File size: 3,211 Bytes
9966e25 | 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 78 79 80 81 82 83 84 85 | ---
language: en
license: apache-2.0
base_model: harsharajkumar273/ProphetNet-Large-Summarization
tags:
- text-generation
- story-generation
- research-paper
- seq2seq
- prophetnet
- lora
- peft
datasets:
- custom
metrics:
- bertscore
- sbert
---
# ProphetNet-Large-Story-Generation
A fine-tuned model for transforming research paper summaries into engaging short stories. This is the second stage of a two-step **Research Paper Simplifier** pipeline, built on top of [harsharajkumar273/ProphetNet-Large-Summarization](https://huggingface.co/harsharajkumar273/ProphetNet-Large-Summarization).
## Model Description
This model takes a summary of a research paper and generates an immersive, narrative-style short story. Fine-tuned using LoRA (PEFT) with 4-bit quantization.
## Pipeline
```
Research Paper ──► [ProphetNet-Large-Summarization] ──► Summary ──► [ProphetNet-Large-Story-Generation] ──► Story
```
## Training Details
| Parameter | Value |
|-----------|-------|
| Base model | harsharajkumar273/ProphetNet-Large-Summarization |
| Task | Story Generation |
| Max input length | 1024 tokens |
| Max target length | 512 tokens |
| Learning rate | 5e-5 |
| Batch size | 2 |
| Gradient accumulation steps | 4 |
| Warmup steps | 1000 |
| Weight decay | 0.01 |
| Fine-tuning method | LoRA (r=16, alpha=64, targets: query_proj, value_proj) |
| Quantization | 4-bit NF4 (bitsandbytes) |
## Usage
```python
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
# Stage 1: Summarize the paper
sum_tokenizer = AutoTokenizer.from_pretrained("harsharajkumar273/ProphetNet-Large-Summarization")
sum_model = AutoModelForSeq2SeqLM.from_pretrained("harsharajkumar273/ProphetNet-Large-Summarization")
paper_text = "Your research paper text here..."
word_count = len(paper_text.split())
sum_prompt = f"Summarize this part of the research paper to less than {word_count // 10} words:\n{paper_text}"
sum_inputs = sum_tokenizer(sum_prompt, return_tensors="pt", max_length=2048, truncation=True)
sum_outputs = sum_model.generate(**sum_inputs, max_length=256, num_beams=4)
summary = sum_tokenizer.decode(sum_outputs[0], skip_special_tokens=True)
# Stage 2: Generate a story from the summary
story_tokenizer = AutoTokenizer.from_pretrained("harsharajkumar273/ProphetNet-Large-Story-Generation")
story_model = AutoModelForSeq2SeqLM.from_pretrained("harsharajkumar273/ProphetNet-Large-Story-Generation")
story_inputs = story_tokenizer(summary, return_tensors="pt", max_length=1024, truncation=True)
story_outputs = story_model.generate(**story_inputs, max_length=512, num_beams=4)
story = story_tokenizer.decode(story_outputs[0], skip_special_tokens=True)
print(story)
```
## Evaluation Metrics
Evaluated using BERTScore and SBERTScore on a held-out 10% test split.
## Related Models
- [harsharajkumar273/ProphetNet-Large-Summarization](https://huggingface.co/harsharajkumar273/ProphetNet-Large-Summarization) — previous stage
- [harsharajkumar273/Bart-Base-Story-Generation](https://huggingface.co/harsharajkumar273/Bart-Base-Story-Generation)
- [harsharajkumar273/T5-Base-Story-Generation](https://huggingface.co/harsharajkumar273/T5-Base-Story-Generation)
|