PKG-5550: Add gguf package & Update to 0.9.1

[jnoller]

gguf 0.9.1

Destination channel: ai-staging

Links

• PKG-5550
• Upstream repository
• Upstream changelog/diff

Explanation of changes:

• Updates the forked gguf-feedstock recipe to match the latest published PyPI package: https://pypi.org/project/gguf/ (0.9.1)
• Pins dependency versions, adds missing sentencepiece dependency

[anaconda-pkg-build]

Linter check found the following problems:

ERROR conda.cli.main_run:execute(125): `conda run conda-lint /tmp/abs_e5fmpk5tml/clone` failed. (See above for error) The following problems have been found:

===== WARNINGS =====

• clone/recipe/meta.yaml:49: missing_description: The recipe is missing a description
• clone/recipe/meta.yaml:20: avoid_noarch: noarch: python packages should be avoided

===== ERRORS =====

• clone/recipe/meta.yaml:49: missing_documentation: The recipe is missing a doc_url or doc_source_url
• clone/recipe/meta.yaml:49: missing_license_family: The recipe is missing the about/license_family key.
• clone/recipe/meta.yaml:49: missing_dev_url: The recipe is missing a dev_url
  ===== Final Report: =====
  3 Errors and 2 Warnings were found

[JeanChristopheMorinPerso]

@jnoller You will need to create a file named abs.yaml at the root of the repo with this content in order to upload the packages to ai-staging:

upload_channels:
- ai-staging

You can add it into this PR.

[jnoller]

@JeanChristopheMorinPerso Added abs.yaml to the PR

[anaconda-pkg-build]

Linter check found the following problems:

ERROR conda.cli.main_run:execute(125): `conda run conda-lint /tmp/abs_0cr6exo34o/clone` failed. (See above for error) The following problems have been found:

===== WARNINGS =====

• clone/recipe/meta.yaml:20: avoid_noarch: noarch: python packages should be avoided
• clone/recipe/meta.yaml:49: missing_description: The recipe is missing a description

===== ERRORS =====

• clone/recipe/meta.yaml:49: missing_documentation: The recipe is missing a doc_url or doc_source_url
• clone/recipe/meta.yaml:49: missing_dev_url: The recipe is missing a dev_url
• clone/recipe/meta.yaml:49: missing_license_family: The recipe is missing the about/license_family key.
  ===== Final Report: =====
  3 Errors and 2 Warnings were found

[anaconda-pkg-build]

Linter check found the following problems:

ERROR conda.cli.main_run:execute(125): `conda run conda-lint /tmp/abs_ffpap9l4ao/clone` failed. (See above for error) The following problems have been found:

===== WARNINGS =====

• clone/recipe/meta.yaml:20: avoid_noarch: noarch: python packages should be avoided
  ===== Final Report: =====
  0 Errors and 1 Warning were found

[anaconda-pkg-build]

Linter check found the following problems:

ERROR conda.cli.main_run:execute(125): `conda run conda-lint /tmp/abs_e380gbzfaw/clone` failed. (See above for error) The following problems have been found:

===== ERRORS =====

• clone/recipe/meta.yaml:31: remove_python_pinning: On arch specific packages, python deps should not be constrained.
  ===== Final Report: =====
  1 Error and 0 Warnings were found

[ryanskeith]

Please add the missing dependency (unless I am looking in the wrong place). Other than that looks good.

[jnoller]

@JeanChristopheMorinPerso @Jrice1317 @ryanskeith please re-review with the latest changes. I added pyyaml, but also moved to using the llama.cpp project's releases directly - the pypi project is pretty out of date and to support adding the llama-cpp-scripts package it makes sense to pin all three (include llama.cpp-feedstock) the same way. The llama-cpp-scripts pyproject.toml in the root of the llama.cpp tree has always had a version of 0.0.0 but it has a dependency on a matching version of the gguf package. This way all 3 packages work from the source repo directly for now.

[jnoller]

I'm closing this PR: After working on packaging the llama-cpp-scripts work and working through the dependency tree, I think it makes more sense collapse the gguf bundling/packaging within the llama-cpp-scripts packaging as the two are tightly coupled. It's better to ignore the upstream pyproject.toml files and follow the instructions to set up quantize and other tools (https://github.com/ggerganov/llama.cpp/blob/master/examples/quantize/README.md). Its unclear if the pyproject.toml files are the recommended path forward (ggml-org/llama.cpp#9107).