From 88eb579be20c36d26d61ac484de07f327c22b30f Mon Sep 17 00:00:00 2001
From: Chizkiyahu <chizkiyahur@altair-semi.com>
Date: Fri, 6 Oct 2023 16:07:21 +0300
Subject: [PATCH] keras fix layer sharing inputs and outputs names

fix layer shring inputs and outputs
---
 source/keras.js | 46 ++++++++++++++++++++++++++++++++++++++++------
 1 file changed, 40 insertions(+), 6 deletions(-)

diff --git a/source/keras.js b/source/keras.js
index 30f765c7db..3f6b088099 100644
--- a/source/keras.js
+++ b/source/keras.js
@@ -788,21 +788,54 @@ keras.Node = class {
 
         const innerType = this.inner ? this.inner.type : null;
         const innerMetadata = innerType ? metadata.type(innerType) : null;
+
+        // handle layer sharing
         let inputIndex = 0;
+        let inputTypes = [];
+        const outputTypes = [];
+        if (this._type && this._type.inputs) {
+            for (let i = 0; i < this._type.inputs.length; i++) {
+                if (layer.inbound_nodes && layer.inbound_nodes.length > 1 && layer.inbound_nodes[0].length > i) {
+                    for (let j = 0; j < layer.inbound_nodes.length; j++) {
+                        const inputType = JSON.parse(JSON.stringify(this._type.inputs[i]));
+                        inputType.name =  j + ": "+ inputType.name;
+                        inputTypes.push(inputType);
+                    }
+                } else {
+                    inputTypes.push(this._type.inputs[i]);
+                }
+            }
+        }
+        if (this._type && this._type.outputs) {
+            for (let i = 0; i < this._type.outputs.length; i++) {
+                if (layer.inbound_nodes && layer.inbound_nodes.length > 1 && layer.inbound_nodes[0].length > i) {
+                    for (let j = 0; j < layer.inbound_nodes.length; j++) {
+                        const outputType = JSON.parse(JSON.stringify(this._type.outputs[i]));
+                        outputType.name = j + ": "+ outputType.name;
+                        outputTypes.push(outputType);
+                    }
+                } else {
+                    outputTypes.push(this._type.inputs[i]);
+                }
+            }
+        }
+        const inbound_nodes_size = layer.inbound_nodes ? layer.inbound_nodes.length : 1;
+        outputs = Array(inbound_nodes_size).fill(outputs).flat();
+
         while (inputs.length > 0) {
             let list = false;
             let name = null;
             let visible = true;
             if (!innerMetadata || inputIndex == 0) {
-                if (this._type && this._type.inputs && inputIndex < this._type.inputs.length) {
-                    const input = this._type.inputs[inputIndex];
+                if (inputTypes && inputIndex < inputTypes.length) {
+                    const input = inputTypes[inputIndex];
                     name = input.name;
                     if (type === 'BatchNormalization' && name === 'gamma' && config.scale === false) {
                         inputIndex++;
                         continue;
                     }
                     visible = input.visible == false ? false : true;
-                    if (this._type.inputs[inputIndex].list) {
+                    if (inputTypes[inputIndex].list) {
                         list = true;
                     }
                 }
@@ -832,7 +865,8 @@ keras.Node = class {
                         break;
                 }
             }
-            const input = !list ? [ inputs.shift() ] : inputs.splice(0, inputs.length);
+            const size = !layer.inbound_nodes  ? 1 : layer.inbound_nodes.length;
+            const input = !list ? [ inputs.shift() ] : inputs.splice(0, size);
             const inputArguments = input.map((input) => {
                 if (input.name) {
                     return values.map(input.name, null, initializers[input.name]);
@@ -861,12 +895,12 @@ keras.Node = class {
 
         for (let i = 0; i < outputs.length; i++) {
             const output = outputs[i];
-            const outputName = (this._type && this._type.outputs && i < this._type.outputs.length && this._type.outputs[i] && this._type.outputs[i].name) ? this._type.outputs[i].name : i.toString();
+            const outputName = (outputTypes[i] && outputTypes[i].name) ? outputTypes[i].name : i.toString();
             const argument = new keras.Argument(outputName, true, output.length === 0 ? [] : [ values.map(output) ]);
             this._outputs.push(argument);
         }
 
-        const inputTypes = new Map((this._type.inputs || []).map((input) => [ input.name, input.type ]));
+        inputTypes = new Map((inputTypes || []).map((input) => [ input.name, input.type ]));
         for (const entry of Object.entries(args)) {
             const name = entry[0];
             const arg = entry[1];