|
--- |
|
license: apache-2.0 |
|
library_name: peft |
|
tags: |
|
- generated_from_trainer |
|
base_model: google/flan-t5-large |
|
model-index: |
|
- name: flan-t5-large-samsum-qlora |
|
results: [] |
|
--- |
|
|
|
# Model description |
|
|
|
flan-t5-large-samsum-qlora is an LLM model adapter and a fine-tuned version of [google/flan-t5-large](https://huggingface.co/google/flan-t5-large) on the |
|
[samsum](https://huggingface.co/datasets/samsum) dataset containing dialoges. Parameter-efficient fine-tuning with QLoRA was employed to fine-tune the base model. |
|
The model is intended for generative summarization tasks and achieved the following scores on the test dataset: |
|
- Rougue1: 49.249596% |
|
- Rouge2: 23.513032% |
|
- RougeL: 39.960812% |
|
- RougeLsum: 39.968438% |
|
|
|
|
|
|
|
## How to use |
|
|
|
Load the model: |
|
|
|
``` python |
|
from peft import PeftModel, PeftConfig |
|
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, BitsAndBytesConfig |
|
|
|
# Load the peft adapter model config |
|
peft_model_id = 'MuntasirHossain/flan-t5-large-samsum-qlora' |
|
peft_config = PeftConfig.from_pretrained(peft_model_id) |
|
|
|
# load the base model and tokenizer |
|
base_model = AutoModelForSeq2SeqLM.from_pretrained(peft_config.base_model_name_or_path, load_in_8bit=True, device_map="auto") |
|
tokenizer = AutoTokenizer.from_pretrained(peft_config.base_model_name_or_path) |
|
|
|
# Load the peft model |
|
model = PeftModel.from_pretrained(base_model, peft_model_id, device_map="auto") |
|
model.eval() |
|
``` |
|
|
|
Example Inference: |
|
|
|
``` python |
|
# random sample text from the samsum test dataset |
|
text = """ |
|
Emma: Hi, we're going with Peter to Amiens tomorrow. |
|
Daniel: oh! Cool. |
|
Emma: Wanna join? |
|
Daniel: Sure, I'm fed up with Paris. |
|
Emma: We're too. The noise, traffic etc. Would be nice to see some countrysides. |
|
Daniel: I don't think Amiens is exactly countrysides though :P |
|
Emma: Nope. Hahahah. But not a megalopolis either! |
|
Daniel: Right! Let's do it! |
|
Emma: But we should leave early. The days are shorter now. |
|
Daniel: Yes, the stupid winter time. |
|
Emma: Exactly! |
|
Daniel: Where should we meet then? |
|
Emma: Come to my place by 9am. |
|
Daniel: oohhh. It means I have to get up before 7! |
|
Emma: Yup. The early bird gets the worm (in Amiens). |
|
Daniel: You sound like my grandmother. |
|
Emma: HAHAHA. I'll even add: no parties tonight, no drinking dear Daniel |
|
Daniel: I really hope Amiens is worth it! |
|
""" |
|
|
|
input = tokenizer(text, return_tensors="pt") |
|
outputs = model.generate(input_ids=input["input_ids"].cuda(), max_new_tokens=40) |
|
print("Summary: ", tokenizer.decode(outputs[0], skip_special_tokens=True)) |
|
|
|
Summary: Emma and Peter are going to Amiens tomorrow. Daniel will join them. They will meet at Emma's place by 9 am. They will not have any parties tonight. |
|
``` |
|
|
|
## Training procedure |
|
|
|
### Training hyperparameters |
|
|
|
The following hyperparameters were used during training: |
|
- learning_rate: 0.001 |
|
- train_batch_size: 8 |
|
- eval_batch_size: 8 |
|
- seed: 42 |
|
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08 |
|
- lr_scheduler_type: linear |
|
- num_epochs: 5 |
|
|
|
### Training results |
|
|
|
|
|
|
|
### Framework versions |
|
|
|
- PEFT 0.8.2 |
|
- Transformers 4.38.1 |
|
- Pytorch 2.1.0+cu121 |
|
- Datasets 2.17.1 |
|
- Tokenizers 0.15.2 |
