Heather Cooper’s Post

Generated this with just a single prompt in seconds using the new Flux model.

What are DeepSeek-R1 distilled models? The DeepSeek-R1 distilled models are smaller, more efficient versions of the larger DeepSeek-R1 model, created through a process called distillation. Distillation involves transferring the knowledge and reasoning capabilities of a larger, more powerful model (in this case, DeepSeek-R1) into smaller models. This allows the smaller models to achieve competitive performance on reasoning tasks while being more computationally efficient and easier to deploy✅

Excited to announce RepairBench - a leaderboard of frontier models for program repair! https://lnkd.in/d4q8bnqN In RepairBench, models are benchmarked on fixing 574 real-world bugs that human developers had to address. As new models are released, RepairBench will provide valuable insights into their performance on this critical task. Claude 3.5 Sonnet is leading with a plausible@1 score of 39.1%, which is +7.4% better than GPT-4o. The best open-source model is Llama3.1 405B, with 27%. However, o1-preview's partial evaluation shows promising results, with a +6% improvement over Claude 3.5 Sonnet. We've also open-sourced everything: - Data: https://lnkd.in/d9qTJpUa - Code: https://lnkd.in/dkN7pn8d For more details, check out our methodology in the technical report: https://lnkd.in/dGWAZGbB

Advanced Sorting Techniques - External sorting for large datasets: - K-way merge implementation - Minimizing disk I/O operations - Handling datasets larger than RAM - Radix Sort and its variations: - LSD vs MSD Radix Sort - Handling strings and floating-point numbers - Parallel Radix Sort implementation - Counting Sort optimizations: - Handling negative numbers - In-place counting sort - Application in Radix Sort - Bucket Sort and its applications: - Choosing optimal bucket size - Handling non-uniform distributions - Geo-spatial data sorting - Hybrid sorting algorithms: - Introsort (QuickSort + HeapSort) - Timsort (Insertion Sort + MergeSort) - Analysis and comparison with standard algorithms What's your go-to sorting algorithm for different scenarios? Any custom sorting solutions you've developed?

"Sorting" , We all related with this term , it's a process to ordering the data element of an array or a list in increasing or decreasing order . There have different types of sorting process , one of them is Radix Sort . Radix sort is a non comparative sorting algorithm that sorts integers by processing their individual digit. It's a very popular sorting algorithm to sort large numbers and every digits sort by using a stable sorting algorithm like counting sort. The time complexity of this algorithm is O(N*K) and Space Complexity is O(N+K) , where K is the no of digits of the large no in the array and N is the linear complexity of the counting sort. It's a very simple algorithm , you can go through this algorithm 👇 https://lnkd.in/gC_iQk2C

First time testing character-based static hand gestures with DenseNet201 using conv3_block9_2_conv! The model identifies static hand gestures by marking these features. ML is interesting. Though before the successful plot, got almost 30 errors of several types 😅

As i had free credits on some SOTA's "closed source" video models, I made a little comparaison without cherry picking anything, just one go for each. The framerate and generated length isn't exactly the same for all so some cut before the end. Top left : Minimax. The car remains in frame all along and it looks really good. Top right : CogVideoX2B, an open source model that runs under 6GB of VRAM. It doesn't do miracles, but in some cases it can give really good results, and as you can run it locally for free you care less if you have to make 10 tries to get a good one :) Bottom left : Runway Gen3. It crops the source, and it looks really slow motion. Bottom right : Kling. The car leaves the frame after a moment, but I find it is the one that feels the most natural : you can see the suspensions working, and when it leaves the frame it feels like it's accelerating. If you're wondering where is Pika, I'm still waiting in queue for now like 5 hours lol, and the credits I have only gives me access to 1.5 so I don't think it will be interesting to put it here.

The Chandy-Lamport algorithm is one of the pioneering jobs that shows decentralized ways to capture the snapshot. Snapshot is useful when you want to implement checkpoints and any stable property detection(deadlock detection) Applying that, however, will involve what this algorithm assumes regarding delivery and failure. https://lnkd.in/gVdnY4Fk