Log inSign up
Log inSign up
vik
20.6K posts
vik profile banner
@vikhyatk

vik

@vikhyatk
teaching computers how to see @moondreamai
vikhyat.net
Joined November 2008
919 Following
29.4K Followers
RepliesRepliesRepostsRepostsMediaMediaArticlesArticles

Log in or sign up for X

See what’s happening and join the conversation

Continue with phone
or
Log in with username or email
Terms·Privacy·Cookies·Accessibility·Ads Info·ÂĐ 2026 X Corp.
  • Pinned
    @vikhyatk
    vik
    @vikhyatk
    Sep 25
    Photon can now speak! Adding support for Qwen3-TTS (0.6B and 1.7B) and Kokoro-82M, with streaming output. First audible audio in under 85 ms at p95, 56–70% sooner than vLLM-Omni on H100 and B200.
    16
  • @vikhyatk
    vik
    @vikhyatk
    10h
    After a few days of using a fast GLM-5.3 endpoint for coding... codex feels unbearably slow
    12
  • @vikhyatk
    vik
    @vikhyatk
    Oct 6
    there is a gpu shortage right now. if your mfu is 9% the government should intervene and redistribute your compute
    15
  • @vikhyatk
    vik
    @vikhyatk
    Oct 6
    am i reading this right? they stole the model weights from one startup and the inference engine from another? lol
    @KanuGulati
    Kanu Gulati | Partner @ Khosla Ventures
    @KanuGulati
    Oct 6
    Hey @ConwayResearch, the fair thing would have been: add @PrismML’s NOTICE file to your releases, tag Bonsai 2 and Qwen3.8 as base models, and credit @inco_ai for Splash, the engine behind @underdogdotai’s speed numbers. tldr; give credit where it's due
    Quote
    @grok
    Grok
    SpaceXAI
    @grok
    Oct 5
    Underdog’s Ternary is a 27B ternary-weight model from Conway Research (Underdog) for efficient private on-device inference. It derives from PrismML’s open-source Ternary Bonsai 2 (near-lossless ternary quantization of Qwen3.8 27B), with post-training applied. Model card and
    13
  • @vikhyatk
    vik
    @vikhyatk
    Oct 4
    Exciting news -- Moondream Parakeet Redux and Ultra are both now supported in parakeet.cpp! github.com/mudler/parakeeâ€Ķ
    @mudler_it
    Ettore Di Giacinto
    @mudler_it
    Oct 4
    parakeet-redux is a really interesting beast. You can extract the VAD head and then, you discover that has better recall than Silero. Despite the footprint compared to it, offline might have other uses. @vikhyatk cooked hard here, congrats for the release, and thanks for working
    8