The parallel tool-calling subagent cut token usage up to 50%.
the model got pulled by microsoft ☹️
good stuff from microsoft: 4B model just to explore code bases, cutting token costs by 10-50% (!!!) while the performance of the big model stays the same :)
Many users criticized Microsoft for pulling its 4B code-exploration model from Hugging Face, viewing the move as part of a pattern of withholding competitive releases while also faulting the model's hallucinations and deployment issues.
No Digg Deeper questions have been answered for this story yet.
@xeophon Classic, just like VibeVoice.
Microsoft hates releasing anything actually above the curve lmao
@xeophon Odd, but there are mirrors up https://huggingface.co/ShaunGves/FastContext-1.0-4B-SFT
and lots of quants https://huggingface.co/models?sort=trending&search=FastContext-1.0-4B-SFT
@xeophon the wizardlm flashbacks are real
@xeophon https://huggingface.co/papers/2606.14066 they have a paper about this, their github link for "code and data" is also 404 now
@xeophon still up here https://www.modelscope.cn/models/microsoft/FastContext-1.0-4B-SFT/summary
@xeophon AGAIN!?!?!?! rip wizardlm
the model got pulled by microsoft ☹️
@xeophon I’ve seen some people report that they couldn’t reproduce the claimed results, so I wonder if it was pulled because it was subpar
@xeophon It didn’t work. It hallucinated everything. It required docker deployment because that’s how it was trained.
@xeophon damn it
@xeophon wtf
@xeophon wait did u have it dld?
@xeophon not really cool of them
@xeophon USG recognized the security threat
@din0s_ @xeophon that's a name i haven't heard in a long time. Legendary model
@xeophon Bruh wtf
@xeophon why does this happen so often with microslop?
@xeophon whos got the torrent?
@xeophon Wah pak @anvie
@xeophon Yep