gpt2-medium-emailgen

Why write the entire email when you can generate (most of) it?

from transformers import pipeline

model_tag = "postbot/gpt2-medium-emailgen"
generator = pipeline(
              'text-generation', 
              model=model_tag, 
            )
            
prompt = """
Hello, 

Following up on the bubblegum shipment."""

result = generator(
    prompt,
    max_length=64,
    do_sample=False,
    early_stopping=True,
) # generate
print(result[0]['generated_text'])

about

This model is a fine-tuned version of gpt2-medium on the postbot/multi-emails-100k dataset. It achieves the following results on the evaluation set:

• Loss: 1.5840

Model description

More information needed

Intended uses & limitations

• this is intended as a tool to save time writing predictable emails and not to write emails without a human-in-the-loop. validate that your email is factually correct before sending it to others.

Training and evaluation data

• the dataset is essentially a hand-curated/augmented expansion to the classic aeslc dataset

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.001
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• distributed_type: multi-GPU
• gradient_accumulation_steps: 8
• total_train_batch_size: 128
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.02
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss
1.8701	1.0	789	1.8378
1.5065	2.0	1578	1.6176
1.1873	3.0	2367	1.5840

Framework versions

• Transformers 4.22.2
• Pytorch 1.10.0+cu113
• Datasets 2.5.1
• Tokenizers 0.12.1

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	25.97
ARC (25-shot)	26.45
HellaSwag (10-shot)	34.31
MMLU (5-shot)	24.1
TruthfulQA (0-shot)	43.96
Winogrande (5-shot)	50.43
GSM8K (5-shot)	0.0
DROP (3-shot)	2.53