi Qianwen product released earlier this year. If you'd like to download and install these models, you can grab them from their Huggingface.
Salesforce Releases DialogStudio to Help Leverage Conversational AI
Building conversational AI has always been a challenge due to the lack of standardization across datasets, due to the sheer unconstrained space that human conversations can wander into. DialogStudio addresses this by aggregating datasets over a variety of different domains such as question-answering, dialogue, and recommendation, with a focus on retaining the original information on each dataset. If you'd like to try out these datasets, you can also get them from the Salesforce Huggingface.
Questions?
Feel free to reach out to Runpod directly if you have any questions about these latest developments, and we'll see what we can do for you!
