XLSCOUT Unveils ParaEmbed 2.0: a Powerful Embedding Model Tailored for Patents and IP with Expert Support from Hugging Face
โข
10
Andrew Reed
andrewrreed
AI & ML interests
Applied ML, Practical AI, Inference & Deployment, LLMs, Multi-modal Models, RAG
Articles
Organizations
andrewrreed's activity
reacted to
clem's
post with ๐๐ฅ
16 days ago
Post
4020
This is no Woodstock AI but will be fun nonetheless haha. Iโll be hosting a live workshop with team members next week about the Enterprise Hugging Face hub.
1,000 spots available first-come first serve with some surprises during the stream!
You can register and add to your calendar here: https://streamyard.com/watch/JS2jHsUP3NDM
1,000 spots available first-come first serve with some surprises during the stream!
You can register and add to your calendar here: https://streamyard.com/watch/JS2jHsUP3NDM
- 4 replies
reacted to
melisa's
post with ๐ฅ
2 months ago
Post
2964
๐ฅ Introducing "Writing in the Margins (WiM)" - better inference pattern for long context LLMs that solves the Lost-in-the-Middle problem ๐ฅ
Paper page: Writing in the Margins: Better Inference Pattern for Long Context Retrieval (2408.14906)
TL;DR
Make your model write "margin notes" as you chunk prefill the KV cache. Then ask it reread all notes before it speaks up.
Works with humans, works with AI ๐ค
WiM leverages the chunked prefill of the key-value cache, which concurrently generates query-based extractive summaries at each step of the prefill that are subsequently reintegrated at the end of the computation. We term these intermediate outputs โmarginsโ, drawing inspiration from the practice of making margin notes for improved comprehension of long contexts in human reading. We show that this technique, which adds only minimal additional computation, significantly improves LLMs long context reasoning capabilities.
Think: Every chunk has a chance to be attended to/ be at the end of the context at least once. ๐
๐ Results:
- An average accuracy boost of 7.5% in multi-hop reasoning tasks like HotpotQA and MultiHop-RAG.
- Even a 30% increase in F1-score for summarisation-like tasks (CWE).
Plus, WiM fits seamlessly into interactive applications (think: progress bar!). It can provide real-time progress updates during data retrieval and integration, making it user-friendly and transparent - a stark contrast to feeding 1mln tokens to an LLMs and waiting 6 min for the first token. ๐คฏ
๐ฉโ๐ป๐งโ๐ป Check it out and contribute to our open-source project here: https://github.com/writer/writing-in-the-margins
๐ง More about chunked prefill: https://docs.vllm.ai/en/latest/models/performance.html#chunked-prefill
Paper page: Writing in the Margins: Better Inference Pattern for Long Context Retrieval (2408.14906)
TL;DR
Make your model write "margin notes" as you chunk prefill the KV cache. Then ask it reread all notes before it speaks up.
Works with humans, works with AI ๐ค
WiM leverages the chunked prefill of the key-value cache, which concurrently generates query-based extractive summaries at each step of the prefill that are subsequently reintegrated at the end of the computation. We term these intermediate outputs โmarginsโ, drawing inspiration from the practice of making margin notes for improved comprehension of long contexts in human reading. We show that this technique, which adds only minimal additional computation, significantly improves LLMs long context reasoning capabilities.
Think: Every chunk has a chance to be attended to/ be at the end of the context at least once. ๐
๐ Results:
- An average accuracy boost of 7.5% in multi-hop reasoning tasks like HotpotQA and MultiHop-RAG.
- Even a 30% increase in F1-score for summarisation-like tasks (CWE).
Plus, WiM fits seamlessly into interactive applications (think: progress bar!). It can provide real-time progress updates during data retrieval and integration, making it user-friendly and transparent - a stark contrast to feeding 1mln tokens to an LLMs and waiting 6 min for the first token. ๐คฏ
๐ฉโ๐ป๐งโ๐ป Check it out and contribute to our open-source project here: https://github.com/writer/writing-in-the-margins
๐ง More about chunked prefill: https://docs.vllm.ai/en/latest/models/performance.html#chunked-prefill
- 2 replies
reacted to
m-ric's
post with ๐ฅ
3 months ago
Post
1100
๐๐น๐ฎ๐บ๐ฎ-๐ฏ.๐ญ ๐บ๐ผ๐ฑ๐ฒ๐น๐ ๐ณ๐ถ๐ป๐ฎ๐น๐น๐ ๐ด๐ฒ๐ ๐๐ต๐ฒ๐ถ๐ฟ ๐๐ต๐ฎ๐๐ฏ๐ผ๐ ๐๐ฟ๐ฒ๐ป๐ฎ ๐ฟ๐ฎ๐ป๐ธ๐ถ๐ป๐ด ๐๏ธ
Given the impressive benchmarks published my Meta for their Llama-3.1 models, I was curious to see how these models would compare to top proprietary models on Chatbot Arena.
Now we've got the results! LMSys released the ELO derived from thousands of user votes for the new models, and here are the rankings:
๐ฅ 405B Model ranks 5th overall, in front of GPT-4-turbo! But behind GPT-4o, Claude-3.5 Sonnet and Gemini-advanced.
๐ 70B Model climbs up to 9th rank ! From 1206 โก๏ธ 1244.
๐ 8B Model improves from 1152 โก๏ธ 1170.
โ This confirms that Llama-3.1 is a good contender for any task: any of its 3 model size is much cheaper to run than equivalent proprietary models!
For instance, here are the inference prices for the top models;
โค GPT-4-Turbo inference price from OpenAI: $5/M input tokens, $15/M output tokens
โค Llama-3.1-405B from HF API (for testing only): 3$/M for input or output tokens (Source linked in the first comment)
โค Llama-3.1-405B from HF API (for testing only): free โจ
Get a head start on the HF API (resource by @andrewrreed ) ๐ https://huggingface.co/learn/cookbook/enterprise_hub_serverless_inference_api
Given the impressive benchmarks published my Meta for their Llama-3.1 models, I was curious to see how these models would compare to top proprietary models on Chatbot Arena.
Now we've got the results! LMSys released the ELO derived from thousands of user votes for the new models, and here are the rankings:
๐ฅ 405B Model ranks 5th overall, in front of GPT-4-turbo! But behind GPT-4o, Claude-3.5 Sonnet and Gemini-advanced.
๐ 70B Model climbs up to 9th rank ! From 1206 โก๏ธ 1244.
๐ 8B Model improves from 1152 โก๏ธ 1170.
โ This confirms that Llama-3.1 is a good contender for any task: any of its 3 model size is much cheaper to run than equivalent proprietary models!
For instance, here are the inference prices for the top models;
โค GPT-4-Turbo inference price from OpenAI: $5/M input tokens, $15/M output tokens
โค Llama-3.1-405B from HF API (for testing only): 3$/M for input or output tokens (Source linked in the first comment)
โค Llama-3.1-405B from HF API (for testing only): free โจ
Get a head start on the HF API (resource by @andrewrreed ) ๐ https://huggingface.co/learn/cookbook/enterprise_hub_serverless_inference_api
- 1 reply
reacted to
dvilasuero's
post with ๐คโค๏ธ๐๐ฅ
5 months ago
Post
7893
Today is a huge day in Argillaโs history. We couldnโt be more excited to share this with the community: weโre joining Hugging Face!
Weโre embracing a larger mission, becoming part of a brilliant and kind team and a shared vision about the future of AI.
Over the past year, weโve been collaborating with Hugging Face on countless projects: launching partner of Docker Spaces, empowering the community to clean Alpaca translations into Spanish and other languages, launching argilla/notus-7b-v1 building on Zephyrโs learnings, the Data is Better Together initiative with hundreds of community contributors, or releasing argilla/OpenHermesPreferences, one of the largest open preference tuning datasets
After more than 2,000 Slack messages and over 60 people collaborating for over a year, it already felt like we were part of the same team, pushing in the same direction. After a week of the smoothest transition you can imagine, weโre now the same team.
To those of you whoโve been following us, this wonโt be a huge surprise, but it will be a big deal in the coming months. This acquisition means weโll double down on empowering the community to build and collaborate on high quality datasets, weโll bring full support for multimodal datasets, and weโll be in a better place to collaborate with the Open Source AI community. For enterprises, this means that the Enterprise Hub will unlock highly requested features like single sign-on and integration with Inference Endpoints.
As a founder, I am proud of the Argilla team. We're now part of something bigger and a larger team but with the same values, culture, and goals. Grateful to have shared this journey with my beloved co-founders Paco and Amรฉlie.
Finally, huge thanks to the Chief Llama Officer @osanseviero for sparking this and being such a great partner during the acquisition process.
Would love to answer any questions you have so feel free to add them below!
Weโre embracing a larger mission, becoming part of a brilliant and kind team and a shared vision about the future of AI.
Over the past year, weโve been collaborating with Hugging Face on countless projects: launching partner of Docker Spaces, empowering the community to clean Alpaca translations into Spanish and other languages, launching argilla/notus-7b-v1 building on Zephyrโs learnings, the Data is Better Together initiative with hundreds of community contributors, or releasing argilla/OpenHermesPreferences, one of the largest open preference tuning datasets
After more than 2,000 Slack messages and over 60 people collaborating for over a year, it already felt like we were part of the same team, pushing in the same direction. After a week of the smoothest transition you can imagine, weโre now the same team.
To those of you whoโve been following us, this wonโt be a huge surprise, but it will be a big deal in the coming months. This acquisition means weโll double down on empowering the community to build and collaborate on high quality datasets, weโll bring full support for multimodal datasets, and weโll be in a better place to collaborate with the Open Source AI community. For enterprises, this means that the Enterprise Hub will unlock highly requested features like single sign-on and integration with Inference Endpoints.
As a founder, I am proud of the Argilla team. We're now part of something bigger and a larger team but with the same values, culture, and goals. Grateful to have shared this journey with my beloved co-founders Paco and Amรฉlie.
Finally, huge thanks to the Chief Llama Officer @osanseviero for sparking this and being such a great partner during the acquisition process.
Would love to answer any questions you have so feel free to add them below!
- 28 replies
reacted to
lunarflu's
post with โค๏ธ
5 months ago
Post
1881
cooking up something....anyone interested in a daily activity tracker for HF?
- 12 replies
reacted to
tomaarsen's
post with ๐๐ฅ
6 months ago
Post
2356
NuMind has just released 3 new state-of-the-art GLiNER models for Named Entity Recognition/Information Extraction. These GLiNER models allow you to specify any label that you want, and it'll find spans in the text corresponding to your label. It's been shown to work quite well on unusual domains, e.g. celestial entities in my picture.
There are 3 models released:
- numind/NuNER_Zero:
The primary model, SOTA & can detect really long entities.
- numind/NuNER_Zero-span:
Slightly better performance than NuNER Zero, but can't detect entities longer than 12 tokens.
- numind/NuNER_Zero-4k:
Slightly worse than NuNER Zero, but has a context length of 4k tokens.
Some more details about these models in general:
- They are *really* small, orders of magnitude smaller than LLMs, which don't reach this level of performance.
- Because they're small - they're fast: <1s per sentence on free GPUs.
- They have an MIT license: free commercial usage.
Try out the demo here: https://huggingface.co/spaces/numind/NuZero
Or check out all of the models here: numind/nunerzero-zero-shot-ner-662b59803b9b438ff56e49e2
If there's ever a need for me to extract some information from any text: I'll be using these. Great work @Serega6678 !
There are 3 models released:
- numind/NuNER_Zero:
The primary model, SOTA & can detect really long entities.
- numind/NuNER_Zero-span:
Slightly better performance than NuNER Zero, but can't detect entities longer than 12 tokens.
- numind/NuNER_Zero-4k:
Slightly worse than NuNER Zero, but has a context length of 4k tokens.
Some more details about these models in general:
- They are *really* small, orders of magnitude smaller than LLMs, which don't reach this level of performance.
- Because they're small - they're fast: <1s per sentence on free GPUs.
- They have an MIT license: free commercial usage.
Try out the demo here: https://huggingface.co/spaces/numind/NuZero
Or check out all of the models here: numind/nunerzero-zero-shot-ner-662b59803b9b438ff56e49e2
If there's ever a need for me to extract some information from any text: I'll be using these. Great work @Serega6678 !
- 3 replies
reacted to
alvarobartt's
post with โค๏ธ๐ฅ
6 months ago
Post
2918
๐ฅ Prometheus 2 was recently released by Kaist AI as an alternative and closely mirroring both human and GPT-4 evaluation, and surpassing the former Prometheus!
prometheus-eval/prometheus-7b-v2.0
prometheus-eval/prometheus-8x7b-v2.0
๐ฌ๏ธFine-tuned on top of mistralai/Mistral-7B-Instruct-v0.2 and mistralai/Mixtral-8x7B-Instruct-v0.1
๐๏ธThe datasets used for fine-tuning have been publicly released i.e. prometheus-eval/Feedback-Collection and prometheus-eval/Preference-Collection
๐ค๐ปUnified LM evaluator for absolute (a single prompt-completion pair) and relative (two completions for a given prompt) due to model merging
โNo longer needs a mandatory reference / golden answer, but can still be provided optionally
๐Surpasses the former version of Prometheus, and has a high correlation with human, GPT-4, and Claude 3 Opus scores when evaluating LMs
๐Apache 2.0 license
Long-story short, an amazing job from Kaist AI bridging the gap with LLM evaluators other than proprietary and bigger models!
This week at Argilla, we decided to add a new task to use Prometheus 2 as an LLM evaluator using
๐ฑ Using
Find the generated dataset and the code at distilabel-internal-testing/instruction-dataset-prometheus
prometheus-eval/prometheus-7b-v2.0
prometheus-eval/prometheus-8x7b-v2.0
๐ฌ๏ธFine-tuned on top of mistralai/Mistral-7B-Instruct-v0.2 and mistralai/Mixtral-8x7B-Instruct-v0.1
๐๏ธThe datasets used for fine-tuning have been publicly released i.e. prometheus-eval/Feedback-Collection and prometheus-eval/Preference-Collection
๐ค๐ปUnified LM evaluator for absolute (a single prompt-completion pair) and relative (two completions for a given prompt) due to model merging
โNo longer needs a mandatory reference / golden answer, but can still be provided optionally
๐Surpasses the former version of Prometheus, and has a high correlation with human, GPT-4, and Claude 3 Opus scores when evaluating LMs
๐Apache 2.0 license
Long-story short, an amazing job from Kaist AI bridging the gap with LLM evaluators other than proprietary and bigger models!
This week at Argilla, we decided to add a new task to use Prometheus 2 as an LLM evaluator using
distilabel, so we implemented PrometheusEval.๐ฑ Using
PrometheusEval running their 7B variant with vLLM in a single L40 on top of
HuggingFaceH4/instruction-dataset, we got the 327 existing prompt-completion pairs evaluated and pushed to the Hub in less than 2 minutes!Find the generated dataset and the code at distilabel-internal-testing/instruction-dataset-prometheus
- 1 reply
reacted to
davanstrien's
post with ๐ฅ
6 months ago
Post
2557
Introducing CosmoChat, a multiturn chat dataset based on Cosmopedia that I'm working on in the open on the Hub.
๐ฏ Goals:
๐ฌ Create multi-turn chats seeded from Cosmopedia
๐ Customize questions for different audience levels
๐ Evaluate the model's ability to elaborate and clarify
๐ค (I want to learn more about creating valuable synthetic datasets, and I learn best by doing stuff rather than reading stuff).
Cosmochat is created using the excellent distilabel library.
๐ Explore the current version of the dataset: davanstrien/cosmochat
๐ Read more: https://huggingface.co/blog/davanstrien/cosmochat
๐ฏ Goals:
๐ฌ Create multi-turn chats seeded from Cosmopedia
๐ Customize questions for different audience levels
๐ Evaluate the model's ability to elaborate and clarify
๐ค (I want to learn more about creating valuable synthetic datasets, and I learn best by doing stuff rather than reading stuff).
Cosmochat is created using the excellent distilabel library.
๐ Explore the current version of the dataset: davanstrien/cosmochat
๐ Read more: https://huggingface.co/blog/davanstrien/cosmochat
- 2 replies
replied to
their
post
6 months ago
Thanks! And yes, several people have pointed out the light mode color issue... will push a fix when I get the chance
posted
an
update
6 months ago
Post
2399
๐ฌ Open LLM Progress Tracker ๐ฌ
Inspired by the awesome work from @mlabonne , I created a Space to monitor the narrowing gap between open and proprietary LLMs as scored by the LMSYS Chatbot Arena ELO ratings ๐ค
The goal is to have a continuously updated place to easily visualize these rapidly evolving industry trends ๐
๐ Open LLM Progress Tracker: andrewrreed/closed-vs-open-arena-elo
๐ Source of Inspiration: https://www.linkedin.com/posts/maxime-labonne_arena-elo-graph-updated-with-new-models-activity-7187062633735368705-u2jB/
Inspired by the awesome work from @mlabonne , I created a Space to monitor the narrowing gap between open and proprietary LLMs as scored by the LMSYS Chatbot Arena ELO ratings ๐ค
The goal is to have a continuously updated place to easily visualize these rapidly evolving industry trends ๐
๐ Open LLM Progress Tracker: andrewrreed/closed-vs-open-arena-elo
๐ Source of Inspiration: https://www.linkedin.com/posts/maxime-labonne_arena-elo-graph-updated-with-new-models-activity-7187062633735368705-u2jB/
-
- 2 replies
reacted to
Pclanglais's
post with ๐ฅ
7 months ago
Post
2320
Announcing that we are on our way to solve a long standing issue of document processing: correction of OCR mistakes. Pleias publishes the largest dataset to date with automated OCR correction, 1 billion words in English, French, German and Italian.
OCR quality is long-standing issue of digitization. Cultural heritage texts are especially concerned due to the primary sources being old documents (with many artifacts, blots, degradation) and to the limitation of OCR technology for historical scripts. When we released Common Corpus, a 500 Billion words corpus in the public domain, this was the primary criticism.
Recent breakthrough in post-OCR correction has been made possible thanks to progress in open LLM research and several months of dedicated training and alignment by Pleias as well as the HPC resources from GENCIโIDRIS (Grant 2023-AD011014736) on Jean-Zay.
Announcement: https://huggingface.co/blog/Pclanglais/post-ocr-correction
Post-OCR-Correction dataset: https://huggingface.co/datasets/PleIAs/Post-OCR-Correction
OCR quality is long-standing issue of digitization. Cultural heritage texts are especially concerned due to the primary sources being old documents (with many artifacts, blots, degradation) and to the limitation of OCR technology for historical scripts. When we released Common Corpus, a 500 Billion words corpus in the public domain, this was the primary criticism.
Recent breakthrough in post-OCR correction has been made possible thanks to progress in open LLM research and several months of dedicated training and alignment by Pleias as well as the HPC resources from GENCIโIDRIS (Grant 2023-AD011014736) on Jean-Zay.
Announcement: https://huggingface.co/blog/Pclanglais/post-ocr-correction
Post-OCR-Correction dataset: https://huggingface.co/datasets/PleIAs/Post-OCR-Correction
reacted to
fdaudens's
post with ๐ฅ
7 months ago
Post
2676
It's been only a week since I joined ๐ค and the community has released a constant flow of content!
Notable models:
- Apple OpenELM apple/openelm-instruct-models-6619ad295d7ae9f868b759ca + apple/openelm-pretrained-models-6619ac6ca12a10bd0d0df89e
- HuggingFaceM4 Idefics2 HuggingFaceM4/idefics2-8b
- Meta Llama 3 meta-llama/meta-llama-3-66214712577ca38149ebb2b6
- Microsoft Phi-3 microsoft/phi-3-6626e15e9585a200d2d761e3
- Snowflake Arctic Snowflake/arctic-66290090abe542894a5ac520
Great datasets:
- HuggingFaceFW FineWeb HuggingFaceFW/fineweb
- HuggingFaceM4/the_cauldron HuggingFaceM4/the_cauldron
- PleIAs/YouTube-Commons PleIAs/YouTube-Commons
Fascinating Spaces
- InstantMesh TencentARC/InstantMesh
- Chat with Llama 3 8B ysharma/Chat_with_Meta_llama3_8b
- Parler-TTS https://huggingface.co/spaces/parler-tts/parler_tts_mini
- AI Jukebox enzostvs/ai-jukebox
- CosXL multimodalart/cosxl
- Singing songstarter nateraw/singing-songstarter
- Play with Idefics2 8B HuggingFaceM4/idefics-8b
- CodeQwen1.5-7B-Chat Bot๐พ
Qwen/CodeQwen1.5-7b-Chat-demo
I expected to be at the center of AI development. I'm not disappointed!
Notable models:
- Apple OpenELM apple/openelm-instruct-models-6619ad295d7ae9f868b759ca + apple/openelm-pretrained-models-6619ac6ca12a10bd0d0df89e
- HuggingFaceM4 Idefics2 HuggingFaceM4/idefics2-8b
- Meta Llama 3 meta-llama/meta-llama-3-66214712577ca38149ebb2b6
- Microsoft Phi-3 microsoft/phi-3-6626e15e9585a200d2d761e3
- Snowflake Arctic Snowflake/arctic-66290090abe542894a5ac520
Great datasets:
- HuggingFaceFW FineWeb HuggingFaceFW/fineweb
- HuggingFaceM4/the_cauldron HuggingFaceM4/the_cauldron
- PleIAs/YouTube-Commons PleIAs/YouTube-Commons
Fascinating Spaces
- InstantMesh TencentARC/InstantMesh
- Chat with Llama 3 8B ysharma/Chat_with_Meta_llama3_8b
- Parler-TTS https://huggingface.co/spaces/parler-tts/parler_tts_mini
- AI Jukebox enzostvs/ai-jukebox
- CosXL multimodalart/cosxl
- Singing songstarter nateraw/singing-songstarter
- Play with Idefics2 8B HuggingFaceM4/idefics-8b
- CodeQwen1.5-7B-Chat Bot๐พ
Qwen/CodeQwen1.5-7b-Chat-demo
I expected to be at the center of AI development. I'm not disappointed!
-
- 4 replies
posted
an
update
7 months ago
Post
2290
IMO, the "grounded generation" feature from Cohere's CommandR+ has flown under the radar...
For RAG use cases, responses directly include inline citations, making source attribution an inherent part of generation rather than an afterthought ๐
Who's working on an open dataset with this for the HF community to fine-tune with??
๐CommandR+ Docs: https://docs.cohere.com/docs/retrieval-augmented-generation-rag
๐Model on the ๐ค Hub: CohereForAI/c4ai-command-r-plus
For RAG use cases, responses directly include inline citations, making source attribution an inherent part of generation rather than an afterthought ๐
Who's working on an open dataset with this for the HF community to fine-tune with??
๐CommandR+ Docs: https://docs.cohere.com/docs/retrieval-augmented-generation-rag
๐Model on the ๐ค Hub: CohereForAI/c4ai-command-r-plus
- 1 reply
reacted to
VictorSanh's
post with โค๏ธ
7 months ago
Post
2742
New open multimodal model in town: Idefics2!
๐ช Strong 8B-parameters model: often on par with open 30B counterparts.
๐Open license: Apache 2.0.
๐ Strong improvement over Idefics1: +12 points on VQAv2, +30 points on TextVQA while having 10x fewer parameters.
๐ Better data: boosting OCR capabilities with 6TB of documents to transcribe, and improving QA capabilities on charts/figures/diagrams.
๐ต๏ธโโ๏ธ Transparent training data: inspect and build upon all the data (10s of TB of data) we trained on.
๐ฒ More natural image processing: Incorporating strategies to treat images in their native resolution and native aspect ratio.
๐ธ High-resolution images: image resolutions up to 980 x 980 and integrating strategies that allow to trade computational efficiency for performance.
๐ 2 checkpoints: Releasing both base checkpoint and instruction fine-tuned checkpoint. Chat version to come.
Ressources: HuggingFaceM4/idefics2-661d1971b7c50831dd3ce0fe
Blogpost: https://huggingface.co/blog/idefics2
๐ช Strong 8B-parameters model: often on par with open 30B counterparts.
๐Open license: Apache 2.0.
๐ Strong improvement over Idefics1: +12 points on VQAv2, +30 points on TextVQA while having 10x fewer parameters.
๐ Better data: boosting OCR capabilities with 6TB of documents to transcribe, and improving QA capabilities on charts/figures/diagrams.
๐ต๏ธโโ๏ธ Transparent training data: inspect and build upon all the data (10s of TB of data) we trained on.
๐ฒ More natural image processing: Incorporating strategies to treat images in their native resolution and native aspect ratio.
๐ธ High-resolution images: image resolutions up to 980 x 980 and integrating strategies that allow to trade computational efficiency for performance.
๐ 2 checkpoints: Releasing both base checkpoint and instruction fine-tuned checkpoint. Chat version to come.
Ressources: HuggingFaceM4/idefics2-661d1971b7c50831dd3ce0fe
Blogpost: https://huggingface.co/blog/idefics2