DeepSeek open source DeepEP – library for MoE training and Inference
1–10 of 77 posts
Re: DeepSeek open source DeepEP – library for MoE training and Inference
#2- Efficient and optimized all-to-all communication
- Both intranode and internode support with NVLink and RDMA
- High-throughput kernels for training and inference prefilling
- Low-latency kernels for inference decoding
- Native FP8 dispatch support
- Flexible GPU resource control for computation-communication overlapping
X: https://x.com/deepseek_ai/status/1894211757604049133
Re: DeepSeek open source DeepEP – library for MoE training and Inference
#3The PTX instructions they talked about in the tech report should be pointing to the code here?
Re: DeepSeek open source DeepEP – library for MoE training and Inference
#4The PTX instructions they talked about in the tech report should be pointing to the code here?
this might help: https://x.com/main_horse/status/1894215779521794058/photo/1
Re: DeepSeek open source DeepEP – library for MoE training and Inference
#5You gotta love these guys, they're really pushing the open source frontier for all of us, thanks for sharing
Re: DeepSeek open source DeepEP – library for MoE training and Inference
#6You gotta love these guys, they're really pushing the open source frontier for all of us, thanks for sharing
Open AI™ (with a space)
Re: DeepSeek open source DeepEP – library for MoE training and Inference
#7I feel like a kid in a candy shop. Some of these tricks would take way too long to reverse engineer correctly based on the papers. I hope that the releases this week start a renaissance of the use of MoE as baseline academic models.
Re: DeepSeek open source DeepEP – library for MoE training and Inference
#8Re: DeepSeek open source DeepEP – library for MoE training and Inference
#9Is the PTX that everyone was looking forward to included this time?