Top Tweets for #TensorFold
@VeyraWithAI @LocalsOnlyAI @MiaAI_lab @ashxhart On a single SGX spark, #tensorfold beats vllm...
And, the newest release is even better!
And, Hermes runs on qwen3.8-flash-next, and can upgrade itself π€―
@MiaAI_lab @ashxhart
@MiaAI_lab What's amazing is that that hasn't been enabled by cheaper faster hardware, the opposite is in fact happening, it's instead been enabled by people like @ashxhart and projects like #tensorfold
On bosse sur le portage XPU et le support de la carte Arc Pro B70 sur #TensorFold version Zig, avec pas mal d'optimisations disponibles sur la table, notamment au niveau XMX et OpenCL. Objectif : 2400 tok/s en prefill et 100+ en decode @ashxhart @fragtalife @IntelFrance #LocalAI
@mfranz_on Itβs worth it. #TensorFold on this thing is rapid. I have two sparks and one of these, and for inference this is the machine. 88tps serving GLM-5.3 Flash. Qwen Next Flash is even faster. Feels like cloud.
Updated V1.4 new NAS Model Cassette Switching Dashboard with metrics out now. #NAS #Qwen #TensorFold #DGX #Spark #DGXSpark #AI #openSource
https://t.co/84Go9lqxks
New TensorFold Studio with Image, Video, MIDI to music generation, script stitching video, voice cloning coming soon!! @ashxhart @MiaAI_lab @machinegenie @ultra_instinct #TensorFold #Qwen #Al
https://t.co/aMwm5tSo9D
New TensorFold Studio with Image, Video, MIDI to music generation, script stitching video, voice cloning coming soon!! @ashxhart @MiaAI_lab @machinegenie @u1tra_instinct #TensorFold #Qwen #AI
Now I have an app on my #SteamFrame from which I can control #Homeassistant stuff.
And the neat thing is it auto detects the entities and puts it on the AR overlay :D.
All done with #LocalLLM on #DGXSpark and #Tensorfold
New TensorFold Video and Image Rendering Studio coming soon!! π₯π₯π₯ #TensorFold #Qwen #ComfyUI #DGXSpark #DGX #Spark #MiniMaxH3 #MiniMax #OpenSouce #AI
Iβm chucking as it was only this morning Ash casually dropped this response. A few hours later he announces the Living Weights concept and before the day is over heβs released it π€© Incredible work to all involved. π€π
#TensorFold
TensorFold 1.0.3 lands with a few cool things.
- Living Weights (beta with Nemotron on Metal) and coming to CUDA in 1.0.4
- Nvidia Ampere (beta) for Nemotron, and native Qwen3.8-27B with DFlash2
- Conversation prefix caching
- Various bugs fixed and improvements
Twelve contributor PRs helped make this happen, so a massive thank you to @niklaslenz_ai and GitHub contributors akol1, jschmied, chaog992, AdrianBinDC and BobClawblaw, and everyone testing, reporting bugs and helping us improve TensorFold.
FYI, Living Weights is in very early beta and has sharp edges, but I think Iβm on to something.
Iβd love community feedback to make it better: https://t.co/pcq8P57cvo
@plotarmordev @petruspennanen @MiaAI_lab @volatilemarkts
https://t.co/H2dBxCSaUV
Turns out Qwen3.8-Flash Next and Deepseek v4 on #Tensorfold are quite good to make @BeatSaber custom levels for the #SteamFrame π.
They are just a bit pushy when they want me to test it and give feedback in time so they can iterate π
Updated new cassette model loader GitHub revision coming soon!! #ollama #vllm #SGLang #TensorFold #EXL3 #llamacpp #Media @MiaAI_lab @ashxhart @machinegenie
I saw people trying to cram Qwen 3.6 35B A3B onto an 8GB card and wondered if I could get it working with #tensorfold ... Suffice it to say not only does it work, but I actually get a usable TPS on my laptop π€― @ashxhart bringing local AI to the masses
@MiaAI_lab Our TensorFold fork:
DeepSeek-V4.1-Flash TP=2 on 2Γ DGX Spark (GB10), EXL3 2.9bpw,
TG (jaybench): code 91 Β· prose 48 Β· structured 132 tok/s
Cold PP: 32K 2.1K Β· 128K 2.1K tok/s
KV: 5.15M tokens (FP4)
Exact spec decode, over HTTP:
https://t.co/rnBXMusdS9
#tensorfold #dualspark
I'm using a priority gate with #tensorfold so my single dgx running Qwen 3.8 flash next can be working on multiple tasks simultaneously even though it's memory bound. So that gives me a bigger KV cache with the same concurrency
#tensorfold is allowing us to run some really interesting experiments on memory - it's been a huge unlock!!
We serve a 125B-param MoE (6B active) on one office GPU: ~2.4k tok/s prefill, 47-100 tok/s decode with MTP drafts, 262k context. Frontier-class inference is going on-prem, and it is fast! Thanks to @ashxhart and #tensorfold
@andersjw_ 's recipe for Kolibri-1 is pretty impressive!
It does have the worker at 96% thing that happened on some past #TensorFold recipes but the speed is preety good! Nice MoE Recipe for this model! https://t.co/qEEw5iIJm5
#TensorFold γmainγγ©γ³γγzigηγ«ε€γγ£γ¦γγζ©
Wow. #TensorFold makes my local model run FAST!
Thanks @ashxhart and @MiaAI_lab for the GLM 5.3 Flash uncensored recipe that uses it!
Last Seen Hashtags on Sotwe
brosis(((((((((())))))))))* filter:videos since:2026-09-15
Seen from Netherlands
anxiouspanda()+filter:native_video
Seen from United States
kazumi
murderdrones #cyn #nsfw
Seen from Argentina
brosis()*(filter:videos)*
Seen from United States
girlmasturbating
Seen from Turkey
stripping
Seen from Turkey
SmilingCrittersOC #NSFW
Seen from Ukraine
candid
Seen from United States
teeNager
Seen from Belgium
Trends for you
Most Popular Users
Elon Musk 
@elonmusk
241.5M followers
Barack Obama
@barackobama
118.9M followers
Cristiano Ronaldo
@cristiano
114.6M followers
Donald J. Trump
@realdonaldtrump
111.8M followers
Narendra Modi
@narendramodi
107.1M followers
Rihanna
@rihanna
98.7M followers
NASA
@nasa
92.3M followers
Justin Bieber
@justinbieber
91.8M followers
KATY PERRY
@katyperry
90.1M followers
Taylor Swift
@taylorswift13
84M followers
Lady Gaga
@ladygaga
75.5M followers
Virat Kohli
@imvkohli
73.5M followers
Kim Kardashian
@kimkardashian
70.9M followers
YouTube
@youtube
68.8M followers
Neymar Jr
@neymarjr
66.6M followers
Bill Gates
@billgates
65.3M followers
Selena Gomez
@selenagomez
63.1M followers
The Ellen Show
@theellenshow
62.2M followers
CNN
@cnn
61.8M followers
X
@x
60.7M followers