Skip to content

Commit 69a0b17

Browse files
committed
Update README
1 parent 71d8492 commit 69a0b17

File tree

1 file changed

+2
-3
lines changed

1 file changed

+2
-3
lines changed

tools/imatrix/README.md

Lines changed: 2 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -41,7 +41,7 @@ For faster computation, make sure to use GPU offloading via the `-ngl | --n-gpu-
4141
```
4242

4343
```bash
44-
# combine Existing imatrices
44+
# combine existing imatrices
4545
./llama-imatrix --in-file imatrix-prev-0.dat --in-file imatrix-prev-1.dat -o imatrix-combined.dat
4646
```
4747

@@ -76,5 +76,4 @@ Weighted averages of Σ(Act²), ZD Score and CosSim are also calculated.
7676
#### Important note on the computed Statistics
7777

7878
When using these statistics, please note that they are computed on the squared activations, **not on the actual (raw) activations**.
79-
Whilst the results are still useful, they're less accurate than using the raw values, and in the case of the cosine similarity, could be misleading if the tensor contains opposite vectors.
80-
This limitation is due to the current implementation of the importance matrix, but a pull request ([use GGUF to store importance matrices](https://github.com/ggml-org/llama.cpp/pull/9400)) aims to address this.
79+
Whilst the results are still useful, they're less reliable than using the raw values, and in the case of the cosine similarity, could be misleading if the tensor contains opposite vectors.

0 commit comments

Comments
 (0)