bugfix: only export per token softmax kernels

[DefTruth]

fix #251

[Copilot]

Copilot reviewed 3 out of 4 changed files in this pull request and generated no comments.

Files not reviewed (1)

• kernels/softmax/softmax.cu: Language not supported

Comments suppressed due to low confidence (2)

kernels/softmax/softmax.py:58

• [nitpick] The comment marker uses a double hashtag, which may be unintentional. Consider using a single '#' for clarity.

# # grid memory fence

README.md:281

• [nitpick] The abbreviation 'per_tok' might be inconsistent with other naming conventions. Consider renaming it to 'per_token' for better clarity.

| ✔️ [softmax_f32_per_tok](./kernels/softmax/softmax.cu)|f32|f32|[link](./kernels/softmax/)|⭐️⭐️|