ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-03-04 02:50:01 +00:00
Code Issues Packages Projects Releases Wiki Activity

imatrix: be able to specify the name of the output tensor

Browse Source 
For some models the same tensor is used for token embeddings and
output. This tensor tends to be named token_embedding.weight rather
than output.weight, which prevernts us from collecting imatrix data
for this tensor. With this commit we can tell the name of the
output tensor to the imatrix tool.
This commit is contained in:
Kawrakow
2024-06-26 17:38:18 +03:00
parent be36ca872f
commit aaec3c1f60
3 changed files with 11 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
examples/imatrix/imatrix.cpp
View File

@@ -83,7 +83,8 @@ bool IMatrixCollector::collect_imatrix(struct ggml_tensor * t, bool ask, void *
if (t->op != GGML_OP_MUL_MAT) return false;
// why are small batches ignored (<16 tokens)?
if (src1->ne[1] < 16 || src1->type != GGML_TYPE_F32) return false;
if (!(wname.substr(0, 4) == "blk." || (m_params.process_output && wname == "output.weight"))) return false;
//printf("wname = %s\n", wname.c_str());
if (!(wname.substr(0, 4) == "blk." || (m_params.process_output && wname == m_params.output_tensor_name))) return false;
return true;
}