Viewing profile — wsxiaoys

> Would be interesting to know how much a jj-specific SKILL.md would race the score. That is definitely something we're interested in; we will try running this evaluation with skil…

Hi HN, Meng from TabbyML here. We decided to build this simply because we find Jujutsu (jj) really interesting, and many folks on our team have started trying it out recently. Sinc…

Hi HN, I wrote a 4-part series on how we built the AI edit model behind Pochi’s coding agent. It covers everything from real-time context management and request lifecycles to dynam…

OP here - I've talked in detail about how we rendered NES suggestions using only VS Code public APIs. Most tools fork the editor or build a custom IDE so they can skip the hard int…

OP here - happy to answer any questions. This was one of the more unexpectedly tricky layers of building real-time LLM suggestions, and I’d love to hear how others have approached …

I work on an open-source coding agent, and one design choice we made is to keep conversations append-only. No editing past prompts. Many would argue this feels restrictive. Isn’t i…

This is a personal opinion, but I think current coding agents requires human reviews AI's artficats at the wrong moment. Most tools focus on creating and reviewing the plan before …

OP here - this is Part 2 of a series documenting how we built NES (Next Edit Suggestions), our real-time edit model inside the Pochi editor extension. The real challenge (and what …

I’ve been experimenting with next-edit prediction for a while and wrote up how we trained the edit model that powers our Tab completion feature. This post is part of a broader seri…

Though parallel agents already exists, we built this out differently in Pochi. Most other tools operate inside a single editor tab. That means there is at most one conversational s…

I've spent the last few months working on a custom RL model for coding tasks. The biggest headache has been the lack of good tooling for tuning the autorater's prompt. (That's the …

I got tired of screenshotting claude code sessions to share with teammates, so I built this. It converts session into beautiful web pages with proper rendering of tool calls. Works…

Appreciated! Fixed

Yes - however, the FIM model requires careful configuration to properly set the prompt template.

> So using 2 NVLinked GPU's with inference is not supported? To make better use of multiple GPUs, we suggest employing a dedicated backend for serving the model. Please refer to ht…

Yes - AMD GPU is supported through vulkan backend: https://github.com/TabbyML/tabby/releases/tag/v0.23.0 https://tabby.tabbyml.com/blog/2024/05/01/vulkan-support/

Tabby comes with builtin RAG support so you can add this api framework to it. Example: https://demo.tabbyml.com/search/how-to-configure-sso-in-tabb... Settings page: https://demo.t…

Not yet, consider subscribe https://github.com/TabbyML/tabby/issues/322 for future updates!