AIBRUH/eden-eve-4d-pipeline
Updated
TJ SPENCER JACQUES
AIBRUH
Β·
AI & ML interests
None yet
Recent Activity
published
a model 4 days ago
AIBRUH/eden-eve-4d-pipeline updated
a Space 8 days ago
AIBRUH/eden-ui-studio published
a Space 8 days ago
AIBRUH/eden-ui-studio Organizations
published a model 4 days ago
updated a Space 8 days ago
published a Space 8 days ago
updated a Space 8 days ago
published a Space 8 days ago
updated a Space 8 days ago
published a Space 8 days ago
updated a Space 9 days ago
published a Space 9 days ago
updated a Space 9 days ago
published a Space 9 days ago
updated a Space 9 days ago
published a Space 9 days ago
liked a Space 9 days ago
Request: Access Request for sesame/csm-1b
2
#21 opened 12 months ago
by
Sadegh69
reacted to Locutusque's post with π₯ 4 months ago
Post
2771
π AutoXLA - Accelerating Large Models on TPU
AutoXLA is an experimental library that automates the distribution, optimization, and quantization of large language models for TPUs using PyTorch/XLA. It extends the Hugging Face Transformers interface with TPU-aware features such as automatic sharding, custom attention kernels, and quantization-aware loading, making large-scale deployment and training both simpler and faster.
With quantization and Splash Attention kernels, AutoXLA achieves up to 4Γ speedups over standard Flash Attention implementations, significantly improving throughput for both inference and training workloads.
Whether youβre experimenting with distributed setups (FSDP, 2D, or 3D sharding) or optimizing memory via LanguageModelQuantizer, AutoXLA is built to make scaling LLMs on TPU seamless.
β οΈ Note: This is an experimental repository. Expect rough edges! Please report bugs or unexpected behavior through GitHub issues.
π GitHub Repository: https://github.com/Locutusque/AutoXLA
AutoXLA is an experimental library that automates the distribution, optimization, and quantization of large language models for TPUs using PyTorch/XLA. It extends the Hugging Face Transformers interface with TPU-aware features such as automatic sharding, custom attention kernels, and quantization-aware loading, making large-scale deployment and training both simpler and faster.
With quantization and Splash Attention kernels, AutoXLA achieves up to 4Γ speedups over standard Flash Attention implementations, significantly improving throughput for both inference and training workloads.
Whether youβre experimenting with distributed setups (FSDP, 2D, or 3D sharding) or optimizing memory via LanguageModelQuantizer, AutoXLA is built to make scaling LLMs on TPU seamless.
β οΈ Note: This is an experimental repository. Expect rough edges! Please report bugs or unexpected behavior through GitHub issues.
π GitHub Repository: https://github.com/Locutusque/AutoXLA
liked a model 4 months ago
