Unsloth AI @UnslothAI Twitter profile | Pikagi

Pikagi

Unsloth AI

@UnslothAI

5,965

Followers

380

Following

30

Media

159

Statuses

Open source fine-tuning of LLMs! 🦥 Github: Discord:

San Francisco

https://t.co/xxBDOI9ApK

Joined November 2023

Don't wanna be here? Send us removal request.

Pinned Tweet

@UnslothAI

Unsloth AI

2 months

Llama 3.1 support is here! Unsloth supports 48K context lengths for Llama 3.1 (70B) on a 80GB GPU - 6x longer than HF+FA2. QLoRA fine-tuning Llama 3.1 (70B) is 1.9x faster, uses 65% less VRAM & Llama 3.1 (8B) is 2.1x faster and fits in a 8GB GPU! Blog:

Tweet media one

5

33

197

Last Seen Profiles

@BradsRxosa

@MajimaGoromi

@El_Gran_Cala

@aaHot8

@RTFKTConne

@r_sologar

@rora_yue

@AnkaraSMMMO

@BethBak17456849

@husnushenel

@sasidharanne

@bokeplokalmalam

@ekinan_lawyer

@Galwaybayfmnews

@Jamesborchard

@elliottszwu

@katiebandz_

@stwmaniax

@AryEcole

@Uchi5Alt

@guardfvsub

@kleinverde

@CHSL1926

@shebetrickin

@nrhztl4ly4

@bokeplokalmalam

@lydy73414197491

@abokeian

@fangheihei11

@NabilAli202

@stw_pdg

@Norinchedelamar

@JeannetteCeja

@EL_CHAREEY

@IsaacMuk

@JoeCody

@UnslothAI

Unsloth AI

2 months

We made a step-by-step tutorial on how to finetune Llama-3 with Google Colab & deploy it to @Ollama Tutorial: Colab notebook: Blog post & video coming soon. 🦥

Tweet media one

Tweet media two

7

122

579

@UnslothAI

Unsloth AI

5 months

Long-context Llama 3 finetuning is here! 🦙 Unsloth supports 48K context lengths for Llama-3 70b on a 80GB GPU - 6x longer than HF+FA2 QLoRA finetuning Llama-3 70b is 1.8x faster, uses 68% less VRAM & Llama-3 8b is 2x faster and fits in a 8GB GPU! Blog:

Tweet media one

4

60

349

@UnslothAI

Unsloth AI

3 months

Unsloth now allows you to do continued pretraining with QLoRA 2x faster and use 50% less VRAM than Hugging Face+FA2. Continued pretraining allows models to train on new domain data. Read our blog:

Tweet media one

7

41

205

@UnslothAI

Unsloth AI

2 months

You can now use CSV/Excel files for fine-tuning & directly export models to @Ollama ! Unsloth also now supports datasets with more than 3 columns, automatically merging them into one prompt for fine-tuning. Blog: Tutorial:

Tweet media one

1

38

197

@UnslothAI

Unsloth AI

4 months

We have resolved issues with training Llama 3, so finetuning is much better now! Unsloth now supports the new Phi-3 models, Mistral v3, Qwen and more! Read our blog:

Tweet media one

1

37

194

@UnslothAI

Unsloth AI

2 months

Gemma 2 support is here! Unsloth supports 50K context lengths for Gemma 2 (9B) on a 80GB GPU - 5x longer than HF+FA2. QLoRA finetuning Gemma 2 (27B) is 1.9x faster, uses 53% less VRAM & Gemma 2 (9B) is 2x faster, 63% less VRAM + fits in a 8GB GPU! Blog:

Tweet media one

2

31

183

@UnslothAI

Unsloth AI

19 days

You can now fine-tune Microsoft's new Phi-3.5 (mini) model 2x faster with 50% less memory with Unsloth! Free Colab notebook: We also 'Llamified' the models for improved accuracy and uploaded them to Hugging Face:

Tweet media one

6

28

181

@UnslothAI

Unsloth AI

2 months

Mistral's new model, NeMo (12B) is now supported! Unsloth makes finetuning NeMo fit in a 12GB GPU! QLoRA training is 2x faster, uses 60% less memory & we support 3-4x longer context lengths than HF+FA2. Read our Blog:

Tweet media one

4

27

163

@UnslothAI

Unsloth AI

6 months

Unsloth is trending on GitHub this week! 🙌🦥 Thanks to everyone & all the ⭐️Stargazers for the support! Check out our repo:

Tweet media one

2

17

128

@UnslothAI

Unsloth AI

6 months

Over 1,000 models have now been trained with Unsloth & shared on @HuggingFace 🤗🦥 Also special thanks to the HF team for the custom icon on our tag! See models trained with Unsloth:

Tweet media one

2

21

119

@UnslothAI

Unsloth AI

5 months

Unsloth now supports fine-tuning of LLMs with 4x longer context windows! We managed to reduce memory usage by a further 30% at the cost of +1.9% extra time overhead. Read our blog:

Tweet media one

5

17

111

@UnslothAI

Unsloth AI

1 month

. @Google releases a new Gemma 2 model with 2B parameters & it's the best performing model for its size! Unsloth makes Gemma 2 (2B) QLoRA fine-tuning 2x faster with 65% less memory. Did you know the Gemma 2 models are Unsloth's 2nd most popular models to fine-tune with (250K

Tweet media one

2

18

109

@UnslothAI

Unsloth AI

4 months

We have hit 1 million monthly downloads on @HuggingFace ! 🥳🦥 Thanks to every single person who has used Unsloth or downloaded our models! ❤️

Tweet media one

6

16

104

@UnslothAI

Unsloth AI

5 months

Unsloth is currently trending on Github! 🙌🦥 If you want to finetune LLMs like Llama 3 or Mistral, now is a good time to try!⭐️

Tweet media one

4

17

101

@UnslothAI

Unsloth AI

28 days

We just hit 2 million monthly downloads on @HuggingFace ! 🦥🥳 Over 13K models trained with Unsloth have also been uploaded to Hugging Face. Huge thanks to the Unsloth community, the model teams and the HF team! 🤗

Tweet media one

5

12

96

@UnslothAI

Unsloth AI

5 months

Unsloth has surpassed 500K+ monthly model downloads on @HuggingFace ! 🥳 🦥 Thanks for the support! 🩷 See our collection of 4bit models + more:

Tweet media one

1

12

93

@UnslothAI

Unsloth AI

3 months

We'll be live on @Ollama 's server at 12pm ET today, to show our new support for Ollama! 🦥🦙 First learn with Sebastien about 'Emotions in AI', then we'll teach & give early access to our guide on how to finetune a model & deploy to Ollama. Join us here:

Tweet media one

2

9

83

@UnslothAI

Unsloth AI

7 months

You can now QLoRA finetune Gemma 7B 2.43x faster & use 58% less VRAM than @HuggingFace + FA2 via Unsloth!🦥 When compared to vanilla HF, we're 2.53x faster & use 70% less VRAM. We have a blog on our learnings (like a RoPE bug, GeGLU) + Colab notebooks!

Tweet media one

2

9

65

@UnslothAI

Unsloth AI

8 months

We're thrilled to share that @HuggingFace has published a blog post about our exciting collab! We've been working hard behind the scenes & are eager for you to learn more about what we've been up to.🦥

Tweet card media

Make LLM Fine-tuning 2x faster with Unsloth and 🤗 TRL

0

14

57

@UnslothAI

Unsloth AI

1 month

Watch a step-by-step video tutorial by @engineerrprompt on how to fine-tune Llama 3.1 using your own data:

Tweet card media

LLAMA-3.1 🦙: EASIET WAY To FINE-TUNE ON YOUR DATA 🙌

Learn how to efficiently fine-tuning the Llama 3.1 model using Unsloth, LoRa, and QLoRa techniques. LINKS:Colab: https://tinyurl.com/bdzxhy5nUnsloth: https:/...

www.youtube.com

0

7

57

@UnslothAI

Unsloth AI

4 months

We're so happy to announce that Unsloth is part of the 2024 @GitHub Accelerator program!🦥 If you want to easily fine-tune LLMs like Llama 3, now is the perfect time!

Tweet card media

2024 GitHub Accelerator: Meet the 11 projects shaping open source AI

Announcing the second cohort, delivering value to projects, and driving a new frontier.

3

7

56

@UnslothAI

Unsloth AI

8 months

• Finetune 387% faster TinyLlama • 600% faster GGUF conversion • 188% faster DPO • 400% faster model downloads All part of the new Unsloth AI release! 🦥 Read our blog:

Tweet media one

1

6

51

@UnslothAI

Unsloth AI

6 months

Unsloth has hit 125K+ monthly model downloads on @HuggingFace ! 🥳🦥 Thanks for the support! 🩷 See our collection of 4-bit models (4x faster downloading) + more:

Tweet media one

1

5

43

@UnslothAI

Unsloth AI

2 months

@danielhanchen We have uploaded 4bit bnb quants for now and are working on Llama 3.1 support! Llama 3.1 (8B) 4bit: Llama 3.1 (8B) Instruct 4bit: Llama 3.1 (70B) 4bit: Llama 3.1 (70B) Instruct 4bit:

Tweet card media

unsloth/Meta-Llama-3.1-70B-Instruct-bnb-4bit · Hugging Face

3

7

42

@UnslothAI

Unsloth AI

6 months

A huge thank you to @UnderscoreTalk and @YKilcher for sharing Unsloth!🦥🩷 Underscore's latest in AI apps video (French): Yannic's latest ML News video: And of course thanks to everyone else for the constant support😊

Tweet media one

1

2

23

@UnslothAI

Unsloth AI

2 months

We're excited to showcase Unsloth for @GitHub 's 2024 Accelerator! 🦥 Watch to learn more about us & discover other incredible open source projects:

Tweet card media

GitHub Accelerator Showcase 2024 | The next wave of AI projects

Get ready to dive into the next wave of AI innovation! 🌐✨ Witness open-source AI projects in action. Our Showcase highlights project demos, the inspiring jo...

www.youtube.com

0

4

23

@UnslothAI

Unsloth AI

3 months

Tomorrow we will be handing out our new stickers for the @aiDotEngineer World's Fair! 🦥 Join us at 9AM, June 25 where we will be doing workshops on LLM analysis + technicals, @Ollama support & more! We'll also be presenting a speech about fixing LLM bugs at 2:20PM, June 26!

Tweet media one

3

5

20

@UnslothAI

Unsloth AI

15 days

@_philschmid @Microsoft Thank you so much Philipp & Hugging Face for the support! And the community loves & appreciates the Microsoft AI team for these models. 🤗

1

1

16

@UnslothAI

Unsloth AI

9 months

Introducing 🦥 You can now train a personal ChatGPT in a few hours instead of 54 days! 30x faster, 60% less memory usage with 0% loss in accuracy. We also released an open source version which finetunes Llama 5x faster and uses 50% less memory!

Tweet media one

1

5

12

@UnslothAI

Unsloth AI

3 months

@MaziyarPanahi Hi there yes, Qwen 110B will work on a H100! You only need 76GB VRAM or so with Unsloth. See our blog for more details:

@danielhanchen

Daniel Han

5 months

Llama-3 70b QLoRA finetuning is 1.83x faster & uses 63% less VRAM than HF+FA2 1. Llama-3 70b + Unsloth can fit 48K context lengths on bsz=1 on A100 80GB (6x longer than FA2) with +1.9% overhead 2. Llama-3 8b QLoRA fits in a 8GB card & is 2x faster, uses 68% less VRAM. Can fit

Tweet media one

7

60

318

1

2

9

@UnslothAI

Unsloth AI

5 months

This works on all model architectures which use gradient checkpointing (ie stable diffusion, Mamba etc) See bar graph for memory saving benchmarks:

Tweet media one

1

0

9

@UnslothAI

Unsloth AI

2 months

@MervinPraison @ollama @huggingface @danielhanchen Thank you Mervin for this incredible and in-depth video as always! 😀

0

1

9

@UnslothAI

Unsloth AI

29 days

@hokazuya Should be fixed now so please update Unsloth. Now you can do xformers & TRL with any version! Please let us know if works. 🙏

1

0

6

@UnslothAI

Unsloth AI

2 months

@rohanpaul_ai @MistralAI @nvidia @danielhanchen Thank you so much Rohan as always for supporting Unsloth! 🦥 Hope you will like Unsloth Studio (our upcoming UI) which will hopefully be out next week. 🥰

1

0

6

@UnslothAI

Unsloth AI

2 months

@mpowers206 @ollama Very soon! We're already rolling out multiGPU beta access for a lot of folks! 👍

0

0

6

@UnslothAI

Unsloth AI

1 month

@maximelabonne @huggingface Thank you for the support Maxime and this guide is just incredible! 🔥🙏

0

0

6

@UnslothAI

Unsloth AI

3 months

@ollama A huge thank you to the @Ollama team for inviting us! Sebastien was amazing & we learnt so much from him! 🙏♥️

1

1

4

@UnslothAI

Unsloth AI

2 months

@hsinskip Thank you so much Henry for the support! We've got a UI coming next week which will make AI even more accessible! :)

1

0

4

@UnslothAI

Unsloth AI

1 month

@onegaspine Thank you so much for using Unsloth! 🥰🦥 And yes, people love the VRAM reductions!

0

0

4

@UnslothAI

Unsloth AI

29 days

@hokazuya Thank you so much hokazuya! Glad it's working now! We're also going to be releasing a UI either next week or the week after to make finetuning even more accessible/easier! 🦥

0

0

4

@UnslothAI

Unsloth AI

2 months

@kaggle @danielhanchen Thank you for supporting open source! ♥️🦥

0

0

4

@UnslothAI

Unsloth AI

3 months

@eugeneyalt @willpienaar @sh_reya @ShreyaR @shrumm @devstein64 @tristanzajonc @danielhanchen @aiDotEngineer Thank you so much for coming! Hopefully you managed to grab some of our stickers! 🦥♥️

0

0

2

@UnslothAI

Unsloth AI

2 months

@MarvinGabler @danielhanchen Thank you so much Marvin for coming and glad you enjoyed our talk! 🦥♥️

0

0

3

@UnslothAI

Unsloth AI

19 days

@op7418 We just saw this but thank you so much for the support! ❤️

0

0

2

@UnslothAI

Unsloth AI

18 days

@gentonje . @maximelabonne wrote a lovely beginner's Guide on using Llama 3.1 with Unsloth so be sure to check it out!

@maximelabonne

Maxime Labonne

1 month

🦥 Fine-tune Llama 3.1 Ultra-Efficiently with @UnslothAI New comprehensive guide about supervised fine-tuning on @huggingface . Over the last year, I've done a lot of fine-tuning and blogging. This guide brings it all together. 📝 Article:

Tweet media one

11

85

452

1

1

2

@UnslothAI

Unsloth AI

2 months

@TheLouisDupont Yes, so we support all the popular models like Llama 3, Gemma 2, Qwen2, Phi-3 and more. We are working on all model support coming soon! Also any model we upload on our @HuggingFace is supported:

Tweet card media

unsloth (Unsloth AI)

0

0

1

@UnslothAI

Unsloth AI

30 days

@hokazuya Hey thanks for using Unsloth! We are currently working on updating TRL dependency to the latest version. Will let you know when it's done! 👌

1

0

1

@UnslothAI

Unsloth AI

1 month

@tuturetom @danielhanchen Thank you so much Tom for the constant support! ♥️🙏

1

0

1

@UnslothAI

Unsloth AI

11 days

@AIMakerspace Thank you so much for inviting us! 🙏 Can't wait to get started! 🦥

0

0

1

@UnslothAI

Unsloth AI

2 months

@lily_xlz @aiDotEngineer @danielhanchen Thank you so much Lily for coming!! ♥️🦥

1

0

1

@UnslothAI

Unsloth AI

28 days

@fouriergalois @huggingface Thank you so much for the constant support Kendrick! 🦥♥️

0

0

1

@UnslothAI

Unsloth AI

26 days

@VContribution @Humbertoblood Thank you so much for using Unsloth and glad Maxime Labonne's guide was useful! 😃

1

0

1

@UnslothAI

Unsloth AI

2 months

@newplatonism @danielhanchen We are working on supporting 8bit for Unsloth. I guess we can still upload them hopefully soon! 🤞

1

0

1

@UnslothAI

Unsloth AI

3 months

@felix_red_panda @aiDotEngineer @ollama @danielhanchen Will let you all know if we can deliver some to you guys instead! 🦥

0

0

1

@UnslothAI

Unsloth AI

1 month

@fahdmirza @YouTube @danielhanchen Thank you so much Fahd as usual for the constant support! 🙏😃

0

1

1

@UnslothAI

Unsloth AI

1 month

@waydegilliam @danielhanchen Hey Wayde, if you are using 2 GPUs, Unsloth is actually faster on 1 GPU than 2. As for multiGPU suppot, we gave a few people early access to our multiGPU support and will be hopefully rolling it out next month via subscription!🙂

0

0

1

@UnslothAI

Unsloth AI

1 month

@RisingSayak @danielhanchen We're working on it. It will take some time though. We're trying to push out a UI first ASAP 🙏

0

0

1

@UnslothAI

Unsloth AI

2 months

@hackwithzach @aiDotEngineer Thank you so much for coming & glad you enjoyed our talk Zach! We really appreciate it! 😀🦥

0

0

1