Distilled GPT-2 Story Generation Model (June 2025)

This is a distilled version of GPT-2, fine-tuned using knowledge distillation from a teacher model (Qwen3-1.7B) on the ROCStories dataset. The model is designed for story generation with constraint words.

Model Details

Base Model: GPT-2
Teacher Model: Qwen3-1.7B
Dataset: ROCStories (Ximing/ROCStories)
Training Objective: Knowledge distillation to match teacher outputs.
Training Date: June 12, 2025
Evaluation: Constraint word inclusion success rate.

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "here4code/distilled-gpt2-story-generation-Qwen3-1.7B"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name)

prompt = "Once upon a time, there was a happy dog"
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=100)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Training Details

Epochs: 2
Batch Size: 128
Training Loss per Epoch: [np.float64(4.636714812309023), np.float64(2.7680806659516835)]

Evaluation Results

Teacher Constraint Word Inclusion Success Rate: 0.8500
Student Constraint Word Inclusion Success Rate: 0.9000

License

This model is released under the MIT License.