batuhan taskaya @isidentical Twitter profile

Pinned Tweet

batuhan taskaya

@isidentical

5 months

dating profile update: works at the hottest ai startup in town. only wants to talk about ai inference.

4

67

Last Seen Profiles

@zaiiwarunya

@penyukastw21

@ometvlovers

@yunjeonghui5

@Jironimaux

@jonjon2125

@stwmaniax

@abe_eon

@gustsan85120

@chachalovenunu

@gazetekars

@DrAlokDixit7

@paulohenri_pt

@putu_ayu19

@sovereign_first

@ifnu_q

@stwmaniax

@xh142830

@BritBoysXXX

@StandByRYU

@ometvlovers

@BojezKang75385

@salmaangel666

@SoleilEvans16

@2700chess

@hinatabar

@Panagit21532655

@ometvlovers

@balorsanford

@CJHoweLab

@kyl1546533

@NetaHewitt60628

@OneZero_Hsia

@keiidakamya

@RzAh68

batuhan taskaya

@isidentical

4 months

what should I try?

162

33

725

batuhan taskaya

@isidentical

4 years

Looks like I'm the #5th most active contributor of CPython for the last 6 months 🥳

18

16

412

batuhan taskaya

@isidentical

4 years

Yesterday I've gained commit privileges and promoted to a CPython Core Developer 🎊🎊

25

15

346

batuhan taskaya

@isidentical

2 months

stop complaining that you don't have a job and build something like this. then your inbox will be full with job offers from startups that you love

naklecha

@naklecha

2 months

day 6 & 7 of automating factorio -- i have massive mod updates. i restructured the mods & added a fuck ton of modding for simple atomic actions that an agent could take. 17 actions implemented so far (example video shows placing a furnace, without collisions). we're cooking :)

5

6

232

4

14

303

batuhan taskaya

@isidentical

12 days

Introducing AuraFace v1: Commercially available & open source identity encoder model for next generation one shot personalization.

Introducing AuraFace: Open-Source Face Recognition and Identity Preservation Models

huggingface.co

7

23

210

batuhan taskaya

@isidentical

2 months

Spent the last 1.5 month building this, finally open source. Can't be more excited about the future (this is just v0.1-B)

Introducing AuraFlow v0.1, an Open Exploration of Large Rectified Flow Models

Open-source AI is in jeopardy. As community interest in AI models skyrocketed over the past year, we noticed that development of new open-source foundational models came to a halt. Some even boldly...

blog.fal.ai

14

21

203

batuhan taskaya

@isidentical

2 months

uhm, does anyone have any idea on how to make model "unlearn" some stuff, if by any chance if it learnt stuff that it shouldn't include in its distribution?

33

3

194

batuhan taskaya

@isidentical

2 months

Anyone born before 2000 is like, old. Idk

31

4

175

batuhan taskaya

@isidentical

4 years

Whoooaaa, I became a PSF Fellow Member! Even though my name is misspelled, this is still amazing news! 🎊🎊🎊🎊

Python Software Foundation

@ThePSF

4 years

Python Software Foundation Fellow Members for Q4 2020

0

13

54

18

7

152

batuhan taskaya

@isidentical

4 months

yoooo, new public dataset drop! 5 million moondream2 (rev=2024-05-08) captioned text to image pairs.

isidentical/moondream2-coyo-5M-captions · Datasets at Hugging Face

huggingface.co

5

18

144

batuhan taskaya

@isidentical

9 days

today i turned 21! legal age for many things 🙈

38

0

142

batuhan taskaya

@isidentical

2 months

For people training their own models and wanna not use SD3's commercial licensed VAE, will be releasing our own 16ch one which is comparable in perf!

8

12

125

batuhan taskaya

@isidentical

4 months

@ClementDelangue stability is a better investment

2

0

111

batuhan taskaya

@isidentical

2 months

This got more attention than I thought it would, so here you go people: . A fully commercially licensed 16 channel VAE.

AuraDiffusion/16ch-vae · Hugging Face

huggingface.co

batuhan taskaya

@isidentical

2 months

For people training their own models and wanna not use SD3's commercial licensed VAE, will be releasing our own 16ch one which is comparable in perf!

8

12

125

8

24

108

batuhan taskaya

@isidentical

2 months

Wait a minute, why is GPT4-o mini has the same price as GPT4-o for vision. One claims it is 255 tokens and the other one is 8500 tokens for the same input image. Like wtf?

10

7

103

batuhan taskaya

@isidentical

5 months

Announcing : a generative arena for text-guided open source image generation models. Like Chatbot Arena -- but more fun (because it is images)!

imgsys.org | an image model arena by fal.ai

A generative AI arena where you can test different prompts and pick the results you like the most. Check-out the model rankings and try it yourself!

imgsys.org

8

23

103

batuhan taskaya

@isidentical

1 month

how can i convince smart people to stop writing rust and work on ML performance?

11

0

99

batuhan taskaya

@isidentical

1 month

OK guys, holy shit, gemini-1.5-pro-exp-0801 might be the captioning king (much better than GPT-4o). I need to caption a couple hundred million images @google where do i get the api keys

10

2

100

batuhan taskaya

@isidentical

16 days

😌HAVE BEEN WAITING AN ETERNITY TO ANNOUNCE THIS. FAL ML TEAM GETS EVEN MORE CRACKED. w/ @jfischoff @_yatharthg @Gothos03 @gokayfem @chengzeyi AND now joining us @cloneofsimo .

Simo Ryu

@cloneofsimo

16 days

Whatup chads, I'll be joining @FAL and lead the research effort to develop open source / proprietary models.Ill work on boundaries of research / product, theory / practice and inference / training. I have couple head counts, dm if you wana join my team. Thx

24

5

302

8

3

85

batuhan taskaya

@isidentical

2 months

you might ask why are we launching this now when the model is clearly undertrained instead of waiting for a few month? the reason is why iterate behind the closed doors and have no community feedback. what do we have to hide? let's build this model together. step by step.

Simo Ryu

@cloneofsimo

2 months

@FAL There is a lot more story to tell, but we are just releasing the intermediate checkpoint for now. What you are seeing is actually now what we have, and is merely a beta release!

1

35

7

2

83

batuhan taskaya

@isidentical

1 month

DeepSeek v2, the AGI for the poor. why isn't it a bigger deal?

9

3

78

batuhan taskaya

@isidentical

6 months

this post is definietly not sponsored by @dylan522p

1

8

73

batuhan taskaya

@isidentical

2 months

RELEASE YOUR MODELS AS OPEN SOURCE PEOPLE!!!!!!

10

6

77

batuhan taskaya

@isidentical

3 months

Last month I promised if our tweet got 200 likes, we’d make whisper 20% faster (and we did it 120% faster while halving the price and maintaining WER). Now I am saying I’ll double the performance of whisper v3 if we get 2000 likes. You go figure what will happen :)

batuhan taskaya

@isidentical

3 months

@gblazex @GroqInc @ArtificialAnlys Did you meant to say one of the fastest? Because i can find someone who is even faster :)

2

0

14

3

9

77

batuhan taskaya

@isidentical

2 months

LETS FUCKING GO MAN. LETS FUCKING GO. THIS IS JUST THE START, AURAGAN, AURAFLOW. If no one is releasing models, WE WILL. WHO CARES, it might not beat the closed source SOTA versions, but it will be our own and we'll keep developing it!!!

Simo Ryu

@cloneofsimo

2 months

@FAL @burkaygur So much of the credit goes to @isidentical who did loads of technical works from data to infra management to make this happen. I dont know how to coordinate NFS amongst nodes nor preprocess massive dataset with ray... This model would have not been here without him, and btw this

2

0

45

2

76

batuhan taskaya

@isidentical

13 days

i am sorry to inform you but clip, in fact, does not work. it is a deeply flawed model. sorry to be the one that is telling this

snats

@snats_xyz

13 days

its a miracle that CLIP actually works btw

3

1

43

7

3

73

batuhan taskaya

@isidentical

1 month

Jonathan Fischoff

@jfischoff

1 month

Thanks for coming to my TED Talk. Make your own here:

4

1

33

2

71

batuhan taskaya

@isidentical

2 months

We at @fal are sponsoring PyTorch Conference'2024! We are big users of torch here, and are continuing to push the frontiers of the provided tooling with PT2

6

4

73

batuhan taskaya

@isidentical

7 months

not sped up. this is real time. try it yourself.

5

7

70

batuhan taskaya

@isidentical

13 days

@t3dotgg @t3dotgg did you see BiRefNet? it is a SOTA background removal model, even better than Bria 1.4, and open source. we also host it serverlessly FYI at

Birefnet Background Removal | AI Playground | fal.ai

bilateral reference framework (BiRefNet) for high-resolution dichotomous image segmentation (DIS)

fal.ai

5

4

67

batuhan taskaya

@isidentical

16 days

have you noticed fine tuned flux models at @fal now run 2.5x faster? and 2.5x cheaper? WTF is going on?

FLUX.1 [dev] with LoRAs | AI Playground | fal.ai

Super fast endpoint for the FLUX.1 [dev] model with LoRA support.

fal.ai

5

1

66

batuhan taskaya

@isidentical

2 months

means a lot coming from comfy themself!!!!

8

0

65

batuhan taskaya

@isidentical

12 days

this lora was trained under 2 minutes

5

0

63

batuhan taskaya

@isidentical

4 months

at @fal , we don't care about your fancy titles or ex-companies you worked with. if you are interested in diffusion models, can deliver high quality stuff, at a flash paced environment, in a consistent manner. DM me. we are hiring the best in their class.

5

61

batuhan taskaya

@isidentical

5 months

A real video -> Moondream description -> Elevenlabs TTS -> Deepgram STT -> SDXL Lightning (by @aconchillo and @ai_meta_agent )

2

7

60

batuhan taskaya

@isidentical

2 months

first native 1024x1024 AuraFlow generation!!! (it has been trained only for a couple hours on this so lots to go)

3

0

60

batuhan taskaya

@isidentical

2 months

Top HF models!

2

5

60

batuhan taskaya

@isidentical

2 months

SOMEONE STOP ME

5

59

batuhan taskaya

@isidentical

6 months

Actively hiring an ML inference performance engineer. If you wanna squeeze the last bit of performance from our wide fleet of GPUs, work with nightly torch APIs and build SOTA tooling on top of it to achieve the extreme levels of performance shoot me a DM or an e-mail.

2

18

58

batuhan taskaya

@isidentical

1 month

holy shit, @JuicedataInc + @TigrisData is just crazy fast (1GB/s+, ~10Gbit)

4

3

54

batuhan taskaya

@isidentical

14 days

Use fal workflows to explore different flux finetunes for the same prompt / seed pairs!!! It's so fun and SO FAST (3 images under 3 seconds!!!)

2

3

61

batuhan taskaya

@isidentical

6 months

@inerati it is not a design decision. It is the intuitive: expected behavior of different language constructs. e.g. how else should def fn(a=A()): … would behave? Do we give a shallow copy at every invocation? Or a deep one? Or maybe disallow it (what makes this different than a

11

0

56

batuhan taskaya

@isidentical

4 months

sorry but if your CEO isn't pushing complex math formulas to the app, you are ngmi. 🔥 @burkaygur

2

56

batuhan taskaya

@isidentical

4 years

🎉🎉

The PyPy Project

@pypyproject

4 years

A big welcome to freshly minted PyPy dev @isidentical , who got his first bug fix merged today and already opened a second feature merge request (f-string debugging expressions). Thank you!

3

49

2

0

55

batuhan taskaya

@isidentical

1 month

> be me > find out that gaudi chips are actually amazing for training, comparable perf between A100 and H100 > think intel is so much undervalued (without having any idea about other stuff) > look at the chart and see it went down 50%, think it can't go any lower > buy intel

GURGAVIN

@gurgavin

1 month

INTEL SHARES ARE NOW DOWN NEARLY 25% TODAY IMO INTEL JUST REPORTED THE WORST EARNINGS OF ANY COMPANY THIS EARNINGS SEASON $INTC

55

67

655

8

1

54

batuhan taskaya

@isidentical

28 days

i want an fully open source video model so bad

12

1

54

batuhan taskaya

@isidentical

22 days

Added FLUX.1 pro/dev/schnell and AuraFlow v0.2 to !!! Go play with it and get us some votez

imgsys.org | an image model arena by fal.ai

A generative AI arena where you can test different prompts and pick the results you like the most. Check-out the model rankings and try it yourself!

imgsys.org

6

52

batuhan taskaya

@isidentical

2 months

Best paper award goes to Scaling Rectified Flow Transformers for High Resolution Image Synthesis by Esser & Rombach et al (aka SD3 paper). Well deserved

0

1

54

batuhan taskaya

@isidentical

21 days

time to blow off some steam. was literally heads down for the last ~18 days, working on FLUX related stuff!

5

0

54

batuhan taskaya

@isidentical

1 month

300ms end to end latency for flux schnell. biggest open weights image model, to date. AND WE CAN RUN IT SO FAST. from DAY ZERO

Gorkem Yurtseven

@gorkemyurt

1 month

really really proud of our inference optimization team with this one. it's a big model - 12B params but we are able to run it incredibly fast!

1

8

81

3

0

52

batuhan taskaya

@isidentical

2 years

Fancy tracebacks in PyPy 3.9 (even before CPython 3.11 🤠) @pypyproject w/ @cfbolz #DusseldorfPyPySprints2022

2

5

52

batuhan taskaya

@isidentical

1 month

If meta didnt release llama, do you guys think mistral would have open sourced (still with NC license) the mistral large? Something tells me they never wanted to release a large model but meta pushed their hands

6

0

51

batuhan taskaya

@isidentical

3 months

any comfy UI users who want their minds to be blown?

16

3

50

batuhan taskaya

@isidentical

2 years

Built with WASM in a matter of hours. The entire site is defined in a single Python script (no CSS/JS, all the widgets/interactions/logic is in Python), and hosted entirely on GitHub pages. I wish somebody would have told me that it was really 'this' easy

5

7

50

batuhan taskaya

@isidentical

2 months

nvidia bros, never fell into the H100 PCIE trap. there is a noticable difference in both the raw flops and memory bw. some providers sell 8x H100 NVLinked and you might think oh this must be SXM but no lol. always double check

8

2

50

batuhan taskaya

@isidentical

3 months

really lucky that i can work at a company who sponsored tens of thousands of hours of H100 compute already and will continue to do so in multiples of that going forward! OPEN SOURCE FTW

fal

@FAL

3 months

Announcing fal Research Grants which provides free compute resources to researchers and developers working on cutting-edge open source initiatives. Learn more here:

9

24

125

0

4

49

batuhan taskaya

@isidentical

25 days

People sometimes compare companies by team sizes and think oh how are you guys supporting more traffic than 2x better funded / 2.5x more staffed companies. A) our whole team is just cracked engineers. No BS roles B) we work till job is done and more C) its either dominate or

4

1

48

batuhan taskaya

@isidentical

16 days

2.5x cheaper, 2.5x faster. was it worth it? absolutely yes. we were already the fastest but that wasn't enough for us.

Jonathan Fischoff

@jfischoff

16 days

We've created an optimized endpoint just for loading multiple FLUX LoRAs. Now running at @fal speed here: Watch how much faster it is compared to our general purpose FLUX endpoint that supports LoRAs and ControlNets. It is also half the price 💸

3

7

77

2

0

47

batuhan taskaya

@isidentical

3 months

GO GO GO

Stable Diffusion V3 | AI Playground | fal.ai

Run SD3 at the speed of light

fal.ai

1

6

48

batuhan taskaya

@isidentical

9 days

did you know you can train a FLUX LoRA cheaper than anywhere else on the internet while having literally higher quality thanks to our own framework? kinda crazy

fal

@FAL

9 days

new updated pricing for our fast FLUX LoRA trainer! enjoy 🙌

1

3

47

6

1

51

batuhan taskaya

@isidentical

1 month

who tf said open source image models were dead after SD3?

fal

@FAL

1 month

Hold on to your seats! Announcing Flux - a text to image model from @bfl_ml , the original team behind Stable Diffusion.

15

27

175

2

0

46

batuhan taskaya

@isidentical

4 months

casually filtering a data on 256 CPUs, ~2T ram.

6

2

46

batuhan taskaya

@isidentical

2 months

this was one of our main goals underneath AuraFlow, MFU as a first class optimization parameter.

finbarr

@finbarrtimbers

2 months

evaluating ml researchers by GPU utilization is honestly a good metric

10

3

120

3

1

46

batuhan taskaya

@isidentical

23 days

just pushed an update to flux lora training in that made it ~2x faster with 0 increase in cost! same price as always.

Train Flux LoRA | AI Playground | fal.ai

General Purpose LORA Training for Flux with advanced options.

fal.ai

14

5

46

batuhan taskaya

@isidentical

3 months

fal is a Stable Diffusion 3 launch partner, see you all at 6PM eastern :)

Stable Diffusion 3 served by fal.ai

Stable Diffusion 3 by stability.ai is coming soon to fal.ai

fal.ai

1

0

46

batuhan taskaya

@isidentical

2 months

Introducing Imagenet.int8, the new MNIST of 2024 :) by none other than @cloneofsimo

cloneofsimo/imagenet.int8 · Datasets at Hugging Face

huggingface.co

Sharif Shameem

@sharifshameem

2 months

I love how replicating GPT-2 has now become a sort of “hello world” for distributed training runs

3

14

522

1

46

batuhan taskaya

@isidentical

7 months

sorry, this is now 24000. just doubled the GPU capacity and people are using it like crazy. 24000 FRAMES A SECOND. ADUDEEEEEEEEEEEEEEE

batuhan taskaya

@isidentical

7 months

generating 12000 IMAGES a SECOND. WHAT THE HELL

3

1

14

5

1

44

batuhan taskaya

@isidentical

13 days

@t3dotgg It is MIT licensed, and can be self served but we are offering it as a serverless optimized endpoint and for ~2000 images per 1$ (literally free).

4

0

45

batuhan taskaya

@isidentical

1 month

Simo cooking while we are at icml. This guy is unhinged cracked.

Simo Ryu

@cloneofsimo

1 month

Ok so here is AuraFlow v0.2 Feature wise, nothing much, its pretrained bit more and spent longer time on highres-fine-tuning. (I undid couple mistakes made during fine-tuning.) Also check out samples & comparisons on VERY COMPLEX PROMPTS: (AuraFlow vs

19

42

273

1

0

44

batuhan taskaya

@isidentical

6 months

@yacineMTB my whole life was a lie. the meme lord i trusted in, believed in turned out to be just another old guy. wow.

3

0

44

batuhan taskaya

@isidentical

2 months

open source ai is in jeporday (or is it?), courtesy of @heyglif @fabianstelzer

batuhan taskaya

@isidentical

2 months

Spent the last 1.5 month building this, finally open source. Can't be more excited about the future (this is just v0.1-B)

14

21

203

0

3

44

batuhan taskaya

@isidentical

2 months

Everyone wants to be rich, but no one is at the office at 7:30. How come?

9

1

43

batuhan taskaya

@isidentical

1 month

try it here:

Black Forest Labs

@bfl_ml

1 month

We are excited to announce the launch of Black Forest Labs. Our mission is to develop and advance state-of-the-art generative deep learning models for media and to push the boundaries of creativity, efficiency and diversity.

123

297

2K

5

2

43

batuhan taskaya

@isidentical

3 months

CVPR take: everyone is debating AR vs Diffusion for image generation. Am very pro on diffusion. Its beautiful. AR is a soulless rule (code) following method. Hate it.

9

0

43

batuhan taskaya

@isidentical

6 months

go follow @fal if you aren't already. they are new in town :) (also some lucky people might receive free compute, I hear)

fal

@FAL

6 months

hello world!

21

6

88

3

5

33

batuhan taskaya

@isidentical

2 months

something is cooking. early results, don't judge yet!

fal

@FAL

2 months

🧑‍🍳

2

5

74

3

0

43

batuhan taskaya

@isidentical

13 days

@snats_xyz no siglip is god tier

4

3

43

batuhan taskaya

@isidentical

1 year

We @fal_ai_data are hiring a senior+ level distributed ML systems engineer! It's a fully remote position w/ extremely competitive pay. But more importantly you get to work with me hand to hand on the intersection of ML problems and computing @edge . DMs are open.

3

14

42

batuhan taskaya

@isidentical

1 month

i have been told (by people who know the music mafia well) not to even think about training a music model. so be it.

3

1

41

batuhan taskaya

@isidentical

21 days

FLUX.1 [dev] is currently leading with a big margin, followed by FLUX.1 [schnell] and the long standing #1 RealVis XL V4.0. Pro model currently needs more samples so go vote!!!

3

0

41

batuhan taskaya

@isidentical

11 days

OMG @ValDotTown 's Townie is crazy. Just showed a single example from @fal , and it built an AI image generation demo app under 1 minute from scratch 🤯🤯🤯🤯

3

40

batuhan taskaya

@isidentical

14 days

i offered to pay. i'd literally pay for uv. am OK. free stuff ain't sustainable. uv changed my life. it is in the same league of software like dingboard and X for me. both of whom i am paying.

vik

@vikhyatk

15 days

@RaghuNC they raised $4M which implies they eventually have to monetize

1

0

19

5

0

40

batuhan taskaya

@isidentical

1 month

those devices are gonna get obsolete <3mo. not sustainable. on device inference is gonna be for super smol utility models but the general intellegince and main models will always be on cloud. network latencies in us is <50ms.

arpit

@arpitingle

1 month

future of compute in a box approach many startups are exploring the idea of providing an independent device for llm inference, like tinybox and truffle. but does this really make sense in the long term? in the future, inference is likely to be predominantly cloud-based and

36

1

141

7

1

37

batuhan taskaya

@isidentical

3 months

we ship at @fal . non-stop.

5

1

40

batuhan taskaya

@isidentical

2 months

new dataset drop! 10M good quality pictures

Ollin Boer Bohan

@madebyollin

2 months

I collected a dataset of links to ~10 million public domain Flickr photos , to hopefully let neural networks learn about the visual world without relying on copyrighted material.

9

30

181

0

4

38

batuhan taskaya

@isidentical

9 days

awaiting birthday presents

6

1

42

batuhan taskaya

@isidentical

2 months

PixArt Sigma, SD3, Stable Cascade, XXX (*512x512). Better prompt following than cascade, better physical semantics than SD3. Aesthetics seem to be bad though, i assume we need a post-training finetuning pass?

4

3

38

batuhan taskaya

@isidentical

1 month

diffusion is beautiful.

vik

@vikhyatk

1 month

reading up on the flux model architecture...

5

12

317

2

0

38

batuhan taskaya

@isidentical

11 days

CogVideoX-5B, the SOTA open weights text-to-video model, is now available at fal with acceleration up to 5x the speed you'd typically see.

2

9

39

batuhan taskaya

@isidentical

26 days

once a swiftie always a swiftie. flux lora training coming today. building on top of @ostrisai 's amazing work!

7

1

38

batuhan taskaya

@isidentical

3 years

I was going to apply for US visa to possibly attend PyCon 2022. But seems like the earliest interview date the embassy can offer is September of 2023 for non-immigrants. 1.5 years from now 😱 Guess I'll be at PyCon 2024 if I apply now?

5

1

36

batuhan taskaya

@isidentical

3 months

Looking for a contractor (1-3 mo gig with the potentially to extend into a full time position) for ML-related data acquisition & processing job. Dealing with billions of images, batch processing them (e.g. feature extraction) on a distributed setup and essentially feeding the

3

6

37

batuhan taskaya

@isidentical

21 days

some tentative results, still need more tho

batuhan taskaya

@isidentical

22 days

Added FLUX.1 pro/dev/schnell and AuraFlow v0.2 to !!! Go play with it and get us some votez

6

52

5

0

36

batuhan taskaya

@isidentical

2 months

Just subscribed to anifusion!!!

Nucleus☕️

@EsotericCofe

2 months

anifusion has its first customer? nani??

49

5

366

2

1

36

batuhan taskaya

@isidentical

2 months

a stylized cartoon image, yellow background, a smiling turtle holding a pizza, the turtle is wearing a white and brown cap, large eyes and a friendly expression, the turtle is standing on its hind legs, the pizza is held in both of its front legs, small clouds and trees on either

6

3

36

batuhan taskaya

@isidentical

8 months

making synonymous with real time diffusion models is one of my proudest achievements

LiveKit

@livekit

8 months

🔌 Plugins Integrations with @elevenlabsio , @DeepgramAI , @fal_ai_data , and @openai make it easy to compose together multimodal AI applications. We’re excited to build more integrations with the LiveKit community!

2

3

19

1

5

35

batuhan taskaya

@isidentical

6 months

@yacineMTB get some vc money and hire frontend people

2

0

36