llama.cpp assets for CSGLite

This Go module packages a version-locked llama.cpp convert_hf_to_gguf.py script and its complete matching gguf-py and conversion packages. Consumers embed the module into their binaries, so end users do not need Git or a llama.cpp source download at runtime.

Usage

import assets "github.com/opencsgs/llama-cpp-assets"

converter, err := fs.ReadFile(assets.FS, assets.ConverterPath)

assets.LlamaCppRef identifies the exact upstream llama.cpp release.

Updating llama.cpp

source ~/.myshrc
./scripts/sync-llama-cpp.sh --tag bXXXX
go test ./...

Review and commit all synchronized files, tag a new semantic module version, and update CSGLite's go.mod. CSGLite must update its llama-server installer tag in the same change.

License

The synchronized llama.cpp files retain their upstream MIT license in data/gguf-py/LICENSE. Repository integration code is available under Apache-2.0; see LICENSE.