add generation prompt enforcement is too severe

[RonanKMcGovern]

Problem:

• When loading a model that does not have add generation prompt in the chat template, it causes a runtime error rather than just a warning. This means - even if one does not want to have a generation prompt - one is forced to do so.

Specific error:

Unsloth: The tokenizer `Llama-3.2-1B-lora-model`
has a {% if add_generation_prompt %} for generation purposes, but wasn't provided correctly.

Recommendations:

1. Make this a warning not a runtime error that stops loading.
2. Provide an example of the exact syntax that unsloth expects.

[danielhanchen]

@RonanKMcGovern Interestingly weirdly I auto fixed it, but it seems like it fails now - I'll check it out

[RonanKMcGovern]

Appreciate it

…

On Mon 25 Nov 2024 at 03:04, Daniel Han ***@***.***> wrote: @RonanKMcGovern <https://github.com/RonanKMcGovern> Interestingly weirdly I auto fixed it, but it seems like it fails now - I'll check it out — Reply to this email directly, view it on GitHub <#1330 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ASVG6CRAFOJNA24NKNFXD4T2CL74HAVCNFSM6AAAAABSLVLFNWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDIOJXGY4TMOBTHE> . You are receiving this because you were mentioned.Message ID: ***@***.***>