This model is based on Llama-2-70b-base-hf, we use share-gpt4-zh dataset to fine-tune it of 3 epochs.
When tuning this model, we use 4-bit quantization and Lora to get a Lora weight and combine it with the original weights to get the final model.
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support