|
|
--- |
|
|
license: apache-2.0 |
|
|
base_model: HuggingFaceTB/SmolLM-360M-Instruct |
|
|
tags: |
|
|
- smollm |
|
|
- customer-support |
|
|
- fine-tuned |
|
|
- lora |
|
|
- instruct |
|
|
language: |
|
|
- en |
|
|
--- |
|
|
|
|
|
# SmolLM-360M-CustomerSupport-Instruct |
|
|
|
|
|
This is a fine-tuned version of [HuggingFaceTB/SmolLM-360M-Instruct](https://huggingface.co/HuggingFaceTB/SmolLM-360M-Instruct) trained on customer support conversations. |
|
|
|
|
|
## Model Details |
|
|
|
|
|
- **Base Model**: SmolLM-360M-Instruct (360M parameters) |
|
|
- **Fine-tuning Method**: LoRA (Low-Rank Adaptation) |
|
|
- **Dataset**: Bitext Customer Support Dataset |
|
|
- **Training Samples**: 2,000 customer support conversations |
|
|
- **Use Case**: Customer service chatbot, support ticket handling |
|
|
|
|
|
## Training Details |
|
|
|
|
|
- **Epochs**: 3 |
|
|
- **Learning Rate**: 1e-4 |
|
|
- **Batch Size**: 8 |
|
|
- **Max Sequence Length**: 512 |
|
|
- **Optimizer**: AdamW with cosine learning rate schedule |
|
|
- **Hardware**: Single GPU (Colab/Kaggle) |
|
|
|
|
|
## Usage |
|
|
|
|
|
```python |
|
|
from transformers import AutoModelForCausalLM, AutoTokenizer |
|
|
|
|
|
model_name = "sweatSmile/SmolLM-360M-CustomerSupport-Instruct" |
|
|
tokenizer = AutoTokenizer.from_pretrained(model_name) |
|
|
model = AutoModelForCausalLM.from_pretrained(model_name) |
|
|
|
|
|
# Format your prompt |
|
|
prompt = "<|im_start|>user\nHow do I reset my password?<|im_end|>\n<|im_start|>assistant\n" |
|
|
|
|
|
inputs = tokenizer(prompt, return_tensors="pt") |
|
|
outputs = model.generate(**inputs, max_new_tokens=150, temperature=0.7) |
|
|
response = tokenizer.decode(outputs[0], skip_special_tokens=True) |
|
|
|
|
|
print(response) |
|
|
``` |
|
|
|
|
|
## Example Outputs |
|
|
|
|
|
**Input**: "My order hasn't arrived yet, what should I do?" |
|
|
|
|
|
**Output**: "I apologize for the delay with your order. Let me help you track it. Could you please provide your order number? Once I have that, I can check the current status and estimated delivery date." |
|
|
|
|
|
## Limitations |
|
|
|
|
|
- Trained on English customer support conversations only |
|
|
- Best for customer service domains (banking, e-commerce, general support) |
|
|
- May not perform well on highly technical or domain-specific queries outside training data |
|
|
- Small model size (360M) - not as capable as larger models but much faster |
|
|
|
|
|
## Intended Use |
|
|
|
|
|
- Customer support chatbots |
|
|
- Automated ticket response systems |
|
|
- FAQ assistance |
|
|
- Initial customer inquiry handling |
|
|
|
|
|
## Training Data |
|
|
|
|
|
The model was fine-tuned on the [Bitext Customer Support Dataset](https://huggingface.co/datasets/bitext/Bitext-customer-support-llm-chatbot-training-dataset), which contains diverse customer service scenarios. |
|
|
|
|
|
## Citation |
|
|
|
|
|
```bibtex |
|
|
@misc{smollm-customer-support-2025, |
|
|
author = {sweatSmile}, |
|
|
title = {SmolLM-360M Fine-tuned for Customer Support}, |
|
|
year = {2025}, |
|
|
publisher = {HuggingFace}, |
|
|
url = {https://huggingface.co/sweatSmile/SmolLM-360M-CustomerSupport-Instruct} |
|
|
} |
|
|
``` |
|
|
|
|
|
## License |
|
|
|
|
|
Apache 2.0 (same as base model) |
|
|
|
