Dampish/Stellar4-FRE_0.7

Model based on Stellar4 trained on french data, the 0.7 signifies for how complete the learning is. the training was cut short so it is not as good as original model. However, this model is to demonstrate the power of Stellar4 and redpajama.

Specs

2.8B parameters
0.7 Epochs training
25K examples
Reached 0.75 Validation loss and 0.6 learning loss.