File size: 3,467 Bytes
3033f6b 2ccf148 3033f6b |
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 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 |
---
tags:
- bert
- transformers
- litert
- tflite
- edge
- on-device
license: mit
base_model: FacebookAI/roberta-base
pipeline_tag: feature-extraction
---
# roberta-base - LiteRT
This is a [LiteRT](https://ai.google.dev/edge/litert) (formerly TensorFlow Lite) conversion of [FacebookAI/roberta-base](https://huggingface.co/FacebookAI/roberta-base) for efficient on-device inference.
## Model Details
| Property | Value |
|----------|-------|
| **Original Model** | [FacebookAI/roberta-base](https://huggingface.co/FacebookAI/roberta-base) |
| **Format** | LiteRT (.tflite) |
| **File Size** | 473.7 MB |
| **Task** | Feature Extraction / Classification Base |
| **Max Sequence Length** | 128 |
| **Output Dimension** | 768 |
| **Pooling Mode** | N/A (Full hidden states) |
## Performance
Benchmarked on AMD CPU (WSL2):
| Metric | Value |
|--------|-------|
| **Inference Latency** | 81.2 ms |
| **Throughput** | 12.3/sec |
| **Cosine Similarity vs Original** | 1.0000 ✅ |
## Quick Start
```python
import numpy as np
from ai_edge_litert.interpreter import Interpreter
from transformers import AutoTokenizer
# Load model and tokenizer
interpreter = Interpreter(model_path="FacebookAI_roberta-base.tflite")
interpreter.allocate_tensors()
input_details = interpreter.get_input_details()
output_details = interpreter.get_output_details()
tokenizer = AutoTokenizer.from_pretrained("FacebookAI/roberta-base")
def get_hidden_states(text: str) -> np.ndarray:
"""Get hidden states for input text."""
encoded = tokenizer(
text,
padding="max_length",
max_length=128,
truncation=True,
return_tensors="np"
)
interpreter.set_tensor(input_details[0]["index"], encoded["input_ids"].astype(np.int64))
interpreter.set_tensor(input_details[1]["index"], encoded["attention_mask"].astype(np.int64))
interpreter.invoke()
return interpreter.get_tensor(output_details[0]["index"])
# Example
hidden = get_hidden_states("Hello, world!")
cls_embedding = hidden[0, 0, :] # CLS token for classification
print(f"Hidden shape: {hidden.shape}") # (1, 128, 768)
```
## Files
- `FacebookAI_roberta-base.tflite` - The LiteRT model file
## Conversion Details
- **Conversion Tool**: [ai-edge-torch](https://github.com/google-ai-edge/ai-edge-torch)
- **Conversion Date**: 2026-01-12
- **Source Framework**: PyTorch → LiteRT
- **Validation**: Cosine similarity 1.0000 vs original
## Intended Use
- **Mobile Applications**: On-device semantic search, RAG systems
- **Edge Devices**: IoT, embedded systems, Raspberry Pi
- **Offline Processing**: Privacy-preserving inference
- **Low-latency Applications**: Real-time processing
## Limitations
- Fixed sequence length (128 tokens)
- CPU inference (GPU delegate requires setup)
- Tokenizer loaded separately from original model
- Float32 precision
## License
This model inherits the license from the original:
- **License**: MIT ([source](https://huggingface.co/FacebookAI/roberta-base))
## Citation
```bibtex
@article{liu2019roberta,
title={RoBERTa: A Robustly Optimized BERT Pretraining Approach},
author={Liu, Yinhan and Ott, Myle and others},
journal={arXiv preprint arXiv:1907.11692},
year={2019}
}
```
## Acknowledgments
- Original model by [FacebookAI](https://huggingface.co/FacebookAI)
- Conversion using [ai-edge-torch](https://github.com/google-ai-edge/ai-edge-torch)
---
*Converted by [Bombek1](https://huggingface.co/Bombek1)*
|