Hugging Face's logo

C4AI Community

community
https://cohere.com/research
CohereForAI
Activity Feed Request to join this org

AI & ML interests

None defined yet.

Recent Activity

roshansk23  authored a paper 5 days ago
INCLUDE: Evaluating Multilingual Language Understanding with Regional Knowledge
roshansk23  authored a paper 5 days ago
Maya: An Instruction Finetuned Multilingual Multimodal Model
jjzha  authored a paper 6 days ago
SnakModel: Lessons Learned from Training an Open Danish Large Language Model
View all activity

C4AI-Community's activity

singhsidhukuldeep 
posted an update 1 day ago
view post
Post
1580
Exciting News in AI: JinaAI Releases JINA-CLIP-v2!

The team at Jina AI has just released a groundbreaking multilingual multimodal embedding model that's pushing the boundaries of text-image understanding. Here's why this is a big deal:

🚀 Technical Highlights:
- Dual encoder architecture combining a 561M parameter Jina XLM-RoBERTa text encoder and a 304M parameter EVA02-L14 vision encoder
- Supports 89 languages with 8,192 token context length
- Processes images up to 512×512 pixels with 14×14 patch size
- Implements FlashAttention2 for text and xFormers for vision processing
- Uses Matryoshka Representation Learning for efficient vector storage

⚡️ Under The Hood:
- Multi-stage training process with progressive resolution scaling (224→384→512)
- Contrastive learning using InfoNCE loss in both directions
- Trained on massive multilingual dataset including 400M English and 400M multilingual image-caption pairs
- Incorporates specialized datasets for document understanding, scientific graphs, and infographics
- Uses hard negative mining with 7 negatives per positive sample

📊 Performance:
- Outperforms previous models on visual document retrieval (52.65% nDCG@5)
- Achieves 89.73% image-to-text and 79.09% text-to-image retrieval on CLIP benchmark
- Strong multilingual performance across 30 languages
- Maintains performance even with 75% dimension reduction (256D vs 1024D)

🎯 Key Innovation:
The model solves the long-standing challenge of unifying text-only and multi-modal retrieval systems while adding robust multilingual support. Perfect for building cross-lingual visual search systems!

Kudos to the research team at Jina AI for this impressive advancement in multimodal AI!
singhsidhukuldeep 
posted an update 3 days ago
view post
Post
1154
Fascinating insights from @Pinterest 's latest research on improving feature interactions in recommendation systems!

Pinterest's engineering team has tackled a critical challenge in their Homefeed ranking system that serves 500M+ monthly active users. Here's what makes their approach remarkable:

>> Technical Deep Dive

Architecture Overview
• The ranking model combines dense features, sparse features, and embedding features to represent users, Pins, and context
• Sparse features are processed using learnable embeddings with size based on feature cardinality
• User sequence embeddings are generated using a transformer architecture processing past engagements

Feature Processing Pipeline
• Dense features undergo normalization for numerical stability
• Sparse and embedding features receive L2 normalization
• All features are concatenated into a single feature embedding

Key Innovations
• Implemented parallel MaskNet layers with 3 blocks
• Used projection ratio of 2.0 and output dimension of 512
• Stacked 4 DCNv2 layers on top for higher-order interactions

Performance Improvements
• Achieved +1.42% increase in Homefeed Save Volume
• Boosted Overall Time Spent by +0.39%
• Maintained memory consumption increase to just 5%

>> Industry Constraints Addressed

Memory Management
• Optimized for 60% GPU memory utilization
• Prevented OOM errors while maintaining batch size efficiency

Latency Optimization
• Removed input-output concatenation before MLP
• Reduced hidden layer sizes in MLP
• Achieved zero latency increase while improving performance

System Stability
• Ensured reproducible results across retraining
• Maintained model stability across different data distributions
• Successfully deployed in production environment

This work brilliantly demonstrates how to balance academic innovations with real-world industrial constraints. Kudos to the Pinterest team!
singhsidhukuldeep 
posted an update 4 days ago
view post
Post
3523
Exciting breakthrough in AI: @Meta 's new Byte Latent Transformer (BLT) revolutionizes language models by eliminating tokenization!

The BLT architecture introduces a groundbreaking approach that processes raw bytes instead of tokens, achieving state-of-the-art performance while being more efficient and robust. Here's what makes it special:

>> Key Innovations
Dynamic Patching: BLT groups bytes into variable-sized patches based on entropy, allocating more compute power where the data is more complex. This results in up to 50% fewer FLOPs during inference compared to traditional token-based models.

Three-Component Architecture:
• Lightweight Local Encoder that converts bytes to patch representations
• Powerful Global Latent Transformer that processes patches
• Local Decoder that converts patches back to bytes

>> Technical Advantages
• Matches performance of Llama 3 at 8B parameters while being more efficient
• Superior handling of non-English languages and rare character sequences
• Remarkable 99.9% accuracy on spelling tasks
• Better scaling properties than token-based models

>> Under the Hood
The system uses an entropy model to determine patch boundaries, cross-attention mechanisms for information flow, and hash n-gram embeddings for improved representation. The architecture allows simultaneous scaling of both patch and model size while maintaining fixed inference costs.

This is a game-changer for multilingual AI and could reshape how we build future language models. Excited to see how this technology evolves!
  • 2 replies
·
jjzha 
authored a paper 6 days ago
takarajordan 
posted an update 7 days ago
view post
Post
1049
I made an RSS feed for HuggingFace Daily Papers!! 🤗

Just Subscribe here: https://papers.takara.ai/api/feed

It updates every 24 hours, completely written as a serverless go script with a Redis cache (to avoid hitting HF all the time).

I'm open sourcing the code, you can check out my repo and deploy it on Vercel extremely easily!
https://github.com/404missinglink/HF-Daily-Papers-Feeds

thanks to @John6666 @p3nGu1nZz for your early support
jinunyachhyon 
authored a paper 7 days ago
peaceAsh 
authored a paper 8 days ago
jjzha 
authored a paper 9 days ago
singhsidhukuldeep 
posted an update 12 days ago
view post
Post
1227
Groundbreaking Research Alert: The 'H' in HNSW Stands for "Hubs", Not "Hierarchy"!

Fascinating new research reveals that the hierarchical structure in the popular HNSW (Hierarchical Navigable Small World) algorithm - widely used for vector similarity search - may be unnecessary for high-dimensional data.

🔬 Key Technical Findings:

• The hierarchical layers in HNSW can be completely removed for vectors with dimensionality > 32, with no performance loss

• Memory savings of up to 38% achieved by removing the hierarchy

• Performance remains identical in both median and tail latency cases across 13 benchmark datasets

🛠️ Under The Hood:
The researchers discovered that "hub highways" naturally form in high-dimensional spaces. These hubs are well-connected nodes that are frequently traversed during searches, effectively replacing the need for explicit hierarchical layers.

The hub structure works because:
• A small subset of nodes appear disproportionately in nearest neighbor lists
• These hub nodes form highly connected subgraphs
• Queries naturally traverse through these hubs early in the search process
• The hubs efficiently connect distant regions of the graph

💡 Industry Impact:
This finding has major implications for vector databases and similarity search systems. Companies can significantly reduce memory usage while maintaining performance by implementing flat navigable small world graphs instead of hierarchical ones.

🚀 What's Next:
The researchers have released FlatNav, an open-source implementation of their flat navigable small world approach, enabling immediate practical applications of these findings.
shayekh 
authored 2 papers 12 days ago
kkr5155 
authored a paper 12 days ago
kkr5155 
authored a paper 13 days ago
singhsidhukuldeep 
posted an update 14 days ago
view post
Post
454
Fascinating new research alert! Just read a groundbreaking paper on understanding Retrieval-Augmented Generation (RAG) systems and their performance factors.

Key insights from this comprehensive study:

>> Architecture Deep Dive
The researchers analyzed RAG systems across 6 datasets (3 code-related, 3 QA-focused) using multiple LLMs. Their investigation revealed critical insights into four key design factors:

Document Types Impact:
• Oracle documents (ground truth) aren't always optimal
• Distracting documents significantly degrade performance
• Surprisingly, irrelevant documents boost code generation by up to 15.6%

Retrieval Precision:
• Performance varies dramatically by task
• QA tasks need 20-100% retrieval recall
• Perfect retrieval still fails up to 12% of the time on previously correct instances

Document Selection:
• More documents ≠ better results
• Adding documents can cause errors on previously correct samples
• Performance degradation increases ~1% per 5 additional documents in code tasks

Prompt Engineering:
• Most advanced prompting techniques underperform simple zero-shot prompts
• Technique effectiveness varies significantly across models and tasks
• Complex prompts excel at difficult problems but struggle with simple ones

>> Technical Implementation
The study utilized:
• Multiple retrievers including BM25, dense retrievers, and specialized models
• Comprehensive corpus of 70,956 unique API documents
• Over 200,000 API calls and 1,000+ GPU hours of computation
• Sophisticated evaluation metrics tracking both correctness and system confidence

💡 Key takeaway: RAG system optimization requires careful balancing of multiple factors - there's no one-size-fits-all solution.
  • 1 reply
·
takarajordan 
posted an update 14 days ago
view post
Post
2200
I'm super excited to release my first open-source text dataset:

WorldScenario 20K is a novel dataset of 20,000 synthetically generated multi-stakeholder scenarios designed to simulate real-world decision-making processes. Each scenario explores a unique environmental, societal, or economic issue.

I used the brand new meta-llama/Llama-3.3-70B-Instruct model to generate this dataset and I put the dataset through some post processing to clean and evaluate the dataset for diversity.

I'd appreciate some feedback and thoughts on my new release! Thanks!

takarajordan/WorldScenario_20K
·
SuryaKrishna02 
authored a paper 14 days ago
chaturvediisha6 
authored a paper 14 days ago
singhsidhukuldeep 
posted an update 15 days ago
view post
Post
1798
Exciting new research alert! 🚀 A groundbreaking paper titled "Understanding LLM Embeddings for Regression" has just been released, and it's a game-changer for anyone working with large language models (LLMs) and regression tasks.

Key findings:

1. LLM embeddings outperform traditional feature engineering in high-dimensional regression tasks.

2. LLM embeddings preserve Lipschitz continuity over feature space, enabling better regression performance.

3. Surprisingly, factors like model size and language understanding don't always improve regression outcomes.

Technical details:

The researchers used both T5 and Gemini model families to benchmark embedding-based regression. They employed a key-value JSON format for string representations and used average-pooling to aggregate Transformer outputs.

The study introduced a novel metric called Normalized Lipschitz Factor Distribution (NLFD) to analyze embedding continuity. This metric showed a high inverse relationship between the skewedness of the NLFD and regression performance.

Interestingly, the paper reveals that applying forward passes of pre-trained models doesn't always significantly improve regression performance for certain tasks. In some cases, using only vocabulary embeddings without a forward pass yielded comparable results.

The research also demonstrated that LLM embeddings are dimensionally robust, maintaining strong performance even with high-dimensional data where traditional representations falter.

This work opens up exciting possibilities for using LLM embeddings in various regression tasks, particularly those with high degrees of freedom. It's a must-read for anyone working on machine learning, natural language processing, or data science!
beyzaermis 
authored 2 papers 16 days ago