|
|
--- |
|
|
library_name: transformers |
|
|
license: apache-2.0 |
|
|
base_model: distilbert-base-uncased |
|
|
tags: |
|
|
- generated_from_trainer |
|
|
metrics: |
|
|
- accuracy |
|
|
- f1 |
|
|
model-index: |
|
|
- name: my-test-model |
|
|
results: [] |
|
|
datasets: |
|
|
- stanfordnlp/imdb |
|
|
--- |
|
|
|
|
|
|
|
|
|
|
|
# my-test-model |
|
|
|
|
|
This model is a fine-tuned version of [distilbert-base-uncased](https://huggingface.co/distilbert-base-uncased) on imdb dataset. |
|
|
|
|
|
|
|
|
## Model description |
|
|
|
|
|
This model is a fine-tuned version of DistilBERT-base-uncased for binary sentiment analysis on movie reviews. Key specifications: |
|
|
|
|
|
Task: Sentiment classification (positive/negative) |
|
|
|
|
|
Base Architecture: 6-layer distilled Transformer model |
|
|
|
|
|
Parameters: ~66 million (standard DistilBERT configuration) |
|
|
|
|
|
Output Labels: |
|
|
|
|
|
0 → "NEGATIVE" |
|
|
|
|
|
1 → "POSITIVE" |
|
|
|
|
|
## Intended uses & limitations |
|
|
|
|
|
Acceptable Use Cases ✅ |
|
|
Sentiment analysis of English movie reviews |
|
|
|
|
|
Educational/research purposes for text classification |
|
|
|
|
|
Baseline model for entertainment industry applications |
|
|
|
|
|
Integration in sentiment analysis pipelines |
|
|
|
|
|
Limitations ⚠️ |
|
|
Language Restriction: Only supports English text |
|
|
|
|
|
Domain Specificity: Optimized for movie reviews - performance degrades on other text types |
|
|
|
|
|
Bias Risks: May reflect demographic/cultural biases in training data |
|
|
|
|
|
Length Constraint: Maximum input length of 256 tokens (longer texts are truncated) |
|
|
|
|
|
Not Suitable For: |
|
|
|
|
|
Multilingual text analysis |
|
|
|
|
|
Sarcasm/irony detection |
|
|
|
|
|
Fine-grained sentiment analysis (e.g., detecting anger, excitement) |
|
|
|
|
|
## Training and evaluation data |
|
|
|
|
|
Training Data |
|
|
Dataset: IMDB Movie Reviews |
|
|
|
|
|
Size: 25,000 labeled examples |
|
|
|
|
|
Class Distribution: |
|
|
|
|
|
Positive: 12,500 (50%) |
|
|
|
|
|
Negative: 12,500 (50%) |
|
|
|
|
|
Preprocessing: |
|
|
|
|
|
Lowercasing |
|
|
|
|
|
DistilBERT tokenization (WordPiece) |
|
|
|
|
|
Dynamic padding |
|
|
|
|
|
Evaluation Data |
|
|
Test Set: Official IMDB test split (25,000 examples) |
|
|
|
|
|
## Training procedure |
|
|
|
|
|
TrainingArguments( |
|
|
num_train_epochs=3, |
|
|
per_device_train_batch_size=16, |
|
|
per_device_eval_batch_size=64, |
|
|
learning_rate=2e-5, |
|
|
weight_decay=0.01, |
|
|
evaluation_strategy="epoch", |
|
|
save_strategy="epoch", |
|
|
metric_for_best_model="accuracy" |
|
|
) |
|
|
|
|
|
### Training hyperparameters |
|
|
|
|
|
The following hyperparameters were used during training: |
|
|
- learning_rate: 2e-05 |
|
|
- train_batch_size: 16 |
|
|
- eval_batch_size: 64 |
|
|
- seed: 42 |
|
|
- optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments |
|
|
- lr_scheduler_type: linear |
|
|
- num_epochs: 3 |
|
|
|
|
|
### Training results |
|
|
|
|
|
| Training Loss | Epoch | Step | Validation Loss | Accuracy | F1 | |
|
|
|:-------------:|:-----:|:----:|:---------------:|:--------:|:------:| |
|
|
| 0.2497 | 1.0 | 1563 | 0.2486 | 0.9026 | 0.9024 | |
|
|
| 0.1496 | 2.0 | 3126 | 0.2896 | 0.9135 | 0.9135 | |
|
|
| 0.1222 | 3.0 | 4689 | 0.3448 | 0.9130 | 0.9130 | |
|
|
|
|
|
|
|
|
### Framework versions |
|
|
|
|
|
- Transformers 4.52.3 |
|
|
- Pytorch 2.7.0+cu128 |
|
|
- Datasets 3.6.0 |
|
|
- Tokenizers 0.21.1 |
