Skip to content

Fix scripts to support cross-platform execution - #285

Closed
4fuu wants to merge 1 commit into
ggml-org:masterfrom
4fuu:master
Closed

4fuu wants to merge 1 commit into
ggml-org:masterfrom
4fuu:master

Conversation

@4fuu

@4fuu 4fuu commented Mar 19, 2023

Copy link
Copy Markdown

This change modifies the quantize.sh script so that it can run properly on different platforms (including the Windows platform in the WSL environment).

@sw sw mentioned this pull request Mar 19, 2023
@ggerganov

ggerganov commented Mar 19, 2023 •

Copy link
Copy Markdown
Member

Should we make it python script instead of BASH?

@sw

sw commented Mar 19, 2023

Copy link
Copy Markdown
Contributor

Should we make it python script instead of BASH?

@SuajCarrot did the work in #222. I have tested that on Linux but not on Windows.

@tomsnunes

Copy link
Copy Markdown

Should we make it python script instead of BASH?

@SuajCarrot did the work in #222. I have tested that on Linux but not on Windows.

I have successfully tested it on Windows.

@ggerganov

Copy link
Copy Markdown
Member

Ok, let's go with #222 then

@ggerganov ggerganov closed this Mar 19, 2023
thedanhoffman pushed a commit to thedanhoffman/llama.cpp that referenced this pull request Aug 15, 2026
update to ov-2026.3, update device drivers
ckti pushed a commit to ckti/llama.cpp-consolidated that referenced this pull request Aug 22, 2026
The roll-up took upstream's tests/test-chat.cpp but kept the fork's older
common_chat_params_init_deepseek_v3_2, so four upstream behaviours were
missing and test-chat aborted:

  - response_format is not passed to the template as additional_context
  - TC_SEPARATOR ("\n\n") is not accepted before a tool-call block
  - the enable_thinking==false continuation path emitted <think></think>
    instead of </think>
  - content_before_tools did not tolerate the separator or trailing space

Restores all four from upstream. The fork's own tolerance for optional
whitespace before <think> is preserved.

test-chat now passes; ctest is 58/59 with the only failure being the
pre-existing TQ3_1S/TQ4_1S large-N sentinel issue tracked in ggml-org#285.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants