Local Large Language Models as Base-R Objects

Documentation for package ‘relm’ version 0.5.0

Help Pages

activations	Select activation features inside a probe formula
as.matrix.relm_trace	Extract one activation slice as a matrix
close.llm	Free a model handle
llm	Load a local large language model
llm_ablate	Ablate neurons in a model
llm_apply_direction	Apply a compatible recorded direction
llm_cancel	Cancel background generation
llm_compare	Compare actual observed states under recorded experimental settings
llm_direction	Construct a recorded contrast direction
llm_download	Download a pinned model, verified by checksum
llm_embed	Embed text with a model
llm_generate	Generate text from a model
llm_logits	Inspect the next-token distribution
llm_probe	Fit grouped statistical probes to activation traces
llm_steer	Steer a model along a direction
llm_timeline	Retain a bounded history of applied live steering
llm_tokens	Tokenize text, or decode token ids back to text
llm_trace	Trace a model's activations
plot.llm	Draw a model block map and configured intervention sites
plot.llm_probe	Summarize and plot a statistical probe set
plot.relm_comparison	Compare actual observed states under recorded experimental settings
plot.relm_timeline	Retain a bounded history of applied live steering
predict.llm_probe	Predict binary probabilities from a fitted probe
print.llm	Load a local large language model
print.llm_probe	Summarize and plot a statistical probe set
print.relm_direction	Construct a recorded contrast direction
print.relm_trace	Trace a model's activations
print.summary.llm	Summarize a model handle
print.summary.llm_probe	Summarize and plot a statistical probe set
print.summary.relm_trace	Trace a model's activations
probe-methods	Summarize and plot a statistical probe set
summary.llm	Summarize a model handle
summary.llm_probe	Summarize and plot a statistical probe set
summary.relm_trace	Trace a model's activations