scripts : add pr2wt.sh by ggerganov · Pull Request #18644 · ggml-org/llama.cpp

ggerganov · 2026-01-06T16:12:04Z

Helper script for quickly creating a git worktree from a PR.

Sample usage:

# create worktree in ../llama.cpp-pr-12345
./scripts/pr2wt.sh 12345

# create worktree and start opencode in it
./scripts/pr2wt.sh 12345 opencode

# chain multiple commands:
./scripts/pr2wt.sh 12345 "cmake -B build && cmake --build build"

ngxson · 2026-01-06T17:44:59Z

I didn't know about this feature in git, seems to be very useful

am17an · 2026-01-06T17:49:43Z

You can also just do gh pr checkout <PR_NUMBER> --worktree if you have the github-cli

EDIT: sorry, that doesn't work

ggerganov · 2026-01-06T17:52:42Z

You can also just do gh pr checkout <PR_NUMBER> --worktree if you have the github-cli

Doesn't work:

  master [?1]  ~/development/github/llama.cpp 
 19:50:59 [1]  git pr checkout 18644 --worktree
checkout doesn't match the pr id pattern.
18644 doesn't match the pr id pattern.
--worktree doesn't match the pr id pattern.
  master [?1]  ~/development/github/llama.cpp 
 19:51:23 [1]  git pr checkout 18644
checkout doesn't match the pr id pattern.
18644 doesn't match the pr id pattern.
  master [?1]  ~/development/github/llama.cpp 
 19:51:24 [1]  git pr 18644 --worktree
error: unknown option `worktree'

Edit: was using git instead of gh ...

Still, it does not work even with gh:

$ gh pr checkout 18644 --worktree
unknown flag: --worktree
Usage:  gh pr checkout [<number> | <url> | <branch>] [flags]
Flags:
  -b, --branch string        Local branch name to use (default [the name of the head branch])
      --detach               Checkout PR with a detached HEAD
  -f, --force                Reset the existing local branch to the latest state of the pull request
      --recurse-submodules   Update all submodules after checkout

$ gh --version
gh version 2.83.2 (2025-12-10)
https://github.com/cli/cli/releases/tag/v2.83.2

elfarolab · 2026-01-06T17:56:33Z

to test a PR I usually do:

cd into the llama.cpp repo dir and:

git fetch origin pull/PR_NUMBER/head:pr-branch

It will create a new branch named: pr-branch, then build as usual.
BTW this script is very useful.

ggerganov · 2026-01-06T18:03:20Z

git fetch origin pull/PR_NUMBER/head:pr-branch

Yup, this almost works. I want to also have the git remote added so I can potentially push stuff to the PR branch.

elfarolab · 2026-01-06T18:17:11Z

Let'see if this other idea will help you,

if I understood correctly your requirements..(https with token or use ssh instead):

git remote add fork https://github.com/ggml-org/llama.cpp.git

git fetch origin pull/PR_NUMBER/head:pr-branch

git checkout pr-branch

git branch --set-upstream-to=fork/pr-branch pr-branch

.. do your changes.. then as usual:

git add .
git commit -m "your message"
git push fork pr-branch

CISC

Nice!

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>

elfarolab · 2026-01-07T14:01:45Z

@ggerganov

sorry, I couldn't help testing this before.

I found an error.

I am running the script with PR #18641 , it is a draft PR, but still, there is work there to be tested.
running:

./scripts/pr2wt.sh 18641

result in this error:

./scripts/pr2wt.sh 18641
org/repo: ggml-org/llama.cpp.git
curl: (22) The requested URL returned error: 404
url:
head_ref:
error: No such remote: 'pr/18641'
usage: git remote add [<options>] <name> <url>

    -f, --fetch           fetch the remote branches
    --tags                import all tags and associated objects when fetching
                          or do not fetch any tag at all (--no-tags)
    -t, --track <branch>  branch(es) to track
    -m, --master <branch>
                          master branch
    --mirror[=(push|fetch)]
                          set up remote as a mirror to push to or fetch from

fatal: 'pr/18641' does not appear to be a git repository
fatal: Could not read from remote repository.

Please make sure you have the correct access rights
and the repository exists.
Preparing worktree (new branch 'pr/18641')
./scripts/pr2wt.sh: line 57: cd: ../llama.cpp-pr-18641: No such file or directory
git worktree created in

Running the curl alone of LOC 40 results in error HTTP 404.

curl -sSf -H "Accept: application/vnd.github+json" "https://api.github.com/repos/ggml-org/llama.cpp.git/pulls/18641"
curl: (22) The requested URL returned error: 404

The error is due because at LOC 36, the results is: 'ggml-org/llama.cpp.git' but it should be: 'ggml-org/llama.cpp' instead.

Also before to continue to next steps, it is worth considering the output error of the curl, just in case.

Running the curl alone like this works:

curl -sSf -H "Accept: application/vnd.github+json" "https://api.github.com/repos/ggml-org/llama.cpp/pulls/18641"

Command Output

{ "url": "https://api.github.com/repos/ggml-org/llama.cpp/pulls/18641", "id": 3149612767, "node_id": "PR_kwDOJH_K4M67u0bf", "html_url": "https://github.com//pull/18641", "diff_url": "https://github.com//pull/18641.diff", "patch_url": "https://github.com//pull/18641.patch", "issue_url": "https://api.github.com/repos//issues/18641", "number": 18641, "state": "open", "locked": false, "title": "[Do Not Merge] model : LFM2.5-Audio-1.5B", "user": { "login": "tdakhran", "id": 20753751, "node_id": "MDQ6VXNlcjIwNzUzNzUx", "avatar_url": "https://avatars.githubusercontent.com/u/20753751?v=4", "gravatar_id": "", "url": "https://api.github.com/users/tdakhran", "html_url": "https://github.com/tdakhran", "followers_url": "https://api.github.com/users/tdakhran/followers", "following_url": "https://api.github.com/users/tdakhran/following{/other_user}", "gists_url": "https://api.github.com/users/tdakhran/gists{/gist_id}", "starred_url": "https://api.github.com/users/tdakhran/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/tdakhran/subscriptions", "organizations_url": "https://api.github.com/users/tdakhran/orgs", "repos_url": "https://api.github.com/users/tdakhran/repos", "events_url": "https://api.github.com/users/tdakhran/events{/privacy}", "received_events_url": "https://api.github.com/users/tdakhran/received_events", "type": "User", "user_view_type": "public", "site_admin": false }, "body": "Liquid AI released [LFM2.5-Audio-1.5B](https://huggingface.co/LiquidAI/LFM2.5-Audio-1.5B). \r\n\r\n> LFM2.5-Audio-1.5B is [Liquid AI](https://www.liquid.ai/)'s updated end-to-end audio foundation model. Key improvements include a custom, LFM based audio detokenizer, llama.cpp compatible GGUFs for CPU inference, and better ASR and TTS performance.\r\n\r\nThis PR is intended to provide a functional implementation in `llama.cpp` until necessary infrastructure is implemented.\r\nThe plan is to split and merge it into upstream in smaller chunks, while keeping and tracking functional implementation here. It will be rebased from time to time.\r\n\r\nGGUFs, precompiled runners, and instructions, live in https://huggingface.co/LiquidAI/LFM2.5-Audio-1.5B-GGUF.\r\n\r\nMerge plan:\r\n\r\n- [x] Support `n_embd_out` https://github.com//pull/18607\r\n- [ ] Support llama_memory_hybrid_iswa https://github.com//pull/18601\r\n- [x] istft for audio output https://github.com//pull/18645\r\n- [ ] reuse llama-model infra for audio tokenizer\r\n- [ ] tbd\r\n\r\nDemo of capabilities (watch with audio on)\r\n\r\nhttps://github.com/user-attachments/assets/bfb7e598-f3bb-4e19-8d93-967bbf912e18\r\n\r\nThank you, @ngxson for the help!", "created_at": "2026-01-06T14:25:07Z", "updated_at": "2026-01-07T13:56:01Z", "closed_at": null, "merged_at": null, "merge_commit_sha": "44159cd83d4631b098d384cb569d83b1fe0e66af", "assignee": null, "assignees": [

],
"requested_reviewers": [

],
"requested_teams": [

  ],
  "visibility": "public",
  "forks": 0,
  "open_issues": 0,
  "watchers": 0,
  "default_branch": "master"
}

],
"requested_reviewers": [

],
"requested_teams": [

  ],
  "visibility": "public",
  "forks": 0,
  "open_issues": 0,
  "watchers": 0,
  "default_branch": "master"
}

If you like I can help fixing this.

Thank you so much for all your work.

elfarolab · 2026-01-07T14:28:51Z

@ggerganov

to get rid of the trailing part (.git) you can do this:

org_repo=$(echo $url_origin | cut -d/ -f4- | sed 's/\..*$//')

to check the output error status of curl:

meta=$(curl -sSf ... ) || { echo "curl failed"; exit 1; }

* scripts : add pr2wt.sh * script : shebang Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com> --------- Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>

scripts : add pr2wt.sh

1abb8b5

CISC approved these changes Jan 6, 2026

View reviewed changes

Comment thread scripts/pr2wt.sh Outdated

github-actions Bot added the script Script related label Jan 6, 2026

script : shebang

5c86cf9

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>

ggerganov merged commit 5642667 into master Jan 7, 2026
2 checks passed

ggerganov deleted the gg/scripts-add-review branch January 7, 2026 13:16

CISC mentioned this pull request Jan 7, 2026

scripts : fix repos cloned with .git extension #18669

Merged

This was referenced Jan 7, 2026

scripts : support chaining commands in pr2wt.sh #18671

Merged

scripts : pr2wt.sh reset to remote head #18695

Merged

ggerganov mentioned this pull request Apr 26, 2026

pr2wt : symlink .pi #22386

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

scripts : add pr2wt.sh#18644

scripts : add pr2wt.sh#18644
ggerganov merged 2 commits into
masterfrom
gg/scripts-add-review

ggerganov commented Jan 6, 2026 •

edited

Loading

Uh oh!

ngxson commented Jan 6, 2026

Uh oh!

am17an commented Jan 6, 2026 •

edited

Loading

Uh oh!

ggerganov commented Jan 6, 2026 •

edited

Loading

Uh oh!

elfarolab commented Jan 6, 2026 •

edited

Loading

Uh oh!

ggerganov commented Jan 6, 2026

Uh oh!

elfarolab commented Jan 6, 2026 •

edited

Loading

Uh oh!

CISC left a comment

Uh oh!

Uh oh!

Uh oh!

elfarolab commented Jan 7, 2026

Uh oh!

elfarolab commented Jan 7, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

Uh oh!

Conversation

ggerganov commented Jan 6, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ngxson commented Jan 6, 2026

Uh oh!

am17an commented Jan 6, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ggerganov commented Jan 6, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

elfarolab commented Jan 6, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ggerganov commented Jan 6, 2026

Uh oh!

elfarolab commented Jan 6, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

CISC left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

elfarolab commented Jan 7, 2026

Uh oh!

elfarolab commented Jan 7, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

ggerganov commented Jan 6, 2026 •

edited

Loading

am17an commented Jan 6, 2026 •

edited

Loading

ggerganov commented Jan 6, 2026 •

edited

Loading

elfarolab commented Jan 6, 2026 •

edited

Loading

elfarolab commented Jan 6, 2026 •

edited

Loading