Open source weights of Lorsa modules introduced in "Lorsa: Low-Rank Sparse Attention Finds Monosemantic Heads in Attention Superposition".
OpenMOSS, Fudan NLP, SII
university
AI & ML interests
LLM
Recent Activity
Organization Card
Joint OpenMOSS gourp from Fudan NLP, SII and MoSi Inc.
Collections
2
The MHA2MLA model published in the paper "Towards Economical Inference: Enabling DeepSeek's Multi-Head Latent Attention in Any Transformer-Based LLMs"
-
Towards Economical Inference: Enabling DeepSeek's Multi-Head Latent Attention in Any Transformer-based LLMs
Paper • 2502.14837 • Published • 4 -
fnlp/Llama-2-7B-MLA-d_kv_16
Text Generation • Updated • 547 -
fnlp/Llama-2-7B-MLA-d_kv_32
Text Generation • Updated • 8 -
fnlp/Llama-2-7B-MLA-d_kv_64
Text Generation • Updated • 9
models
73
fnlp/Lorsa
Updated
fnlp/Lorsa-Llama-3.1-8B
Updated
fnlp/Lorsa-Pythia-160M
Updated
fnlp/Llama-2-7B-MLA-d_kv_16
Text Generation
•
Updated
•
547
fnlp/Llama-2-7B-MLA-d_kv_32
Text Generation
•
Updated
•
8
fnlp/Llama-2-7B-MLA-d_kv_64
Text Generation
•
Updated
•
9
fnlp/Llama-2-7B-MHA-d_kv_256
Text Generation
•
Updated
•
13
fnlp/SmolLM-1B7-MLA-d_kv_8
Text Generation
•
Updated
•
8
fnlp/SmolLM-1B7-MLA-d_kv_16
Text Generation
•
Updated
•
49
fnlp/SmolLM-1B7-MLA-d_kv_32
Text Generation
•
Updated
•
35
datasets
15
fnlp/MHA2MLA-corpus-qwen1_5
Updated
•
36
fnlp/MHA2MLA-corpus-qwen2
Updated
•
71
fnlp/MHA2MLA-corpus-mistral-v0_1
Updated
•
48
fnlp/MHA2MLA-corpus-smollm_v1
Updated
•
26
fnlp/MHA2MLA-corpus-llama2
Updated
•
27
fnlp/Ultra-Innerthought
Viewer
•
Updated
•
2.09M
•
113
•
2
fnlp/case2code-data
Viewer
•
Updated
•
887k
•
22
fnlp/AnyInstruct-resolution-1024
Updated
•
185
fnlp/AnyInstruct
Updated
•
47
•
40
fnlp/moss-003-sft-data
Preview
•
Updated
•
83
•
65