Dev Jhawar, writing https://devjhawar.com/ Notes on LLM internals, agents and AI infrastructure. en Bigger AI models aren't always better: how a 70B model beat a 280B one by reading more https://dev.to/dev_jhawar/bigger-ai-models-arent-always-better-how-a-70b-model-beat-a-280b-one-by-reading-more-2jmh https://dev.to/dev_jhawar/bigger-ai-models-arent-always-better-how-a-70b-model-beat-a-280b-one-by-reading-more-2jmh Tue, 20 Oct 2026 19:02:31 GMT For a few years, the way to build a better AI model was simple. Make it bigger. Then in 2022... ai llm machinelearning What it actually costs to train an LLM: GPUs, days and dollars, worked out step by step https://dev.to/dev_jhawar/what-it-actually-costs-to-train-an-llm-gpus-days-and-dollars-worked-out-step-by-step-1g5j https://dev.to/dev_jhawar/what-it-actually-costs-to-train-an-llm-gpus-days-and-dollars-worked-out-step-by-step-1g5j Tue, 20 Oct 2026 17:47:59 GMT You've seen the headlines. "Training a big model costs tens of millions of dollars." Okay, but why... ai llm python Why Meta trained a small 8B model on 15 trillion tokens, and why that's cheaper than it sounds https://dev.to/dev_jhawar/why-meta-trained-a-small-8b-model-on-15-trillion-tokens-and-why-thats-cheaper-than-it-sounds-1oep https://dev.to/dev_jhawar/why-meta-trained-a-small-8b-model-on-15-trillion-tokens-and-why-thats-cheaper-than-it-sounds-1oep Tue, 20 Oct 2026 17:47:35 GMT There's a rule of thumb from a famous paper: for the best result per dollar of training, feed a model... ai llm python The KV cache is eating your GPU. MQA and GQA are how models fight back https://dev.to/dev_jhawar/the-kv-cache-is-eating-your-gpu-mqa-and-gqa-are-how-models-fight-back-3809 https://dev.to/dev_jhawar/the-kv-cache-is-eating-your-gpu-mqa-and-gqa-are-how-models-fight-back-3809 Tue, 20 Oct 2026 17:16:12 GMT Quick recap from the last post: the KV cache saves you from recomputing old tokens, and it's the... ai python llm machinelearning Tokenization is why your agent bill looks weird https://dev.to/dev_jhawar/tokenization-is-why-your-agent-bill-looks-weird-4n47 https://dev.to/dev_jhawar/tokenization-is-why-your-agent-bill-looks-weird-4n47 Tue, 20 Oct 2026 17:15:47 GMT Models don't read text. They read integers. Between your string and the model there's a tokenizer,... ai python llm machinelearning Why your agent gets slower and pricier with every tool call https://dev.to/dev_jhawar/why-your-agent-gets-slower-and-pricier-with-every-tool-call-1l6n https://dev.to/dev_jhawar/why-your-agent-gets-slower-and-pricier-with-every-tool-call-1l6n Tue, 20 Oct 2026 17:09:05 GMT Okay so here's something that bugged me for a while. You build an agent. Step 1 feels snappy. By... ai python llm machinelearning Running 100 agents on a free tier (Groq 30 RPM, Tavily credits) https://dev.to/dev_jhawar/running-100-agents-on-a-free-tier-groq-30-rpm-tavily-credits-pgi https://dev.to/dev_jhawar/running-100-agents-on-a-free-tier-groq-30-rpm-tavily-credits-pgi Tue, 20 Oct 2026 16:27:27 GMT GridMind is a spreadsheet where every cell is its own research agent. You type "Stripe" in a row,... ai python backend opensource