Author
Shang Tingyu
Papers that credit Shang Tingyu in their source bylines. Matching names may refer to different researchers.
1 results
Papers that credit Shang Tingyu in their source bylines. Matching names may refer to different researchers.
1 results
BASS compresses Mamba-2 states by preserving joint write-and-read structure, with lower average WikiText-2 perplexity cost than GHOST at half retention.