Post
220
we now have more than 2000 public AI models using ModelHubMixin๐ค
Nerdy Face
Enterprise
company
AI & ML interests
None defined yet.
Recent Activity
View all activity
nerdyface's activity
Post
735
๐ ๐ถ๐ป๐ถ๐ ๐ฎ๐
'๐ ๐ป๐ฒ๐ ๐ ๐ผ๐ ๐๐๐ ๐ฟ๐ฒ๐ฎ๐ฐ๐ต๐ฒ๐ ๐๐น๐ฎ๐๐ฑ๐ฒ-๐ฆ๐ผ๐ป๐ป๐ฒ๐ ๐น๐ฒ๐๐ฒ๐น ๐๐ถ๐๐ต ๐ฐ๐ ๐๐ผ๐ธ๐ฒ๐ป๐ ๐ฐ๐ผ๐ป๐๐ฒ๐
๐ ๐น๐ฒ๐ป๐ด๐๐ต ๐ฅ
This work from Chinese startup @MiniMax-AI introduces a novel architecture that achieves state-of-the-art performance while handling context windows up to 4 million tokens - roughly 20x longer than current models. The key was combining lightning attention, mixture of experts (MoE), and a careful hybrid approach.
๐๐ฒ๐ ๐ถ๐ป๐๐ถ๐ด๐ต๐๐:
๐๏ธ MoE with novel hybrid attention:
โฃ Mixture of Experts with 456B total parameters (45.9B activated per token)
โฃ Combines Lightning attention (linear complexity) for most layers and traditional softmax attention every 8 layers
๐ Outperforms leading models across benchmarks while offering vastly longer context:
โฃ Competitive with GPT-4/Claude-3.5-Sonnet on most tasks
โฃ Can efficiently handle 4M token contexts (vs 256K for most other LLMs)
๐ฌ Technical innovations enable efficient scaling:
โฃ Novel expert parallel and tensor parallel strategies cut communication overhead in half
โฃ Improved linear attention sequence parallelism, multi-level padding and other optimizations achieve 75% GPU utilization (that's really high, generally utilization is around 50%)
๐ฏ Thorough training strategy:
โฃ Careful data curation and quality control by using a smaller preliminary version of their LLM as a judge!
Overall, not only is the model impressive, but the technical paper is also really interesting! ๐
It has lots of insights including a great comparison showing how a 2B MoE (24B total) far outperforms a 7B model for the same amount of FLOPs.
Read it in full here ๐ MiniMax-01: Scaling Foundation Models with Lightning Attention (2501.08313)
Model here, allows commercial use <100M monthly users ๐ MiniMaxAI/MiniMax-Text-01
This work from Chinese startup @MiniMax-AI introduces a novel architecture that achieves state-of-the-art performance while handling context windows up to 4 million tokens - roughly 20x longer than current models. The key was combining lightning attention, mixture of experts (MoE), and a careful hybrid approach.
๐๐ฒ๐ ๐ถ๐ป๐๐ถ๐ด๐ต๐๐:
๐๏ธ MoE with novel hybrid attention:
โฃ Mixture of Experts with 456B total parameters (45.9B activated per token)
โฃ Combines Lightning attention (linear complexity) for most layers and traditional softmax attention every 8 layers
๐ Outperforms leading models across benchmarks while offering vastly longer context:
โฃ Competitive with GPT-4/Claude-3.5-Sonnet on most tasks
โฃ Can efficiently handle 4M token contexts (vs 256K for most other LLMs)
๐ฌ Technical innovations enable efficient scaling:
โฃ Novel expert parallel and tensor parallel strategies cut communication overhead in half
โฃ Improved linear attention sequence parallelism, multi-level padding and other optimizations achieve 75% GPU utilization (that's really high, generally utilization is around 50%)
๐ฏ Thorough training strategy:
โฃ Careful data curation and quality control by using a smaller preliminary version of their LLM as a judge!
Overall, not only is the model impressive, but the technical paper is also really interesting! ๐
It has lots of insights including a great comparison showing how a 2B MoE (24B total) far outperforms a 7B model for the same amount of FLOPs.
Read it in full here ๐ MiniMax-01: Scaling Foundation Models with Lightning Attention (2501.08313)
Model here, allows commercial use <100M monthly users ๐ MiniMaxAI/MiniMax-Text-01
Post
2144
๐ช๐ฒ'๐๐ฒ ๐ท๐๐๐ ๐ฟ๐ฒ๐น๐ฒ๐ฎ๐๐ฒ๐ฑ ๐๐บ๐ผ๐น๐ฎ๐ด๐ฒ๐ป๐๐ ๐๐ญ.๐ฏ.๐ฌ ๐, and it comes with a major feature: you can now log agent runs using OpenTelemetry to inspect them afterwards! ๐
This interactive format is IMO much easier to inspect big multi-step runs than endless console logs.
The setup is very easy, in a few lines of code.
Find a tutorial here ๐ https://huggingface.co/docs/smolagents/tutorials/inspect_runs
This interactive format is IMO much easier to inspect big multi-step runs than endless console logs.
The setup is very easy, in a few lines of code.
Find a tutorial here ๐ https://huggingface.co/docs/smolagents/tutorials/inspect_runs
- 4 replies
Post
3629
Published a new blogpost ๐
In this blogpost I have gone through the transformers' architecture emphasizing how shapes propagate throughout each layer.
๐ https://huggingface.co/blog/not-lain/tensor-dims
some interesting takeaways :
In this blogpost I have gone through the transformers' architecture emphasizing how shapes propagate throughout each layer.
๐ https://huggingface.co/blog/not-lain/tensor-dims
some interesting takeaways :
Post
559
๐ข๐ฆ-๐๐ฒ๐ป๐ฒ๐๐ถ๐: ๐ป๐ฒ๐ ๐ฟ๐ฒ๐๐ฒ๐ฎ๐ฟ๐ฐ๐ต ๐ฝ๐ฎ๐ฝ๐ฒ๐ฟ ๐ฝ๐ฟ๐ผ๐ฝ๐ผ๐๐ฒ๐ ๐ฎ ๐ป๐ผ๐๐ฒ๐น ๐๐ฟ๐ฎ๐ถ๐ป๐ถ๐ป๐ด ๐ฑ๐ฎ๐๐ฎ ๐ด๐ฒ๐ป๐ฒ๐ฟ๐ฎ๐๐ถ๐ผ๐ป ๐บ๐ฒ๐๐ต๐ผ๐ฑ ๐ณ๐ผ๐ฟ ๐๐น๐ฎ๐๐ฑ๐ฒ-๐๐ผ๐บ๐ฝ๐๐๐ฒ๐ฟ-๐จ๐๐ฒ-๐น๐ถ๐ธ๐ฒ ๐ฎ๐ด๐ฒ๐ป๐๐, ๐๐ถ๐๐ต ๐ถ๐บ๐ฝ๐ฟ๐ฒ๐๐๐ถ๐๐ฒ ๐ฟ๐ฒ๐๐๐น๐๐! ๐ฅ
The main bottleneck in building GUI agents it to find training data.
GUI Agent trajectories are not easy to get by. Crowdsourcing trajectories, then manually annotating them, could be an option, but at scale, it's hard to do
You could use synthetic data generation (ask 1000s small existing GUI agents to solve tasks, keep only successful runs). But then it's hard to come up with many high level-tasks.
โก๏ธ Well, a novel technique was just published that creates a new promising paradigm for synthetic data generation: Shanghai AI Lab researchers propose OS-Genesis, a novel way to create training data for GUI agents that flips the traditional approach on its head. Instead of starting with predefined tasks and having humans or machines execute them, OS-Genesis first explores the interface naturally, then derives meaningful tasks from those interactions.
๐ Exploration-driven vs task-driven approach:
โฃ Instead of starting with tasks, OS-Genesis first explores GUIs by clicking and interacting
โฃ It then reverse-engineers high-level tasks from successful interaction patterns
โฃ This leads to more natural and diverse training data than predefined tasks
๐ฏ Novel reward model for trajectory quality:
โฃ Rather than discarding incomplete trajectories, OS-Genesis scores them based on coherence and completion
โฃ This preserves valuable partial successes that would otherwise be wasted
๐ Superior results across environments:
โฃ Nearly doubles performance on AndroidWorld (9.8% โ 17.4%)
By the way, this field of GUI agents is still in infancy, so you can still make a difference with "low-cost" setups: their paper gets SOTA results with only 8xA100!
Read the paper here ๐ OS-Genesis: Automating GUI Agent Trajectory Construction via Reverse Task Synthesis (2412.19723)
The main bottleneck in building GUI agents it to find training data.
GUI Agent trajectories are not easy to get by. Crowdsourcing trajectories, then manually annotating them, could be an option, but at scale, it's hard to do
You could use synthetic data generation (ask 1000s small existing GUI agents to solve tasks, keep only successful runs). But then it's hard to come up with many high level-tasks.
โก๏ธ Well, a novel technique was just published that creates a new promising paradigm for synthetic data generation: Shanghai AI Lab researchers propose OS-Genesis, a novel way to create training data for GUI agents that flips the traditional approach on its head. Instead of starting with predefined tasks and having humans or machines execute them, OS-Genesis first explores the interface naturally, then derives meaningful tasks from those interactions.
๐ Exploration-driven vs task-driven approach:
โฃ Instead of starting with tasks, OS-Genesis first explores GUIs by clicking and interacting
โฃ It then reverse-engineers high-level tasks from successful interaction patterns
โฃ This leads to more natural and diverse training data than predefined tasks
๐ฏ Novel reward model for trajectory quality:
โฃ Rather than discarding incomplete trajectories, OS-Genesis scores them based on coherence and completion
โฃ This preserves valuable partial successes that would otherwise be wasted
๐ Superior results across environments:
โฃ Nearly doubles performance on AndroidWorld (9.8% โ 17.4%)
By the way, this field of GUI agents is still in infancy, so you can still make a difference with "low-cost" setups: their paper gets SOTA results with only 8xA100!
Read the paper here ๐ OS-Genesis: Automating GUI Agent Trajectory Construction via Reverse Task Synthesis (2412.19723)
Post
4983
Since I published it on GitHub a few days ago,
Hugging Face's new agentic library ๐๐บ๐ผ๐น๐ฎ๐ด๐ฒ๐ป๐๐ has gathered nearly 4k stars ๐คฏ
โก๏ธ But we are just getting started on agents: so we are hiring an ML Engineer to join me and double down on this effort!
The plan is to build GUI agents: agents that can act on your computer with mouse & keyboard, like Claude Computer Use.
We will make it work better, and fully open. โจ
Sounds like something you'd like to do? Apply here ๐ https://apply.workable.com/huggingface/j/AF1D4E3FEB/
Hugging Face's new agentic library ๐๐บ๐ผ๐น๐ฎ๐ด๐ฒ๐ป๐๐ has gathered nearly 4k stars ๐คฏ
โก๏ธ But we are just getting started on agents: so we are hiring an ML Engineer to join me and double down on this effort!
The plan is to build GUI agents: agents that can act on your computer with mouse & keyboard, like Claude Computer Use.
We will make it work better, and fully open. โจ
Sounds like something you'd like to do? Apply here ๐ https://apply.workable.com/huggingface/j/AF1D4E3FEB/
- 3 replies
jeffboudierย
posted
an
update
11 days ago
Post
521
NVIDIA just announced the Cosmos World Foundation Models, available on the Hub:
nvidia/cosmos-6751e884dc10e013a0a0d8e6
Cosmos is a family of pre-trained models purpose-built for generating physics-aware videos and world states to advance physical AI development.
The release includes Tokenizers nvidia/cosmos-tokenizer-672b93023add81b66a8ff8e6
Learn more in this great community article by @mingyuliutw and @PranjaliJoshi https://huggingface.co/blog/mingyuliutw/nvidia-cosmos
Cosmos is a family of pre-trained models purpose-built for generating physics-aware videos and world states to advance physical AI development.
The release includes Tokenizers nvidia/cosmos-tokenizer-672b93023add81b66a8ff8e6
Learn more in this great community article by @mingyuliutw and @PranjaliJoshi https://huggingface.co/blog/mingyuliutw/nvidia-cosmos
- 1 reply
Post
3996
Cool to see
@ylecun
joining the top 10 of most followed on HF!
(and leaderboard by @mvaloatto is here: mvaloatto/TCTF)
(and leaderboard by @mvaloatto is here: mvaloatto/TCTF)
- 2 replies
Post
715
Hey everyone ๐ค!
Check out this new Virtual Try Off model (based on SD1.5): 1aurent/TryOffAnyone
This model isn't as accurate as others (e.g. xiaozaa/cat-try-off-flux based on FLUX.1) but it sure is fast!
Check out this new Virtual Try Off model (based on SD1.5): 1aurent/TryOffAnyone
This model isn't as accurate as others (e.g. xiaozaa/cat-try-off-flux based on FLUX.1) but it sure is fast!
suayptalhaย
posted
an
update
21 days ago
Post
1855
๐ Introducing ๐
๐ข๐ซ๐ฌ๐ญ ๐๐ฎ๐ ๐ ๐ข๐ง๐ ๐
๐๐๐ ๐๐ง๐ญ๐๐ ๐ซ๐๐ญ๐ข๐จ๐ง ๐จ๐ ๐ฆ๐ข๐ง๐๐๐ ๐๐จ๐๐๐ฅ๐ฌ from the paper ๐๐๐ซ๐ ๐๐๐๐ฌ ๐๐ฅ๐ฅ ๐๐ ๐๐๐๐๐๐?
๐ฅ I have integrated ๐ง๐๐ฑ๐ญ-๐ ๐๐ง๐๐ซ๐๐ญ๐ข๐จ๐ง ๐๐๐๐ฌ, specifically minGRU, which offer faster performance compared to Transformer architectures, into HuggingFace. This allows users to leverage the lighter and more efficient minGRU models with the "๐ญ๐ซ๐๐ง๐ฌ๐๐จ๐ซ๐ฆ๐๐ซ๐ฌ" ๐ฅ๐ข๐๐ซ๐๐ซ๐ฒ for both usage and training.
๐ป I integrated two main tasks: ๐๐ข๐ง๐๐๐๐ ๐จ๐ซ๐๐๐ช๐ฎ๐๐ง๐๐๐๐ฅ๐๐ฌ๐ฌ๐ข๐๐ข๐๐๐ญ๐ข๐จ๐ง and ๐๐ข๐ง๐๐๐๐ ๐จ๐ซ๐๐๐ฎ๐ฌ๐๐ฅ๐๐.
๐๐ข๐ง๐๐๐๐ ๐จ๐ซ๐๐๐ช๐ฎ๐๐ง๐๐๐๐ฅ๐๐ฌ๐ฌ๐ข๐๐ข๐๐๐ญ๐ข๐จ๐ง:
You can use this class for ๐๐๐ช๐ฎ๐๐ง๐๐ ๐๐ฅ๐๐ฌ๐ฌ๐ข๐๐ข๐๐๐ญ๐ข๐จ๐ง tasks. I also trained a Sentiment Analysis model with stanfordnlp/imdb dataset.
๐๐ข๐ง๐๐๐๐ ๐จ๐ซ๐๐๐ฎ๐ฌ๐๐ฅ๐๐:
You can use this class for ๐๐๐ฎ๐ฌ๐๐ฅ ๐๐๐ง๐ ๐ฎ๐๐ ๐ ๐๐จ๐๐๐ฅ tasks such as GPT, Llama. I also trained an example model with roneneldan/TinyStories dataset. You can fine-tune and use it!
๐ ๐๐ข๐ง๐ค๐ฌ:
Models: suayptalha/mingru-676fe8d90760d01b7955d7ab
GitHub: https://github.com/suayptalha/minGRU-hf
LinkedIn Post: https://www.linkedin.com/posts/suayp-talha-kocabay_mingru-a-suayptalha-collection-activity-7278755484172439552-wNY1
๐ฐ ๐๐ซ๐๐๐ข๐ญ๐ฌ:
Paper Link: https://arxiv.org/abs/2410.01201
I am thankful to Leo Feng, Frederick Tung, Mohamed Osama Ahmed, Yoshua Bengio and Hossein Hajimirsadeghi for their papers.
๐ฅ I have integrated ๐ง๐๐ฑ๐ญ-๐ ๐๐ง๐๐ซ๐๐ญ๐ข๐จ๐ง ๐๐๐๐ฌ, specifically minGRU, which offer faster performance compared to Transformer architectures, into HuggingFace. This allows users to leverage the lighter and more efficient minGRU models with the "๐ญ๐ซ๐๐ง๐ฌ๐๐จ๐ซ๐ฆ๐๐ซ๐ฌ" ๐ฅ๐ข๐๐ซ๐๐ซ๐ฒ for both usage and training.
๐ป I integrated two main tasks: ๐๐ข๐ง๐๐๐๐ ๐จ๐ซ๐๐๐ช๐ฎ๐๐ง๐๐๐๐ฅ๐๐ฌ๐ฌ๐ข๐๐ข๐๐๐ญ๐ข๐จ๐ง and ๐๐ข๐ง๐๐๐๐ ๐จ๐ซ๐๐๐ฎ๐ฌ๐๐ฅ๐๐.
๐๐ข๐ง๐๐๐๐ ๐จ๐ซ๐๐๐ช๐ฎ๐๐ง๐๐๐๐ฅ๐๐ฌ๐ฌ๐ข๐๐ข๐๐๐ญ๐ข๐จ๐ง:
You can use this class for ๐๐๐ช๐ฎ๐๐ง๐๐ ๐๐ฅ๐๐ฌ๐ฌ๐ข๐๐ข๐๐๐ญ๐ข๐จ๐ง tasks. I also trained a Sentiment Analysis model with stanfordnlp/imdb dataset.
๐๐ข๐ง๐๐๐๐ ๐จ๐ซ๐๐๐ฎ๐ฌ๐๐ฅ๐๐:
You can use this class for ๐๐๐ฎ๐ฌ๐๐ฅ ๐๐๐ง๐ ๐ฎ๐๐ ๐ ๐๐จ๐๐๐ฅ tasks such as GPT, Llama. I also trained an example model with roneneldan/TinyStories dataset. You can fine-tune and use it!
๐ ๐๐ข๐ง๐ค๐ฌ:
Models: suayptalha/mingru-676fe8d90760d01b7955d7ab
GitHub: https://github.com/suayptalha/minGRU-hf
LinkedIn Post: https://www.linkedin.com/posts/suayp-talha-kocabay_mingru-a-suayptalha-collection-activity-7278755484172439552-wNY1
๐ฐ ๐๐ซ๐๐๐ข๐ญ๐ฌ:
Paper Link: https://arxiv.org/abs/2410.01201
I am thankful to Leo Feng, Frederick Tung, Mohamed Osama Ahmed, Yoshua Bengio and Hossein Hajimirsadeghi for their papers.
suayptalhaย
posted
an
update
24 days ago
Post
2469
๐ Introducing Substitution Cipher Solvers!
As @suayptalha and @Synd209 we are thrilled to share an update!
๐ This project contains a text-to-text model designed to decrypt English and Turkish text encoded using a substitution cipher. In a substitution cipher, each letter in the plaintext is replaced by a corresponding, unique letter to form the ciphertext. The model leverages statistical and linguistic properties of English to make educated guesses about the letter substitutions, aiming to recover the original plaintext message.
These models were fine-tuned on T5-base. The models are for monoalphabetic English and Turkish substitution ciphers, and they output decoded text and the alphabet with an accuracy that has never been achieved before!
Example:
Encoded text: Z hztwgx tstcsf qf z ulooqfe osfuqb tzx uezx awej z ozewsbe vlfwby fsmqisfx.
Decoded text: A family member or a support person may stay with a patient during recovery.
Model Collection Link: Cipher-AI/substitution-cipher-solvers-6731ebd22f0f0d8e0e2e2e00
Organization Link: https://huggingface.co/Cipher-AI
As @suayptalha and @Synd209 we are thrilled to share an update!
๐ This project contains a text-to-text model designed to decrypt English and Turkish text encoded using a substitution cipher. In a substitution cipher, each letter in the plaintext is replaced by a corresponding, unique letter to form the ciphertext. The model leverages statistical and linguistic properties of English to make educated guesses about the letter substitutions, aiming to recover the original plaintext message.
These models were fine-tuned on T5-base. The models are for monoalphabetic English and Turkish substitution ciphers, and they output decoded text and the alphabet with an accuracy that has never been achieved before!
Example:
Encoded text: Z hztwgx tstcsf qf z ulooqfe osfuqb tzx uezx awej z ozewsbe vlfwby fsmqisfx.
Decoded text: A family member or a support person may stay with a patient during recovery.
Model Collection Link: Cipher-AI/substitution-cipher-solvers-6731ebd22f0f0d8e0e2e2e00
Organization Link: https://huggingface.co/Cipher-AI
-
- 3 replies
suayptalhaย
posted
an
update
29 days ago
Post
1630
๐ FastLlama Series is Live!
๐ฆพ Experience faster, lighter, and smarter language models! The new FastLlama makes Meta's LLaMA models work with smaller file sizes, lower system requirements, and higher performance. The model supports 8 languages, including English, German, and Spanish.
๐ค Built on the LLaMA 3.2-1B-Instruct model, fine-tuned with Hugging Face's SmolTalk and MetaMathQA-50k datasets, and powered by LoRA (Low-Rank Adaptation) for groundbreaking mathematical reasoning.
๐ป Its compact size makes it versatile for a wide range of applications!
๐ฌ Chat with the model:
๐ Chat Link: suayptalha/Chat-with-FastLlama
๐ Model Link: suayptalha/FastLlama-3.2-1B-Instruct
๐ฆพ Experience faster, lighter, and smarter language models! The new FastLlama makes Meta's LLaMA models work with smaller file sizes, lower system requirements, and higher performance. The model supports 8 languages, including English, German, and Spanish.
๐ค Built on the LLaMA 3.2-1B-Instruct model, fine-tuned with Hugging Face's SmolTalk and MetaMathQA-50k datasets, and powered by LoRA (Low-Rank Adaptation) for groundbreaking mathematical reasoning.
๐ป Its compact size makes it versatile for a wide range of applications!
๐ฌ Chat with the model:
๐ Chat Link: suayptalha/Chat-with-FastLlama
๐ Model Link: suayptalha/FastLlama-3.2-1B-Instruct
Post
2316
After 6 years, BERT, the workhorse of encoder models, finally gets a replacement: ๐ช๐ฒ๐น๐ฐ๐ผ๐บ๐ฒ ๐ ๐ผ๐ฑ๐ฒ๐ฟ๐ป๐๐๐ฅ๐ง! ๐ค
We talk a lot about โจGenerative AIโจ, meaning "Decoder version of the Transformers architecture", but this is only one of the ways to build LLMs: encoder models, that turn a sentence in a vector, are maybe even more widely used in industry than generative models.
The workhorse for this category has been BERT since its release in 2018 (that's prehistory for LLMs).
It's not a fancy 100B parameters supermodel (just a few hundred millions), but it's an excellent workhorse, kind of a Honda Civic for LLMs.
Many applications use BERT-family models - the top models in this category cumulate millions of downloads on the Hub.
โก๏ธ Now a collaboration between Answer.AI and LightOn just introduced BERT's replacement: ModernBERT.
๐ง๐;๐๐ฅ:
๐๏ธ Architecture changes:
โ First, standard modernizations:
- Rotary positional embeddings (RoPE)
- Replace GeLU with GeGLU,
- Use Flash Attention 2
โจ The team also introduced innovative techniques like alternating attention instead of full attention, and sequence packing to get rid of padding overhead.
๐ฅ As a result, the model tops the game of encoder models:
It beats previous standard DeBERTaV3 for 1/5th the memory footprint, and runs 4x faster!
Read the blog post ๐ https://huggingface.co/blog/modernbert
We talk a lot about โจGenerative AIโจ, meaning "Decoder version of the Transformers architecture", but this is only one of the ways to build LLMs: encoder models, that turn a sentence in a vector, are maybe even more widely used in industry than generative models.
The workhorse for this category has been BERT since its release in 2018 (that's prehistory for LLMs).
It's not a fancy 100B parameters supermodel (just a few hundred millions), but it's an excellent workhorse, kind of a Honda Civic for LLMs.
Many applications use BERT-family models - the top models in this category cumulate millions of downloads on the Hub.
โก๏ธ Now a collaboration between Answer.AI and LightOn just introduced BERT's replacement: ModernBERT.
๐ง๐;๐๐ฅ:
๐๏ธ Architecture changes:
โ First, standard modernizations:
- Rotary positional embeddings (RoPE)
- Replace GeLU with GeGLU,
- Use Flash Attention 2
โจ The team also introduced innovative techniques like alternating attention instead of full attention, and sequence packing to get rid of padding overhead.
๐ฅ As a result, the model tops the game of encoder models:
It beats previous standard DeBERTaV3 for 1/5th the memory footprint, and runs 4x faster!
Read the blog post ๐ https://huggingface.co/blog/modernbert
- 1 reply
Post
2477
๐๐ฎ๐ ๐ ๐ข๐ง๐ ๐
๐๐๐ ๐ซ๐๐ฅ๐๐๐ฌ๐๐ฌ ๐๐ข๐๐จ๐ญ๐ซ๐จ๐ง, ๐ ๐ฆ๐ข๐๐ซ๐จ๐ฌ๐๐จ๐ฉ๐ข๐ ๐ฅ๐ข๐ ๐ญ๐ก๐๐ญ ๐ฌ๐จ๐ฅ๐ฏ๐๐ฌ ๐๐๐ ๐ญ๐ซ๐๐ข๐ง๐ข๐ง๐ ๐๐ ๐ฉ๐๐ซ๐๐ฅ๐ฅ๐๐ฅ๐ข๐ณ๐๐ญ๐ข๐จ๐ง ๐ฅณ
๐ฐ๏ธ Llama-3.1-405B took 39 million GPU-hours to train, i.e. about 4.5 thousand years.
๐ด๐ป If they had needed all this time, we would have GPU stories from the time of Pharaoh ๐: "Alas, Lord of Two Lands, the shipment of counting-stones arriving from Cathay was lost to pirates, this shall delay the building of your computing temple by many moons "
๐ ๏ธ But instead, they just parallelized the training on 24k H100s, which made it take just a few months.
This required parallelizing across 4 dimensions: data, tensor, context, pipeline.
And it is infamously hard to do, making for bloated code repos that hold together only by magic.
๐ค ๐๐๐ ๐ป๐ผ๐ ๐๐ฒ ๐ฑ๐ผ๐ป'๐ ๐ป๐ฒ๐ฒ๐ฑ ๐ต๐๐ด๐ฒ ๐ฟ๐ฒ๐ฝ๐ผ๐ ๐ฎ๐ป๐๐บ๐ผ๐ฟ๐ฒ! Instead of building mega-training codes, Hugging Face colleagues cooked in the other direction, towards tiny 4D parallelism libs. A team has built Nanotron, already widely used in industry.
And now a team releases Picotron, a radical approach to code 4D Parallelism in just a few hundred lines of code, a real engineering prowess, making it much easier to understand what's actually happening!
โก ๐๐'๐ ๐๐ถ๐ป๐, ๐๐ฒ๐ ๐ฝ๐ผ๐๐ฒ๐ฟ๐ณ๐๐น:
Counting in MFU (Model FLOPs Utilization, how much the model actually uses all the compute potential), this lib reaches ~50% on SmolLM-1.7B model with 8 H100 GPUs, which is really close to what huge libs would reach. (Caution: the team is leading further benchmarks to verify this)
Go take a look ๐ https://github.com/huggingface/picotron/tree/main/picotron
๐ฐ๏ธ Llama-3.1-405B took 39 million GPU-hours to train, i.e. about 4.5 thousand years.
๐ด๐ป If they had needed all this time, we would have GPU stories from the time of Pharaoh ๐: "Alas, Lord of Two Lands, the shipment of counting-stones arriving from Cathay was lost to pirates, this shall delay the building of your computing temple by many moons "
๐ ๏ธ But instead, they just parallelized the training on 24k H100s, which made it take just a few months.
This required parallelizing across 4 dimensions: data, tensor, context, pipeline.
And it is infamously hard to do, making for bloated code repos that hold together only by magic.
๐ค ๐๐๐ ๐ป๐ผ๐ ๐๐ฒ ๐ฑ๐ผ๐ป'๐ ๐ป๐ฒ๐ฒ๐ฑ ๐ต๐๐ด๐ฒ ๐ฟ๐ฒ๐ฝ๐ผ๐ ๐ฎ๐ป๐๐บ๐ผ๐ฟ๐ฒ! Instead of building mega-training codes, Hugging Face colleagues cooked in the other direction, towards tiny 4D parallelism libs. A team has built Nanotron, already widely used in industry.
And now a team releases Picotron, a radical approach to code 4D Parallelism in just a few hundred lines of code, a real engineering prowess, making it much easier to understand what's actually happening!
โก ๐๐'๐ ๐๐ถ๐ป๐, ๐๐ฒ๐ ๐ฝ๐ผ๐๐ฒ๐ฟ๐ณ๐๐น:
Counting in MFU (Model FLOPs Utilization, how much the model actually uses all the compute potential), this lib reaches ~50% on SmolLM-1.7B model with 8 H100 GPUs, which is really close to what huge libs would reach. (Caution: the team is leading further benchmarks to verify this)
Go take a look ๐ https://github.com/huggingface/picotron/tree/main/picotron
- 1 reply
Post
1886
Coming back to Paris Friday to open our new Hugging Face office!
We're at capacity for the party but add your name in the waiting list as we're trying to privatize the passage du Caire for extra space for robots ๐ค๐ฆพ๐ฆฟ
https://t.co/enkFXjWndJ
We're at capacity for the party but add your name in the waiting list as we're trying to privatize the passage du Caire for extra space for robots ๐ค๐ฆพ๐ฆฟ
https://t.co/enkFXjWndJ
- 1 reply
Post
2207
๐ฃ๐ผ๐๐ฒ๐ป๐๐ถ๐ฎ๐น ๐ฝ๐ฎ๐ฟ๐ฎ๐ฑ๐ถ๐ด๐บ ๐๐ต๐ถ๐ณ๐ ๐ถ๐ป ๐๐๐ ๐: ๐ป๐ฒ๐ ๐ฝ๐ฎ๐ฝ๐ฒ๐ฟ ๐ฏ๐ ๐ ๐ฒ๐๐ฎ ๐ฐ๐น๐ฎ๐ถ๐บ๐ ๐๐ต๐ฎ๐ ๐๐ฒ ๐ฐ๐ฎ๐ป ๐ด๐ฒ๐ ๐ฟ๐ถ๐ฑ ๐ผ๐ณ ๐๐ผ๐ธ๐ฒ๐ป๐ถ๐๐ฒ๐ฟ๐! ๐ฅณ
Current LLMs process text by first splitting it into tokens. They use a module named "tokenizer", that -spl-it-s- th-e- te-xt- in-to- arbitrary tokens depending on a fixed dictionnary.
On the Hub you can find this dictionary in a model's files under tokenizer.json.
โก๏ธ This process is called BPE tokenization. It is suboptimal, everyone says it. It breaks text into predefined chunks that often fail to capture the nuance of language. But it has been a necessary evil in language models since their inception.
๐ฅ In Byte Latent Transformer (BLT), Meta researchers propose an elegant solution by eliminating tokenization entirely, working directly with raw bytes while maintaining efficiency through dynamic "patches."
This had been tried before with different byte-level tokenizations, but it's the first time that an architecture of this type scales as well as BPE tokenization. And it could mean a real paradigm shift! ๐๐
๐๏ธ ๐๐ฟ๐ฐ๐ต๐ถ๐๐ฒ๐ฐ๐๐๐ฟ๐ฒ:
Instead of a lightweight tokenizer, BLT has a lightweight encoder that process raw bytes into patches. Then the patches are processed by the main heavy-duty transformers as we do normally (but for patches of bytes instead of tokens), before converting back to bytes.
๐งฉ ๐๐๐ป๐ฎ๐บ๐ถ๐ฐ ๐ฃ๐ฎ๐๐ฐ๐ต๐ถ๐ป๐ด:
Instead of fixed tokens, BLT groups bytes based on their predictability (measured by entropy) - using more compute for complex sequences and efficiently handling simple ones. This allows efficient processing while maintaining byte-level understanding.
I hope this breakthrough is confirmed and we can get rid of all the tokenizer stuff, it will make model handling easier!
Read their paper here ๐ https://dl.fbaipublicfiles.com/blt/BLT__Patches_Scale_Better_Than_Tokens.pdf
Current LLMs process text by first splitting it into tokens. They use a module named "tokenizer", that -spl-it-s- th-e- te-xt- in-to- arbitrary tokens depending on a fixed dictionnary.
On the Hub you can find this dictionary in a model's files under tokenizer.json.
โก๏ธ This process is called BPE tokenization. It is suboptimal, everyone says it. It breaks text into predefined chunks that often fail to capture the nuance of language. But it has been a necessary evil in language models since their inception.
๐ฅ In Byte Latent Transformer (BLT), Meta researchers propose an elegant solution by eliminating tokenization entirely, working directly with raw bytes while maintaining efficiency through dynamic "patches."
This had been tried before with different byte-level tokenizations, but it's the first time that an architecture of this type scales as well as BPE tokenization. And it could mean a real paradigm shift! ๐๐
๐๏ธ ๐๐ฟ๐ฐ๐ต๐ถ๐๐ฒ๐ฐ๐๐๐ฟ๐ฒ:
Instead of a lightweight tokenizer, BLT has a lightweight encoder that process raw bytes into patches. Then the patches are processed by the main heavy-duty transformers as we do normally (but for patches of bytes instead of tokens), before converting back to bytes.
๐งฉ ๐๐๐ป๐ฎ๐บ๐ถ๐ฐ ๐ฃ๐ฎ๐๐ฐ๐ต๐ถ๐ป๐ด:
Instead of fixed tokens, BLT groups bytes based on their predictability (measured by entropy) - using more compute for complex sequences and efficiently handling simple ones. This allows efficient processing while maintaining byte-level understanding.
I hope this breakthrough is confirmed and we can get rid of all the tokenizer stuff, it will make model handling easier!
Read their paper here ๐ https://dl.fbaipublicfiles.com/blt/BLT__Patches_Scale_Better_Than_Tokens.pdf
- 2 replies
Post
2596
๐ฅ ๐๐ผ๐ผ๐ด๐น๐ฒ ๐ฟ๐ฒ๐น๐ฒ๐ฎ๐๐ฒ๐ ๐๐ฒ๐บ๐ถ๐ป๐ถ ๐ฎ.๐ฌ, ๐๐๐ฎ๐ฟ๐๐ถ๐ป๐ด ๐๐ถ๐๐ต ๐ฎ ๐๐น๐ฎ๐๐ต ๐บ๐ผ๐ฑ๐ฒ๐น ๐๐ต๐ฎ๐ ๐๐๐ฒ๐ฎ๐บ๐ฟ๐ผ๐น๐น๐ ๐๐ฃ๐ง-๐ฐ๐ผ ๐ฎ๐ป๐ฑ ๐๐น๐ฎ๐๐ฑ๐ฒ-๐ฏ.๐ฒ ๐ฆ๐ผ๐ป๐ป๐ฒ๐! And they start a huge effort on agentic capabilities.
๐ The performance improvements are crazy for such a fast model:
โฃ Gemini 2.0 Flash outperforms the previous 1.5 Pro model at twice the speed
โฃ Now supports both input AND output of images, video, audio and text
โฃ Can natively use tools like Google Search and execute code
โก๏ธ If the price is on par with previous Flash iteration ($0.30 / M tokens, to compare with GPT-4o's $1.25) the competition will have a big problem with this 4x cheaper model that gets better benchmarks ๐คฏ
๐ค What about the agentic capabilities?
โฃ Project Astra: A universal AI assistant that can use Google Search, Lens and Maps
โฃ Project Mariner: A Chrome extension that can complete complex web tasks (83.5% success rate on WebVoyager benchmark, this is really impressive!)
โฃ Jules: An AI coding agent that integrates with GitHub workflows
I'll be eagerly awaiting further news from Google!
Read their blogpost here ๐ https://blog.google/technology/google-deepmind/google-gemini-ai-update-december-2024/
๐ The performance improvements are crazy for such a fast model:
โฃ Gemini 2.0 Flash outperforms the previous 1.5 Pro model at twice the speed
โฃ Now supports both input AND output of images, video, audio and text
โฃ Can natively use tools like Google Search and execute code
โก๏ธ If the price is on par with previous Flash iteration ($0.30 / M tokens, to compare with GPT-4o's $1.25) the competition will have a big problem with this 4x cheaper model that gets better benchmarks ๐คฏ
๐ค What about the agentic capabilities?
โฃ Project Astra: A universal AI assistant that can use Google Search, Lens and Maps
โฃ Project Mariner: A Chrome extension that can complete complex web tasks (83.5% success rate on WebVoyager benchmark, this is really impressive!)
โฃ Jules: An AI coding agent that integrates with GitHub workflows
I'll be eagerly awaiting further news from Google!
Read their blogpost here ๐ https://blog.google/technology/google-deepmind/google-gemini-ai-update-december-2024/
Post
1818
๐๐๐๐ฅ๐ข๐ง๐ ๐ฅ๐๐ฐ๐ฌ ๐๐ซ๐ ๐ง๐จ๐ญ ๐๐๐๐ ๐ฒ๐๐ญ! New blog post suggests Anthropic might have an extremely strong Opus-3.5 already available, but is not releasing it to keep their edge over the competition. ๐ง
โSince the release of Opus-3.5 has been delayed indefinitely, there have been lots of rumors and articles about LLMs plateauing. Scaling laws, the main powering factor of the LLM competence increase, could have stopped, according to these rumors, being the cause of this stalling of progress.
These rumors were quickly denied by many people at the leading LLM labs, including OpenAI and Anthropic. But these people would be expected to hype the future of LLMs even if scaling laws really plateaued, so the jury is still out.
๐๏ธ This new article by Semianalysis (generally a good source, specifically on hardware) provides a counter-rumor that I find more convincing:
โก๏ธ Maybe scaling laws still work, Opus-3.5 is ready and as good as planned, but they just don't release it because the synthetic data it helps provide can bring cheaper/smaller models Claude and Haiku up in performance, without risking to leak this precious high-quality synthetic data to competitors.
Time will tell! I feel like we'll know more soon.
Read the article: https://semianalysis.com/2024/12/11/scaling-laws-o1-pro-architecture-reasoning-infrastructure-orion-and-claude-3-5-opus-failures/
โSince the release of Opus-3.5 has been delayed indefinitely, there have been lots of rumors and articles about LLMs plateauing. Scaling laws, the main powering factor of the LLM competence increase, could have stopped, according to these rumors, being the cause of this stalling of progress.
These rumors were quickly denied by many people at the leading LLM labs, including OpenAI and Anthropic. But these people would be expected to hype the future of LLMs even if scaling laws really plateaued, so the jury is still out.
๐๏ธ This new article by Semianalysis (generally a good source, specifically on hardware) provides a counter-rumor that I find more convincing:
โก๏ธ Maybe scaling laws still work, Opus-3.5 is ready and as good as planned, but they just don't release it because the synthetic data it helps provide can bring cheaper/smaller models Claude and Haiku up in performance, without risking to leak this precious high-quality synthetic data to competitors.
Time will tell! I feel like we'll know more soon.
Read the article: https://semianalysis.com/2024/12/11/scaling-laws-o1-pro-architecture-reasoning-infrastructure-orion-and-claude-3-5-opus-failures/
- 1 reply
Post
2239
Last week was crazy in OS AI, with important models and datasets releases every day.
Here are the most important ones I've pinned:
๐ Cohere relased GLobal-MMLU, a multilingual version of MMLU, to evaluate AI models' world knowledge in many languages!
๐ฆ Meta released Llama-3.3-70B-Instruct, a 70B model that's on par with Llama-3.1-405B-Instruct, GPT-4o and Claude. Probably my new go-to for agentic workflows.
๐ FishAudio released fish-speech-1.5, multilingual text to speech model
๐จ Microsoft Research released TRELLIS, an extremely impressive image-to-3D model, which you can try here: JeffreyXiang/TRELLIS
๐ Yesterday, Hugging Face release FineWeb 2, a new version that extends the previous FineWeb to over 1000 languages, including extended coverage in Russina, Mandarin, German, Japanese, Spanish, French, so a huge, high-quality dataset of > 3 trillion words! HuggingFaceFW/fineweb-2
Now let's go build to make this week as productive as last one!
Here are the most important ones I've pinned:
๐ Cohere relased GLobal-MMLU, a multilingual version of MMLU, to evaluate AI models' world knowledge in many languages!
๐ฆ Meta released Llama-3.3-70B-Instruct, a 70B model that's on par with Llama-3.1-405B-Instruct, GPT-4o and Claude. Probably my new go-to for agentic workflows.
๐ FishAudio released fish-speech-1.5, multilingual text to speech model
๐จ Microsoft Research released TRELLIS, an extremely impressive image-to-3D model, which you can try here: JeffreyXiang/TRELLIS
๐ Yesterday, Hugging Face release FineWeb 2, a new version that extends the previous FineWeb to over 1000 languages, including extended coverage in Russina, Mandarin, German, Japanese, Spanish, French, so a huge, high-quality dataset of > 3 trillion words! HuggingFaceFW/fineweb-2
Now let's go build to make this week as productive as last one!