Update Vulkan RoPE implementation (llama/7818)

* Update Vulkan RoPE implementation

* Return nullptr on alloc_buffer when allocation fails, instead of throwing an exception

Minor fixes

* Fix segfault when running out of VRAM

Co-authored-by: slaren <slarengh@gmail.com>

---------

Co-authored-by: slaren <slarengh@gmail.com>
This commit is contained in:
0cc4m
2024-06-11 21:20:29 +02:00
committed by Georgi Gerganov
parent a99e213a82
commit f100b3b523
2 changed files with 35 additions and 60 deletions

View File

@ -886,7 +886,7 @@ static bool alloc_tensor_range(struct ggml_context * ctx,
fprintf(stderr, "%s: failed to allocate %s buffer of size %zu\n", __func__, ggml_backend_buft_name(buft), size);
#endif
for (size_t i = 0; i < *n_buffers; i++) {
ggml_backend_buffer_free(*buffers[i]);
ggml_backend_buffer_free((*buffers)[i]);
}
free(*buffers);
return false;