Found bug that I've been working on all week. Required leveling up conceptual understanding of a particular area of the stack, building new observability tooling, and running many iterative experiments to isolate the issue. Incredible feeling now that it's fixed.
COMPUTING
-
FX Tool Enables Model Portability Across IR Frameworks
By
–
i heard someone built this thing called FX that lets you move models from one IR to another. the rest is an easy exercise left to the reader
-
Retrieval Systems Fetch Parent Documents for Enhanced Context
By
–
Usually retrieval system then fetch each chunk individually. The idea behind this is you fetch the parent document they come from (for more context)
-
Python ecosystem shifting to Rust over C++ for performance
By
–
Or more like a competitor to Rust since the Python ecosystem is moving to that instead of C++ (eg see polars and some others)? And for GPU & AI stuff a competitor to Triton/CUDA?
-
BlackBerry Scales Cybersecurity Services Using Lakehouse Technology
By
–
Detecting #cyberattacks through large data sets — what like it’s hard? Not with Databricks Discover why @BlackBerry chose #Lakehouse over other cloud data warehouse vendors to help globally scale its #cybersecurity services
-
Analytics Cloud Roadshow San Francisco September 14
By
–
The #AnalyticsCloudRoadshow is coming to San Francisco September 14! Join Alteryx, @databricks and @PwC for an afternoon of fun, learning and expert discussions. See how organizations use the power of the cloud for the fastest insights. Limited spots: https://
ow.ly/CMYv50PEa1U -
Dendrites: A Novel Approach to Computer Chip Design
By
–
Could dendrites, the spindly protrusions that neurons use to detect signals, offer a novel way of thinking about computer chips?
-
FT-GPT-3.5 Latency Reduced 4-5x Faster Inference Speed
By
–
We’ve reduced the model latency by 4-5x, serving results on average in 0.65 seconds instead of 3.15 seconds (FT-GPT-3.5 compared to GPT-4). You may notice the speedup when Copilot prompts you for user input. Every second counts, and we’re here to make them all productive. pic.twitter.com/dGTu8aYtXw
— Perplexity (@perplexity_ai) 25 août 2023We’ve reduced the model latency by 4-5x, serving results on average in 0.65 seconds instead of 3.15 seconds (FT-GPT-3.5 compared to GPT-4). You may notice the speedup when Copilot prompts you for user input. Every second counts, and we’re here to make them all productive.
-
Industrial Metaverse Revolutionizes Manufacturing with Digital Twins
By
–
Discover how the #industrial #metaverse revolutionizes #manufacturing with astonishing capabilities like improving real-world actions with synthetic data, simulating entire factories and individual products with #digitaltwins, and #immersive #VR. https://
forbes.com/sites/bernardm
arr/2023/08/25/the-future-of-factories-3-ways-to-navigate-the-industrial-metaverse/
… -
PyTorch Python overhead and realistic performance gains discussion
By
–
2/2
PyTorch's Python overhead is usually quoted to be at ~10% (compared to PyTorch's C++ API). I think for PyTorch users (or CUDA-dependent packages), the speed advantage is probably more going to be in the single digit range, not 35,000x.
Still super exciting though!