Added warning and recommendation about possible loss of multilingual and coding capabilities.
Browse files
README.md
CHANGED
@@ -21,6 +21,11 @@ source data you are using: with random tokens or short text a small context make
|
|
21 |
context matching the size of the texts might be a better choice. Remember that the size is in tokens, which roughly translates
|
22 |
to number of words, not characters.
|
23 |
|
|
|
|
|
|
|
|
|
|
|
24 |
You will find below descriptions for the various input files provided, to help you choose the correct one.
|
25 |
|
26 |
## Community provided files
|
|
|
21 |
context matching the size of the texts might be a better choice. Remember that the size is in tokens, which roughly translates
|
22 |
to number of words, not characters.
|
23 |
|
24 |
+
Another important factor to consider is, **an importance matrix based on english language only will degrade
|
25 |
+
the model multingual capabilities**. Therefore the use of multilingual matrices are especially important when
|
26 |
+
considering models like Mistral or Command-R. The same applies to coding capabilities. Choose carefully based
|
27 |
+
on the model strength.
|
28 |
+
|
29 |
You will find below descriptions for the various input files provided, to help you choose the correct one.
|
30 |
|
31 |
## Community provided files
|