[goliaro]

Summary

• The notebooks called SinusoidalPositionalEmbedding(d_model, max_seq_len), so a correct implementation printed (1, 50, 100) instead of (1, 50, 64). The test now passes (max_seq_len, d_model) and an input of shape (1, 50, 64).
• The write-up now states the Attention Is All You Need formula: frequency index i runs over 0 … d_model/2 - 1, and dimensions 2i and 2i+1 share 1 / 10000**(2i/d_model). forward is specified to return (1, seq_len, d_model).
• The grader checks those sine/cosine values and the forward shape. The problem is marked auto-graded, and the published hints match.

Test plan

• check("sinusoidal-positional-embedding", SinusoidalPositionalEmbedding) passes on the solution notebook
• An implementation that gives sine and cosine different frequencies fails the new check
• The notebook test prints torch.Size([1, 50, 64])

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[vercel]

@goliaro is attempting to deploy a commit to the exorust's projects Team on Vercel.

A member of the Team first needs to authorize it.