Repeting this experiemnt of teaching LLM to follow chat structure and reply with all CAPS. This time with Gemma 2B as the base model. Compared to Stable LM 1.6B this model took 68 minutes (vs 11) and didn't learn the capability for RU language.

image/png

Downloads last month
50
Safetensors
Model size
2.51B params
Tensor type
BF16
·
Inference Providers NEW
This model is not currently available via any of the supported Inference Providers.