| import hashlib |
| import json |
| import os |
| from pathlib import Path |
| import time |
| ROOT = Path(os.environ.get('RSI_GUI_ROOT', str(Path(__file__).resolve().parents[1]))) |
| CACHE = Path(os.environ.get('RSI_GUI_CACHE', str(Path.home() / '.cache/rsi_anything/gui'))) |
| CACHE.mkdir(parents=True, exist_ok=True) |
| PORT = int(os.environ.get('RSI_GUI_POLICY_PORT', '8767')) |
| BASE_MODELS = Path(os.environ.get('RSI_BASE_MODELS_DIR', str(Path.home() / '.cache/rsi_anything/base_models'))).expanduser() |
| BASE_WORLD = BASE_MODELS / 'Qwen2.5-Coder-3B-Instruct' |
| BASE_POLICY = BASE_MODELS / 'ReasonFlux-Coder-7B' |
|
|
| def digest(value): |
| return hashlib.sha256(json.dumps(value, sort_keys=True, ensure_ascii=False).encode()).hexdigest() |
|
|
| def atomic(path, value): |
| path = Path(path) |
| path.parent.mkdir(parents=True, exist_ok=True) |
| temp = path.with_name(path.name + f'.{os.getpid()}.tmp') |
| temp.write_text(json.dumps(value, ensure_ascii=False, indent=2)) |
| temp.replace(path) |
|
|
| def context(obs, action): |
| return json.dumps({'instruction': obs['instruction'], 'visible_elements': obs['elements'], 'remaining_actions': obs['remaining_actions'], 'candidate_action': action}, sort_keys=True, ensure_ascii=False) |
|
|
| def file_hash(path): |
| h = hashlib.sha256() |
| with Path(path).open('rb') as f: |
| for chunk in iter(lambda : f.read(8 * 1024 * 1024), b''): |
| h.update(chunk) |
| return h.hexdigest() |
|
|
| def atomic_jsonl(path, rows): |
| p = Path(path) |
| p.parent.mkdir(parents=True, exist_ok=True) |
| tmp = p.with_name(p.name + f'.{os.getpid()}.tmp') |
| with tmp.open('w') as f: |
| for row in rows: |
| f.write(json.dumps(row, ensure_ascii=False) + '\n') |
| tmp.replace(p) |
|
|
| def checkpoint_identity(path): |
| if path is None: |
| return 'base' |
| p = Path(path) |
| meta = p / 'inference.json' |
| weights = p / 'gui_state.pt' if (p / 'gui_state.pt').exists() else p / 'adapter_model.safetensors' |
| return digest([str(p.resolve()), file_hash(meta), file_hash(weights)]) |
|
|
| def effective_cost(rows): |
| return sum((r['policy_nominal_seconds'] + r['world_seconds'] + r['browser_seconds'] for r in rows)) / len(rows) |
|
|
| def goal_bits(obs): |
| sentence = obs['instruction'] |
| (prefix, suffix) = ('Select ', ' and click Submit.') |
| if not sentence.startswith(prefix) or not sentence.endswith(suffix): |
| raise ValueError('Unexpected official instruction: ' + sentence) |
| wanted = sentence[len(prefix):-len(suffix)] |
| names = set() if wanted == 'nothing' else set(wanted.split(', ')) |
| labels = [x['label'] for x in obs['elements']] |
| if not names.issubset(set(labels)): |
| raise ValueError('Target names absent from visible labels') |
| return [x in names for x in labels] |
|
|
| def correctness(obs): |
| desired = goal_bits(obs) |
| return sum((e['checked'] == g for (e, g) in zip(obs['elements'], desired))) / len(desired) |
|
|
| def legal_actions(obs): |
| return [{'kind': 'click', 'slot': e['slot']} for e in obs['elements']] + [{'kind': 'submit'}] |
|
|
