Robbyant/lingbot-map

Largest clones in the code

• 287 lines in 2 placespython1,708 tokens
  lingbot_map/models/gct_stream_window.py:527–813·lingbot_map/models/gct_stream_window_v2.py:590–876

  527# Phase 1: Process scale frames together
  528# These frames get bidirectional attention among themselves via scale token
  529logger.info(f'Processing {scale_frames} scale frames...')
  530scale_images = images[:, :scale_frames].to(_model_device, non_blocking=True)
  531scale_output = self.forward(
  532    scale_images,
  533    num_frame_for_scale=scale_frames,
  534    num_frame_per_block=scale_frames,  # Process all scale frames as one block
  535    causal_inference=True,
  536)
  

  First 10 of 287 lines.

• 180 lines in 2 placespython1,411 tokens
  lingbot_map/models/gct_stream_window.py:820–999·lingbot_map/models/gct_stream_window_v2.py:883–1062

  820# Scale from depth — aggregate across all paired keyframes in overlap.
  821s_ab = unit_s.clone()
  822da = prev_pred.get("depth")
  823db = curr_pred.get("depth")
  824if (da is not None and db is not None
  825        and int(prev_depth_idx.max().item()) < da.shape[1]
  826        and int(curr_depth_idx.max().item()) < db.shape[1]):
  827    s_ab = GCTStream._depth_ratio_scale(
  828        da[:, prev_depth_idx, ..., 0],
  829        db[:, curr_depth_idx, ..., 0],
  

  First 10 of 180 lines.

• 163 lines in 2 placespython1,293 tokens
  lingbot_map/models/gct_stream_window.py:346–508·lingbot_map/models/gct_stream_window_v2.py:409–571

  346non-keyframe processing in keyframe-based streaming inference.
  347
  348Args:
  349    skip: If True, subsequent forward passes will not append KV to cache.
  350"""
  351if hasattr(self.aggregator, 'kv_cache') and self.aggregator.kv_cache is not None:
  352    self.aggregator.kv_cache["_skip_append"] = skip
  353# FlashInfer manager
  354if hasattr(self.aggregator, 'kv_cache_manager') and self.aggregator.kv_cache_manager is not None:
  355    self.aggregator.kv_cache_manager._skip_append = skip
  

  First 10 of 163 lines.