5 articles

📰 Dev.to · Mayank Ketkar

Articles from Dev.to · Mayank Ketkar · 5 articles · Updated every 3 hours · View all reads

All ⚡ AI Lessons (9011) ArXiv cs.AIDev.to · FORUM WEBForbes InnovationOpenAI NewsDev.to AIHugging Face Blog
One Week in Ray: 21 Bugs Between Us and a Production ML Pipeline
Dev.to · Mayank Ketkar 1mo ago
One Week in Ray: 21 Bugs Between Us and a Production ML Pipeline
The pipeline finished. All four checkpoints evaluated. Metrics written to cloud storage. Zero errors...
Read article → + Playlist
The Ghost in the Batch: How vLLM Silently Switches Algorithms
Dev.to · Mayank Ketkar 1mo ago
The Ghost in the Batch: How vLLM Silently Switches Algorithms
You run Qwen3-VL on a single prompt. You record the output logprobs to full precision. Then you run...
Read article → + Playlist
Two Ways to Move Tensors Without Stopping: Inside vLLM's Async GPU Transfer Patterns
Dev.to · Mayank Ketkar 1mo ago
Two Ways to Move Tensors Without Stopping: Inside vLLM's Async GPU Transfer Patterns
A single torch.cuda.synchronize() in the wrong place can erase every optimization you spent weeks...
Read article → + Playlist
Your Ray Data Pipeline Works at 10K Samples. Here's Why It Crashes at 1M.
Dev.to · Mayank Ketkar 2mo ago
Your Ray Data Pipeline Works at 10K Samples. Here's Why It Crashes at 1M.
There's a moment every ML infrastructure engineer knows: the evaluation pipeline that worked...
Read article → + Playlist
Compiling the Vision Encoder: Squeezing 3% More Throughput from Qwen3-VL on Hopper GPUs
Dev.to · Mayank Ketkar 2mo ago
Compiling the Vision Encoder: Squeezing 3% More Throughput from Qwen3-VL on Hopper GPUs
When you run a vision-language model through vLLM, the framework does something clever: it compiles...
Read article → + Playlist