Eval Request (Loki Re-Test)

#441
by pix2pix - opened

Hello, I would like to ask you to retest this model https://huggingface.co/CrucibleLab/M3.2-24B-Loki-V1.3

Based on the table, you used a Mistral-V7-Takken template, but this model has been heavily modified. It easily digests the template from Llama 3 (the author of the model recommends his custom version, but I preferred to use a clean template from Llama 3). The results are quite different from the standard Mistral template, for the better. I wanted to know if the template really improves her work, or if it’s just a placebo effect?

Thank you for your time.

DontPlanToEnd changed discussion status to closed

Sign up or log in to comment