Online soft compression models for RAG. We release the models with llama-1B as compressor.
AI & ML interests
3D vision, speech, NLP, neural retrieval, computer vision, LLM, FMs, visual localization, robotics
Recent Activity
View all activity
PISCO models are compression models for RAG. They are intended as plug-in replacement for RAG systems with x5 faster inference
A Hub for Human-Centric 3D Vision
Our SPLADE models
Provence: lightweight context pruning models for retrieval-augmented generation. Blogpost: https://huggingface.co/blog/nadiinchi/provence
Compression Models for RAG
-
Context Embeddings for Efficient Answer Generation in RAG
Paper • 2407.09252 • Published • 2 -
naver/cocom-v1-128-mistral-7b
Feature Extraction • 7B • Updated • 41 • 4 -
naver/cocom-v1-16-mistral-7b
Feature Extraction • 7B • Updated • 5 -
naver/cocom-v1-4-mistral-7b
Feature Extraction • 7B • Updated • 18 • 2
Multilingual Distilwhisper allows for better ASR performance in target languages by adding lightweight CLSR modules on top of whisper-small.
DUSt3R/MASt3R models
-
naver/MASt3R_ViTLarge_BaseDecoder_512_catmlpdpt_metric
Image-to-3D • 0.7B • Updated • 259k • 16 -
naver/DUSt3R_ViTLarge_BaseDecoder_512_dpt
Image-to-3D • 0.6B • Updated • 11.1k • 15 -
naver/DUSt3R_ViTLarge_BaseDecoder_512_linear
Image-to-3D • Updated • 699 -
naver/DUSt3R_ViTLarge_BaseDecoder_224_linear
Image-to-3D • Updated • 174
Online soft compression models for RAG. We release the models with llama-1B as compressor.
Provence: lightweight context pruning models for retrieval-augmented generation. Blogpost: https://huggingface.co/blog/nadiinchi/provence
PISCO models are compression models for RAG. They are intended as plug-in replacement for RAG systems with x5 faster inference
Compression Models for RAG
-
Context Embeddings for Efficient Answer Generation in RAG
Paper • 2407.09252 • Published • 2 -
naver/cocom-v1-128-mistral-7b
Feature Extraction • 7B • Updated • 41 • 4 -
naver/cocom-v1-16-mistral-7b
Feature Extraction • 7B • Updated • 5 -
naver/cocom-v1-4-mistral-7b
Feature Extraction • 7B • Updated • 18 • 2
A Hub for Human-Centric 3D Vision
Multilingual Distilwhisper allows for better ASR performance in target languages by adding lightweight CLSR modules on top of whisper-small.
Our SPLADE models
DUSt3R/MASt3R models
-
naver/MASt3R_ViTLarge_BaseDecoder_512_catmlpdpt_metric
Image-to-3D • 0.7B • Updated • 259k • 16 -
naver/DUSt3R_ViTLarge_BaseDecoder_512_dpt
Image-to-3D • 0.6B • Updated • 11.1k • 15 -
naver/DUSt3R_ViTLarge_BaseDecoder_512_linear
Image-to-3D • Updated • 699 -
naver/DUSt3R_ViTLarge_BaseDecoder_224_linear
Image-to-3D • Updated • 174