neph1
/

llama-3-instruct-bellman-8b-swe-preview

Model card Files Files and versions Community

neph1 commited on Apr 21

Commit

5c71dd0

•

1 Parent(s): d2d00cb

Update README.md

Files changed (1) hide show

README.md +4 -2

README.md CHANGED Viewed

@@ -9,9 +9,11 @@ License: https://llama.meta.com/llama3/license
 240421: Bellman's back!
 This is a preliminary test run, run on mostly the same settings and dataset as bellman-mistral-instruct (But not the dpo pass). Context length is 3072.
-I've done some basic testing, and it's not a total mess. Whether it's an improvement to llama-3-instruct, I'm not sure, because it's REALLY good.
-I'll try to make a pass on full context length soon. And hopefully squeeze improve the results more.
 ![image/png](https://cdn-uploads.huggingface.co/production/uploads/653cd3049107029eb004f968/IDGX3d9lGe6yx-yHjsrav.png)

 240421: Bellman's back!
 This is a preliminary test run, run on mostly the same settings and dataset as bellman-mistral-instruct (But not the dpo pass). Context length is 3072.
+I've done some basic testing, and it's not a total mess. Whether it's an improvement to llama-3-instruct, I'm not sure, because that's REALLY good.
+I'll try to make a pass on full context length soon. And hopefully improve the results more.
+Make sure to use the correct chat template (llama-3) for best results.
 ![image/png](https://cdn-uploads.huggingface.co/production/uploads/653cd3049107029eb004f968/IDGX3d9lGe6yx-yHjsrav.png)