| #!/bin/bash |
| |
| |
| |
| |
| |
|
|
| set -e |
|
|
| |
| |
| |
| EXPERIMENT_NAME_PREFIX="circle_packing_v3_flash" |
| NUM_GENERATIONS=200 |
| MAX_PARALLEL_JOBS=4 |
| META_INTERVAL=10 |
| PERSISTENT_AGENTS="" |
| PERSISTENT_CONTEXT_REFRESH_INTERVAL=10 |
|
|
| LLM_MODELS="native-gemini-3-flash-preview" |
| |
| LLM_SELECTION="ucb1" |
| LLM_TEMPERATURES="0.7 1.0" |
| LLM_MAX_TOKENS=65536 |
|
|
| USE_EVAL_SERVICE="--use-eval-service" |
| EVAL_SERVICE_URL="http://localhost:8733" |
| EVAL_TRIGGER_MODE="periodic" |
| EVAL_TRIGGER_INTERVAL=5 |
|
|
| EXPERIMENT_NAME=${EXPERIMENT_NAME_PREFIX}_g${NUM_GENERATIONS}_${EVAL_TRIGGER_MODE}${EVAL_TRIGGER_INTERVAL} |
|
|
| USE_WANDB="--use-wandb" |
| WANDB_PROJECT="ev2" |
| WANDB_TAGS="circle_packing eval-service full-experiment ${EVAL_TRIGGER_MODE}" |
|
|
| |
| |
| |
| .venv/bin/python tasks/circle_packing/run_experiment.py \ |
| --experiment-name "$EXPERIMENT_NAME" \ |
| --num-generations "$NUM_GENERATIONS" \ |
| --max-parallel-jobs "$MAX_PARALLEL_JOBS" \ |
| --meta-interval "$META_INTERVAL" \ |
| $PERSISTENT_AGENTS \ |
| --persistent-context-refresh-interval "$PERSISTENT_CONTEXT_REFRESH_INTERVAL" \ |
| --llm-models $LLM_MODELS \ |
| --llm-selection "$LLM_SELECTION" \ |
| --llm-temperatures $LLM_TEMPERATURES \ |
| --llm-max-tokens "$LLM_MAX_TOKENS" \ |
| $USE_EVAL_SERVICE \ |
| --eval-service-url "$EVAL_SERVICE_URL" \ |
| --eval-trigger-mode "$EVAL_TRIGGER_MODE" \ |
| --eval-trigger-interval "$EVAL_TRIGGER_INTERVAL" \ |
| $USE_WANDB \ |
| --wandb-project "$WANDB_PROJECT" \ |
| --wandb-tags $WANDB_TAGS \ |
| --verbose \ |
| --trajectory-log |
|
|