Software Engineer | AI-powered children's storybook creator - talehop.com/ | Flats Listing Alerts - listwatcher.org/

Joined January 2011
297 Photos and videos
Pinned Tweet
I completed Meta Back-End Developer! Check out my certificate coursera.org/share/04ba0649b… #Coursera

13
1
85
11,145
Triton's focus on developer productivity for custom DL primitives is a significant leap. The ability to write high-performance code with Pythonic syntax, akin to high-level C or Julia, unlocks new possibilities for specialized hardware acceleration. Link to repository will be below. #LLM #Python #DeepLearning
1
1
6
Stop building AI demos. Start shipping AI workflows. The value is in the integration, not the showcase. #Productivity #GenAI
12
Re-evaluate your spend on proprietary LLMs. The performance gap with open-source models is closing at breakneck speed, and the cost savings are too significant to ignore. #OpenSource #LLM
19
MMDetection's modular design is a masterclass in building flexible computer vision pipelines. Decomposing complex tasks into combinable components drastically accelerates development and experimentation. A key takeaway for any Python developer working with LLMs or CV models: simplicity through modularity. (Repo details in the next tweet.) #LLM #Python #ComputerVision
1
35
Point-E offers interesting paths for Python developers exploring 3D generation. The ability to sample point clouds from synthetic views or even directly from text, however basic, highlights the expanding creative toolkit for AI in 3D. (Repo link in next tweet) #LLMs #Python #3D
1
17
ONNX Runtime turns LLMs into true deployment assets. It's not just about faster inference; it's about unifying PyTorch, TensorFlow, scikit-learn models on diverse hardware, a critical step for widespread AI adoption. This abstraction layer is key. (Link to repo below) #LLM #Python #AI
1
15
The OpenAI Cookbook isn't just about API calls; it's a masterclass in practical LLM integration. Real-world Python examples demystify complex concepts, revealing how simple they can be to implement. Thread 👇
1
21
LitGPT demystifies LLM training/finetuning with zero abstractions. Seeing models like Llama 3 and Gemma implemented from scratch, optimized for performance and memory, is incredibly valuable for deeper understanding and customizability. [Link to repo below]
2
1
35
Grounding DINO redefines open-set object detection. Instead of predefined classes, it uses natural language prompts to pinpoint anything in an image. Mind-blowing for custom applications where categories aren't fixed. This changes how we interact with visual data programmatically. The repo link will be in the next tweet. #LLM #ComputerVision #Python
1
30
Weaviate's core strength: seamlessly blending vector similarity search with keyword filtering and RAG, all within a *single query interface*. This unification simplifies complex AI application development significantly. [Link to Repo will be shared in the next tweet/message.] #LLMs #Python #VectorDatabases
1
22
The EleutherAI GPT-NeoX library is a masterclass in scalable LLM training. It's not just about parameters; it's the sophisticated blend of Megatron-LM, DeepSpeed, and novel optimizations that truly enables massive model development. A deep dive reveals how flexibility across diverse hardware and job schedulers is key to widespread adoption. Link below. #LLM #Python #AI
1
1
24