You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
The default Mistral Small 2409 chat template (after hitting reset) errors, even though it appears to be correct. And fixing the supposed error (removing the colon in messages[1:]) drastically degrades performance.
According to AI (I'm not a technical user) the colon is needed to subsequently alternate between the user and assistant while skipping the system prompt, yet the colon isn't supported in the included code block prior to the bos token. So it's required, but unsupported, if that makes any sense.
Steps to Reproduce
Just load Mistral Small Instruct 2409
Hit reset chat template
See error message
Expected Behavior
No red box with an error message.
Possible Fix: At the end of this message is the chat template taken from Mistral Small's tokenizer_config.json, which appears to work, although there may be an issue with the system prompt and minor performance degradation, which is probably why you guys added the code that errors out before the bos token, plus a couple other changes.
Your Environment
GPT4All version: 3.5.1
Operating System: Windows 11
Below is an apparently working chat template taken from Mistral Small's tokenizer_config.json
Bug Report
The default Mistral Small 2409 chat template (after hitting reset) errors, even though it appears to be correct. And fixing the supposed error (removing the colon in messages[1:]) drastically degrades performance.
According to AI (I'm not a technical user) the colon is needed to subsequently alternate between the user and assistant while skipping the system prompt, yet the colon isn't supported in the included code block prior to the bos token. So it's required, but unsupported, if that makes any sense.
Steps to Reproduce
Expected Behavior
No red box with an error message.
Possible Fix: At the end of this message is the chat template taken from Mistral Small's tokenizer_config.json, which appears to work, although there may be an issue with the system prompt and minor performance degradation, which is probably why you guys added the code that errors out before the bos token, plus a couple other changes.
Your Environment
Below is an apparently working chat template taken from Mistral Small's tokenizer_config.json
Below is the default chat format in GPT4All (when hitting reset) that errors out and I can't figure out how to fix.
The text was updated successfully, but these errors were encountered: