README.md

metadata

license: apache-2.0
base_model: distilbert-base-uncased
tags:
  - generated_from_trainer
datasets:
  - HuggingFaceFW/fineweb-edu-llama3-annotations
language:
  - en

distilbert-base-uncased: edu classifier

This is a (rare) encoder that supports flash attention 2! Use attn_implementation="flash_attention_2" when loading w/ FA2 installed for faster inference.

This model is a fine-tuned version of distilbert-base-uncased on the HuggingFaceFW/fineweb-edu-llama3-annotations dataset. It achieves the following results on the evaluation set:

• Loss: 0.2324
• Mse: 0.2324

Usage

Note this is for CPU, for GPU you will need to make some (small) changes.

from transformers import AutoTokenizer, AutoModelForSequenceClassification

tokenizer = AutoTokenizer.from_pretrained("pszemraj/mpnet-base-edu-classifier")
model = AutoModelForSequenceClassification.from_pretrained("pszemraj/mpnet-base-edu-classifier")

text = "This is a test sentence."
inputs = tokenizer(text, return_tensors="pt", padding="longest", truncation=True)
outputs = model(**inputs)
logits = outputs.logits.squeeze(-1).float().detach().numpy()
score = logits.item()
result = {
    "text": text,
    "score": score,
    "int_score": int(round(max(0, min(score, 5)))),
}

print(result)
# {'text': 'This is a test sentence.', 'score': 0.3350256383419037, 'int_score': 0}

Intended uses & limitations

Refer to the hf classifier's model card for more details

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-05
• train_batch_size: 16
• eval_batch_size: 16
• seed: 90085
• gradient_accumulation_steps: 8
• total_train_batch_size: 128
• optimizer: Adam with betas=(0.9,0.98) and epsilon=1e-09
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.05
• num_epochs: 1.0