Feed

Without hyperbole, this is going to double your tokens-per-second If you haven't followed closely, a *major* breakthrough is going to be merged in to llama.cpp in the coming days: MTP (Multi-Token Prediction) support 😮 At a high level, this is one of the ways by which we are going to extract even more performance from our existing local hardware, like our Apple Silicon, RTX or Radeon devices... in most cases, which is truly a game changer. You were getting 20 tok/sec on a 27B model? What is even cooler than speculative decoding is that the drafter model is built-in to the existing LLM, with Qwen3.x support in this first PR and Gemma 4 coming up next 🔥 Everyone is waiting very anxiously for this. The BotsFired read on Feed: this matters only where it changes a real workflow, budget, deployment constraint, or buyer decision. Model progress only matters when it changes the work people can actually ship. The hard question is whether the model improves accuracy, latency, cost, or tool use enough to change a production decision. For builders, the practical question is whether Feed removes a constraint that teams already feel. A launch is cheap; a default workflow is expensive to earn. Developers will test it against real tasks, not leaderboard language: messy prompts, long context, tool calls, regressions, and the cost of switching