Implemented PEFT feature in CSModel class by MKHarsha03 · Pull Request #28 · vandijklab/cell2sentence

MKHarsha03 · 2026-03-14T12:08:10Z

Added the LoRAConfig function from peft into CSModel
Added new parameters/arguments to CSModel init constructor to support the newly added LoRAConfig
Added an argument to init function in CSModel class to accept the huggingface token
Added tutorials 7 to 10 in the readme file and change the version from 3.8 to 3.10. Also changed the pip command for cell2sentence to use version 1.2.0
Added new requirements: sentencepiece, tiktoken and protobuf to support the tokenizer of gemma models
Commented out the old test script for CSModel and added a new test script to check if model is loading with LoRA modules correctly and validate save path

MKHarsha03 · 2026-03-14T12:14:00Z

@aakashdp6548 @SyedA5688 I am an engineering student and this is my first contribution. I am working on a project that uses your library and added the LoRA feature as I needed it for my project. Please review my code and mention any sorts of corrections.

Thank you

Copilot

Pull request overview

This PR adds initial PEFT/LoRA support to the CSModel wrapper and updates packaging/docs to support newer model/tokenizer requirements and a newer Python baseline.

Changes:

Add optional LoRA (PEFT) configuration to CSModel initialization.
Update packaging requirements (Python >= 3.10; add tokenizer-related deps; add peft/bitsandbytes).
Update documentation + README tutorials list and adjust ReadTheDocs configuration.

Reviewed changes

Copilot reviewed 9 out of 10 changed files in this pull request and generated 13 comments.

Show a summary per file

File	Description
`src/cell2sentence/csmodel.py`	Adds PEFT/LoRA parameters, HF login option, and updates `Trainer` initialization.
`src/cell2sentence/tests/test_csmodel.py`	Replaces prior CSModel test with a new PEFT-loading-focused test (currently problematic).
`src/cell2sentence/tests/small_data_diffgenes.csv`	Modifies a test fixture CSV (currently breaks existing CSData tests).
`setup.cfg`	Raises minimum Python version and adds new dependencies.
`docs/source/csmodel.rst`	Updates CSModel docs page (currently contains an unresolved conflict marker).
`README.md`	Updates Python version, install instructions, and tutorials list (contains a broken tutorial link and an incomplete row).
`.readthedocs.yaml`	Attempts to configure RTD installs (currently duplicated blocks).
`docs/Makefile`, `docs/make.bat`	Adds trailing newline only.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

.readthedocs.yaml

MKHarsha03 · 2026-03-14T12:42:11Z

docs/source/csmodel.rst

+
+A CSModel object is a wrapper around a Cell2Sentence model, which tracks the path of the model 
+saved on disk. When needed, the model is loaded from the path on disk for inference or finetuning.
+The class contains utilities for model generation and cell embedding with a Huggingface backend.
+
+.. autofunction:: csmodel.CSModel
+
+.. autofunction:: csmodel.CSModel.__init__
+
+.. autofunction:: csmodel.CSModel.__str__
+
+.. autofunction:: csmodel.CSModel.fine_tune
+
+.. autofunction:: csmodel.CSModel.generate_from_prompt
+
+.. autofunction:: csmodel.CSModel.generate_from_prompt_batched
+
+.. autofunction:: csmodel.CSModel.embed_cell
+
+.. autofunction:: csmodel.CSModel.embed_cells_batched
+
+.. autofunction:: csmodel.CSModel.push_model_to_hub
+=======


@copilot This is resolved in latest commit

MKHarsha03 · 2026-03-14T12:43:33Z

src/cell2sentence/tests/small_data_diffgenes.csv

@@ -2,4 +2,3 @@
 g1,0,3,0,1,3
 g2,0,0,1,1,2
 g3,3,1,0,0,1


MKHarsha03 · 2026-03-14T12:46:38Z

src/cell2sentence/csmodel.py

+        if not os.path.exists(save_dir):
+            os.mkdir(save_dir)


src/cell2sentence/csmodel.py

+        if huggingface_token:
+            login(huggingface_token)
+


MKHarsha03 · 2026-03-15T01:39:33Z

src/cell2sentence/csmodel.py

+            data_collator=data_collator,
+            train_dataset=train_dataset,
+            eval_dataset=eval_dataset,
+            processing_class=self.tokenizer #changed argument from tokenizer to processing_class as per modern documentation


MKHarsha03 · 2026-03-15T01:39:12Z

src/cell2sentence/tests/test_csmodel.py

+class TestCSModelPeftModelLoadingAndErrorHandling:
+    @classmethod
+    def setup_class(self):
+        self.save_dir = "/mnt/c/Users/khmam/Desktop/c2s_model_directory"
+        self.save_name = "lora_gemma_model"
+        hf_model_path = "vandijklab/C2S-Scale-Gemma-2-2B"
+        self.csmodel = CSModel(
+            model_name_or_path=hf_model_path,
+            save_dir=self.save_dir,
+            save_name=self.save_name,
+            peft = True,
+        )
+
+    def test_csmodel_created_correctly(self):
+        assert self.csmodel.save_path == os.path.join(self.save_dir, self.save_name)
+
+    def test_layers_are_created_correctly(self):
+        model = AutoModelForCausalLM.from_pretrained(self.csmodel.save_path, trust_remote_code = True)
+        print(model)


MKHarsha03 · 2026-03-15T01:38:50Z

README.md

 | [c2s_tutorial_4_cell_type_prediction.ipynb](tutorials/c2s_tutorial_4_cell_type_prediction.ipynb) | Cell type prediction using C2S models
 | [c2s_tutorial_5_cell_generation.ipynb](tutorials/c2s_tutorial_5_cell_generation.ipynb) | Cell generation conditioned on cell type
 | [c2s_tutorial_6_cell_annotation_with_foundation_model.ipynb](tutorials/c2s_tutorial_6_cell_annotation_with_foundation_model.ipynb) | Cell type annotation with foundation model
+| [c2s_tutorial_7_custom_prompt_templates.ipynb](tutorials/c2s_tutorials_7_custom_prompt_templates.ipynb) | Custom Prompt Templates with C2S PromptFormatter class


README.md

setup.cfg

+install_requires =
+    torch
+    transformers
+    peft
+    bitsandbytes
+    datasets
+    anndata


MKHarsha03 · 2026-03-14T13:17:00Z

Added the LoRAConfig function from peft into CSModel
Added new parameters/arguments to CSModel init constructor to support the newly added LoRAConfig
Added an argument to init function in CSModel class to accept the huggingface token
Added tutorials 7 to 10 in the readme file and change the version from 3.8 to 3.10. Also changed the pip command for cell2sentence to use version 1.2.0
Added new requirements: sentencepiece, tiktoken and protobuf to support the tokenizer of gemma models
Commented out the old test script for CSModel and added a new test script to check if model is loading with LoRA modules correctly and validate save path
Changed the tokenizer argument in Trainer to processing_class (new versions of transformer accepts this argument)
Modified the test code to check if lora layers are created

…the peft model correctly

MKHarsha03 added 9 commits March 14, 2026 11:24

Peft feature implemented

4a7bffe

Minor updates

fccd40d

Removing errors from merge conflicts

7e4a8e1

Cleared merge conflicts in docs/source

2d2d598

merge resolved

d8131c3

merge resolved

ab78081

merge resolved

afdfef7

merge resolved

2b02223

merge resolved

a37661e

Copilot AI review requested due to automatic review settings March 14, 2026 12:08

Copilot started reviewing on behalf of MKHarsha03 March 14, 2026 12:08 View session

Copilot AI reviewed Mar 14, 2026

View reviewed changes

MKHarsha03 added 6 commits March 14, 2026 12:30

Added the description for the 10th tutorial in README file

19299ff

Fixed the tutorial 7 path in readme

e7a2e06

removed duplication in readthedocs.yaml file

0c03c1c

Removed duplication in csmodel.rst

01dcb47

Removed duplication in csmodel.rst

0f5bd2e

changed line 87 in csmodel.py to os.makedirs

34e5122

MKHarsha03 closed this Mar 14, 2026

MKHarsha03 deleted the peft_lora branch March 14, 2026 13:09

Created proper test to check if peft layers are loading correctly

9216bf8

MKHarsha03 reopened this Mar 14, 2026

MKHarsha03 added 3 commits March 14, 2026 13:21

Modified the model loading in fine_tune method under CSModel to load …

fc90705

…the peft model correctly

Changed the Trainer argument type back to tokenizer

881dda4

Fixed the broken link for python3.10.20 in README.md

d9d62db

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Implemented PEFT feature in CSModel class#28

Implemented PEFT feature in CSModel class#28
MKHarsha03 wants to merge 19 commits intovandijklab:masterfrom
MKHarsha03:peft_lora

MKHarsha03 commented Mar 14, 2026 •

edited

Loading

Uh oh!

MKHarsha03 commented Mar 14, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

MKHarsha03 Mar 14, 2026

Uh oh!

MKHarsha03 Mar 14, 2026

Uh oh!

MKHarsha03 Mar 14, 2026

Uh oh!

MKHarsha03 Mar 15, 2026

Uh oh!

MKHarsha03 Mar 15, 2026

Uh oh!

MKHarsha03 Mar 15, 2026

Uh oh!

Uh oh!

MKHarsha03 commented Mar 14, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

MKHarsha03 commented Mar 14, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

MKHarsha03 commented Mar 14, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

MKHarsha03 Mar 14, 2026

Choose a reason for hiding this comment

Uh oh!

MKHarsha03 Mar 14, 2026

Choose a reason for hiding this comment

Uh oh!

MKHarsha03 Mar 14, 2026

Choose a reason for hiding this comment

Uh oh!

MKHarsha03 Mar 15, 2026

Choose a reason for hiding this comment

Uh oh!

MKHarsha03 Mar 15, 2026

Choose a reason for hiding this comment

Uh oh!

MKHarsha03 Mar 15, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

MKHarsha03 commented Mar 14, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

MKHarsha03 commented Mar 14, 2026 •

edited

Loading