# Evidence of Fortran boosting GenAI

**URL:** <https://fortran-lang.discourse.group/t/evidence-of-fortran-boosting-genai/7750>\
**Category:** Uncategorized\
**Created:** [April 2, 2024, 7:32am UTC](https://fortran-lang.discourse.group/t/evidence-of-fortran-boosting-genai/7750 "2024-04-02T07:32:09Z")\
**Posts on this page:** 5\
**Page:** 1

<div class="post-metadata">

**Author:** ![manuel.arenaz](https://yyz2.discourse-cdn.com/free1/user_avatar/fortran-lang.discourse.group/manuel.arenaz/32/3217_2.png) [@manuel.arenaz](https://fortran-lang.discourse.group/u/manuel.arenaz)\
**Post date:** [April 2, 2024, 7:32am UTC](https://fortran-lang.discourse.group/t/evidence-of-fortran-boosting-genai/7750/1 "2024-04-02T07:32:09Z")

</div>

Hi all,

Recently I came across this post in LinkedIn and it is interesting that they show how NumPy leverages Fortran code for compute-intensive kernels like MATMUL:

> **[Manuel Arenaz on LinkedIn: Interesting post by Mitko showing evidence of...](https://www.linkedin.com/posts/manuelarenaz_i-find-it-interesting-how-early-we-are-in-activity-7180828419465625600-Ryu0?utm_source=share&utm_medium=member_android)**
>
> Interesting post by Mitko showing evidence of Fortran enabling a significant boost in GenAI speed.

I did not find the link to the technical article yet. I will post it here!

Are you aware of other evidence of Fortran powering the GenAI tools?

---

<div class="post-metadata">

**Author:** ![gnikit](https://yyz2.discourse-cdn.com/free1/user_avatar/fortran-lang.discourse.group/gnikit/32/1213_2.png) [@gnikit](https://fortran-lang.discourse.group/u/gnikit)\
**Post date:** [April 2, 2024, 7:40am UTC](https://fortran-lang.discourse.group/t/evidence-of-fortran-boosting-genai/7750/2 "2024-04-02T07:40:52Z")

</div>

I believe the more technical write up is hosted here:

> **[LLaMA Now Goes Faster on CPUs](https://justine.lol/matmul)**
>
> I wrote 84 new matmul kernels to improve llamafile CPU performance.

---

<div class="post-metadata">

**Author:** ![manuel.arenaz](https://yyz2.discourse-cdn.com/free1/user_avatar/fortran-lang.discourse.group/manuel.arenaz/32/3217_2.png) [@manuel.arenaz](https://fortran-lang.discourse.group/u/manuel.arenaz)\
**Post date:** [April 2, 2024, 9:01am UTC](https://fortran-lang.discourse.group/t/evidence-of-fortran-boosting-genai/7750/3 "2024-04-02T09:01:33Z")

</div>

Thanks for sharing!

---

<div class="post-metadata">

**Author:** ![Beliavsky](https://avatars.discourse-cdn.com/v4/letter/b/ba8739/32.png) [@Beliavsky](https://fortran-lang.discourse.group/u/Beliavsky)\
**Post date:** [April 2, 2024, 10:26am UTC](https://fortran-lang.discourse.group/t/evidence-of-fortran-boosting-genai/7750/4 "2024-04-02T10:26:45Z")

</div>

The post was [much discussed](https://news.ycombinator.com/item?id=39890262) at Hacker News (currently 1233 points and 401 comments).

---

<div class="post-metadata">

**Author:** ![rbitr](https://avatars.discourse-cdn.com/v4/letter/r/f1d935/32.png) [@rbitr](https://fortran-lang.discourse.group/u/rbitr)\
**Post date:** [April 2, 2024, 9:42pm UTC](https://fortran-lang.discourse.group/t/evidence-of-fortran-boosting-genai/7750/5 "2024-04-02T21:42:30Z")

</div>

See also [GitHub - rbitr/llm.f90: LLM inference in Fortran](https://github.com/rbitr/llm.f90) this Fortran implementations of CPU language model inference for a few models including llama. There is a branch for PHI-2 that is a smaller model better suited to CPU. (Disclaimer, I am the author)