lutzroeder/netron · error · Error

Expected 'precision'.

Error message

Expected 'precision'.

What it means

Generated protobuf requirement check: after binary-decoding a caffe2.QTensorProto, the required field 'precision' was not present. Quantized tensor protos must carry precision, scale, bias and is_signed; decode() throws on the first missing one in declaration order.

Source

Thrown at source/caffe2-proto.js:268

                case 9:
                    message.scales = reader.doubles(message.scales, tag);
                    break;
                case 10:
                    message.biases = reader.doubles(message.biases, tag);
                    break;
                case 11:
                    message.axis = reader.int32();
                    break;
                case 12:
                    message.is_multiparam = reader.bool();
                    break;
                default:
                    reader.skipType(tag & 7);
                    break;
            }
        }
        if (!Object.prototype.hasOwnProperty.call(message, 'precision')) {
            throw new Error("Expected 'precision'.");
        }
        if (!Object.prototype.hasOwnProperty.call(message, 'scale')) {
            throw new Error("Expected 'scale'.");
        }
        if (!Object.prototype.hasOwnProperty.call(message, 'bias')) {
            throw new Error("Expected 'bias'.");
        }
        if (!Object.prototype.hasOwnProperty.call(message, 'is_signed')) {
            throw new Error("Expected 'is_signed'.");
        }
        return message;
    }

    static decodeText(reader) {
        const message = new caffe2.QTensorProto();
        reader.start();
        while (!reader.end()) {
            const tag = reader.tag();

View on GitHub (pinned to d8a543f5f8)

Solutions

1. Re-export the quantized model with a caffe2 build matching the library's schema
2. Verify file integrity to rule out truncation
3. Make sure the decoder is applied to actual quantized tensor data, not ordinary TensorProto bytes
4. Inspect the message with protoc --decode_raw to see which fields are actually present

const REQUIRED = ['precision','scale','bias','is_signed'];
const missing = REQUIRED.filter((k) => !(k in qt));
if (missing.length) throw new Error(`QTensorProto missing: ${missing.join(', ')}`);

function isQTensor(m) {
  return m != null && ['precision','scale','bias','is_signed']
    .every((k) => Object.prototype.hasOwnProperty.call(m, k));
}

try {
  const qt = caffe2.QTensorProto.decode(reader);
} catch (e) {
  if (/^Expected '(precision|scale|bias|is_signed)'\.$/.test(e.message)) {
    // invalid quantized tensor: skip and report, do not retry
  } else throw e;
}

When it happens

Trigger: Decoding a quantized tensor blob from a caffe2 model where the serialized QTensorProto omits the precision field — truncated message, wrong bytes being parsed as QTensorProto, or a producer using a different schema revision.

Common situations: Loading quantized caffe2 models exported by a different (older/newer) caffe2 version; corrupt downloads; accidentally pointing the QTensor decoder at raw float tensor bytes.

Related errors

• Expected 'scale'.
• Expected 'bias'.
• Expected 'begin'.
• Expected 'end'.
• Expected 'key'.

AI-assisted analysis of lutzroeder/netron@d8a543f5f8 (2026-08-27). Data as JSON: /api/errors/1e17a1cd062957a2. Report an issue: GitHub.