Share and discover more about AI with social posts from the community.huggingface/OpenAi
๐๐ฟ๐ฒ ๐๐ด๐ฒ๐ป๐๐ ๐ฐ๐ฎ๐ฝ๐ฎ๐ฏ๐น๐ฒ ๐ฒ๐ป๐ผ๐๐ด๐ต ๐ณ๐ผ๐ฟ ๐๐ฎ๐๐ฎ ๐ฆ๐ฐ๐ถ๐ฒ๐ป๐ฐ๐ฒ? โ ๐ ๐ฒ๐ฎ๐๐๐ฟ๐ฒ ๐๐ต๐ฒ๐ถ๐ฟ ๐ฝ๐ฒ๐ฟ๐ณ๐ผ๐ฟ๐บ๐ฎ๐ป๐ฐ๐ฒ ๐๐ถ๐๐ต ๐๐ฆ๐๐ฒ๐ป๐ฐ๐ต ๐
A team from Tencent AI wanted to evaluate agentic systems on data science (DS) tasks : but they noticed that existing agentic benchmarks were severely limited in several aspects: they were limited to text and did not include tables or images, were only specific to certain packages, only performed exact match evaluationโฆ
โก๏ธ So they set out to build a much more exhaustive approach, to finally make the definitive DS agent benchmark.
๐ง๐ต๐ฒ ๐๐ฆ๐๐ฒ๐ป๐ฐ๐ต ๐ฑ๐ฎ๐๐ฎ๐๐ฒ๐
โช๏ธDS bench has 466 data analysis tasks and 74 data modelling tasks
โช๏ธThe tasks are sourced from ModelOff and Kaggle, the platforms hosting the most popular data science competitions
โช๏ธDifference with previous DS benchmarks:
โถ This benchmark leverages various modalities on top of text: images, Excel files, tables
โท Complex tables: sometimes several tables should be leveraged to answer one question
โธ The context is richer, with longer descriptions.
โช๏ธ Evaluation metrics : the benchmark is scored with an LLM as a judge, using a specific prompt.
๐๐ป๐๐ถ๐ด๐ต๐๐ ๐ณ๐ฟ๐ผ๐บ ๐ฒ๐๐ฎ๐น๐๐ฎ๐๐ถ๐ป๐ด ๐ฎ๐ด๐ฒ๐ป๐๐
โช๏ธ Their evaluation confirms that using LLMs in an agent setup, for instance by allowing them to run a single step of code execution, is more costly (especially with multi-turn frameworks like autogen) but also much more performant than the vanilla LLM.
โช๏ธ The sets of tasks solved by different models (like GPT-3.5 vs Llama-3-8B) has quite low overlap, which suggests that different models tend to try very different approches.
This new benchmark is really welcome, can't wait to try transformers agents on it! ๐ค
Read their full paper ๐
DSBench: How Far Are Data Science Agents to Becoming Data Science Experts? (2409.07703)https://huggingface.co/papers/2409.07703
A team from Tencent AI wanted to evaluate agentic systems on data science (DS) tasks : but they noticed that existing agentic benchmarks were severely limited in several aspects: they were limited to text and did not include tables or images, were only specific to certain packages, only performed exact match evaluationโฆ
โก๏ธ So they set out to build a much more exhaustive approach, to finally make the definitive DS agent benchmark.
๐ง๐ต๐ฒ ๐๐ฆ๐๐ฒ๐ป๐ฐ๐ต ๐ฑ๐ฎ๐๐ฎ๐๐ฒ๐
โช๏ธDS bench has 466 data analysis tasks and 74 data modelling tasks
โช๏ธThe tasks are sourced from ModelOff and Kaggle, the platforms hosting the most popular data science competitions
โช๏ธDifference with previous DS benchmarks:
โถ This benchmark leverages various modalities on top of text: images, Excel files, tables
โท Complex tables: sometimes several tables should be leveraged to answer one question
โธ The context is richer, with longer descriptions.
โช๏ธ Evaluation metrics : the benchmark is scored with an LLM as a judge, using a specific prompt.
๐๐ป๐๐ถ๐ด๐ต๐๐ ๐ณ๐ฟ๐ผ๐บ ๐ฒ๐๐ฎ๐น๐๐ฎ๐๐ถ๐ป๐ด ๐ฎ๐ด๐ฒ๐ป๐๐
โช๏ธ Their evaluation confirms that using LLMs in an agent setup, for instance by allowing them to run a single step of code execution, is more costly (especially with multi-turn frameworks like autogen) but also much more performant than the vanilla LLM.
โช๏ธ The sets of tasks solved by different models (like GPT-3.5 vs Llama-3-8B) has quite low overlap, which suggests that different models tend to try very different approches.
This new benchmark is really welcome, can't wait to try transformers agents on it! ๐ค
Read their full paper ๐
DSBench: How Far Are Data Science Agents to Becoming Data Science Experts? (2409.07703)https://huggingface.co/papers/2409.07703
huggingface.co
Paper page - DSBench: How Far Are Data Science Agents to Becoming Data Science
Experts?
Experts?
Join the discussion on this paper page
Bringing Open-Source Text-to-Speech to French! ๐ฃ๐ซ๐ท
Hugging Face's Parler TTS mini can now speak French! ๐ซ๐ท๐
You can try it here:
PHBJT/french_parler_tts
Key highlights:
Transform the English TTS model to speak French ๐ฌ๐งโก๏ธ๐ซ๐ท
Fully open source (code, weights, and datasets) ๐
It can be replicated for every language ๐
Read more about it in this article: https://huggingface.co/blog/PHBJT/french-parler-tts
Special thanks to FlexAI and their dedicated team for providing the computing power that made this possible and of course to all of the Parler TTS community ๐ค
Hugging Face's Parler TTS mini can now speak French! ๐ซ๐ท๐
You can try it here:
PHBJT/french_parler_tts
Key highlights:
Transform the English TTS model to speak French ๐ฌ๐งโก๏ธ๐ซ๐ท
Fully open source (code, weights, and datasets) ๐
It can be replicated for every language ๐
Read more about it in this article: https://huggingface.co/blog/PHBJT/french-parler-tts
Special thanks to FlexAI and their dedicated team for providing the computing power that made this possible and of course to all of the Parler TTS community ๐ค
huggingface.co
Fine-tuning Parler TTS on a Specific Language
A Blog post by Paul Henri Biojout on Hugging Face
OpenAI's latest model, "o1", has demonstrated remarkable performance on the Norway Mensa IQ test, scoring an estimated IQ of 120.
Everyone should think before answering!
Key findings:
โข o1 correctly answered 25 out of 35 IQ questions, surpassing average human performance
โข The model excelled at pattern recognition and logical reasoning tasks
โข Performance was validated on both public and private test sets to rule out training data bias
Technical details:
โข o1 utilizes advanced natural language processing and visual reasoning capabilities
โข The model likely employs transformer architecture with billions of parameters
โข Improved few-shot learning allows o1 to tackle novel problem types
Implications:
โข This represents a significant leap in AI reasoning abilities
โข We may see AIs surpassing 140 IQ by 2026 if the trend continues
โข Raises important questions about the nature of intelligence and cognitionhttps://cdn-uploads.huggingface.co/production/uploads/662bf5bfe93bb73804ef9344/Vk04-meRDfz9ay8YaMrLT.png
Everyone should think before answering!
Key findings:
โข o1 correctly answered 25 out of 35 IQ questions, surpassing average human performance
โข The model excelled at pattern recognition and logical reasoning tasks
โข Performance was validated on both public and private test sets to rule out training data bias
Technical details:
โข o1 utilizes advanced natural language processing and visual reasoning capabilities
โข The model likely employs transformer architecture with billions of parameters
โข Improved few-shot learning allows o1 to tackle novel problem types
Implications:
โข This represents a significant leap in AI reasoning abilities
โข We may see AIs surpassing 140 IQ by 2026 if the trend continues
โข Raises important questions about the nature of intelligence and cognitionhttps://cdn-uploads.huggingface.co/production/uploads/662bf5bfe93bb73804ef9344/Vk04-meRDfz9ay8YaMrLT.png
๐๐ปโโ๏ธHey there folks,
Nvidia just released a small 4B Nemotron-mini model , and it works surprisingly well !
you can check it out here :
base :
nvidia/Minitron-4B-Base
instruct :
nvidia/Nemotron-Mini-4B-Instruct
demo :
Tonic/Nemotron-Mini-4B
hoep you like it ๐ค๐ค
Nvidia just released a small 4B Nemotron-mini model , and it works surprisingly well !
you can check it out here :
base :
nvidia/Minitron-4B-Base
instruct :
nvidia/Nemotron-Mini-4B-Instruct
demo :
Tonic/Nemotron-Mini-4B
hoep you like it ๐ค๐ค
๐ฌ Chat as a way to query SQL! The Airtrain AI team is happy to share a new Hugging Face Space that lets you interact with Hugging Face Hub datasets using a natural language chatbot. ๐ค
Start Exploring ๐
airtrain-ai/hf-dataset-chat-to-sql
This Space is forked from
davidberenstein1957/text-to-sql-hub-datasets
by @davidberenstein1957 and features chat capability with improved table naming. The tool works with Hugging Faceโs recently released in-browser DuckDB-based SQL query engine for datasets.
Start Exploring ๐
airtrain-ai/hf-dataset-chat-to-sql
This Space is forked from
davidberenstein1957/text-to-sql-hub-datasets
by @davidberenstein1957 and features chat capability with improved table naming. The tool works with Hugging Faceโs recently released in-browser DuckDB-based SQL query engine for datasets.
Could someone please give me a screenshot of their fine tuning/training space form before they initiate the training? I have no idea the format column mapping field.
Column1,column2,column3
"Column1","column2","column3"
๐คท
Column1,column2,column3
"Column1","column2","column3"
๐คท
For all the Muslims out there who are interested in Quran and its tafsir (explanations). This humble dataset consists of 84 different books of tafsir for nearly all the ayat in the Quran:
MohamedRashad/Quran-Tafseer
I hope it helps someone to build something nice and useful with it ^_^
MohamedRashad/Quran-Tafseer
I hope it helps someone to build something nice and useful with it ^_^
Anybody ever play Final Fantasy: Crystal Chronicles?
Like, *really* play it?
Mag Mell has been in my head recently. What a place that was.
Those cocoons looked like I could lay down inside of one, and it would be the most powerful sleep of a lifetime, with dreams that would last one thousand years, and I'd wake up with the wisdom of generations.
...Hey, anybody like text adventures?
Like, *really* play it?
Mag Mell has been in my head recently. What a place that was.
Those cocoons looked like I could lay down inside of one, and it would be the most powerful sleep of a lifetime, with dreams that would last one thousand years, and I'd wake up with the wisdom of generations.
...Hey, anybody like text adventures?
Last Week in Medical AI: Top Research Papers/Models
๐ (September 7 - September 14, 2024)
๐ Medical AI Paper of the week
Chai-1 Foundation model molecular structure prediction
Medical LLMs & Benchmarks
- BrainWave: A Brain Signal Foundation Model
- DS-ViT: Vision Transformer for Alzheimerโs Diagnosis
- EyeCLIP: Visualโlanguage model for ophthalmic
- Segment Anything Model for Tumor Segmentation
- MEDIC: Evaluating LLMs in Clinical Applications
Medical LLM Applications
- KARGEN: Radiology Report Generation LLMs
- DrugAgent: Explainable Drug Repurposing Agents
- Improving RAG in Medicine with Follow-up Questions
Frameworks and Methodologies
- Infrastructure for Automatic Cell Segmentation
- Data Alignment for Dermatology AI
- Diagnostic Reasoning in Natural Language
- Two-Stage Instruction Fine-tuning Approach for Med
AI in Healthcare Ethics
- Concerns and Choices of Using LLMs for Healthcare
- Understanding Fairness in Recommender Systems
- Towards Fairer Health Recommendations
Check the full thread: https://x.com/OpenlifesciAI/status/1832476252260712788
Thank you for your continued support and love for this series! Stay up-to-date with weekly updates on Medical LLMs, datasets, and top research papers by following @aaditya ๐ค
๐ (September 7 - September 14, 2024)
๐ Medical AI Paper of the week
Chai-1 Foundation model molecular structure prediction
Medical LLMs & Benchmarks
- BrainWave: A Brain Signal Foundation Model
- DS-ViT: Vision Transformer for Alzheimerโs Diagnosis
- EyeCLIP: Visualโlanguage model for ophthalmic
- Segment Anything Model for Tumor Segmentation
- MEDIC: Evaluating LLMs in Clinical Applications
Medical LLM Applications
- KARGEN: Radiology Report Generation LLMs
- DrugAgent: Explainable Drug Repurposing Agents
- Improving RAG in Medicine with Follow-up Questions
Frameworks and Methodologies
- Infrastructure for Automatic Cell Segmentation
- Data Alignment for Dermatology AI
- Diagnostic Reasoning in Natural Language
- Two-Stage Instruction Fine-tuning Approach for Med
AI in Healthcare Ethics
- Concerns and Choices of Using LLMs for Healthcare
- Understanding Fairness in Recommender Systems
- Towards Fairer Health Recommendations
Check the full thread: https://x.com/OpenlifesciAI/status/1832476252260712788
Thank you for your continued support and love for this series! Stay up-to-date with weekly updates on Medical LLMs, datasets, and top research papers by following @aaditya ๐ค
Trained Myself With 256 Images on FLUX โ Results Mind Blowing
Detailed Full Workflow
Medium article : https://medium.com/@furkangozukara/ultimate-flux-lora-training-tutorial-windows-and-cloud-deployment-abb72f21cbf8
Windows main tutorial : https://youtu.be/nySGu12Y05k
Cloud tutorial for GPU poor or scaling : https://youtu.be/-uhL2nW7Ddw
Full detailed results and conclusions : https://www.patreon.com/posts/111891669
Full config files and details to train : https://www.patreon.com/posts/110879657
SUPIR Upscaling (default settings are now perfect) : https://youtu.be/OYxVEvDf284
I used my Poco X6 Camera phone and solo taken images
My dataset is far from being ready, thus I have used so many repeating and almost same images, but this was rather experimental
Hopefully I will continue taking more shots and improve dataset and reduce size in future
I trained Clip-L and T5-XXL Text Encoders as well
Since there was too much push from community that my workflow wonโt work with expressions, I had to take a break from research and use whatever I have
I used my own researched workflow for training with Kohya GUI and also my own self developed SUPIR app batch upscaling with face upscaling and auto LLaVA captioning improvement
Download images to see them in full size, the last provided grid is 50% downscaled
Workflow
Gather a dataset that has expressions and perspectives that you like after training, this is crucial, whatever you add, it can generate perfect
Follow one of the LoRA training tutorials / guides
After training your LoRA, use your favorite UI to generate images
I prefer SwarmUI and here used prompts (you can add specific expressions to prompts) including face inpainting :
https://gist.github.com/FurkanGozukara/ce72861e52806c5ea4e8b9c7f4409672
After generating images, use SUPIR to upscale 2x with maximum resemblance
Short Conclusions
Using 256 images certainly caused more overfitting than necessary
Detailed Full Workflow
Medium article : https://medium.com/@furkangozukara/ultimate-flux-lora-training-tutorial-windows-and-cloud-deployment-abb72f21cbf8
Windows main tutorial : https://youtu.be/nySGu12Y05k
Cloud tutorial for GPU poor or scaling : https://youtu.be/-uhL2nW7Ddw
Full detailed results and conclusions : https://www.patreon.com/posts/111891669
Full config files and details to train : https://www.patreon.com/posts/110879657
SUPIR Upscaling (default settings are now perfect) : https://youtu.be/OYxVEvDf284
I used my Poco X6 Camera phone and solo taken images
My dataset is far from being ready, thus I have used so many repeating and almost same images, but this was rather experimental
Hopefully I will continue taking more shots and improve dataset and reduce size in future
I trained Clip-L and T5-XXL Text Encoders as well
Since there was too much push from community that my workflow wonโt work with expressions, I had to take a break from research and use whatever I have
I used my own researched workflow for training with Kohya GUI and also my own self developed SUPIR app batch upscaling with face upscaling and auto LLaVA captioning improvement
Download images to see them in full size, the last provided grid is 50% downscaled
Workflow
Gather a dataset that has expressions and perspectives that you like after training, this is crucial, whatever you add, it can generate perfect
Follow one of the LoRA training tutorials / guides
After training your LoRA, use your favorite UI to generate images
I prefer SwarmUI and here used prompts (you can add specific expressions to prompts) including face inpainting :
https://gist.github.com/FurkanGozukara/ce72861e52806c5ea4e8b9c7f4409672
After generating images, use SUPIR to upscale 2x with maximum resemblance
Short Conclusions
Using 256 images certainly caused more overfitting than necessary
Medium
Ultimate FLUX LoRA Training Tutorial: Windows and Cloud Deployment
I have done total 104 different LoRA trainings and compared each one of them to find the very best hyper parameters and the workflow forโฆ
Researchers from Tencent have developed DepthCrafter, a novel method for generating temporally consistent long depth sequences for open-world videos using video diffusion models.
It leverages a pre-trained image-to-video diffusion model (SVD) as the foundation and uses a 3-stage training strategy on paired video-depth datasets:
1. Train on a large realistic dataset (1-25 frames)
2. Fine-tune temporal layers on realistic data (1-110 frames)
3. Fine-tune spatial layers on synthetic data (45 frames)
It adapts SVD's conditioning mechanism for frame-by-frame video input and employs latent diffusion in VAE space for efficiency.
Sprinkle some intelligent inference strategy for extremely long videos:
- Segment-wise processing (up to 110 frames)
- Noise initialization to anchor depth distributions
- Latent interpolation for seamless stitching
And outperforms SOTA methods on multiple datasets (Sintel, ScanNet, KITTI, Bonn).
Read here: https://depthcrafter.github.io
It leverages a pre-trained image-to-video diffusion model (SVD) as the foundation and uses a 3-stage training strategy on paired video-depth datasets:
1. Train on a large realistic dataset (1-25 frames)
2. Fine-tune temporal layers on realistic data (1-110 frames)
3. Fine-tune spatial layers on synthetic data (45 frames)
It adapts SVD's conditioning mechanism for frame-by-frame video input and employs latent diffusion in VAE space for efficiency.
Sprinkle some intelligent inference strategy for extremely long videos:
- Segment-wise processing (up to 110 frames)
- Noise initialization to anchor depth distributions
- Latent interpolation for seamless stitching
And outperforms SOTA methods on multiple datasets (Sintel, ScanNet, KITTI, Bonn).
Read here: https://depthcrafter.github.io
nanoGPT with Sigmoid Self-Attention
I couldnโt resist had to give it a try:)
Some observations on M2:
SSA was ~5-10% faster in training with similar final loss values, slightly less coherent text generation, marginally higher perplexity, and lower memory usage compared to softmax.
Code: https://github.com/Jaykef/ai-algorithms/blob/main/sigmoid_attn.ipynb
I couldnโt resist had to give it a try:)
Some observations on M2:
SSA was ~5-10% faster in training with similar final loss values, slightly less coherent text generation, marginally higher perplexity, and lower memory usage compared to softmax.
Code: https://github.com/Jaykef/ai-algorithms/blob/main/sigmoid_attn.ipynb
GitHub
ai-algorithms/sigmoid_attn.ipynb at main ยท Jaykef/ai-algorithms
First-principle implementations of various AI algorithms using a wide range of deep learning frameworks, accompanied by relevant research papers - Jaykef/ai-algorithms
How much VRAM will you need for training your AI model? ๐พ๐ง
Check out this app where you convert:
Pytorch/tensorflow summary -> required VRAM
or
Parameter count -> required VRAM
Use it in: http://howmuchvram.com
And everything is open source! Ask for new functionalities or contribute in:
https://github.com/AlexBodner/How_Much_VRAM
If it's useful to you leave a star ๐and share it to someone that will find the tool useful!
More discussion in: https://x.com/AlexBodner_/status/1832054850294812679
Check out this app where you convert:
Pytorch/tensorflow summary -> required VRAM
or
Parameter count -> required VRAM
Use it in: http://howmuchvram.com
And everything is open source! Ask for new functionalities or contribute in:
https://github.com/AlexBodner/How_Much_VRAM
If it's useful to you leave a star ๐and share it to someone that will find the tool useful!
More discussion in: https://x.com/AlexBodner_/status/1832054850294812679
inflatebot/MN-12B-Mag-Mell-R1
https://huggingface.co/inflatebot/MN-12B-Mag-Mell-R1
MN-12B-Mag-Mell is a multi-stage merge, inspired by hypermerges like Tiefighter and Umbral Mind, intended for use as a general-purpose "Best of Nemo" model for co-writing, roleplay, and text adventures.
Consistently, Mag Mell produced prose that shocked testers, with a minimum of "slop". It also exhibited a unique sense of humor, and a propensity for inserting bespoke details into adventuring scenarios.
https://huggingface.co/inflatebot/MN-12B-Mag-Mell-R1
MN-12B-Mag-Mell is a multi-stage merge, inspired by hypermerges like Tiefighter and Umbral Mind, intended for use as a general-purpose "Best of Nemo" model for co-writing, roleplay, and text adventures.
Consistently, Mag Mell produced prose that shocked testers, with a minimum of "slop". It also exhibited a unique sense of humor, and a propensity for inserting bespoke details into adventuring scenarios.
huggingface.co
inflatebot/MN-12B-Mag-Mell-R1 ยท Hugging Face
Weโre on a journey to advance and democratize artificial intelligence through open source and open science.
Have you tried the new SQL Console yet?
Would love to know any queries you've tried or general feedback! If you haven't go try it out and let us know ๐ค
If you have some interesting queries feel free to share the URLs as well!
Would love to know any queries you've tried or general feedback! If you haven't go try it out and let us know ๐ค
If you have some interesting queries feel free to share the URLs as well!
๐๐
๐๐ฟ๐ฎ๐ฐ๐๐ถ๐ป๐ด ๐๐ผ๐๐ฟ ๐๐ง๐ ๐ ๐๐ฒ๐ฏ๐ฝ๐ฎ๐ด๐ฒ๐ ๐๐ผ ๐บ๐ฎ๐ฟ๐ธ๐ฑ๐ผ๐๐ป ๐ถ๐ ๐ป๐ผ๐ ๐ฝ๐ผ๐๐๐ถ๐ฏ๐น๐ฒ ๐ฒ๐ป๐ฑ-๐๐ผ-๐ฒ๐ป๐ฑ ๐๐ถ๐๐ต ๐ฎ ๐๐ถ๐บ๐ฝ๐น๐ฒ ๐๐๐ ! ๐
Jina just released Reader-LM, that handles the whole pipeline of extracting markdown from HTML webpages.
A while ago, Jina had released a completely code-based deterministic program to do this extraction, based on some heuristics : e.g., โif the text is in a <p> tag, keep it, but if itโs hidden behind another, remove itโ.
๐ค But they received complaints from readers: some found it too detailed, other not enough, depending on the pages.
โก๏ธ So they decided, ๐บ๐ฎ๐๐ฏ๐ฒ ๐ต๐ฒ๐๐ฟ๐ถ๐๐๐ถ๐ฐ๐ ๐๐ฒ๐ฟ๐ฒ ๐ป๐ผ๐ ๐ฒ๐ป๐ผ๐๐ด๐ต: ๐ถ๐ป๐๐๐ฒ๐ฎ๐ฑ, ๐๐ต๐ฒ๐ ๐๐ฟ๐ถ๐ฒ๐ฑ ๐๐ผ ๐๐ฟ๐ฎ๐ถ๐ป ๐ฎ ๐๐๐ ๐๐ผ ๐ฑ๐ผ ๐๐ต๐ฒ ๐ฐ๐ผ๐บ๐ฝ๐น๐ฒ๐๐ฒ ๐ฒ๐ ๐๐ฟ๐ฎ๐ฐ๐๐ถ๐ผ๐ป. This LLM does not need to be very strong,but it should handle a very long context: itโs a challenging, โshallow-but-wideโ architecture.
๐ง๐ฒ๐ฐ๐ต๐ป๐ถ๐ฐ๐ฎ๐น ๐ถ๐ป๐๐ถ๐ด๐ต๐๐:
2๏ธโฃ models: Reader-LM-0.5B and 1.5B
โ๏ธ Two stages of training: first, short and simple HTML to get the basics, then ramp up to longer and harder HTML up to 128k tokens
๐ Use contrastive search for decoding: this empirically reduces โrepeating outputโ issues
โก๏ธ Their models beat much larger models at HTML extraction ๐ฅ
๐ค Weights available on HF (sadly cc-by-nc license):
jinaai/reader-lm-1.5b
Jina just released Reader-LM, that handles the whole pipeline of extracting markdown from HTML webpages.
A while ago, Jina had released a completely code-based deterministic program to do this extraction, based on some heuristics : e.g., โif the text is in a <p> tag, keep it, but if itโs hidden behind another, remove itโ.
๐ค But they received complaints from readers: some found it too detailed, other not enough, depending on the pages.
โก๏ธ So they decided, ๐บ๐ฎ๐๐ฏ๐ฒ ๐ต๐ฒ๐๐ฟ๐ถ๐๐๐ถ๐ฐ๐ ๐๐ฒ๐ฟ๐ฒ ๐ป๐ผ๐ ๐ฒ๐ป๐ผ๐๐ด๐ต: ๐ถ๐ป๐๐๐ฒ๐ฎ๐ฑ, ๐๐ต๐ฒ๐ ๐๐ฟ๐ถ๐ฒ๐ฑ ๐๐ผ ๐๐ฟ๐ฎ๐ถ๐ป ๐ฎ ๐๐๐ ๐๐ผ ๐ฑ๐ผ ๐๐ต๐ฒ ๐ฐ๐ผ๐บ๐ฝ๐น๐ฒ๐๐ฒ ๐ฒ๐ ๐๐ฟ๐ฎ๐ฐ๐๐ถ๐ผ๐ป. This LLM does not need to be very strong,but it should handle a very long context: itโs a challenging, โshallow-but-wideโ architecture.
๐ง๐ฒ๐ฐ๐ต๐ป๐ถ๐ฐ๐ฎ๐น ๐ถ๐ป๐๐ถ๐ด๐ต๐๐:
2๏ธโฃ models: Reader-LM-0.5B and 1.5B
โ๏ธ Two stages of training: first, short and simple HTML to get the basics, then ramp up to longer and harder HTML up to 128k tokens
๐ Use contrastive search for decoding: this empirically reduces โrepeating outputโ issues
โก๏ธ Their models beat much larger models at HTML extraction ๐ฅ
๐ค Weights available on HF (sadly cc-by-nc license):
jinaai/reader-lm-1.5b
Hugging face presents FineVideo ๐ฅ! Unlocking the next generation of Video understanding ๐
๐คฏ3400 hours of annotated Creative Common videos with rich character descriptions, scene splits, mood, and content descriptions per scene as well as QA pairs.
๐ฅ
@mfarre processed over 2M videos of Youtube-CC to make this incredibly powerful selection.
Very psyched to fine-tune idefics on this dataset. โก๏ธ
Explore the videos:
HuggingFaceFV/FineVideo-Explorer
๐คฏ3400 hours of annotated Creative Common videos with rich character descriptions, scene splits, mood, and content descriptions per scene as well as QA pairs.
๐ฅ
@mfarre processed over 2M videos of Youtube-CC to make this incredibly powerful selection.
Very psyched to fine-tune idefics on this dataset. โก๏ธ
Explore the videos:
HuggingFaceFV/FineVideo-Explorer
Hugging face presents FineVideo ๐ฅ! Unlocking the next generation of Video understanding ๐
๐คฏ3400 hours of annotated Creative Common videos with rich character descriptions, scene splits, mood, and content descriptions per scene as well as QA pairs.
๐ฅ
@mfarre processed over 2M videos of Youtube-CC to make this incredibly powerful selection.
Very psyched to fine-tune idefics on this dataset. โก๏ธ
Explore the videos:
HuggingFaceFV/FineVideo-Explorer
๐คฏ3400 hours of annotated Creative Common videos with rich character descriptions, scene splits, mood, and content descriptions per scene as well as QA pairs.
๐ฅ
@mfarre processed over 2M videos of Youtube-CC to make this incredibly powerful selection.
Very psyched to fine-tune idefics on this dataset. โก๏ธ
Explore the videos:
HuggingFaceFV/FineVideo-Explorer
๐๐ฉ๐๐ง๐๐ ๐๐ข๐ง๐๐ฅ๐ฅ๐ฒ ๐ซ๐๐ฏ๐๐๐ฅ๐ฌ โ๐โ: ๐๐ซ๐๐ณ๐ฒ ๐๐ก๐๐ข๐ง-๐จ๐-๐ญ๐ก๐จ๐ฎ๐ ๐ก๐ญ-๐ญ๐ฎ๐ง๐๐ ๐ฆ๐จ๐๐๐ฅ >> ๐๐๐-๐๐จ ๐ฅ
OpenAI had hinted at a mysterious โproject strawberryโ for a long time: ๐๐ต๐ฒ๐ ๐ฝ๐๐ฏ๐น๐ถ๐๐ต๐ฒ๐ฑ ๐๐ต๐ถ๐ ๐ป๐ฒ๐ ๐บ๐ผ๐ฑ๐ฒ๐น ๐ฐ๐ฎ๐น๐น๐ฒ๐ฑ โ๐ผ๐ญโ ๐ญ๐ต๐ผ๐๐ฟ ๐ฎ๐ด๐ผ, ๐ฎ๐ป๐ฑ ๐๐ต๐ฒ ๐ฝ๐ฒ๐ฟ๐ณ๐ผ๐ฟ๐บ๐ฎ๐ป๐ฐ๐ฒ ๐ถ๐ ๐ท๐๐๐ ๐บ๐ถ๐ป๐ฑ-๐ฏ๐น๐ผ๐๐ถ๐ป๐ด.
๐คฏ Ranks among the top 500 students in the US in a qualifier for the USA Math Olympiad
๐คฏ Beats human-PhD-level accuracy by 8% on GPQA, hard science problems benchmark where the previous best was Claude 3.5 Sonnet with 59.4.
๐คฏ Scores 78.2% on vision benchmark MMMU, making it the first model competitive w/ human experts
๐คฏ GPT-4o on MATH scored 60% โ o1 scores 95%
How did they pull this? Sadly OpenAI keeps increasing their performance in โmaking cryptic AF reports to not reveal any real infoโ, so here are excerpts:
๐ฌ โ๐ผ๐ญ ๐๐๐ฒ๐ ๐ฎ ๐ฐ๐ต๐ฎ๐ถ๐ป ๐ผ๐ณ ๐๐ต๐ผ๐๐ด๐ต๐ ๐๐ต๐ฒ๐ป ๐ฎ๐๐๐ฒ๐บ๐ฝ๐๐ถ๐ป๐ด ๐๐ผ ๐๐ผ๐น๐๐ฒ ๐ฎ ๐ฝ๐ฟ๐ผ๐ฏ๐น๐ฒ๐บ. ๐ง๐ต๐ฟ๐ผ๐๐ด๐ต ๐ฟ๐ฒ๐ถ๐ป๐ณ๐ผ๐ฟ๐ฐ๐ฒ๐บ๐ฒ๐ป๐ ๐น๐ฒ๐ฎ๐ฟ๐ป๐ถ๐ป๐ด, ๐ผ๐ญ ๐น๐ฒ๐ฎ๐ฟ๐ป๐ ๐๐ผ ๐ต๐ผ๐ป๐ฒ ๐ถ๐๐ ๐ฐ๐ต๐ฎ๐ถ๐ป ๐ผ๐ณ ๐๐ต๐ผ๐๐ด๐ต๐ ๐ฎ๐ป๐ฑ ๐ฟ๐ฒ๐ณ๐ถ๐ป๐ฒ ๐๐ต๐ฒ ๐๐๐ฟ๐ฎ๐๐ฒ๐ด๐ถ๐ฒ๐ ๐ถ๐ ๐๐๐ฒ๐. It learns to recognize and correct its mistakes.โ
And of course, they decide to hide the content of this precious Chain-of-
Thought. Would it be for maximum profit? Of course not, you awful capitalist, itโs to protect users:
๐ฌ โWe also do not want to make an unaligned chain of thought directly visible to users.โ
Theyโre right, it would certainly have hurt my feelings to see the internal of this model tearing apart math problems.
๐ค I suspect it could be not only CoT, but also some agentic behaviour where the model can just call a code executor. The kind of score improvement the show certainly looks like the ones you see with agents.
This model will be immediately released for ChatGPT and some โtrusted API usersโ.
Letโs start cooking to release the same thing in 6 months! ๐
OpenAI had hinted at a mysterious โproject strawberryโ for a long time: ๐๐ต๐ฒ๐ ๐ฝ๐๐ฏ๐น๐ถ๐๐ต๐ฒ๐ฑ ๐๐ต๐ถ๐ ๐ป๐ฒ๐ ๐บ๐ผ๐ฑ๐ฒ๐น ๐ฐ๐ฎ๐น๐น๐ฒ๐ฑ โ๐ผ๐ญโ ๐ญ๐ต๐ผ๐๐ฟ ๐ฎ๐ด๐ผ, ๐ฎ๐ป๐ฑ ๐๐ต๐ฒ ๐ฝ๐ฒ๐ฟ๐ณ๐ผ๐ฟ๐บ๐ฎ๐ป๐ฐ๐ฒ ๐ถ๐ ๐ท๐๐๐ ๐บ๐ถ๐ป๐ฑ-๐ฏ๐น๐ผ๐๐ถ๐ป๐ด.
๐คฏ Ranks among the top 500 students in the US in a qualifier for the USA Math Olympiad
๐คฏ Beats human-PhD-level accuracy by 8% on GPQA, hard science problems benchmark where the previous best was Claude 3.5 Sonnet with 59.4.
๐คฏ Scores 78.2% on vision benchmark MMMU, making it the first model competitive w/ human experts
๐คฏ GPT-4o on MATH scored 60% โ o1 scores 95%
How did they pull this? Sadly OpenAI keeps increasing their performance in โmaking cryptic AF reports to not reveal any real infoโ, so here are excerpts:
๐ฌ โ๐ผ๐ญ ๐๐๐ฒ๐ ๐ฎ ๐ฐ๐ต๐ฎ๐ถ๐ป ๐ผ๐ณ ๐๐ต๐ผ๐๐ด๐ต๐ ๐๐ต๐ฒ๐ป ๐ฎ๐๐๐ฒ๐บ๐ฝ๐๐ถ๐ป๐ด ๐๐ผ ๐๐ผ๐น๐๐ฒ ๐ฎ ๐ฝ๐ฟ๐ผ๐ฏ๐น๐ฒ๐บ. ๐ง๐ต๐ฟ๐ผ๐๐ด๐ต ๐ฟ๐ฒ๐ถ๐ป๐ณ๐ผ๐ฟ๐ฐ๐ฒ๐บ๐ฒ๐ป๐ ๐น๐ฒ๐ฎ๐ฟ๐ป๐ถ๐ป๐ด, ๐ผ๐ญ ๐น๐ฒ๐ฎ๐ฟ๐ป๐ ๐๐ผ ๐ต๐ผ๐ป๐ฒ ๐ถ๐๐ ๐ฐ๐ต๐ฎ๐ถ๐ป ๐ผ๐ณ ๐๐ต๐ผ๐๐ด๐ต๐ ๐ฎ๐ป๐ฑ ๐ฟ๐ฒ๐ณ๐ถ๐ป๐ฒ ๐๐ต๐ฒ ๐๐๐ฟ๐ฎ๐๐ฒ๐ด๐ถ๐ฒ๐ ๐ถ๐ ๐๐๐ฒ๐. It learns to recognize and correct its mistakes.โ
And of course, they decide to hide the content of this precious Chain-of-
Thought. Would it be for maximum profit? Of course not, you awful capitalist, itโs to protect users:
๐ฌ โWe also do not want to make an unaligned chain of thought directly visible to users.โ
Theyโre right, it would certainly have hurt my feelings to see the internal of this model tearing apart math problems.
๐ค I suspect it could be not only CoT, but also some agentic behaviour where the model can just call a code executor. The kind of score improvement the show certainly looks like the ones you see with agents.
This model will be immediately released for ChatGPT and some โtrusted API usersโ.
Letโs start cooking to release the same thing in 6 months! ๐
I believe Hugging Face should have something similar to Hacktoberfest. I miss the days when there were events like this every 3 months for audio, deep reinforcement learning, gradio themes, but it turns out everything slowed down. There are no more Hugging Face events.