This is a LoRa finetuning of Bloom-7b1 using the Alpaca instruction dataset.
It really highlights how the Bloom models are undertrained with ~400M tokens as opposed to 1 Trillion in the smaller LLaMa models.
This is a LoRa finetuning of Bloom-7b1 using the Alpaca instruction dataset.
It really highlights how the Bloom models are undertrained with ~400M tokens as opposed to 1 Trillion in the smaller LLaMa models.