llama : store non-RoPEd K cache (WIP) - SemanticDiff

Commit

2 years ago

llama : store non-RoPEd K cache (WIP)

References

custom-attention-mask-no-roped-cache

#3234 - llama : store non-RoPEd K cache

Author

ggerganov

ggerganov

Committer

ggerganov

ggerganov

Parents

Loading