Quantum-Enhanced Language Model Fine-Tuning with Merlin

[bsiegelwax]

This is from running https://merlinquantum.ai/notebooks/QLLM_fine_tuning.html as-is. I have made no modifications. I used Google Colab.

Training Quantum Layer heads...

Training Quantum Head: 2 modes, 1 photons

Input state: [1, 0]

Number of parameters in Quantum head: 76914

Epoch 50/200, Train Loss: 0.7964

Epoch 100/200, Train Loss: 0.7366

Epoch 150/200, Train Loss: 0.6766

Epoch 200/200, Train Loss: 0.6186

---

TypeError Traceback (most recent call last)

/tmp/ipython-input-21-1197100152.py in <cell line: 0>()

 31         # Evaluate

 32         q_val_predictions = model.model_head.predict(

---> 33 sentence_transformer.encode(eval_dataset["sentence"], convert_to_tensor=True).cpu().numpy()

 34         )

 35         q_val_accuracy = accuracy_score(eval_dataset["label"], q_val_predictions)

7 frames

/usr/local/lib/python3.11/dist-packages/datasets/formatting/formatting.py in _raise_bad_key_type(key)

 42 

 43 def _raise_bad_key_type(key: Any):

---> 44 raise TypeError(

 45         f"Wrong key type: '{key}' of type '{type(key)}'. Expected one of int, slice, range, str or Iterable."

 46     )

TypeError: Wrong key type: '79' of type '<class 'numpy.int64'>'. Expected one of int, slice, range, str or Iterable.

[CassNot]

Dear @bsiegelwax ,

Thank you for your issue ! I have been trying to reproduce the same settings as in your environment, in 3 different Google Colab here:

• directly from https://merlinquantum.ai/notebooks/QLLM_fine_tuning.html (ctrl+s on the source) : https://drive.google.com/file/d/1ITE6bfYyCUP5IjDP5T6oT8p1HXwOXro9/view?usp=sharinghttps://drive.google.com/file/d/1ITE6bfYyCUP5IjDP5T6oT8p1HXwOXro9/view?usp=sharing
• current version on our main branch : https://drive.google.com/file/d/1sreQwpwctDwxVT2E4Ir5EbbA7z9wig1r/view?usp=sharing (you can find the notebook here)
• the version from the release: https://drive.google.com/file/d/1RxmFIEPz9Cz3zo6oWhRBu69FKAp1v2Kl/view?usp=sharing (you can find the notebook here)

However, I cannot reproduce your error and my code runs smoothly:

Training Quantum Head: 2 modes, 1 photons
   Input state: [1, 0]
Number of parameters in Quantum head: 76914
Epoch 50/200, Train Loss: 0.6458
Epoch 100/200, Train Loss: 0.5887
Epoch 150/200, Train Loss: 0.5334
Epoch 200/200, Train Loss: 0.4817
   Quantum 2-1 - Val: 0.8480, Test: 0.8537

I had further questions to be sure not to miss anything here:

• what is the version of numpy installed in your Google Colab ?
• could you share your google colab so I can be sure we are working with the same version of the notebook ?

Looking forward to solve this type issue 😃

Cassandre

[bsiegelwax]

• I used the version of NumPy that Colab is currently using by default.
• I used your notebook as-is.

I just tried it again and I see the same error.

[CassNot]

Dear @bsiegelwax

In my google colab, I run a !pip install merlinquantum datasets setfit and then proceed with a !pip freeze > requirements_colab.txt that you will find attached. Could you send me the similar file from your google colab so I can compare our environment ? This way, I can be sure to have the same environment as you do.
requirements_colab.txt
In this google colab, I am using the notebook as-is and do not have this issue, that is why I would like to compare the environments : https://drive.google.com/file/d/1ITE6bfYyCUP5IjDP5T6oT8p1HXwOXro9/view?usp=sharing

Is this an issue you face with all approaches (Logistic Regression, SVM, MLP) or just with the QuantumLayer ?

Thank you !

Cassandre

[bsiegelwax]

Uploading your notebook results in:
Bad control character in string literal in JSON at position 13307 (line 1 column 13308) SyntaxError: Bad control character in string literal in JSON at position 13307 (line 1 column 13308) at JSON.parse (<anonymous>) at qa.program_ (https://ssl.gstatic.com/colaboratory-static/common/0a3a17aacb832914ee118c937a88b3e5/external_binary.js:4011:262) at ra (https://ssl.gstatic.com/colaboratory-static/common/0a3a17aacb832914ee118c937a88b3e5/external_binary.js:17:57) at qa.next_ (https://ssl.gstatic.com/colaboratory-static/common/0a3a17aacb832914ee118c937a88b3e5/external_binary.js:15:196) at qaa.next (https://ssl.gstatic.com/colaboratory-static/common/0a3a17aacb832914ee118c937a88b3e5/external_binary.js:18:48) at b (https://ssl.gstatic.com/colaboratory-static/common/0a3a17aacb832914ee118c937a88b3e5/external_binary.js:18:309)

That said, MY notebook + the requirements + the pip installs executed.