Environment variables

The variables that yzma reads.

For your programs

YZMA_LIB

The directory that holds the llama.cpp shared library files. This is the only variable that a normal program needs.

export YZMA_LIB=/path/to/lib

set YZMA_LIB=C:\yzma\lib

The yzma install command and the yzma verify command also read it, so the --lib flag is not necessary when it is set.

A program does not have to use this variable. llama.Load takes any path:

llama.Load("/opt/myapp/lib")

YZMA_VERIFY

How to check the digest of each download.

Value	What it does
`available`	Checks a digest when there is one. This is the default.
`require`	Fails when an asset has no digest.
`off`	Does no check.

export YZMA_VERIFY=require

See Verify an installation.

For the tests of yzma

These variables name the model files that the test suite uses. make test sets them.

Variable	What it names
`YZMA_TEST_MODEL`	A text model.
`YZMA_TEST_MMMODEL`	A multimodal model.
`YZMA_TEST_MMPROJ`	The projector of that multimodal model.
`YZMA_TEST_QUANTIZE_MODEL`	A model for the quantization tests.
`YZMA_TEST_ENCODER_MODEL`	A model that has an encoder.
`YZMA_TEST_LORA_MODEL`	A model for the LoRA tests.
`YZMA_TEST_LORA_ADAPTER`	A LoRA adapter file.
`YZMA_TEST_SPLIT_MODELS`	A model that comes in splits.
`YZMA_TEST_VIDEO`	A video file for the mtmd video tests.
`YZMA_TEST_LLAMA_TAG`	The `llama.cpp` release tag to test against.

For the benchmarks

Variable	What it names
`YZMA_BENCHMARK_MODEL`	The text model for the benchmarks.
`YZMA_BENCHMARK_MMMODEL`	The multimodal model for the benchmarks.
`YZMA_BENCHMARK_MMPROJ`	The projector of that model.

yzma model get -u https://huggingface.co/QuantFactory/SmolLM-135M-GGUF/resolve/main/SmolLM-135M.Q2_K.gguf
export YZMA_BENCHMARK_MODEL=~/models/SmolLM-135M.Q2_K.gguf

For the Makefile

Variable	Default	What it does
`YZMA_LIB`	`./lib`	Where the Makefile puts the libraries.
`MODELS_DIR`	`$HOME/models`	Where the Makefile puts the models.
`WASM_DIR`	`build/wasm`	Where the WebAssembly build goes.