Chuan L

• — ☻

[Chuan L @chuan_l]

— Some experimental #diffusion work ,
Using " #Tobii #eye #tracking " data to generate photoreal portraits. Part o' th' " #Imaginary #Friends " exhibition with Danielle Freakley at " #WA #Museum " mid august 2023 :

[Chuan L @chuan_l]

@Dillon_Valdez @michaeljburry @Fortune @Bloomberg Pretty soon human memory will be more valuable than vram. Do you remember “ FSD by end of 2017 “ “ 1 million robotaxis by 2022 “ , and Open AI “ AGI by December 24 “ ..

[Chuan L @chuan_l]

@jasteinerman 1. The " whatsapp " calls were received. He was just unable to pick up the call. Likely an os bug with priorities and selection ..

2. The guy in the kitchen also had no issues getting a response from the back end. Though gave the incorrect canned response out of order ..

[Chuan L @chuan_l]

@NVIDIAGeForce “ GeForce RTX 5090 “

[Unsloth AI @UnslothAI]

Introducing 1.58bit DeepSeek-R1 GGUFs! 🐋

DeepSeek-R1 can now run in 1.58-bit, while being fully functional. We shrank the 671B parameter model from 720GB to just 131GB - a 80% size reduction.

Naively quantizing all layers breaks the model entirely, causing endless loops & gibberish outputs. Our dynamic quants solve this.

The 1.58-bit quant fits in 160GB VRAM (2x H100 80GB) for fast inference at ~140 tokens/sec.

By studying DeepSeek-R1’s architecture, we selectively quantized certain layers to higher bits (like 4-bit), and leave most MoE layers to 1.5-bit.

Benchmarks + Blog: https://t.co/99I4VOQPYC
Dynamic GGUFs (131GB–212GB) on Hugging Face: https://t.co/0cZkamD9iL

[Chuan L @chuan_l]

[ #MIFF2024 ] #Jia #Zhangke " Caught By Th' Tides " — More an epilogue made from : 2 - parts " #Still #Life " mixed with 1 - part " #Platform " / sift _ d through a hazy gauze / #Hu #Bo feels more relevant ..

[Chuan L @chuan_l]

[ #MIFF2024 ] #Frederick #Wiseman " Menus Plaisirs Les Troisgros " — Masterful 4 - hour #documentary into th' care & #feeding o' investment bankers. Holistic an' complete with #terroir tantrums ..

[Chuan L @chuan_l]

[ #MIFF2024 ] #Cristóbal #León " The Hyperboreans " — Th' re : invention o' #Miguel #Serrano / santiago via satan / no " #wolf #house " but th' doors are shut + mind still #blown ..

[Chuan L @chuan_l]

[ #MIFF2024 ] #N #Santos #Arias " Pepe " — Brilliant multi layer _d #filmmaking re : Pablo Escobar's #cocaine #hippos. Like " Big River Man " but from th' #perspective o' th' river ..

[Chuan L @chuan_l]

[ #MIFF2024 ] #Harmony #Korine " Aggro Dr1ft " — Going off piste ! Pure cinema + images from : a #FLIR #T1020 thermal camera. Stunning rothko - like brilliance with his usual #dumb #shit ..

[Chuan L @chuan_l]

[ #MIFF2024 ] #Miguel #Gomes " Grand Tour " — Inane muddl _d junket o' #oriental fetish. " Tabu " had real #murk and romance. This has pointless context switching to : colour ..

[Chuan L @chuan_l]

@xiaolonw @OpenAI Its most likely that " sora " is using depth estimation to help with 3d coherence like " control nets " over successive frames. That leads to being able to re - construct + remap these scenes ..

[Chuan L @chuan_l]

@tonyzzhao @zipengfu @chelseabfinn Hm methinks a - lot more people are " going to love machines in the year 2029 ad " , really nice work + thank you for #open #source !

[Chuan L @chuan_l]

@ketkijadhav1705 This is awesome !

[Chuan L @chuan_l]

#Meta " #Reality #Labs " codec avatars 2.x demo from th' Lex Fridman 398 podcast. Up close an' personal with #realtime #3d facial reconstruction + #encode using : eye - track _d data , spatial audio & relighting for #telepresence :

[Chuan L @chuan_l]

#Meta " Emu " #image #diffusion from 1.1 Bn pre - train _d pairs + 2k " quality " fine - tune : https://t.co/YzWpmVz3uD / 16 - channel auto encoder reduces loss due to : #latent #compression ..

[Chuan L @chuan_l]

#Meta " Quest 3 " has much better 2 x 4 MP RGB #pass #through cameras ! 8 Gb ram with built - in depth sensor / mo' overhead & new #pancake #lens at 97 ° ..

[Chuan L @chuan_l]

#Reality #Labs — Yaser Sheikh on : " codec avatars " , re : creating tele - presence , affordances for #human #interaction & th' " concert " demo : https://t.co/Hg4rn4ockB 59 mins ..

[Jonathon Luiten @JonathonLuiten]

Dynamic 3D Gaussians: Tracking by Persistent Dynamic View Synthesis

https://t.co/rTiKZhEjXM

We model the world as a set of 3D Gaussians that move & rotate over time. This extends Gaussian Splatting to dynamic scenes, with accurate novel-view synthesis and dense 3D trajectories.

[Chuan L @chuan_l]

#Argentum #Studio — #Unreal 5.1 / 5.2 Houdini #OpenVDB rendering tool / replaces default #volumetric #shaders + enables self - shadows , pyro " heat " attributes ..

[Chuan L @chuan_l]

#Sony #Japan — Demos new #haptic #vr #controller prototype in : a versatile 2 - part design / looks expensive ! Though happy to see it : https://t.co/YgNbnC1WhD ..