Package: llamaR 0.2.5

llamaR: Interface for Large Language Models via 'llama.cpp'

Provides R bindings to 'llama.cpp' for running large language models locally, with optional GPU acceleration via 'ggmlR'. Supports text generation, embeddings, chat-based workflows, tool calling, and multimodal (vision) inference. Includes 'OpenAI'- and 'Anthropic'-compatible HTTP servers for serving local models, along with device selection and multi-GPU support.

Authors:Yuri Baramykov [aut, cre], Georgi Gerganov [cph]

llamaR_0.2.5.tar.gz
llamaR_0.2.5.zip(r-4.7-x86_64)llamaR_0.2.5.zip(r-4.6-x86_64)llamaR_0.2.5.zip(r-4.5-x86_64)
llamaR_0.2.5.tgz(r-4.6-x86_64)llamaR_0.2.5.tgz(r-4.6-arm64)llamaR_0.2.5.tgz(r-4.5-x86_64)llamaR_0.2.5.tgz(r-4.5-arm64)
llamaR_0.2.5.tar.gz(r-4.7-arm64)llamaR_0.2.5.tar.gz(r-4.7-x86_64)llamaR_0.2.5.tar.gz(r-4.6-arm64)llamaR_0.2.5.tar.gz(r-4.6-x86_64)
manual.pdf |manual.html✨
DESCRIPTION |NEWS
card.svg |card.png
llamaR/json (API)

# Install 'llamaR' in R:
install.packages('llamaR', repos = c('https://zabis13.r-universe.dev', 'https://cloud.r-project.org'))

Bug tracker:https://github.com/zabis13/llamar/issues

Uses libs:
  • c++– GNU Standard C++ Library v3
  • openmp– GCC OpenMP (GOMP) support library

On CRAN:

Conda:

cppopenmp

6.93 score 6 stars 72 scripts 364 downloads 99 exports 4 dependencies

Last updated from:7af1093184. Checks:12 OK, 1 FAIL. Indexed: yes.

TargetResultTimeFilesSyslog
linux-devel-arm64OK450
linux-devel-x86_64OK561
source / vignettesOK842
linux-release-arm64OK460
linux-release-x86_64OK423
macos-release-arm64OK321
macos-release-x86_64OK625
macos-oldrel-arm64OK322
macos-oldrel-x86_64OK664
windows-develOK645
windows-releaseOK679
windows-oldrelOK699
wasm-releaseFAIL355

Exports:chat_llamarchat_llamar_stopembed_llamarllama_backend_devicesllama_batch_freellama_batch_initllama_chat_apply_templatellama_chat_buildllama_chat_builtin_templatesllama_chat_parsellama_chat_templatellama_detokenizellama_embed_batchllama_embeddingsllama_encodellama_free_contextllama_free_modelllama_gen_beginllama_gen_begin_atllama_gen_endllama_gen_nextllama_generatellama_generate_batchllama_get_embeddingsllama_get_embeddings_ithllama_get_embeddings_seqllama_get_logitsllama_get_logits_ithllama_get_modelllama_get_verbosityllama_hf_cache_clearllama_hf_cache_dirllama_hf_cache_infollama_hf_downloadllama_hf_listllama_image_evalllama_image_loadllama_load_modelllama_load_model_hfllama_lora_applyllama_lora_clearllama_lora_loadllama_lora_removellama_max_devicesllama_memory_breakdown_printllama_memory_can_shiftllama_memory_clearllama_memory_seq_addllama_memory_seq_cpllama_memory_seq_divllama_memory_seq_keepllama_memory_seq_pos_rangellama_memory_seq_rmllama_model_infollama_model_metallama_model_meta_valllama_mtmd_loadllama_mtmd_markerllama_mtmd_set_verbosityllama_mtmd_support_audiollama_mtmd_support_visionllama_n_batchllama_n_ctxllama_n_ctx_seqllama_n_seq_maxllama_n_threadsllama_n_threads_batchllama_n_ubatchllama_new_contextllama_numa_initllama_perfllama_perf_printllama_perf_resetllama_pooling_typellama_serve_anthropicllama_serve_openaillama_set_abort_callbackllama_set_causal_attnllama_set_threadsllama_set_verbosityllama_set_warmupllama_state_get_sizellama_state_loadllama_state_savellama_supports_gpullama_supports_mlockllama_supports_mmapllama_supports_rpcllama_synchronizellama_system_infollama_time_usllama_token_to_piecellama_tokenizellama_vocab_get_scorellama_vocab_get_textllama_vocab_infollama_vocab_is_controlllama_vocab_is_eogllama_vocab_type

Dependencies:genericsggmlRjsonliteR6

Multi-GPU: Splits, Replicas and Benchmarking
1. The one rule that matters | 2. Data parallelism: replicas | 3. When the model does not fit: layer vs row | 4. The hybrid: TP × DP | 5. Measuring it yourself | 6. Serving | 7. Summary
Rendered frommulti-gpu.Rmdusingknitr::rmarkdown

Last update: 2026-07-12
Started: 2026-07-12

Chat and Agents
1. The chat object: chat_llamar() | Mode A — spawn a server for a model | Mode B — connect to a running server | System prompt | 2. The server: llama_serve_openai() | Connecting OpenCode | 3. The command-line example | 4. ragnar: retrieval-augmented chat | 5. Tool calling: llama_chat_build() / llama_chat_parse() | 6. Claude Code on a local model: llama_serve_anthropic() | 7. Concurrency | See also
Rendered fromchat-and-agents.Rmdusingknitr::rmarkdown

Last update: 2026-06-27
Started: 2026-05-27

Getting Started with llamaR
1. Getting a model | 2. Loading a model and creating a context | 3. Generating text | 4. Chat models and templates | 5. Tokenization | 6. Embeddings | ragnar-compatible provider | 7. Serving and chatting | See also
Rendered fromgetting-started.Rmdusingknitr::rmarkdown

Last update: 2026-05-27
Started: 2026-05-27

Readme and manuals

Help Manual

Help pageTopics
Chat with a local model through an ellmer::Chat objectchat_llamar
Stop the server spawned by chat_llamar()chat_llamar_stop
Embedding provider for ragnar / standalone useembed_llamar
List available backend devicesllama_backend_devices
Free a llama batch allocated with 'llama_batch_init()'llama_batch_free
Initialise a llama batchllama_batch_init
Apply chat template to messagesllama_chat_apply_template
Build a tool-aware chat prompt and its parsing grammarllama_chat_build
List built-in chat templatesllama_chat_builtin_templates
Parse raw model output into content and tool callsllama_chat_parse
Get model's built-in chat templatellama_chat_template
Detokenize token IDs back to textllama_detokenize
Batch embeddings for multiple textsllama_embed_batch
Extract embeddings for a textllama_embeddings
Encode tokens using the encoder (encoder-decoder models only)llama_encode
Free an inference contextllama_free_context
Free a loaded modelllama_free_model
Begin a streaming (token-by-token) generationllama_gen_begin
Begin streaming generation from an already-prefilled contextllama_gen_begin_at
Finish a streaming generationllama_gen_end
Pull the next chunk of a streaming generationllama_gen_next
Generate text from a promptllama_generate
Generate completions for multiple prompts in parallelllama_generate_batch
Get all output token embeddings as a matrixllama_get_embeddings
Get embeddings for the i-th token in the batchllama_get_embeddings_ith
Get pooled embeddings for a sequencellama_get_embeddings_seq
Get logits from the last decode stepllama_get_logits
Get logits for a specific token positionllama_get_logits_ith
Get the model associated with a contextllama_get_model
Get current verbosity levelllama_get_verbosity
Clear the model cachellama_hf_cache_clear
Get the cache directory for downloaded modelsllama_hf_cache_dir
Show information about the model cachellama_hf_cache_info
Download a GGUF model from Hugging Facellama_hf_download
List GGUF files in a Hugging Face repositoryllama_hf_list
Evaluate an image + prompt into a llama contextllama_image_eval
Load an image file into an mtmd bitmapllama_image_load
Load a GGUF model filellama_load_model
Load a model directly from Hugging Facellama_load_model_hf
Apply a LoRA adapter to contextllama_lora_apply
Remove all LoRA adapters from contextllama_lora_clear
Load a LoRA adapterllama_lora_load
Remove a LoRA adapter from contextllama_lora_remove
Get maximum number of devicesllama_max_devices
Print memory breakdown by devicellama_memory_breakdown_print
Check if the KV cache supports shiftingllama_memory_can_shift
Clear the KV cachellama_memory_clear
Shift token positions in a sequencellama_memory_seq_add
Copy a sequence in the KV cachellama_memory_seq_cp
Integer-divide token positions in a sequencellama_memory_seq_div
Keep only one sequence in the KV cachellama_memory_seq_keep
Get position range for a sequencellama_memory_seq_pos_range
Remove tokens from a sequence in the KV cachellama_memory_seq_rm
Get model metadatallama_model_info
Get all model metadata as a named character vectorllama_model_meta
Get a single model metadata value by keyllama_model_meta_val
Load a multimodal projector (mmproj)llama_mtmd_load
Media marker string for multimodal promptsllama_mtmd_marker
Set verbosity of the multimodal subsystemllama_mtmd_set_verbosity
Does this multimodal context support audio?llama_mtmd_support_audio
Does this multimodal context support vision (images)?llama_mtmd_support_vision
Get logical batch sizellama_n_batch
Get context window sizellama_n_ctx
Get per-sequence context window sizellama_n_ctx_seq
Get maximum number of sequencesllama_n_seq_max
Get number of threads for single-token generationllama_n_threads
Get number of threads for batch processingllama_n_threads_batch
Get physical micro-batch sizellama_n_ubatch
Create an inference contextllama_new_context
Initialize NUMA optimizationllama_numa_init
Get performance statisticsllama_perf
Print performance statistics to the consolellama_perf_print
Reset performance countersllama_perf_reset
Get pooling typellama_pooling_type
Serve an Anthropic Messages API-compatible endpoint for a local modelllama_serve_anthropic
Serve an OpenAI-compatible HTTP API for a local modelllama_serve_openai
Set or clear the abort callbackllama_set_abort_callback
Set causal attention modellama_set_causal_attn
Set the number of threads for a contextllama_set_threads
Set logging verbosity levelllama_set_verbosity
Set warmup modellama_set_warmup
Get the size of the serialized context state in bytesllama_state_get_size
Load context state from filellama_state_load
Save context state to filellama_state_save
Check whether GPU offloading is availablellama_supports_gpu
Check whether memory locking is supportedllama_supports_mlock
Check whether memory-mapped file I/O is supportedllama_supports_mmap
Check whether RPC backend is availablellama_supports_rpc
Synchronize asynchronous computationllama_synchronize
Get system information stringllama_system_info
Get current time in microsecondsllama_time_us
Convert a single token ID to its text piecellama_token_to_piece
Tokenize text into token IDsllama_tokenize
Get the score of a tokenllama_vocab_get_score
Get the text representation of a tokenllama_vocab_get_text
Get vocabulary special token IDsllama_vocab_info
Check if a token is a control tokenllama_vocab_is_control
Check if a token is an end-of-generation tokenllama_vocab_is_eog
Get vocabulary typellama_vocab_type