| --- |
| language: en |
| license: apache-2.0 |
| tags: |
| - summarization |
| - conversational-ai |
| - text2text-generation |
| - t5 |
| datasets: |
| - knkarthick/samsum |
| metrics: |
| - rouge |
| base_model: |
| - google-t5/t5-small |
| --- |
| |
| # 470 Final Project Model -> Summary Model |
|
|
| ## Model Overview |
| This repository contains a fine-tuned **T5-small** model for **abstractive conversational text summarization**. |
| Given a multi-speaker dialogue, the model generates a concise natural-language summary that captures the main points of the conversation. |
|
|
| - **Base model:** google-t5/t5-small |
| - **Task:** Abstractive text summarization |
| - **Model type:** Encoder–decoder transformer (T5) |
|
|
| --- |
|
|
| ## Dataset |
| The model was fine-tuned on the **SAMSum** dataset, which consists of chat-style conversations paired with human-written summaries. |
|
|
| - **Dataset name:** knkarthick/samsum |
| - **Fields:** |
| - `dialogue`: conversation text (input) |
| - `summary`: reference summary (target) |
| - **Splits:** train / validation / test |
|
|
| --- |
|
|
| ## Training Details |
| - **Epochs:** 3 |
| - **Learning rate:** 3e-4 |
| - **Batch size:** 8 |
| - **Max input length:** 512 tokens |
| - **Max target length:** 128 tokens |
| - **Training framework:** Hugging Face Transformers (`Seq2SeqTrainer`) |
| - **Hardware:** GPU (Google Colab) |
|
|
| --- |
|
|
| ## Evaluation |
| The model was evaluated on the test split of the SAMSum dataset using ROUGE metrics. |
|
|
| - **ROUGE-1:** 0.4538 |
| - **ROUGE-2:** 0.2123 |
| - **ROUGE-L:** 0.3762 |
|
|
| (Replace the values with the scores obtained in the notebook.) |
|
|
| --- |
|
|
| ## Intended Uses |
| This model can be used for: |
| - Summarizing chat conversations or dialogues |
| - Demonstrations of abstractive summarization |
| - Educational purposes in NLP and machine learning |
|
|
| --- |
|
|
| ## Limitations |
| - The model may omit important details in long or complex conversations. |
| - Generated summaries may occasionally be imprecise or incomplete. |
| - The model is trained on informal dialogue and may not generalize well to other domains. |
|
|
| --- |
|
|
| ## How to Use |
|
|
| ```python |
| from transformers import AutoTokenizer, AutoModelForSeq2SeqLM |
| |
| repo_id = "marvingoenner/470finalprojectmodel" |
| |
| tokenizer = AutoTokenizer.from_pretrained(repo_id) |
| model = AutoModelForSeq2SeqLM.from_pretrained(repo_id) |
| |
| dialogue = "Amanda: I baked cookies. Jerry: Sounds great! Amanda: I will bring some tomorrow." |
| inputs = tokenizer("summarize: " + dialogue, return_tensors="pt", truncation=True) |
| output_ids = model.generate(**inputs, max_new_tokens=64) |
| |
| print(tokenizer.decode(output_ids[0], skip_special_tokens=True)) |
