| from collections import deque |
|
|
| from .defs.samplers import SAMPLERS |
| from .defs.combo import SAMPLER_SELECTION_METHOD |
|
|
|
|
| class Trace: |
| @classmethod |
| def trace(cls, start_node_id, prompt): |
| class_type = prompt[start_node_id]["class_type"] |
| Q = deque() |
| Q.append((start_node_id, 0)) |
| trace_tree = {start_node_id: (0, class_type)} |
| while len(Q) > 0: |
| current_node_id, distance = Q.popleft() |
| input_fields = prompt[current_node_id]["inputs"] |
| for value in input_fields.values(): |
| if isinstance(value, list): |
| nid = value[0] |
| class_type = prompt[nid]["class_type"] |
| trace_tree[nid] = (distance + 1, class_type) |
| Q.append((nid, distance + 1)) |
| return trace_tree |
|
|
| @classmethod |
| def find_sampler_node_id(cls, trace_tree, sampler_selection_method, node_id): |
| if sampler_selection_method == SAMPLER_SELECTION_METHOD[2]: |
| node_id = str(node_id) |
| _, class_type = trace_tree.get(node_id, (-1, None)) |
| if class_type in SAMPLERS.keys(): |
| return node_id |
| return -1 |
|
|
| sorted_by_distance_trace_tree = sorted( |
| [(k, v[0], v[1]) for k, v in trace_tree.items()], |
| key=lambda x: x[1], |
| reverse=(sampler_selection_method == SAMPLER_SELECTION_METHOD[0]), |
| ) |
| for nid, _, class_type in sorted_by_distance_trace_tree: |
| if class_type in SAMPLERS.keys(): |
| return nid |
| return -1 |
|
|
| @classmethod |
| def filter_inputs_by_trace_tree(cls, inputs, trace_tree): |
| filtered_inputs = {} |
| for meta, inputs_list in inputs.items(): |
| for node_id, input_value in inputs_list: |
| trace = trace_tree.get(node_id) |
| if trace is not None: |
| distance = trace[0] |
| if meta not in filtered_inputs: |
| filtered_inputs[meta] = [] |
| filtered_inputs[meta].append((node_id, input_value, distance)) |
|
|
| |
| for k, v in filtered_inputs.items(): |
| filtered_inputs[k] = sorted(v, key=lambda x: x[2]) |
| return filtered_inputs |
|
|
