Aymeric Roucher's picture

Aymeric Roucher

m-ric

AI & ML interests

Leading Agents at Hugging Face ๐Ÿค—

Recent Activity

updated a Space about 7 hours ago
m-ric/hf-model-downloads
updated a dataset 2 days ago
huggingface/documentation-images
updated a dataset 2 days ago
m-ric/agents_medium_benchmark
View all activity

Articles

Organizations

Hugging Face's profile picture Atmos Bank's profile picture Hugging Test Lab's profile picture Tools's profile picture HuggingFaceM4's profile picture lecocqassociate's profile picture huggingPartyParis's profile picture Supreme's profile picture Propulse Lab's profile picture FactSet's profile picture Leaderboard Organization's profile picture FactSet's profile picture CGIAR's profile picture Aperture Laboratories's profile picture AI Energy Score Project's profile picture C&A's profile picture Social Post Explorers's profile picture Dev Mode Explorers's profile picture Agent Collab's profile picture SLLHF's profile picture Data Agents's profile picture Hugging Face Party @ PyTorch Conference's profile picture Nerdy Face's profile picture Hugging Face Science's profile picture Agents Leaderboard's profile picture

m-ric's activity

posted an update 6 days ago
view post
Post
1632
After 6 years, BERT, the workhorse of encoder models, finally gets a replacement: ๐—ช๐—ฒ๐—น๐—ฐ๐—ผ๐—บ๐—ฒ ๐— ๐—ผ๐—ฑ๐—ฒ๐—ฟ๐—ป๐—•๐—˜๐—ฅ๐—ง! ๐Ÿค—

We talk a lot about โœจGenerative AIโœจ, 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.

โžก๏ธ Now a collaboration between Answer.AI and LightOn just introduced BERT's replacement: ModernBERT.

๐—ง๐—Ÿ;๐——๐—ฅ:
๐Ÿ›๏ธ Architecture changes:
โ‡’ First, standard modernizations:
- Rotary positional embeddings (RoPE)
- Replace GeLU with GeGLU,
- Use Flash Attention 2
โœจ The team also introduced innovative techniques like alternating attention instead of full attention, and sequence packing to get rid of padding overhead.

๐Ÿฅ‡ 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 ๐Ÿ‘‰ https://huggingface.co./blog/modernbert
  • 1 reply
ยท
posted an update 7 days ago
view post
Post
2028
๐‡๐ฎ๐ ๐ ๐ข๐ง๐  ๐…๐š๐œ๐ž ๐ซ๐ž๐ฅ๐ž๐š๐ฌ๐ž๐ฌ ๐๐ข๐œ๐จ๐ญ๐ซ๐จ๐ง, ๐š ๐ฆ๐ข๐œ๐ซ๐จ๐ฌ๐œ๐จ๐ฉ๐ข๐œ ๐ฅ๐ข๐› ๐ญ๐ก๐š๐ญ ๐ฌ๐จ๐ฅ๐ฏ๐ž๐ฌ ๐‹๐‹๐Œ ๐ญ๐ซ๐š๐ข๐ง๐ข๐ง๐  ๐Ÿ’๐ƒ ๐ฉ๐š๐ซ๐š๐ฅ๐ฅ๐ž๐ฅ๐ข๐ณ๐š๐ญ๐ข๐จ๐ง ๐Ÿฅณ

๐Ÿ•ฐ๏ธ Llama-3.1-405B took 39 million GPU-hours to train, i.e. about 4.5 thousand years.

๐Ÿ‘ด๐Ÿป 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 "

๐Ÿ› ๏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.

๐Ÿค ๐—•๐˜‚๐˜ ๐—ป๐—ผ๐˜„ ๐˜„๐—ฒ ๐—ฑ๐—ผ๐—ป'๐˜ ๐—ป๐—ฒ๐—ฒ๐—ฑ ๐—ต๐˜‚๐—ด๐—ฒ ๐—ฟ๐—ฒ๐—ฝ๐—ผ๐˜€ ๐—ฎ๐—ป๐˜†๐—บ๐—ผ๐—ฟ๐—ฒ! 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!

โšก ๐—œ๐˜'๐˜€ ๐˜๐—ถ๐—ป๐˜†, ๐˜†๐—ฒ๐˜ ๐—ฝ๐—ผ๐˜„๐—ฒ๐—ฟ๐—ณ๐˜‚๐—น:
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 ๐Ÿ‘‰ https://github.com/huggingface/picotron/tree/main/picotron
  • 1 reply
ยท
posted an update 12 days ago
view post
Post
2155
๐—ฃ๐—ผ๐˜๐—ฒ๐—ป๐˜๐—ถ๐—ฎ๐—น ๐—ฝ๐—ฎ๐—ฟ๐—ฎ๐—ฑ๐—ถ๐—ด๐—บ ๐˜€๐—ต๐—ถ๐—ณ๐˜ ๐—ถ๐—ป ๐—Ÿ๐—Ÿ๐— ๐˜€: ๐—ป๐—ฒ๐˜„ ๐—ฝ๐—ฎ๐—ฝ๐—ฒ๐—ฟ ๐—ฏ๐˜† ๐— ๐—ฒ๐˜๐—ฎ ๐—ฐ๐—น๐—ฎ๐—ถ๐—บ๐˜€ ๐˜๐—ต๐—ฎ๐˜ ๐˜„๐—ฒ ๐—ฐ๐—ฎ๐—ป ๐—ด๐—ฒ๐˜ ๐—ฟ๐—ถ๐—ฑ ๐—ผ๐—ณ ๐˜๐—ผ๐—ธ๐—ฒ๐—ป๐—ถ๐˜‡๐—ฒ๐—ฟ๐˜€! ๐Ÿฅณ

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.

โžก๏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.

๐Ÿ’ฅ 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! ๐Ÿ‘๐Ÿ‘

๐Ÿ—๏ธ ๐—”๐—ฟ๐—ฐ๐—ต๐—ถ๐˜๐—ฒ๐—ฐ๐˜๐˜‚๐—ฟ๐—ฒ:
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.

๐Ÿงฉ ๐——๐˜†๐—ป๐—ฎ๐—บ๐—ถ๐—ฐ ๐—ฃ๐—ฎ๐˜๐—ฐ๐—ต๐—ถ๐—ป๐—ด:
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 ๐Ÿ‘‰ https://dl.fbaipublicfiles.com/blt/BLT__Patches_Scale_Better_Than_Tokens.pdf
  • 2 replies
ยท
posted an update 14 days ago
view post
Post
2420
๐Ÿ’ฅ ๐—š๐—ผ๐—ผ๐—ด๐—น๐—ฒ ๐—ฟ๐—ฒ๐—น๐—ฒ๐—ฎ๐˜€๐—ฒ๐˜€ ๐—š๐—ฒ๐—บ๐—ถ๐—ป๐—ถ ๐Ÿฎ.๐Ÿฌ, ๐˜€๐˜๐—ฎ๐—ฟ๐˜๐—ถ๐—ป๐—ด ๐˜„๐—ถ๐˜๐—ต ๐—ฎ ๐—™๐—น๐—ฎ๐˜€๐—ต ๐—บ๐—ผ๐—ฑ๐—ฒ๐—น ๐˜๐—ต๐—ฎ๐˜ ๐˜€๐˜๐—ฒ๐—ฎ๐—บ๐—ฟ๐—ผ๐—น๐—น๐˜€ ๐—š๐—ฃ๐—ง-๐Ÿฐ๐—ผ ๐—ฎ๐—ป๐—ฑ ๐—–๐—น๐—ฎ๐˜‚๐—ฑ๐—ฒ-๐Ÿฏ.๐Ÿฒ ๐—ฆ๐—ผ๐—ป๐—ป๐—ฒ๐˜! And they start a huge effort on agentic capabilities.

๐Ÿš€ The performance improvements are crazy for such a fast model:
โ€ฃ Gemini 2.0 Flash outperforms the previous 1.5 Pro model at twice the speed
โ€ฃ Now supports both input AND output of images, video, audio and text
โ€ฃ Can natively use tools like Google Search and execute code

โžก๏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 ๐Ÿคฏ

๐Ÿค– What about the agentic capabilities?

โ€ฃ Project Astra: A universal AI assistant that can use Google Search, Lens and Maps
โ€ฃ Project Mariner: A Chrome extension that can complete complex web tasks (83.5% success rate on WebVoyager benchmark, this is really impressive!)
โ€ฃ Jules: An AI coding agent that integrates with GitHub workflows

I'll be eagerly awaiting further news from Google!

Read their blogpost here ๐Ÿ‘‰ https://blog.google/technology/google-deepmind/google-gemini-ai-update-december-2024/
posted an update 15 days ago
view post
Post
1791
๐’๐œ๐š๐ฅ๐ข๐ง๐  ๐ฅ๐š๐ฐ๐ฌ ๐š๐ซ๐ž ๐ง๐จ๐ญ ๐๐ž๐š๐ ๐ฒ๐ž๐ญ! 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. ๐Ÿง

โ“Since 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.

๐Ÿ—ž๏ธ This new article by Semianalysis (generally a good source, specifically on hardware) provides a counter-rumor that I find more convincing:

โžก๏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
ยท
reacted to julien-c's post with โค๏ธ๐Ÿ”ฅ 15 days ago
view post
Post
7611
After some heated discussion ๐Ÿ”ฅ, we clarify our intent re. storage limits on the Hub

TL;DR:
- public storage is free, and (unless blatant abuse) unlimited. We do ask that you consider upgrading to PRO and/or Enterprise Hub if possible
- private storage is paid above a significant free tier (1TB if you have a paid account, 100GB otherwise)

docs: https://huggingface.co./docs/hub/storage-limits

We optimize our infrastructure continuously to scale our storage for the coming years of growth in Machine learning, to the benefit of the community ๐Ÿ”ฅ

cc: @reach-vb @pierric @victor and the HF team
ยท
posted an update 16 days ago
view post
Post
2221
Last week was crazy in OS AI, with important models and datasets releases every day.

Here are the most important ones I've pinned:

๐ŸŒŽ Cohere relased GLobal-MMLU, a multilingual version of MMLU, to evaluate AI models' world knowledge in many languages!

๐Ÿฆ™ 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.

๐Ÿ”‰ FishAudio released fish-speech-1.5, multilingual text to speech model

๐ŸŽจ Microsoft Research released TRELLIS, an extremely impressive image-to-3D model, which you can try here: JeffreyXiang/TRELLIS

๐Ÿ“š 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!
reacted to merve's post with โค๏ธ 17 days ago
view post
Post
5508
This week in open-source AI was insane ๐Ÿค  A small recap๐Ÿ•บ๐Ÿป merve/dec-6-releases-67545caebe9fc4776faac0a3

Multimodal ๐Ÿ–ผ๏ธ
> Google shipped a PaliGemma 2, new iteration of PaliGemma with more sizes: 3B, 10B and 28B, with pre-trained and captioning variants ๐Ÿ‘
> OpenGVLab released InternVL2, seven new vision LMs in different sizes, with sota checkpoint with MIT license โœจ
> Qwen team at Alibaba released the base models of Qwen2VL models with 2B, 7B and 72B ckpts

LLMs ๐Ÿ’ฌ
> Meta released a new iteration of Llama 70B, Llama3.2-70B trained further
> EuroLLM-9B-Instruct is a new multilingual LLM for European languages with Apache 2.0 license ๐Ÿ”ฅ
> Dataset: CohereForAI released GlobalMMLU, multilingual version of MMLU with 42 languages with Apache 2.0 license
> Dataset: QwQ-LongCoT-130K is a new dataset to train reasoning models
> Dataset: FineWeb2 just landed with multilinguality update! ๐Ÿ”ฅ nearly 8TB pretraining data in many languages!

Image/Video Generation ๐Ÿ–ผ๏ธ
> Tencent released HunyuanVideo, a new photorealistic video generation model
> OminiControl is a new editing/control framework for image generation models like Flux

Audio ๐Ÿ”Š
> Indic-Parler-TTS is a new text2speech model made by community
posted an update 22 days ago
view post
Post
1468
๐—ฆ๐—ต๐—ผ๐˜„๐—จ๐—œ: ๐—ฎ ๐˜€๐—บ๐—ฎ๐—น๐—น ๐—ฒ๐—ป๐—ฑ-๐˜๐—ผ-๐—ฒ๐—ป๐—ฑ ๐—ฎ๐—ด๐—ฒ๐—ป๐˜ ๐˜๐—ต๐—ฎ๐˜ ๐—ฐ๐—ฎ๐—ป ๐—ป๐—ฎ๐˜ƒ๐—ถ๐—ด๐—ฎ๐˜๐—ฒ ๐—ฎ๐—ป๐˜† ๐—จ๐—œ ๐—ฎ๐—ป๐—ฑ ๐—ผ๐˜‚๐˜๐—ฝ๐—ฒ๐—ฟ๐—ณ๐—ผ๐—ฟ๐—บ๐˜€ ๐—บ๐˜‚๐—ฐ๐—ต ๐—ฏ๐—ถ๐—ด๐—ด๐—ฒ๐—ฟ ๐˜€๐˜†๐˜€๐˜๐—ฒ๐—บ๐˜€! ๐Ÿ“ฒ

A team from NUS and Microsoft just released an agent that can act on any UI (Desktop, Android, Web) without needing additional text information. It works extremely well : they applied their method on a tiny Qwen2-VL-2B, and they managed to beat methods that use either much more powerful vision models (like GPT-4V) without using any additional info (e.g. leveraging the DOM of a webpage) like previous methods did ! ๐Ÿ‘๐Ÿ‘

They started from the idea that most existing methods rely heavily on text, which makes them less generalizable, while letting aside rich UI structure that user actually rely on when navigating this interfaces.

โš™๏ธ They put several good ideas to work:

๐Ÿ’ก Simplify screenshots to the max:
They prune a lot the heavy visual content of UI screenshots, by removing cloned image patches (like any vast patch of the same color will be reduced to a small patch, while maintaining positional embeddings), then group patches from the same GUI elements together to simplify even further

๐Ÿ’ก Build a truly generalist dataset:
To train a general UI agent, you need trajectories from each possible UI, and express them in a common language. Authors merge datasets like OmniAct for Desktop, Mind2Web for websites, AMEX for Android trajectories to create a high-quality and diverse dataset.

โžก๏ธ Nice results ensued:
They fine-tune a tiny Qwen-2-VL-2B on their method, and it reaches SOTA on several task (element identification, web navigation), even beating methods that either use additional info from the DOM or use much bigger VLMS like GPT-4v! ๐Ÿ†

And performance could certainly jump with a slightly bigger vision model. Let's hope the community builds this soon! ๐Ÿš€

Paper added to my "Agents" collection ๐Ÿ‘‰ m-ric/agents-65ba776fbd9e29f771c07d4e
posted an update 22 days ago
view post
Post
1207
Need a measurement for traction of a GitHub repo, a more reliable one than Github star history? (which is a bit too hype-driven) ๐Ÿ“ˆ

โžก๏ธ I've made a Space to visualize PyPI downloads.

Try it here ๐Ÿ‘‰ m-ric/package-download-history
  • 1 reply
ยท
posted an update 23 days ago
view post
Post
1269
๐Ÿค– ๐—”๐—ฑ๐—ผ๐—ฏ๐—ฒ'๐˜€ ๐—ฐ๐—ผ๐—ฑ๐—ฒ-๐—ด๐—ฒ๐—ป๐—ฒ๐—ฟ๐—ฎ๐˜๐—ถ๐—ป๐—ด ๐—ฎ๐—ด๐—ฒ๐—ป๐˜ ๐—ฟ๐—ฒ๐—ฎ๐—ฐ๐—ต๐—ฒ๐˜€ ๐˜๐—ต๐—ฒ ๐˜๐—ผ๐—ฝ ๐—ผ๐—ณ ๐—š๐—”๐—œ๐—” ๐—น๐—ฒ๐—ฎ๐—ฑ๐—ฒ๐—ฟ๐—ฏ๐—ผ๐—ฎ๐—ฟ๐—ฑ - and their paper cites my work!

๐Ÿ’ก Reminder:ย In short, Agentic systems are a vehicle in which you put your LLM to allow it access to the outside world.

โžก๏ธ The team of researchers at Adobe started from the idea that current agentic systems lack the ability to define their own tools. So they decided to make an agent that writes actions as code, thus allowing it to write python functions that can be re-used later as tools!

Here's what the LLM generations can look like with the proper prompt:

Thought: I need to access the excel file using a different method.
Action:
def access_excel_file(file_path)
	... # rest of the code (the agent does writes it, but I don't have room in this post)
	return rows


Then your system executes this and appends the observation to the agent's memory.

Why is this code formulation better than classical tool use formulation as JSON? The paper explains:

"Most existing work uses text or JSON as the representation of actions, which significantly lacks the two criteria mentioned earlier: generality and composability. In contrast, DynaSaur can utilize available actions or create new ones if necessary, using code as a unified representation. In principle, acting with code enables agents to solve any Turing-complete problem."

The idea of using code is not new: in fact, we do it in transformers.agents (thus the citation that I got). They implementation adds further refinements, like using RAG to retrieve relevant functions before generating an action, which increases performance further.

And they observe that code agents perform much better, reaching the top of GAIA leaderboard! ๐Ÿฅ‡

Go take a look, it's really clear and informative!

Paper added to my agents collection ๐Ÿ‘‰ m-ric/agents-65ba776fbd9e29f771c07d4e
posted an update 26 days ago
view post
Post
2376
Single most important thing to do today: ๐—ด๐—ผ ๐˜๐—ฟ๐˜† ๐—ค๐˜„๐—ค ๐—ผ๐—ป ๐—›๐˜‚๐—ด๐—ด๐—ถ๐—ป๐—ด ๐—–๐—ต๐—ฎ๐˜!

๐Ÿ‘‰ https://huggingface.co./chat/models/Qwen/QwQ-32B-Preview
  • 2 replies
ยท
posted an update about 1 month ago
view post
Post
1079
๐Ÿ—ž๏ธ ๐—ฆ๐˜๐—ฎ๐˜๐—ฒ ๐—ผ๐—ณ ๐—˜๐—ป๐˜๐—ฒ๐—ฟ๐—ฝ๐—ฟ๐—ถ๐˜€๐—ฒ ๐—”๐—œ ๐Ÿฎ๐Ÿฌ๐Ÿฎ๐Ÿฐ: ๐—”๐—ป๐˜๐—ต๐—ฟ๐—ผ๐—ฝ๐—ถ๐—ฐ ๐—ฒ๐—ฎ๐˜๐—ถ๐—ป๐—ด ๐˜‚๐—ฝ ๐—ข๐—ฝ๐—ฒ๐—ป๐—”๐—œ, ๐—”๐—ด๐—ฒ๐—ป๐˜๐˜€ ๐—ฟ๐—ฎ๐—บ๐—ฝ ๐˜‚๐—ฝ ๐˜๐—ผ ๐Ÿญ๐Ÿฎ% ๐—ผ๐—ณ ๐˜‚๐˜€๐—ฒ-๐—ฐ๐—ฎ๐˜€๐—ฒ๐˜€, ๐—ผ๐—ฝ๐—ฒ๐—ป ๐—บ๐—ผ๐—ฑ๐—ฒ๐—น๐˜€ ๐—บ๐—ฎ๐—ธ๐—ฒ ๐Ÿญ๐Ÿต% ๐—ผ๐—ณ ๐˜‚๐˜€๐—ฎ๐—ด๐—ฒ

Menlo Ventures surveyed 600 enterprise IT decision-makers for their 2024 report. They reveal that AI spending surged to $13.8 billion this year, more than 6x the $2.3 billion spent in 2023!

Companies are shifting from experimentation to serious implementation.

๐Ÿ‘ท Top enterprise use cases by adoption:
โ€ฃ Code copilots (51%)
- GitHub Copilot hit $300M revenue run rate
โ€ฃ Support chatbots (31%)
โ€ฃ RAG (28%)
โ€ฃ Data extraction/transformation (27%)
โ€ฃ Meeting summarization (25%)

๐Ÿ“ˆ Market dynamics:
โ€ฃ OpenAI's enterprise share dropped from 50% to 34% ๐Ÿ‘Ž
โ€ฃ Anthropic doubled presence from 12% to 24% ๐Ÿš€
โ€ฃ Open-source makes up 19% of usage ๐Ÿค—

๐Ÿ˜ฌ Implementation challenges:
โ€ฃ 26% failed due to unexpected implementation costs
โ€ฃ 21% failed due to data privacy issues
โ€ฃ 18% failed due to disappointing ROI
โ€ฃ 15% failed due to hallucinations

Read the full report here ๐Ÿ‘‰ https://menlovc.com/2024-the-state-of-generative-ai-in-the-enterprise/
posted an update about 1 month ago
view post
Post
1211
Made a new app to visualize the LLM race โ‡’ ๐—ก๐—ผ ๐—˜๐˜‚๐—ฟ๐—ผ๐—ฝ๐—ฒ๐—ฎ๐—ป ๐—ฐ๐—ผ๐—บ๐—ฝ๐—ฎ๐—ป๐˜† ๐—ถ๐—ป ๐˜๐—ต๐—ฒ ๐˜๐—ผ๐—ฝ ๐Ÿญ๐Ÿฌ ๐Ÿ‡ช๐Ÿ‡บโŒ

See the app here ๐Ÿ‘‰ m-ric/llm-race-to-the-top

I've adapted an app by @andrewrreed that tracks progress of LLMs ( andrewrreed/closed-vs-open-arena-elo), on the Chatbot Arena leaderboard, to compare companies from different countries.

The outcome is quite sad, as a Frenchman and European.

The top 10 is exclusively US ๐Ÿ‡บ๐Ÿ‡ธ and Chinese ๐Ÿ‡จ๐Ÿ‡ณ companies (after great Chinese LLM releases recently, like the Qwen2.5 series), with the notable exception of Mistral AI ๐Ÿ‡ซ๐Ÿ‡ท.

American companies are making fast progress, Chinese ones even faster. Europe is at risk of being left behind. And the EU AI Act hasn't even come into force yet to slow down the EU market. We need to wake up ๐Ÿ˜ฌ

โš ๏ธ Caution: This Chatbot Arena ELO ranking is not the most accurate, especially at high scores like this, because LLM makers can game it to some extent.
  • 1 reply
ยท
posted an update about 1 month ago
view post
Post
796
๐—ก๐—ฒ๐˜„ ๐—น๐—ฒ๐—ฎ๐—ฑ๐—ฒ๐—ฟ๐—ฏ๐—ผ๐—ฎ๐—ฟ๐—ฑ ๐—ฟ๐—ฎ๐—ป๐—ธ๐˜€ ๐—Ÿ๐—Ÿ๐— ๐˜€ ๐—ณ๐—ผ๐—ฟ ๐—Ÿ๐—Ÿ๐— -๐—ฎ๐˜€-๐—ฎ-๐—ท๐˜‚๐—ฑ๐—ด๐—ฒ: ๐—Ÿ๐—น๐—ฎ๐—บ๐—ฎ-๐Ÿฏ.๐Ÿญ-๐Ÿณ๐Ÿฌ๐—• ๐˜๐—ผ๐—ฝ๐˜€ ๐˜๐—ต๐—ฒ ๐—ฟ๐—ฎ๐—ป๐—ธ๐—ถ๐—ป๐—ด๐˜€! ๐Ÿง‘โ€โš–๏ธ

Evaluating systems is critical during prototyping and in production, and LLM-as-a-judge has become a standard technique to do it.

First, what is "LLM-as-a-judge"?
๐Ÿ‘‰ It's a very useful technique for evaluating LLM outputs. If anything you're evaluating cannot be properly evaluated with deterministic criteria, like the "politeness" of an LLM output, or how faithful it is to an original source, you can use LLM-judge instead : prompt another LLM with "Here's an LLM output, please rate this on criterion {criterion} on a scale of 1 to 5", then parse the number from its output, and voilร , you get your score.

๐Ÿง But who judges the judge?
How can you make sure your LLM-judge is reliable? You can have a specific dataset annotated with scores provided by human judges, and compare how LLM-judge scores correlate with human judge scores.

๐Ÿ“Š Before even running that benchmark, to get you started, there's a new option to get you started: a leaderboard that measures how well different model perform as judges!

And the outcome is surprising, models come in quite different orders from what we're used to in general rankings: probably some have much better bias mitigation than others!

Take a deeper look here ๐Ÿ‘‰ https://huggingface.co./blog/arena-atla
posted an update about 1 month ago
view post
Post
270
Lifehack of the day:
Adding "r.jina.ai/" before any url transforms it in Markdown using Jina AI's Reader! Here with @cyrilzakka 's blog post.
reacted to cfahlgren1's post with โค๏ธ about 1 month ago
view post
Post
3091
You can clean and format datasets entirely in the browser with a few lines of SQL.

In this post, I replicate the process @mlabonne used to clean the new microsoft/orca-agentinstruct-1M-v1 dataset.

The cleaning process consists of:
- Joining the separate splits together / add split column
- Converting string messages into list of structs
- Removing empty system prompts

https://huggingface.co./blog/cfahlgren1/the-beginners-guide-to-cleaning-a-dataset

Here's his new cleaned dataset: mlabonne/orca-agentinstruct-1M-v1-cleaned
  • 1 reply
ยท
posted an update about 1 month ago
view post
Post
786
๐Ÿ” Meta teams use a fine-tuned Llama model to fix production issues in seconds

One of Meta's engineering teams shared how they use a fine-tuned small Llama (Llama-2-7B, so not even a very recent model) to identify the root cause of production issues with 42% accuracy.

๐Ÿค” 42%, is that not too low?
โžก๏ธ Usually, whenever there's an issue in production, engineers dive into recent code changes to find the offending commit. At Meta's scale (thousands of daily changes), this is like finding a needle in a haystack.
๐Ÿ’ก So when the LLM-based suggestion is right, it cuts incident resolution time from hours to seconds!

How did they do it?

๐Ÿ”„ Two-step approach:
โ€ฃ Heuristics (code ownership, directory structure, runtime graphs) reduce thousands of potential changes to a manageable set
โ€ฃ Fine-tuned Llama 2 7B ranks the most likely culprits

๐ŸŽ“ Training pipeline:
โ€ฃ Continued pre-training on Meta's internal docs and wikis
โ€ฃ Supervised fine-tuning on past incident investigations
โ€ฃ Training data mimicked real-world constraints (2-20 potential changes per incident)

๐Ÿ”ฎ Now future developments await:
โ€ฃ Language models could handle more of the incident response workflow (runbooks, mitigation, post-mortems)
โ€ฃ Improvements in model reasoning should boost accuracy further

Read it in full ๐Ÿ‘‰ https://www.tryparity.com/blog/how-meta-uses-llms-to-improve-incident-response
reacted to reach-vb's post with ๐Ÿ”ฅ about 1 month ago
view post
Post
4329
What a brilliant week for Open Source AI!

Qwen 2.5 Coder by Alibaba - 0.5B / 1.5B / 3B / 7B / 14B/ 32B (Base + Instruct) Code generation LLMs, with 32B tackling giants like Gemnini 1.5 Pro, Claude Sonnet
Qwen/qwen25-coder-66eaa22e6f99801bf65b0c2f

LLM2CLIP from Microsoft - Leverage LLMs to train ultra-powerful CLIP models! Boosts performance over the previous SOTA by ~17%
microsoft/llm2clip-672323a266173cfa40b32d4c

Athene v2 Chat & Agent by NexusFlow - SoTA general LLM fine-tuned from Qwen 2.5 72B excels at Chat + Function Calling/ JSON/ Agents
Nexusflow/athene-v2-6735b85e505981a794fb02cc

Orca Agent Instruct by Microsoft - 1 million instruct pairs covering text editing, creative writing, coding, reading comprehension, etc - permissively licensed
microsoft/orca-agentinstruct-1M-v1

Ultravox by FixieAI - 70B/ 8B model approaching GPT4o level, pick any LLM, train an adapter with Whisper as Audio Encoder
reach-vb/ultravox-audio-language-model-release-67373b602af0a52b2a88ae71

JanusFlow 1.3 by DeepSeek - Next iteration of their Unified MultiModal LLM Janus with RectifiedFlow
deepseek-ai/JanusFlow-1.3B

Common Corpus by Pleais - 2,003,039,184,047 multilingual, commercially permissive and high quality tokens!
PleIAs/common_corpus

I'm sure I missed a lot, can't wait for the next week!

Put down in comments what I missed! ๐Ÿค—