llama.cpp
fcca0a70 - refact : fix convert script + zero out KV cache to avoid nans (#3523)

Commit

1 year ago

refact : fix convert script + zero out KV cache to avoid nans (#3523) * refact : fix convert script + zero out KV cache to avoid nans * ggml : silu(-inf) should never happen * metal : assert various kernel requirements

References

#3523 - refact : fix convert script + zero out KV cache to avoid nans

Author

ggerganov

Parents

dcc09d25

Files6

convert-refact-hf-to-gguf.py
examples/parallel
- parallel.cpp
ggml-metal.m
ggml-metal.metal
ggml.c
llama.cpp

llama.cpp fcca0a70 - refact : fix convert script + zero out KV cache to avoid nans (#3523)

llama.cpp
fcca0a70 - refact : fix convert script + zero out KV cache to avoid nans (#3523)