File size: 2,438 Bytes
3779932
 
49b57fd
 
3779932
49b57fd
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
---
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:

```bash
pip install transformers
```

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

```python
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.