This directory contains the release-facing batch runner for SabreCoder benchmarks.

run_sabrecoder_main_benchmark_parallel.sh:

• discovers prepared LCC and CCEval prompt files
• launches one benchmark job per dataset
• assigns jobs to the GPU list you provide
• writes logs and results into timestamped subdirectories

Example:

bash scripts/run_sabrecoder_main_benchmark_parallel.sh \
  --model_name deepseek-ai/deepseek-coder-1.3b-base \
  --budgets "14336 12288" \
  --gpus "0 1"

Prepared prompt directories are expected at:

• data/_cceval_rag_prompts
• data/_lcc_budget_prompts

If you built prompt sets with tokenizer-specific suffixes, the runner will try to select the matching directory automatically.