Package 'llamacppR'

Create an ellmer-style llama.cpp chat

Description

Creates a local chat object backed by native llama.cpp inference while following the ellmer chat API style.

Usage

chat_llamacpp(
  system_prompt = NULL,
  model,
  seed = NULL,
  params = ellmer::params(),
  echo = c("none", "output", "all"),
  n_ctx = 2048L,
  n_batch = n_ctx,
  n_threads = 0L,
  n_gpu_layers = 0L
)

Arguments

system_prompt	Optional system prompt.
model	Path to a local GGUF model file.
seed	Optional seed forwarded to llama.cpp sampling.
params	An ellmer::params() list.
echo	Whether to echo generated output.
n_ctx	Context size.
n_batch	Batch size used for prompt evaluation.
n_threads	CPU threads used by llama.cpp.
n_gpu_layers	Number of layers to offload to GPU when supported.

---

Get the cache path for a curated default model

Description

Returns the local cache path used by llamacppR for one of the curated default GGUF model presets.

Usage

llamacpp_default_model_path(model = c("3b", "0.5b", "starcoder", "deepseek"))

Arguments

model

Which curated default model path to return.

---

Download a curated default GGUF model

Description

Downloads a curated GGUF model from Hugging Face and returns the local path.

Usage

llamacpp_download_default_model(
  model = c("3b", "0.5b", "starcoder", "deepseek"),
  path = NULL,
  force = FALSE
)

Arguments

model	Which curated default model to download.
path	Destination path for the downloaded model.
force	Whether to overwrite an existing file.

---

Download a curated model preset

Description

Downloads one of the curated model presets shipped with llamacppR.

Usage

llamacpp_download_model(
  model = c("qwen_3b", "qwen_0_5b", "starcoder", "deepseek"),
  path = NULL,
  force = FALSE
)

Arguments

model	Preset id or alias.
path	Destination path for the downloaded model.
force	Whether to overwrite an existing file.

---

Check whether a file is GGUF

Description

Validates the magic bytes at the start of a file to determine whether it looks like a GGUF model.

Usage

llamacpp_is_gguf(path)

Arguments

path	Path to inspect.

---

List local GGUF models in the llamacppR cache

Description

Lists GGUF files found in the local llamacppR cache directory and marks whether they match one of the curated default model presets.

Usage

llamacpp_list_models(path = llamacpp_cache_dir(), recursive = TRUE)

Arguments

path	Directory to scan for GGUF files.
recursive	Whether to scan subdirectories recursively.

---

Inspect a GGUF model through llama.cpp

Description

Loads a GGUF model through native llama.cpp bindings and returns basic metadata.

Usage

llamacpp_model_info(
  model,
  n_ctx = 2048L,
  n_batch = n_ctx,
  n_threads = 0L,
  n_gpu_layers = 0L
)

Arguments

model	Path to a GGUF file.
n_ctx	Context size used when opening the model.
n_batch	Batch size used when opening the model.
n_threads	Number of CPU threads.
n_gpu_layers	Number of GPU layers to offload when supported.

---

Get the cache path for a curated model preset

Description

Returns the local cache path used by llamacppR for a curated model preset.

Usage

llamacpp_model_path(model = c("qwen_3b", "qwen_0_5b", "starcoder", "deepseek"))

Arguments

model

Preset id or alias.

---

List curated llama.cpp model presets

Description

Returns the curated model catalog shipped with llamacppR, including stable preset ids, aliases, filenames, approximate sizes, and short descriptions.

Usage

llamacpp_model_presets()

---

Unload a llama.cpp session

Description

Explicitly releases the native llama.cpp model and context associated with a chat or session object.

Usage

llamacpp_unload(x)

Arguments

x	A chat object created by chat_llamacpp() or a native session pointer.

---