Ruckus-PyAssi-13b

This model is a fine-tuned version of meta-llama/Llama-2-13b-hf on a 10 000 examples from flytech/llama-python-codes-30k dataset.

Model description

Model trained in 4-bit architecture using SFT (Supervised Fine Tuning) and LoRA (Low-Rank Adaptation) methods, fine-tuning further is possible.

Intended uses & limitations

Code-generation, but as like all Ruckus models

• Created to serve as an executional layer
• Rich in Python codes and instructional tasks
• Specially formatted for chat (see inference)

Training procedure

Model was being trained for 13 hours of A6000 single 48GB vRAM GPU

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0002
• train_batch_size: 32
• eval_batch_size: 32 * 2
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: constant
• num_epochs: 5

Inference

• Make sure to format your prompt: [INST]This is my prompt[/INST]

[INST]Ruckus, open google[/INST]

Framework versions

• Transformers 4.34.0
• Pytorch 2.0.1+cu118
• Datasets 2.14.5
• Tokenizers 0.14.1