Open evelinehong opened 2 months ago
The core implementation of these two trainers is essentially the same.
The pg_19trainer is the code we use to calculate experimental metrics on the pg19 dataset.
Compared to pg_19trainer , the complete_trainer has added the following process:
Generation -> Training the generated chunk -> Updating LoRa -> Continuing generation
This process can be considered the minimum viable product of Temp-Lora in real-world applications.
The core implementation of these two trainers is essentially the same.
The pg_19trainer is the code we use to calculate experimental metrics on the pg19 dataset.
Compared to pg_19trainer , the complete_trainer has added the following process:
Generation -> Training the generated chunk -> Updating LoRa -> Continuing generation
This process can be considered the minimum viable product of Temp-Lora in real-world applications.