add ascend_310b deployment (inference_manifest.json)

inference_manifest.json

@@ -1,68 +1,52 @@
 {
-  "schema_version": 3,
-  "model": {
-    "interface": "tensor_model",
-    "model_type": "zipvoice",
-    "operation": "synthesize",
-    "inputs": [
+  "bundle": {
+    "digest": {
+      "algorithm": "sha256",
+      "scope": "structure",
+      "value": "004a3f46987950104238f5f2f35868ba2459574064960627a1c6f9b8d5b81a50"
+    },
+    "files": [
       {
-        "semantic": "tts.text",
-        "dtype": "uint8",
-        "shape": [
-          -1
-        ]
+        "path": "assets/prompts/default.npz"
       },
       {
-        "semantic": "tts.prompt_audio",
-        "dtype": "float32",
-        "shape": [
-          -1
-        ]
+        "path": "assets/tokens.txt"
       },
       {
-        "semantic": "tts.prompt_sample_rate",
-        "dtype": "int64",
-        "shape": []
+        "path": "assets/vocos/pytorch_model.bin"
       },
       {
-        "semantic": "tts.prompt_text",
-        "dtype": "uint8",
-        "shape": [
-          -1
-        ]
-      }
-    ],
-    "outputs": [
+        "path": "assets/zipvoice_310p.json"
+      },
       {
-        "semantic": "tts.audio",
-        "dtype": "float32",
-        "shape": [
-          -1
-        ]
+        "path": "assets/zipvoice_onnx.json"
       }
     ],
-    "semantic_identity": {
-      "logical_model_revision": "zipvoice-distill-onnx-dynamic-2026-08-15",
-      "preprocessing_contract": "emilia-zh-cn2an-jieba-pypinyin-fixed-golden-prompt-v1",
-      "output_semantics": "mono-float32-pcm-24000hz-vocos-cpu"
-    }
+    "name": "zipvoice",
+    "revision": 1,
+    "uuid": "4a5dc0c0-6ba3-4db9-9de0-ef25689871f1"
   },
   "deployments": {
-    "ascend_310p": {
-      "uuid": "3775e8d7-e1d4-4f5c-9ec0-c2a062605b4c",
-      "revision": 1,
+    "ascend_310b": {
       "artifacts": {
         "flow_decoder_1537": {
           "format": "om",
-          "path": "artifacts/ascend/ascend_310p/fm_decoder_f1537_mask_noif_mixed.om",
-          "sha256": "22be611d858e667e0a0d39111fa44c772f2fbdb7ff0b0ea55d708f0a469f4c6c"
+          "path": "artifacts/ascend_310b/fm_decoder_f1537_mask_310b_origin.om",
+          "sha256": "d9f5a36f5e0d538c6ffc98f5ebd398264b3e36da4536dd5ba0b0618e40f3c23d"
         },
         "text_encoder": {
           "format": "om",
-          "path": "artifacts/ascend/ascend_310p/text_encoder_t256_p29_f3072.om",
-          "sha256": "31a7aa1c06d06abbb9e73d566cc63b38b25b987f13bafeee63f01241a979d287"
+          "path": "artifacts/ascend_310b/text_encoder_t256_p29_f3072_310b_origin.om",
+          "sha256": "25e4c9795466a7b549d890827725a99946a9b621a6be18bfffeb17ce31772099"
         }
       },
+      "audio_contract": {
+        "channel_semantics": "mono",
+        "channels": 1,
+        "execution_mode": "offline",
+        "sample_dtype": "float32",
+        "sample_rate_hz": 24000
+      },
       "bindings": {
         "flow_decoder_1537": {
           "inputs": [
@@ -128,7 +112,7 @@
             {
               "dtype": "float32",
               "index": 0,
-              "runtime_name": "PartitionedCall_/fm_decoder/Transpose_1_Transpose_2643:0:v",
+              "runtime_name": "PartitionedCall_/fm_decoder/Transpose_1_Transpose_2566:0:velocity",
               "semantic": "host.zipvoice.velocity",
               "shape": [
                 1,
@@ -186,7 +170,7 @@
             {
               "dtype": "float32",
               "index": 0,
-              "runtime_name": "/Where_5:0:text_condition",
+              "runtime_name": "/Mul_3:0:text_condition",
               "semantic": "host.zipvoice.text_condition",
               "shape": [
                 1,
@@ -197,14 +181,14 @@
             {
               "dtype": "int64",
               "index": 1,
-              "runtime_name": "/Reshape_7:0:features_len",
+              "runtime_name": "/Add_1:0:features_len",
               "semantic": "host.zipvoice.features_len",
               "shape": []
             },
             {
               "dtype": "bool",
               "index": 2,
-              "runtime_name": "/Unsqueeze_3:0:padding_mask",
+              "runtime_name": "/Unsqueeze_2:0:padding_mask",
               "semantic": "host.zipvoice.padding_mask",
               "shape": [
                 1,
@@ -219,163 +203,365 @@
         "flow_decoder_1537"
       ],
       "execution_contract": {
-        "state_scope": "request",
+        "cancellation_granularity": "checkpoint",
         "execution_structure": "iterative",
         "orchestration_visibility": "session",
-        "cancellation_granularity": "checkpoint"
+        "state_scope": "request"
       },
+      "revision": 1,
       "runtime_profile": {
         "backend": "ascend",
-        "target": {
-          "runtime": "acl",
-          "runtime_abi": "cann-8.1.RC1",
-          "soc": "Ascend310P1"
-        },
         "profile": {
           "device_id": 0
+        },
+        "target": {
+          "runtime": "acl",
+          "runtime_abi": "cann-8.3.RC1",
+          "soc": "Ascend310B1"
         }
-      }
+      },
+      "uuid": "4198c9a0-75c9-4559-bd80-c5ac9dc72970"
     },
-    "ubuntu_onnx": {
-      "uuid": "3ddcb022-42a9-4835-b2cd-636ed8ec72c4",
-      "revision": 1,
+    "ascend_310p": {
       "artifacts": {
-        "text_encoder": {
-          "format": "onnx",
-          "path": "artifacts/onnx/text_encoder.onnx",
-          "sha256": "495eca2d5f8a911f5c361bcce5bd55cdd2508ccdd26ce3e9bf1d3c29eb974861"
+        "flow_decoder_1537": {
+          "format": "om",
+          "path": "artifacts/ascend/ascend_310p/fm_decoder_f1537_mask_noif_mixed.om",
+          "sha256": "22be611d858e667e0a0d39111fa44c772f2fbdb7ff0b0ea55d708f0a469f4c6c"
         },
-        "flow_decoder": {
-          "format": "onnx",
-          "path": "artifacts/onnx/fm_decoder.onnx",
-          "sha256": "4510d4f5f049f14ef80207fca695e13c820e2cea61635f402954950bc62b1e3c"
+        "text_encoder": {
+          "format": "om",
+          "path": "artifacts/ascend/ascend_310p/text_encoder_t256_p29_f3072.om",
+          "sha256": "31a7aa1c06d06abbb9e73d566cc63b38b25b987f13bafeee63f01241a979d287"
         }
       },
+      "audio_contract": {
+        "channel_semantics": "mono",
+        "channels": 1,
+        "execution_mode": "offline",
+        "sample_dtype": "float32",
+        "sample_rate_hz": 24000
+      },
       "bindings": {
+        "flow_decoder_1537": {
+          "inputs": [
+            {
+              "dtype": "float32",
+              "index": 0,
+              "runtime_name": "t",
+              "semantic": "host.zipvoice.t",
+              "shape": []
+            },
+            {
+              "dtype": "float32",
+              "index": 1,
+              "runtime_name": "x",
+              "semantic": "host.zipvoice.flow_x",
+              "shape": [
+                1,
+                1537,
+                100
+              ]
+            },
+            {
+              "dtype": "float32",
+              "index": 2,
+              "runtime_name": "text_condition",
+              "semantic": "host.zipvoice.flow_text_condition",
+              "shape": [
+                1,
+                1537,
+                100
+              ]
+            },
+            {
+              "dtype": "float32",
+              "index": 3,
+              "runtime_name": "speech_condition",
+              "semantic": "host.zipvoice.speech_condition",
+              "shape": [
+                1,
+                1537,
+                100
+              ]
+            },
+            {
+              "dtype": "bool",
+              "index": 4,
+              "runtime_name": "padding_mask",
+              "semantic": "host.zipvoice.flow_padding_mask",
+              "shape": [
+                1,
+                1537
+              ]
+            },
+            {
+              "dtype": "float32",
+              "index": 5,
+              "runtime_name": "guidance_scale",
+              "semantic": "host.zipvoice.guidance_scale",
+              "shape": []
+            }
+          ],
+          "outputs": [
+            {
+              "dtype": "float32",
+              "index": 0,
+              "runtime_name": "PartitionedCall_/fm_decoder/Transpose_1_Transpose_2643:0:v",
+              "semantic": "host.zipvoice.velocity",
+              "shape": [
+                1,
+                1537,
+                100
+              ]
+            }
+          ]
+        },
         "text_encoder": {
           "inputs": [
             {
-              "semantic": "host.zipvoice.tokens",
               "dtype": "int64",
               "index": 0,
+              "runtime_name": "tokens",
+              "semantic": "host.zipvoice.tokens",
               "shape": [
                 1,
                 256
               ]
             },
             {
-              "semantic": "host.zipvoice.tokens_len",
               "dtype": "int64",
               "index": 1,
+              "runtime_name": "tokens_len",
+              "semantic": "host.zipvoice.tokens_len",
               "shape": []
             },
             {
-              "semantic": "host.zipvoice.prompt_tokens",
               "dtype": "int64",
               "index": 2,
+              "runtime_name": "prompt_tokens",
+              "semantic": "host.zipvoice.prompt_tokens",
               "shape": [
                 1,
                 29
               ]
             },
             {
-              "semantic": "host.zipvoice.prompt_features_len",
               "dtype": "int64",
               "index": 3,
+              "runtime_name": "prompt_features_len",
+              "semantic": "host.zipvoice.prompt_features_len",
               "shape": []
             },
             {
-              "semantic": "host.zipvoice.speed",
               "dtype": "float32",
               "index": 4,
+              "runtime_name": "speed",
+              "semantic": "host.zipvoice.speed",
               "shape": []
             }
           ],
           "outputs": [
             {
-              "semantic": "host.zipvoice.text_condition",
               "dtype": "float32",
               "index": 0,
+              "runtime_name": "/Where_5:0:text_condition",
+              "semantic": "host.zipvoice.text_condition",
               "shape": [
-                -1,
+                1,
+                3072,
                 100
               ]
             },
             {
-              "semantic": "host.zipvoice.features_len",
               "dtype": "int64",
               "index": 1,
+              "runtime_name": "/Reshape_7:0:features_len",
+              "semantic": "host.zipvoice.features_len",
               "shape": []
             },
             {
-              "semantic": "host.zipvoice.padding_mask",
-              "dtype": "float32",
+              "dtype": "bool",
               "index": 2,
+              "runtime_name": "/Unsqueeze_3:0:padding_mask",
+              "semantic": "host.zipvoice.padding_mask",
               "shape": [
-                -1
+                1,
+                3072
               ]
             }
           ]
+        }
+      },
+      "execution": [
+        "text_encoder",
+        "flow_decoder_1537"
+      ],
+      "execution_contract": {
+        "cancellation_granularity": "checkpoint",
+        "execution_structure": "iterative",
+        "orchestration_visibility": "session",
+        "state_scope": "request"
+      },
+      "revision": 2,
+      "runtime_profile": {
+        "backend": "ascend",
+        "profile": {
+          "device_id": 0
         },
+        "target": {
+          "runtime": "acl",
+          "runtime_abi": "cann-8.1.RC1",
+          "soc": "Ascend310P1"
+        }
+      },
+      "uuid": "3775e8d7-e1d4-4f5c-9ec0-c2a062605b4c"
+    },
+    "ubuntu_onnx": {
+      "artifacts": {
+        "flow_decoder": {
+          "format": "onnx",
+          "path": "artifacts/onnx/fm_decoder.onnx",
+          "sha256": "4510d4f5f049f14ef80207fca695e13c820e2cea61635f402954950bc62b1e3c"
+        },
+        "text_encoder": {
+          "format": "onnx",
+          "path": "artifacts/onnx/text_encoder.onnx",
+          "sha256": "495eca2d5f8a911f5c361bcce5bd55cdd2508ccdd26ce3e9bf1d3c29eb974861"
+        }
+      },
+      "audio_contract": {
+        "channel_semantics": "mono",
+        "channels": 1,
+        "execution_mode": "offline",
+        "sample_dtype": "float32",
+        "sample_rate_hz": 24000
+      },
+      "bindings": {
         "flow_decoder": {
           "inputs": [
             {
-              "semantic": "host.zipvoice.t",
               "dtype": "float32",
               "index": 0,
+              "semantic": "host.zipvoice.t",
               "shape": []
             },
             {
-              "semantic": "host.zipvoice.flow_x",
               "dtype": "float32",
               "index": 1,
+              "semantic": "host.zipvoice.flow_x",
               "shape": [
                 -1,
                 100
               ]
             },
             {
-              "semantic": "host.zipvoice.flow_text_condition",
               "dtype": "float32",
               "index": 2,
+              "semantic": "host.zipvoice.flow_text_condition",
               "shape": [
                 -1,
                 100
               ]
             },
             {
-              "semantic": "host.zipvoice.speech_condition",
               "dtype": "float32",
               "index": 3,
+              "semantic": "host.zipvoice.speech_condition",
               "shape": [
                 -1,
                 100
               ]
             },
             {
-              "semantic": "host.zipvoice.flow_padding_mask",
               "dtype": "float32",
               "index": 4,
+              "semantic": "host.zipvoice.flow_padding_mask",
               "shape": [
                 -1
               ]
             },
             {
-              "semantic": "host.zipvoice.guidance_scale",
               "dtype": "float32",
               "index": 5,
+              "semantic": "host.zipvoice.guidance_scale",
               "shape": []
             }
           ],
           "outputs": [
             {
+              "dtype": "float32",
+              "index": 0,
               "semantic": "host.zipvoice.velocity",
+              "shape": [
+                -1,
+                100
+              ]
+            }
+          ]
+        },
+        "text_encoder": {
+          "inputs": [
+            {
+              "dtype": "int64",
+              "index": 0,
+              "semantic": "host.zipvoice.tokens",
+              "shape": [
+                1,
+                256
+              ]
+            },
+            {
+              "dtype": "int64",
+              "index": 1,
+              "semantic": "host.zipvoice.tokens_len",
+              "shape": []
+            },
+            {
+              "dtype": "int64",
+              "index": 2,
+              "semantic": "host.zipvoice.prompt_tokens",
+              "shape": [
+                1,
+                29
+              ]
+            },
+            {
+              "dtype": "int64",
+              "index": 3,
+              "semantic": "host.zipvoice.prompt_features_len",
+              "shape": []
+            },
+            {
+              "dtype": "float32",
+              "index": 4,
+              "semantic": "host.zipvoice.speed",
+              "shape": []
+            }
+          ],
+          "outputs": [
+            {
               "dtype": "float32",
               "index": 0,
+              "semantic": "host.zipvoice.text_condition",
               "shape": [
                 -1,
                 100
               ]
+            },
+            {
+              "dtype": "int64",
+              "index": 1,
+              "semantic": "host.zipvoice.features_len",
+              "shape": []
+            },
+            {
+              "dtype": "float32",
+              "index": 2,
+              "semantic": "host.zipvoice.padding_mask",
+              "shape": [
+                -1
+              ]
             }
           ]
         }
@@ -385,47 +571,70 @@
         "flow_decoder"
       ],
       "execution_contract": {
-        "state_scope": "request",
+        "cancellation_granularity": "checkpoint",
         "execution_structure": "iterative",
         "orchestration_visibility": "session",
-        "cancellation_granularity": "checkpoint"
+        "state_scope": "request"
       },
+      "revision": 2,
       "runtime_profile": {
         "backend": "torch",
-        "target": {
-          "runtime": "torch"
-        },
         "profile": {
           "device": "cpu"
+        },
+        "target": {
+          "runtime": "torch"
         }
-      }
+      },
+      "uuid": "3ddcb022-42a9-4835-b2cd-636ed8ec72c4"
     }
   },
-  "bundle": {
-    "uuid": "4a5dc0c0-6ba3-4db9-9de0-ef25689871f1",
-    "revision": 1,
-    "name": "zipvoice",
-    "files": [
+  "model": {
+    "inputs": [
       {
-        "path": "assets/prompts/default.npz"
+        "dtype": "uint8",
+        "semantic": "tts.text",
+        "shape": [
+          -1
+        ]
       },
       {
-        "path": "assets/tokens.txt"
+        "dtype": "float32",
+        "semantic": "tts.prompt_audio",
+        "shape": [
+          -1
+        ]
       },
       {
-        "path": "assets/vocos/pytorch_model.bin"
+        "dtype": "int64",
+        "semantic": "tts.prompt_sample_rate",
+        "shape": []
       },
       {
-        "path": "assets/zipvoice_310p.json"
-      },
+        "dtype": "uint8",
+        "semantic": "tts.prompt_text",
+        "shape": [
+          -1
+        ]
+      }
+    ],
+    "interface": "tensor_model",
+    "model_type": "zipvoice",
+    "operation": "synthesize",
+    "outputs": [
       {
-        "path": "assets/zipvoice_onnx.json"
+        "dtype": "float32",
+        "semantic": "tts.audio",
+        "shape": [
+          -1
+        ]
       }
     ],
-    "digest": {
-      "algorithm": "sha256",
-      "scope": "structure",
-      "value": "004a3f46987950104238f5f2f35868ba2459574064960627a1c6f9b8d5b81a50"
+    "semantic_identity": {
+      "logical_model_revision": "zipvoice-distill-onnx-dynamic-2026-08-15",
+      "output_semantics": "mono-float32-pcm-24000hz-vocos-cpu",
+      "preprocessing_contract": "emilia-zh-cn2an-jieba-pypinyin-fixed-golden-prompt-v1"
     }
-  }
+  },
+  "schema_version": 3
 }