Author
Minshen Zhang
Papers that credit Minshen Zhang in their source bylines. Matching names may refer to different researchers.
2 results
Separating expert work helps some rack-scale language models
On one GB200 rack, separating expert work matched or beat tuned colocated serving on some decoding workloads.
A tree-based method extends language models to million token contexts
Recursive Sparse Attention uses a tree of cache summaries to extend a 32K-trained model to million-token tests.
