nvidia-cosmos/cosmos-rl
Dec 2025 – Dec 2025
1 Month active
Languages Used
Python
Technical Skills
Deep LearningGPU programmingPythondeep learningmachine learningmodel optimization
bastefaniak
PROFILE
Bastefaniak
Worked on the nvidia-cosmos/cosmos-rl repository to enhance model reliability and hardware compatibility for deep learning workloads. Addressed stability issues on legacy GPUs by disabling the DeepEP feature for architectures older than Hopper, ensuring broader support across diverse hardware. Improved the accuracy of Mixture of Experts (MoE) routing by correcting the n_local_experts computation for DeepseekV3 and Qwen3 models, which increased performance and efficiency in parallelized environments. Leveraged Python, deep learning, and GPU programming expertise to align these changes with the hardware support matrix, reducing edge-case failures and enabling smoother deployment for machine learning models on a wider range of GPU architectures.
Overall Statistics
Feature vs Bugs
50%Features
Repository Contributions
2Total
Bugs
1
Commits
2
Features
1
Lines of code
28
Activity Months1
Your Network
1647 peopleSame Organization
@nvidia.com1624
Work History
December 2025
2 Commits • 1 Features
Dec 1, 2025December 2025 (nvidia-cosmos/cosmos-rl): Key features delivered and bugs fixed with a focus on hardware compatibility and MoE reliability. Achieved stability for legacy GPUs by disabling DeepEP on architectures older than Hopper, and corrected MoE routing by fixing n_local_experts computation for DeepseekV3 and Qwen3. These changes reduce edge-case failures, improve performance and efficiency, and support broader deployment across GPU architectures.
2 Commits • 1 Features
Dec 1, 2025December 2025 (nvidia-cosmos/cosmos-rl): Key features delivered and bugs fixed with a focus on hardware compatibility and MoE reliability. Achieved stability for legacy GPUs by disabling DeepEP on architectures older than Hopper, and corrected MoE routing by fixing n_local_experts computation for DeepseekV3 and Qwen3. These changes reduce edge-case failures, improve performance and efficiency, and support broader deployment across GPU architectures.
December 2025
Activity
Loading activity data...
Quality Metrics
Correctness100.0%
Maintainability80.0%
Architecture80.0%
Performance80.0%
AI Usage30.0%
Skills & Technologies
Programming Languages
Python
Technical Skills
Deep LearningGPU programmingPythondeep learningmachine learningmodel optimization
Repositories Contributed To
Overview of all repositories you've contributed to across your timeline