keras-team/keras · error · ValueError

Weight count mismatch for layer #{k} (named {layer.name} in

Error message

Weight count mismatch for layer #{k} (named {layer.name} in the current model, {name} in the save file). Layer expects {len(symbolic_weights)} weight(s). Received {len(weight_values)} saved weight(s)

What it means

After matching layer #k between file and model, load_weights_from_hdf5_group compares the number of weight tensors in the saved layer group against the layer's symbolic weights. A per-layer count mismatch raises this ValueError naming both the current and saved layer names.

Source

Thrown at keras/src/legacy/saving/legacy_h5_format.py:387

            layer_group, "weight_names"
        )
        if weight_names:
            filtered_layer_names.append(name)
    layer_names = filtered_layer_names
    if len(layer_names) != len(filtered_layers):
        raise ValueError(
            "Layer count mismatch when loading weights from file. "
            f"Model expected {len(filtered_layers)} layers, found "
            f"{len(layer_names)} saved layers."
        )

    for k, name in enumerate(layer_names):
        layer_group = safe_get_h5_group(group, name)
        layer = filtered_layers[k]
        symbolic_weights = _legacy_weights(layer)
        weight_values = load_subset_weights_from_hdf5_group(layer_group)
        if len(weight_values) != len(symbolic_weights):
            raise ValueError(
                f"Weight count mismatch for layer #{k} (named {layer.name} in "
                f"the current model, {name} in the save file). "
                f"Layer expects {len(symbolic_weights)} weight(s). Received "
                f"{len(weight_values)} saved weight(s)"
            )
        _set_weights(
            layer,
            symbolic_weights,
            weight_values,
            skip_mismatch=skip_mismatch,
            name=f"layer #{k} (named {layer.name})",
        )

    if "top_level_model_weights" in group:
        symbolic_weights = list(
            # model.weights
            v
            for v in model._trainable_variables + model._non_trainable_variables

View on GitHub (pinned to 7a34a03db6)

Solutions

1. Align the layer definition with the checkpoint (same use_bias, units, etc.)
2. Ensure the model is built (call build(input_shape) or run a forward pass) before load_weights
3. Use by_name=True so only matching layers are loaded and inspect the mismatched layer named in the message

Example fix

# before
dense = layers.Dense(64, use_bias=False)  # saved with use_bias=True
model.load_weights('w.h5')
# after
dense = layers.Dense(64, use_bias=True)
model.load_weights('w.h5')

model.build(input_shape)  # ensure symbolic weights exist before loading

try:
    model.load_weights(p)
except ValueError as e:
    if 'Weight count mismatch' not in str(e):
        raise
    model.load_weights(p, by_name=True)

When it happens

Trigger: A layer whose weight count changed between save and load: toggling use_bias off, changing units, replacing a layer with a similar one that has fewer/more weights (e.g. BatchNormalization vs LayerNormalization), even when overall layer counts match.

Common situations: Fine-tuning setups where layers were rebuilt with different hyperparameters; loading checkpoints across model refactors; loading before model.build so symbolic weights do not exist yet.

Related errors

• Layer count mismatch when loading weights from file. Model e
• You called `set_weights(weights)` on layer '{self.name}' wit
• Layer '{self.name}' expected {len(all_vars)} variables, but
• `load_model()` using h5 format requires h5py. Could not impo
• No model config found in the file at {filepath}.

AI-assisted analysis of keras-team/keras@7a34a03db6 (2026-08-25). Data as JSON: /api/errors/d60553954b84776b. Report an issue: GitHub.