From 1d0029cf9a8d45d036ba157d39630844ed27fde8 Mon Sep 17 00:00:00 2001
From: Matthew Wood <62712722+mattwoodx@users.noreply.github.com>
Date: Fri, 26 Sep 2025 16:24:24 +0200
Subject: [PATCH 1/3] scGPT Binning Randomness (#269)

* Revert "Hot fix remove notebook from release run"

This reverts commit 52bba026f1cb6149cf7e6a7f57f1ee261a6c4769.

* Use generator instead of seed state for reproducible embeddings independent of ordering of cells
---
 .github/workflows/release.yml         | 1 -
 ci/tests/test_scgpt/test_binning.py   | 4 ++--
 helical/models/scgpt/binning.py       | 3 ++-
 helical/models/scgpt/data_collator.py | 5 ++++-
 4 files changed, 8 insertions(+), 5 deletions(-)

diff --git a/.github/workflows/release.yml b/.github/workflows/release.yml
index 8fed8c75..c31b24c5 100644
--- a/.github/workflows/release.yml
+++ b/.github/workflows/release.yml
@@ -175,7 +175,6 @@ jobs:
           sed -i 's/list(np.array(train_dataset.obs\[\\"LVL1\\"].tolist()))/list(np.array(train_dataset.obs\[\\"LVL1\\"].tolist()))[:100]/g' ./examples/notebooks/Cell-Type-Classification-Fine-Tuning.ipynb
           sed -i 's/list(np.array(test_dataset.obs\[\\"LVL1\\"].tolist()))/list(np.array(test_dataset.obs\[\\"LVL1\\"].tolist()))[:10]/g' ./examples/notebooks/Cell-Type-Classification-Fine-Tuning.ipynb
           rm ./examples/notebooks/Evo-2.ipynb
-          rm ./examples/notebooks/Geneformer-Series-Comparison.ipynb
 
       - name: Run Notebooks
         run: |
diff --git a/ci/tests/test_scgpt/test_binning.py b/ci/tests/test_scgpt/test_binning.py
index b93526a7..524b7ad0 100644
--- a/ci/tests/test_scgpt/test_binning.py
+++ b/ci/tests/test_scgpt/test_binning.py
@@ -38,7 +38,7 @@ def test_digitize_identical_bins():
     x = np.array([1, 2, 3, 4, 5])
     bins = np.array([2, 2, 4, 4])
     result = _digitize(x, bins)
-    expected = np.array([0, 1, 2, 3, 4])
+    expected = np.array([0, 2, 2, 3, 4])
     assert np.array_equal(result, expected)
 
 
@@ -51,7 +51,7 @@ def test_digitize_identical_bins():
             np.array([1, 1, 1, 2, 2, 2, 3, 3, 4]),
         ),
         # distrubution of the bins depends on the distribution of the data
-        (np.array([1, 1, 1, 1, 1, 1, 1, 8, 9]), np.array([2, 1, 1, 2, 3, 3, 3, 3, 4])),
+        (np.array([1, 1, 1, 1, 1, 1, 1, 8, 9]), np.array([1, 2, 1, 1, 3, 1, 1, 3, 4])),
         (np.array([1, 2, 1, 1, 9, 6, 7, 8, 9]), np.array([1, 2, 1, 1, 4, 2, 2, 3, 4])),
     ),
 )
diff --git a/helical/models/scgpt/binning.py b/helical/models/scgpt/binning.py
index 9c457fb1..d029da6c 100644
--- a/helical/models/scgpt/binning.py
+++ b/helical/models/scgpt/binning.py
@@ -32,7 +32,8 @@ def _digitize(x: np.ndarray, bins: np.ndarray, side="both") -> np.ndarray:
 
     right_digits = np.digitize(x, bins, right=True)
 
-    rands = np.random.rand(len(x))  # uniform random numbers
+    rng = np.random.default_rng(42)
+    rands = rng.random(len(x))
 
     digits = rands * (right_digits - left_digits) + left_digits
     digits = np.ceil(digits).astype(np.int64)
diff --git a/helical/models/scgpt/data_collator.py b/helical/models/scgpt/data_collator.py
index 5dd3a634..5fdc1882 100644
--- a/helical/models/scgpt/data_collator.py
+++ b/helical/models/scgpt/data_collator.py
@@ -163,7 +163,10 @@ def _sample(
 
         # keep the first n tokens unchanged
         _n = self.keep_first_n_tokens
-        indices = torch.randperm(len(genes) - _n, device=device)[: max_length - _n]
+        g = torch.Generator().manual_seed(0)
+        indices = torch.randperm(len(genes) - _n, device=device, generator=g)[
+            : max_length - _n
+        ]
         indices = torch.cat([torch.arange(_n), indices + _n], dim=0)
         return genes[indices], expressions[indices]
 

From b4442824a0e22facf57d1d9d0e9ada2abd1dab0c Mon Sep 17 00:00:00 2001
From: Benoit Putzeys <157973952+bputzeys@users.noreply.github.com>
Date: Fri, 26 Sep 2025 16:31:37 +0200
Subject: [PATCH 2/3] Update pyproject.toml

---
 pyproject.toml | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/pyproject.toml b/pyproject.toml
index f6747baa..3c3b9be7 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "helical"
-version = "1.4.4"
+version = "1.4.5"
 authors = [
   { name="Helical Team", email="support@helical-ai.com" },
 ]

From 805c30cac613b3ad7897e91294783574d65114aa Mon Sep 17 00:00:00 2001
From: Matthew Wood <62712722+mattwoodx@users.noreply.github.com>
Date: Mon, 29 Sep 2025 09:51:59 +0200
Subject: [PATCH 3/3] Release workflow fix (#271)

* Revert "Hot fix remove notebook from release run"

This reverts commit 52bba026f1cb6149cf7e6a7f57f1ee261a6c4769.

* Remove large notebook from workflow

* fixup! Remove large notebook from workflow
---
 .github/workflows/release.yml | 1 +
 1 file changed, 1 insertion(+)

diff --git a/.github/workflows/release.yml b/.github/workflows/release.yml
index c31b24c5..8fed8c75 100644
--- a/.github/workflows/release.yml
+++ b/.github/workflows/release.yml
@@ -175,6 +175,7 @@ jobs:
           sed -i 's/list(np.array(train_dataset.obs\[\\"LVL1\\"].tolist()))/list(np.array(train_dataset.obs\[\\"LVL1\\"].tolist()))[:100]/g' ./examples/notebooks/Cell-Type-Classification-Fine-Tuning.ipynb
           sed -i 's/list(np.array(test_dataset.obs\[\\"LVL1\\"].tolist()))/list(np.array(test_dataset.obs\[\\"LVL1\\"].tolist()))[:10]/g' ./examples/notebooks/Cell-Type-Classification-Fine-Tuning.ipynb
           rm ./examples/notebooks/Evo-2.ipynb
+          rm ./examples/notebooks/Geneformer-Series-Comparison.ipynb
 
       - name: Run Notebooks
         run: |