Opensource Lorsas and Transcoders
AI & ML interests
LLM
Recent Activity
View all activity
Papers
BandPO: Bridging Trust Regions and Ratio Clipping via Probability-Aware Bounds for LLM Reinforcement Learning
MOSS-Audio-Tokenizer: Scaling Audio Tokenizers for Future Audio Foundation Models
A unified multimodal large language model for end-to-end speaker-attributed, time-stamped transcription.
Evaluating Agentic Backend Coding Capabilities in Real-World Development Scenarios
-
ABC-Bench: Benchmarking Agentic Backend Coding in Real-World Development
Paper ā¢ 2601.11077 ā¢ Published ā¢ 65 -
OpenMOSS-Team/ABC-Bench
Viewer ā¢ Updated ā¢ 224 ā¢ 172 ā¢ 3 -
OpenMOSS-Team/Qwen3-32B-ABC
Text Generation ā¢ 33B ā¢ Updated ā¢ 5 ā¢ 1 -
OpenMOSS-Team/Qwen3-8B-ABC
Text Generation ā¢ 8B ā¢ Updated ā¢ 5 ā¢ 2
[ICLR 2026] Game-RL: Synthesizing Multimodal Verifiable Game Data to Boost VLMs' General Reasoning
An Efficient Training Framework for Diffusion Language Models
-
World Modeling Makes a Better Planner: Dual Preference Optimization for Embodied Task Planning
Paper ā¢ 2503.10480 ā¢ Published ā¢ 56 -
Unleashing Embodied Task Planning Ability in LLMs via Reinforcement Learning
Paper ā¢ 2506.23127 ā¢ Published ā¢ 1 -
World-aware Planning Narratives Enhance Large Vision-Language Model Planner
Paper ā¢ 2506.21230 ā¢ Published -
OpenMOSS-Team/Embodied_R1-ScienceWorld
8B ā¢ Updated ā¢ 3
The MHA2MLA model published in the paper "Towards Economical Inference: Enabling DeepSeek's Multi-Head Latent Attention in Any Transformer-Based LLMs"
-
OpenMOSS-Team/SmolLM-135M-MLA-d_kv_8-refactor
Text Generation ā¢ 0.1B ā¢ Updated ā¢ 4 -
OpenMOSS-Team/SmolLM-135M-MLA-d_kv_32-refactor
Text Generation ā¢ 0.1B ā¢ Updated ā¢ 1 -
OpenMOSS-Team/SmolLM-135M-MLA-d_kv_16-refactor
Text Generation ā¢ 0.1B ā¢ Updated ā¢ 4 -
OpenMOSS-Team/SmolLM-360M-MLA-d_kv_8-refactor
Text Generation ā¢ 0.3B ā¢ Updated
-
OpenMOSS-Team/moss-moon-003-sft-plugin
Text Generation ā¢ Updated ā¢ 25 ā¢ 69 -
OpenMOSS-Team/moss-moon-003-sft
Text Generation ā¢ Updated ā¢ 63 ā¢ 127 -
OpenMOSS-Team/moss-moon-003-base
Text Generation ā¢ Updated ā¢ 136 ā¢ 131 -
OpenMOSS-Team/moss-moon-003-sft-int4
Text Generation ā¢ Updated ā¢ 34 ā¢ 40
-
OpenMOSS-Team/MOSS-TTS
Text-to-Speech ā¢ 8B ā¢ Updated ā¢ 106k ā¢ 340 -
OpenMOSS-Team/MOSS-TTS-Local-Transformer
Text-to-Speech ā¢ 3B ā¢ Updated ā¢ 58.7k ā¢ 20 -
OpenMOSS-Team/MOSS-TTS-Realtime
Text-to-Speech ā¢ 2B ā¢ Updated ā¢ 90.9k ā¢ 65 -
OpenMOSS-Team/MOSS-Audio-Tokenizer
Feature Extraction ā¢ 2B ā¢ Updated ā¢ 83.4k ā¢ 37
-
OpenMOSS-Team/MOSS-TTSD-v1.0
Text-to-Speech ā¢ 8B ā¢ Updated ā¢ 23.1k ā¢ 49 -
OpenMOSS-Team/MOSS-TTSD-v0.7
Text-to-Speech ā¢ 2B ā¢ Updated ā¢ 296 ā¢ 17 -
OpenMOSS-Team/MOSS-TTSD-v0.5
Text-to-Speech ā¢ 2B ā¢ Updated ā¢ 1.39k ā¢ 53 -
OpenMOSS-Team/MOSS-TTSD-v0
Text-to-Speech ā¢ 2B ā¢ Updated ā¢ 4 ā¢ 27
True Speech-to-Speech Langugage Model
First Omni-modal Future Forecasting Benchmark
https://github.com/OpenMOSS/FRoM-W1
Proactive Robot Manipulation in Omni-modal Context
Open source weights of Lorsa modules introduced in "Towards Understanding the Nature of Attention with Low-Rank Sparse Decomposition".
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 ā¢ 2502.14837 ā¢ Published ā¢ 3 -
OpenMOSS-Team/Llama-2-7B-MLA-d_kv_16
Text Generation ā¢ 6B ā¢ Updated ā¢ 1 -
OpenMOSS-Team/Llama-2-7B-MLA-d_kv_32
Text Generation ā¢ 6B ā¢ Updated ā¢ 1 -
OpenMOSS-Team/Llama-2-7B-MLA-d_kv_64
Text Generation ā¢ 7B ā¢ Updated ā¢ 4
Opensource Lorsas and Transcoders
-
OpenMOSS-Team/MOSS-TTS
Text-to-Speech ā¢ 8B ā¢ Updated ā¢ 106k ā¢ 340 -
OpenMOSS-Team/MOSS-TTS-Local-Transformer
Text-to-Speech ā¢ 3B ā¢ Updated ā¢ 58.7k ā¢ 20 -
OpenMOSS-Team/MOSS-TTS-Realtime
Text-to-Speech ā¢ 2B ā¢ Updated ā¢ 90.9k ā¢ 65 -
OpenMOSS-Team/MOSS-Audio-Tokenizer
Feature Extraction ā¢ 2B ā¢ Updated ā¢ 83.4k ā¢ 37
-
OpenMOSS-Team/MOSS-TTSD-v1.0
Text-to-Speech ā¢ 8B ā¢ Updated ā¢ 23.1k ā¢ 49 -
OpenMOSS-Team/MOSS-TTSD-v0.7
Text-to-Speech ā¢ 2B ā¢ Updated ā¢ 296 ā¢ 17 -
OpenMOSS-Team/MOSS-TTSD-v0.5
Text-to-Speech ā¢ 2B ā¢ Updated ā¢ 1.39k ā¢ 53 -
OpenMOSS-Team/MOSS-TTSD-v0
Text-to-Speech ā¢ 2B ā¢ Updated ā¢ 4 ā¢ 27
A unified multimodal large language model for end-to-end speaker-attributed, time-stamped transcription.
True Speech-to-Speech Langugage Model
Evaluating Agentic Backend Coding Capabilities in Real-World Development Scenarios
-
ABC-Bench: Benchmarking Agentic Backend Coding in Real-World Development
Paper ā¢ 2601.11077 ā¢ Published ā¢ 65 -
OpenMOSS-Team/ABC-Bench
Viewer ā¢ Updated ā¢ 224 ā¢ 172 ā¢ 3 -
OpenMOSS-Team/Qwen3-32B-ABC
Text Generation ā¢ 33B ā¢ Updated ā¢ 5 ā¢ 1 -
OpenMOSS-Team/Qwen3-8B-ABC
Text Generation ā¢ 8B ā¢ Updated ā¢ 5 ā¢ 2
First Omni-modal Future Forecasting Benchmark
[ICLR 2026] Game-RL: Synthesizing Multimodal Verifiable Game Data to Boost VLMs' General Reasoning
https://github.com/OpenMOSS/FRoM-W1
An Efficient Training Framework for Diffusion Language Models
Proactive Robot Manipulation in Omni-modal Context
-
World Modeling Makes a Better Planner: Dual Preference Optimization for Embodied Task Planning
Paper ā¢ 2503.10480 ā¢ Published ā¢ 56 -
Unleashing Embodied Task Planning Ability in LLMs via Reinforcement Learning
Paper ā¢ 2506.23127 ā¢ Published ā¢ 1 -
World-aware Planning Narratives Enhance Large Vision-Language Model Planner
Paper ā¢ 2506.21230 ā¢ Published -
OpenMOSS-Team/Embodied_R1-ScienceWorld
8B ā¢ Updated ā¢ 3
Open source weights of Lorsa modules introduced in "Towards Understanding the Nature of Attention with Low-Rank Sparse Decomposition".
The MHA2MLA model published in the paper "Towards Economical Inference: Enabling DeepSeek's Multi-Head Latent Attention in Any Transformer-Based LLMs"
-
OpenMOSS-Team/SmolLM-135M-MLA-d_kv_8-refactor
Text Generation ā¢ 0.1B ā¢ Updated ā¢ 4 -
OpenMOSS-Team/SmolLM-135M-MLA-d_kv_32-refactor
Text Generation ā¢ 0.1B ā¢ Updated ā¢ 1 -
OpenMOSS-Team/SmolLM-135M-MLA-d_kv_16-refactor
Text Generation ā¢ 0.1B ā¢ Updated ā¢ 4 -
OpenMOSS-Team/SmolLM-360M-MLA-d_kv_8-refactor
Text Generation ā¢ 0.3B ā¢ Updated
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 ā¢ 2502.14837 ā¢ Published ā¢ 3 -
OpenMOSS-Team/Llama-2-7B-MLA-d_kv_16
Text Generation ā¢ 6B ā¢ Updated ā¢ 1 -
OpenMOSS-Team/Llama-2-7B-MLA-d_kv_32
Text Generation ā¢ 6B ā¢ Updated ā¢ 1 -
OpenMOSS-Team/Llama-2-7B-MLA-d_kv_64
Text Generation ā¢ 7B ā¢ Updated ā¢ 4
-
OpenMOSS-Team/moss-moon-003-sft-plugin
Text Generation ā¢ Updated ā¢ 25 ā¢ 69 -
OpenMOSS-Team/moss-moon-003-sft
Text Generation ā¢ Updated ā¢ 63 ā¢ 127 -
OpenMOSS-Team/moss-moon-003-base
Text Generation ā¢ Updated ā¢ 136 ā¢ 131 -
OpenMOSS-Team/moss-moon-003-sft-int4
Text Generation ā¢ Updated ā¢ 34 ā¢ 40