metadata

license: apache-2.0
datasets:
  - sail/symbolic-instruction-tuning

gemma-2B Fine-Tuning on SAIL/Symbolic-Instruction-Tuning

This repository contains the gemma-2B model fine-tuned on the sail/symbolic-instruction-tuning dataset. The model is designed to interpret and execute symbolic instructions with improved accuracy and efficiency.

Overview

The gemma-2B model, originally known for its robust language understanding capabilities, has been fine-tuned to enhance its performance on symbolic instruction data. This involves retraining the model on the sail/symbolic-instruction-tuning dataset, which comprises a diverse range of instructional data that tests a model's ability to follow abstract and complex directives.

Motivation

The motivation behind fine-tuning gemma-2B on this particular dataset is to bridge the gap between language understanding and execution in a symbolic context. This has wide applications in areas such as code generation, automated reasoning, and more sophisticated AI instruction following.

Getting Started

To use this model, you'll need to have an account on Hugging Face and the transformers library installed. You can install the library using pip:

pip install transformers

Once installed, you can use the following code to load and use the model:

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "your-huggingface-username/gemma-2B-fine-tuned"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name)

# Now you can use the model for inference
input_text = "Your symbolic instruction here"
input_ids = tokenizer.encode(input_text, return_tensors='pt')

# Generate the output
output = model.generate(input_ids)
print(tokenizer.decode(output[0], skip_special_tokens=True))

Fine-Tuning Process

The model was fine-tuned using the following process:

Preprocessing: The sail/symbolic-instruction-tuning dataset was preprocessed to conform with the input format required by gemma-2B.
Training: The model was fine-tuned using a custom training loop that monitors loss and evaluates on a held-out validation set.
Hyperparameters: The fine-tuning used specific hyperparameters, which you can find in the training_script.py file.
Evaluation: The fine-tuned model was evaluated against a benchmark to ensure that it meets our performance standards.