File size: 2,527 Bytes
73219f5 7fca23a 73219f5 7fca23a 23c0391 7fca23a |
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 |
# My Dummy Model
---
language: fr
license: apache-2.0
tags:
- masked-lm
- camembert
- transformers
- tf
- french
- fill-mask
---
# CamemBERT MLM - Fine-tuned Model
This is a TensorFlow-based masked language model (MLM) based on the [camembert-base](https://huggingface.co/camembert-base) checkpoint, a RoBERTa-like model trained on French text.
## Model description
This model uses the CamemBERT architecture, which is a RoBERTa-based transformer trained on large-scale French corpora (e.g., OSCAR, CCNet). It's designed to perform Masked Language Modeling (MLM) tasks.
It was loaded and saved using the `transformers` library in TensorFlow (`TFAutoModelForMaskedLM`). It can be used for fill-in-the-blank tasks in French.
## Intended uses & limitations
### Intended uses
- Fill-mask predictions in French
- Feature extraction for NLP tasks
- Fine-tuning on downstream tasks like text classification, NER, etc.
### Limitations
- Works best with French text
- May not generalize well to other languages
- Cannot be used for generative tasks (e.g., translation, text generation)
## How to use
```python
from transformers import TFAutoModelForMaskedLM, AutoTokenizer
import tensorflow as tf
model = TFAutoModelForMaskedLM.from_pretrained("Mhammad2023/my-dummy-model")
tokenizer = AutoTokenizer.from_pretrained("Mhammad2023/my-dummy-model")
inputs = tokenizer("J'aime le [MASK] rouge.", return_tensors="tf")
outputs = model(**inputs)
logits = outputs.logits
masked_index = tf.argmax(inputs.input_ids == tokenizer.mask_token_id, axis=1)[0]
predicted_token_id = tf.argmax(logits[0, masked_index])
predicted_token = tokenizer.decode([predicted_token_id])
print(f"Predicted word: {predicted_token}")
```
## Limitations and bias
This model inherits the limitations and biases from the camembert-base checkpoint, including:
Potential biases from the training data (e.g., internet corpora)
## Inappropriate predictions for sensitive topics
Use with caution in production or sensitive applications.
## Training data
The model was not further fine-tuned; it is based directly on camembert-base, which was trained on:
OSCAR (Open Super-large Crawled ALMAnaCH coRpus)
CCNet (Common Crawl News)
## Training procedure
No additional training was applied for this version. You can load and fine-tune it on your task using Trainer or Keras API.
## Evaluation results
This version has not been evaluated on downstream tasks. For evaluation metrics and benchmarks, refer to the original camembert-base model card. |