YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
Mistral-7B QLoRA Adapter — ECTSum
This repository contains QLoRA adapters fine-tuned on the ECTSum dataset for executive summarization of earnings call transcripts.
The adapters are trained on top of Mistral-7B (4-bit) using Unsloth.
Base Model
- Base model:
mistralai/Mistral-7B-v0.1 - Quantization: 4-bit (bnb)
- Fine-tuning method: QLoRA (LoRA adapters only)
Training Data
The adapters were trained using the ECTSum dataset, introduced in:
ECTSum: A New Benchmark for Executive Summarization of Earnings Calls
Notes:
- The full dataset is not redistributed here.
- Only processed subsets were used during experimentation.
- All rights to the dataset belong to the original authors.
Intended Use
This adapter is intended for:
- research and experimentation
- financial document summarization
- studying lightweight fine-tuning with long-context inputs
It is not intended for production use without further validation.
Evaluation Summary
Evaluation was performed on a fixed sampled subset of the test split.
- ROUGE/BLEU improvements were marginal due to the highly abstractive nature of the task.
- BERTScore indicates that the base model already captures general financial semantics.
- Qualitative inspection suggests improvements in summary structure and tone.
See the GitHub repository for full details.
GitHub Repository
Code, notebooks, and evaluation details are available here:
https://github.com/devanshpursnanii/finetune-mistral-ectsum
Usage
Load this adapter on top of the base Mistral-7B model using PEFT / Unsloth.