[PR #1695] fix(tinytorch): reuse Sigmoid class in GELU solution #10237

New Issue

GiteaMirror · 2026-05-08T11:21:46-05:00

GiteaMirror commented

2026-05-08 11:21:46 -05:00

📋 Pull Request Information

Original PR: https://github.com/harvard-edge/cs249r_book/pull/1695
Author: @Shashank-Tripathi-07
Created: 5/7/2026
Status: 🔄 Open

Base: dev ← Head: fix/tinytorch-gelu-reuse-sigmoid

📝 Commits (1)

4257203 fix(tinytorch): reuse Sigmoid class in GELU solution

📊 Changes

1 file changed (+1 additions, -6 deletions)

View changed files

📝 tinytorch/src/02_activations/02_activations.py (+1 -6)

📄 Description

Summary

GELU's forward() solution used raw np.exp to compute the sigmoid, bypassing the Sigmoid class students built earlier in the same module.
The docstring explicitly says "reuse sigmoid (which you just built!)" -- the old solution contradicted this hint.
Replaced the 6-line implementation with return Sigmoid()(x * 1.702) * x, which is numerically identical and demonstrates class composition.

Fix

Before:

### BEGIN SOLUTION
# GELU approximation: x * sigmoid(1.702 * x)
# First compute sigmoid part
sigmoid_part = 1.0 / (1.0 + np.exp(-1.702 * x.data))
# Then multiply by x
result = x.data * sigmoid_part
return Tensor(result)
### END SOLUTION

After:

### BEGIN SOLUTION
return Sigmoid()(x * 1.702) * x
### END SOLUTION

Verified numerically: both produce identical outputs across positive and negative inputs.

Fixes #1692

Test plan

Confirm np.allclose(Sigmoid()(x * 1.702) * x, old_impl(x)) passes for representative inputs
Run TinyTorch test suite: pytest tests/
Regenerate notebook via tito src export 02_activations and confirm the student-facing cell is still a blank stub

_{🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.}

## 📋 Pull Request Information **Original PR:** https://github.com/harvard-edge/cs249r_book/pull/1695 **Author:** [@Shashank-Tripathi-07](https://github.com/Shashank-Tripathi-07) **Created:** 5/7/2026 **Status:** 🔄 Open **Base:** `dev` ← **Head:** `fix/tinytorch-gelu-reuse-sigmoid` --- ### 📝 Commits (1) - [`4257203`](https://github.com/harvard-edge/cs249r_book/commit/42572031c71d04af54ef90570e742180db416801) fix(tinytorch): reuse Sigmoid class in GELU solution ### 📊 Changes **1 file changed** (+1 additions, -6 deletions) <details> <summary>View changed files</summary> 📝 `tinytorch/src/02_activations/02_activations.py` (+1 -6) </details> ### 📄 Description ## Summary - GELU's `forward()` solution used raw `np.exp` to compute the sigmoid, bypassing the `Sigmoid` class students built earlier in the same module. - The docstring explicitly says "reuse sigmoid (which you just built!)" -- the old solution contradicted this hint. - Replaced the 6-line implementation with `return Sigmoid()(x * 1.702) * x`, which is numerically identical and demonstrates class composition. ## Fix **Before:** ```python ### BEGIN SOLUTION # GELU approximation: x * sigmoid(1.702 * x) # First compute sigmoid part sigmoid_part = 1.0 / (1.0 + np.exp(-1.702 * x.data)) # Then multiply by x result = x.data * sigmoid_part return Tensor(result) ### END SOLUTION ``` **After:** ```python ### BEGIN SOLUTION return Sigmoid()(x * 1.702) * x ### END SOLUTION ``` Verified numerically: both produce identical outputs across positive and negative inputs. Fixes #1692 ## Test plan - [ ] Confirm `np.allclose(Sigmoid()(x * 1.702) * x, old_impl(x))` passes for representative inputs - [ ] Run TinyTorch test suite: `pytest tests/` - [ ] Regenerate notebook via `tito src export 02_activations` and confirm the student-facing cell is still a blank stub --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

GiteaMirror added the pull-request label 2026-05-08 11:21:47 -05:00

Sign in to join this conversation.

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: github-starred/cs249r_book#10237