File size: 1,750 Bytes
993d37d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
54
55
56
57
58
59
60
61
---
license: mit
tags:
- llm
- tinyllama
- function-calling
- question-answering
- finetuned
---

# TinyLlama Fine-tuned for Function Calling

This is a fine-tuned version of the [TinyLlama](https://huggingface.co/jzhang38/TinyLlama) model optimized for function calling tasks.

## Model Details

- **Base Model**: [Successmove/tinyllama-function-calling-cpu-optimized](https://huggingface.co/Successmove/tinyllama-function-calling-cpu-optimized)
- **Fine-tuning Data**: [Successmove/combined-function-calling-context-dataset](https://huggingface.co/datasets/Successmove/combined-function-calling-context-dataset)
- **Training Method**: LoRA (Low-Rank Adaptation)
- **Training Epochs**: 3
- **Final Training Loss**: ~0.05

## Usage

```python
from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel

# Load base model
base_model_name = "Successmove/tinyllama-function-calling-cpu-optimized"
model = AutoModelForCausalLM.from_pretrained(base_model_name)

# Load the LoRA adapters
model = PeftModel.from_pretrained(model, "path/to/this/model")

# Load tokenizer
tokenizer = AutoTokenizer.from_pretrained("path/to/this/model")

# Generate text
input_text = "Set a reminder for tomorrow at 9 AM"
inputs = tokenizer(input_text, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=100)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
```

## Training Details

This model was fine-tuned using:
- LoRA with r=8
- Learning rate: 2e-4
- Batch size: 4
- Gradient accumulation steps: 2
- 3 training epochs

## Limitations

This is a research prototype and may not be suitable for production use without further evaluation and testing.

## License

This model is licensed under the MIT License.