Add missing references to ik_llama.cpp

[ikawrakow]

The modifications added in PR #754 by @moonshadow-25 were copied from ik_llama.cpp. The file iqk_mul_mat.inc to which the changes were added comes from the llamafile project, which just like KTRansformers itself is Apache-2.0 licensed. But the code copied from without attribution has been published under a MIT License.

I left a comment in PR #754 5 days ago, and also had this exchange with @moonshadow-25 3 days ago, which I think has given @moonshadow-25 enough time to correct the problem themself.

This PR corrects the problem created by @moonshadow-25 by adding links to the original source code in ik_llama.cpp along with MIT copyright notices.

As additional piece of information, user @godrosev, who is the author of the first two commits in PR #754, came to the ik_llama.cpp repository to ask for "1.58-bit" support with this issue. As I already had support for optimized IQ1_S matrix multiplications in row-interleaved format, I wasn't planning to also support plain IQ1_S, but then decided to help @godrosev, and added support in this PR.

[lin72h]

@ikawrakow Your contributions to llama.cpp—especially the foundational quantization work—have been instrumental and deserve more credit. It's essential for contributors to respect original work and licensing, as overlooking this can discourage valuable innovations like yours. Thanks for addressing this clearly and helping maintain community standards.

[KMSorSMS]

First and foremost, we would like to express our sincerest apologies for the oversight during the merging of contributor code in PR #754. Regrettably, we failed to thoroughly verify the sources from which parts of the code might have been derived. We now understand that this led to the unintentional incorporation of code originally published under the MIT License without proper attribution, thereby creating an infringement issue. We take full responsibility for this mistake and are truly sorry for any inconvenience or concerns it may have caused.

We will merge your corrections, which corrects the problem by adding links to the original source code in ik_llama.cpp, along with the appropriate MIT copyright notices. Additionally, we will acknowledge your valuable work in our project's acknowledgement section as a testament to your significant contributions and as a token of our gratitude for having pointed out these issues and for taking the time to help us correct them.

Furthermore, after reviewing your implementation of ik_llama.cpp, we recognized several improvements in your version. As such, we are seriously considering adopting your version of llama.cpp in future updates to our project, as we believe it offers enhanced functionality and optimization.

Once again, please accept our heartfelt apologies and thanks. Your feedback has been immensely valuable, and we look forward to any future opportunities for collaboration.

Best regards,
KTransformers team

[ikawrakow]

Thank you!

If you decide to adopt the improvements in ik_llama.cpp in your project, let me know if I can help.

[KMSorSMS]

Sure! Thanks for your generous help. 👍