Live Ollama comparisons for a tibble of pairs

submit_ollama_pairs_live() is a robust row-wise wrapper around ollama_compare_pair_live(). It takes a tibble of pairs (ID1 / text1 / ID2 / text2), submits each pair to a local (or remote) Ollama server, and collects the results.

Usage

submit_ollama_pairs_live(
  pairs,
  model,
  trait_name,
  trait_description,
  prompt_template = set_prompt_template(),
  host = getOption("pairwiseLLM.ollama_host", "http://127.0.0.1:11434"),
  verbose = TRUE,
  status_every = 1,
  progress = TRUE,
  think = FALSE,
  num_ctx = 8192L,
  include_raw = FALSE,
  save_path = NULL,
  parallel = FALSE,
  workers = 1,
  ...
)

Arguments

pairs: Tibble or data frame with at least columns ID1, text1, ID2, text2. Typically created by make_pairs(), sample_pairs(), and randomize_pair_order().
model: Ollama model name (for example "mistral-small3.2:24b", "qwen3:32b", "gemma3:27b").
trait_name: Trait name to pass to ollama_compare_pair_live().
trait_description: Trait description to pass to ollama_compare_pair_live().
prompt_template: Prompt template string, typically from set_prompt_template().
host: Base URL of the Ollama server. Defaults to the option getOption("pairwiseLLM.ollama_host", "http://127.0.0.1:11434").
verbose: Logical; if TRUE, prints status, timing, and result summaries.
status_every: Integer; print status and timing for every status_every-th pair. Defaults to 1 (every pair). Errors are always printed.
progress: Logical; if TRUE, shows a textual progress bar.
think: Logical; see ollama_compare_pair_live() for behavior. When TRUE and the model name starts with "qwen", the temperature is set to 0.6; otherwise the temperature remains 0.
num_ctx: Integer; context window to use via options$num_ctx. The default is 8192L.
include_raw: Logical; if TRUE, each row of the returned tibble will include a raw_response list-column with the parsed JSON body from Ollama. Note: Raw responses are not saved to the incremental CSV file.
save_path: Character string; optional file path (e.g., "output.csv") to save results incrementally. If the file exists, the function reads it to identify and skip pairs that have already been processed (resume mode). Requires the readr package.
parallel: Logical; if TRUE, enables parallel processing using future.apply. Requires the future and future.apply packages. Defaults to FALSE.
workers: Integer; the number of parallel workers (threads) to use if parallel = TRUE. Defaults to 1.
...: Reserved for future extensions and forwarded to ollama_compare_pair_live().

Value

A list containing two elements:

results: A tibble with one row per successfully processed pair.
failed_pairs: A tibble containing the rows from pairs that failed to process (due to API errors or timeouts), along with an error_message column.

Details

This function offers:

Incremental Saving: Writes results to a CSV file as they complete. If the process is interrupted, re-running the function with the same save_path will automatically skip pairs that were already successfully processed.
Parallel Processing: Uses the future package to process multiple pairs simultaneously. Note: Since Ollama typically runs locally on the GPU, parallel processing may degrade performance or cause out-of-memory errors unless the hardware can handle concurrent requests. Defaults are set to sequential processing.

Temperature and context length are controlled as follows:

By default, temperature = 0 for all models.
For Qwen models (model names beginning with "qwen") and think = TRUE, temperature is set to 0.6.
The context window is set via options$num_ctx, which defaults to 8192 but may be overridden via the num_ctx argument.

In most user-facing workflows, it is more convenient to call submit_llm_pairs() with backend = "ollama" rather than using submit_ollama_pairs_live() directly.

As with ollama_compare_pair_live(), this function assumes that:

An Ollama server is running and reachable at host.
The requested models have been pulled in advance (for example ollama pull mistral-small3.2:24b).

Examples

if (FALSE) { # \dontrun{
# Requires a running Ollama server and locally available models.

data("example_writing_samples", package = "pairwiseLLM")

pairs <- example_writing_samples |>
  make_pairs() |>
  sample_pairs(n_pairs = 5, seed = 123) |>
  randomize_pair_order(seed = 456)

td <- trait_description("overall_quality")
tmpl <- set_prompt_template()

# Live comparisons with incremental saving
res_mistral <- submit_ollama_pairs_live(
  pairs             = pairs,
  model             = "mistral-small3.2:24b",
  trait_name        = td$name,
  trait_description = td$description,
  prompt_template   = tmpl,
  save_path         = "ollama_results.csv",
  verbose           = TRUE
)

# Access results
res_mistral$results
} # }

Usage

Arguments

Value

Details

See also

Examples