Spaces:
Running
Running
Update README.md
Browse files
README.md
CHANGED
|
@@ -37,4 +37,14 @@ We do accept merge requests if they address an issue.
|
|
| 37 |
|
| 38 |
# Report Issues!
|
| 39 |
|
| 40 |
-
Please report issues with our quants/models! We don't do extensive testing because the volume of stuff we push out is massive.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 37 |
|
| 38 |
# Report Issues!
|
| 39 |
|
| 40 |
+
Please report issues with our quants/models! We don't do extensive testing because the volume of stuff we push out is massive.
|
| 41 |
+
|
| 42 |
+
# Chat Completion Issues?
|
| 43 |
+
|
| 44 |
+
Chat Completion (not to get confused with text completion) typically is not supported by our EXL2 quants.
|
| 45 |
+
|
| 46 |
+
This is not something we cause. It's because the model creator didn't design/test their model for chat completion.
|
| 47 |
+
|
| 48 |
+
You can fix this by replacing the chat_template in the tokenizer_config.json (in some cases it's better to replace all .json files other then the config.json).
|
| 49 |
+
|
| 50 |
+
You can replace it with any model that works and is the same architecture. tl;dr if your model is based on llama 3.3 find one that works and copy/paste.
|