| #!/bin/bash |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
|
|
| set -euo pipefail |
| REPO_ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/../.." && pwd)" |
|
|
| MODEL_NAME="llava-hf/llava-1.5-7b-hf" |
| DEVICE_ID="${DEVICE_ID:-0}" |
| DTYPE="${DTYPE:-bfloat16}" |
| NUM_IMAGES="${NUM_IMAGES:-100}" |
| SEED="${SEED:-0}" |
| METHODS="${METHODS:-base lora efuf nullu}" |
| SCORE="${SCORE:-logit}" |
| HOOK_TYPE="${HOOK_TYPE:-post}" |
| MAX_NEW_TOKENS="${MAX_NEW_TOKENS:-64}" |
|
|
| BATH_PROBE="${BATH_PROBE:-/data/caotue/latent_probes/seqprobes_4variant_bathroom/post/seqprobe.pth}" |
| TOILET_PROBE="${TOILET_PROBE:-/data/caotue/latent_probes/seqprobes_4variant_toilet/post/seqprobe.pth}" |
| LORA_PATH="${LORA_PATH:-/data/caotue/multilayer-sae/adv_gen_outputs/run_bathroom_toilet_v2/lora_adapter}" |
| EFUF_PATH="${EFUF_PATH:-/data/caotue/multilayer-sae/EFUF/efuf/checkpoints/llava_vicuna_7b/bathroom_toilet_paper_10ep/epoch_002.pth}" |
| NULLU_PATH="${NULLU_PATH:-/data/caotue/nullu/edited_models/LLaVA-7B-top4-0-32-bathroom_toilet}" |
| NULLU_LOWEST="${NULLU_LOWEST:-8}" |
| NULLU_HIGHEST="${NULLU_HIGHEST:-32}" |
|
|
| OUT="${OUT:-${REPO_ROOT}/mechanistic_interp/graph/correlation_baselines.png}" |
| OUT_JSON="${OUT_JSON:-${REPO_ROOT}/mechanistic_interp/graph/correlation_baselines.json}" |
|
|
| cd "${REPO_ROOT}" |
| export PYTHONPATH="$(cd .. && pwd):$(pwd):${PYTHONPATH:-}" |
| export PYTHONUNBUFFERED=1 |
|
|
| python -m mechanistic_interp.correlation_baselines \ |
| --model_name "${MODEL_NAME}" \ |
| --device_id "${DEVICE_ID}" \ |
| --dtype "${DTYPE}" \ |
| --methods ${METHODS} \ |
| --lora_path "${LORA_PATH}" \ |
| --efuf_path "${EFUF_PATH}" \ |
| --nullu_path "${NULLU_PATH}" \ |
| --nullu_lowest "${NULLU_LOWEST}" \ |
| --nullu_highest "${NULLU_HIGHEST}" \ |
| --bath_probe "${BATH_PROBE}" \ |
| --toilet_probe "${TOILET_PROBE}" \ |
| --num_images "${NUM_IMAGES}" \ |
| --seed "${SEED}" \ |
| --score "${SCORE}" \ |
| --hook_type "${HOOK_TYPE}" \ |
| --max_new_tokens "${MAX_NEW_TOKENS}" \ |
| --out "${OUT}" \ |
| --out_json "${OUT_JSON}" \ |
| "$@" |
|
|