neuml / txtinstruct Goto Github PK

Review txtinstruct is a great library for instruction fine tuning.

Just wondering about saving the instruction fine-tuned model? Please help me with this.

Currently, the logic to generate unanswerable questions selects a random statement or template. This logic should add a check to ensure the template selected isn't for the current row. Otherwise, the same question could be generated with an answer and as unanswerable which would confuse a model.

Can you please help with a tutorial or example that generates instruction set on custom local data?

ImportError Traceback (most recent call last)
in <cell line: 10>()
8 # Train model
9 generator = StatementGenerator()
---> 10 model, tokenizer = generator(
11 "google/flan-t5-small",
12 dataset,

7 frames
/usr/local/lib/python3.10/dist-packages/transformers/training_args.py in _setup_devices(self)
1799 if not is_sagemaker_mp_enabled():
1800 if not is_accelerate_available(min_version="0.20.1"):
-> 1801 raise ImportError(
1802 "Using the Trainer with PyTorch requires accelerate>=0.20.1: Please run pip install transformers[torch] or pip install accelerate -U"
1803 )

ImportError: Using the Trainer with PyTorch requires accelerate>=0.20.1: Please run pip install transformers[torch] or pip install accelerate -U

NOTE: If your import is failing due to a missing package, you can
manually install dependencies using either !pip or !apt.

To view examples of installing some common dependencies, click the
"Open Examples" button below.

It is a Colab error. how can i handle this error

is possible to use custom embeddings ??

in the Notebook 01_Introducing_txtinstruct in chapter "Train an instruction-tuned model", there is

with open("data.json", encoding="utf-8") as f:
    data = json.load(f)

What is this JSON file? What is the description of it?