summaryrefslogtreecommitdiff
path: root/llama.h
diff options
context:
space:
mode:
authorKerfuffle <44031344+KerfuffleV2@users.noreply.github.com>2023-09-01 08:02:48 -0600
committerGitHub <noreply@github.com>2023-09-01 08:02:48 -0600
commit5d6f19f16b2173afe2d5c6aee2f5c9fc31038eba (patch)
tree5e4ea604f958f75f7c63e2ecdd315b934c0c661b /llama.h
parent0d5893668625456c94bbadfddc53fc69cd51c223 (diff)
Allow quantize to only copy tensors, some other improvements (#2931)
* Allow quantize tool to only copy tensors to allow repackaging models. * Slightly better logic when requantizing. * Change help message to go to `stdout`.
Diffstat (limited to 'llama.h')
-rw-r--r--llama.h1
1 files changed, 1 insertions, 0 deletions
diff --git a/llama.h b/llama.h
index 6e5e1df6..422f2852 100644
--- a/llama.h
+++ b/llama.h
@@ -164,6 +164,7 @@ extern "C" {
enum llama_ftype ftype; // quantize to this llama_ftype
bool allow_requantize; // allow quantizing non-f32/f16 tensors
bool quantize_output_tensor; // quantize output.weight
+ bool only_copy; // only copy tensors - ftype, allow_requantize and quantize_output_tensor are ignored
} llama_model_quantize_params;
// grammar types