d73e36a44ab3fa760d385c881db0fa393a557a65,torch/quantization/ns/numeric_suite_core_apis_fx.py,,prepare_model_outputs,#Any#Any#Any#Any#Any#,195

Before Change

    gm_b: GraphModule,
    logger_cls: Callable,
) -> Tuple[GraphModule, GraphModule]:
    matched_subgraph_pairs = get_matching_subgraph_pairs(gm_a, gm_b)
    nodes_and_names_to_instrument_a = []
    nodes_and_names_to_instrument_b = []
    for match_name, (subgraph_a, subgraph_b) in matched_subgraph_pairs.items():
        node_start_a, node_end_a = subgraph_a
        node_start_b, node_end_b = subgraph_b
        // Note: for matching activations we always use the end nodes,
        // such as observing the output of relu in linear-relu
        nodes_and_names_to_instrument_a.append((node_end_a, match_name))
        nodes_and_names_to_instrument_b.append((node_end_b, match_name))

    gm_a = prepare_single_model_output(
        name_a, gm_a, nodes_and_names_to_instrument_a, logger_cls)
    gm_b = prepare_single_model_output(
        name_b, gm_b, nodes_and_names_to_instrument_b, logger_cls)

After Change

    model_b: nn.Module,
    logger_cls: Callable,
) -> Tuple[nn.Module, nn.Module]:
    tracer_a, tracer_b = NSTracer(), NSTracer()
    gm_a = GraphModule(model_a, tracer_a.trace(model_a))
    gm_b = GraphModule(model_b, tracer_b.trace(model_b))
    return _prepare_model_outputs_impl(name_a, gm_a, name_b, gm_b, logger_cls)


def add_activation_info_to_dict(
    model: nn.Module,

Instances