Upload folder using huggingface_hub

config.json

@@ -42,6 +42,7 @@
   "max_length": 448,
   "max_source_positions": 1500,
   "max_target_positions": 448,
+  "median_filter_width": 7,
   "model_type": "whisper",
   "num_hidden_layers": 6,
   "num_mel_bins": 80,
@@ -139,7 +140,7 @@
     50360,
     50361
   ],
-  "transformers_version": "4.30.2",
+  "transformers_version": "4.33.0.dev0",
   "use_cache": true,
   "use_weighted_layer_sum": false,
   "vocab_size": 51864

generation_config.json

@@ -1,5 +1,26 @@
 {
-  "alignment_heads": [[3, 3], [4, 7], [5, 1], [5, 5], [5, 7]],
+  "alignment_heads": [
+    [
+      3,
+      3
+    ],
+    [
+      4,
+      7
+    ],
+    [
+      5,
+      1
+    ],
+    [
+      5,
+      5
+    ],
+    [
+      5,
+      7
+    ]
+  ],
   "begin_suppress_tokens": [
     220,
     50256
@@ -111,5 +132,5 @@
     50360,
     50361
   ],
-  "transformers_version": "4.30.2"
+  "transformers_version": "4.33.0.dev0"
 }

onnx/decoder_model.onnx

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d00e18e2dc652ed56cd85984146d59ccc35acf72a3462d418b703bb15f7f03d3
-size 314487019
+oid sha256:803d14c05a61b48b3993d2be821e39eb6e4de2205348ec31408776f7cf1cfef9
+size 208302147

onnx/decoder_model_merged.onnx

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b86d63f0644a5df4c335edddfc8c52feb1377bdffa6927f587506c609e409077
-size 314706485
+oid sha256:e2b3d8e71a16e9462c2849ffb79b2966da800997aab299bfb9a754ba3caa54c1
+size 208558935

onnx/decoder_model_merged_quantized.onnx

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:95c3939f7b64dedfea7bff57264c1635bbcf6dace6837bafc6da3167bcdc8211
-size 80171680
+oid sha256:a25afc5858a20aabb7652cb2d555996ebe10691a69bbdb423d5073d52f060325
+size 53707027

onnx/decoder_model_quantized.onnx

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6cc9146e51bebedcb6f2fdbdb3e6f4544aa4d99d49032fe3421c87497d340f95
-size 79821844
+oid sha256:295b8ae738ab8b39612e7ee1ee8cef31e865733f3ecf33d5d319a11df66bf8d0
+size 53309365

onnx/decoder_with_past_model.onnx

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0a54d2939d2f1ea7c998f8e997639a1e78968049af91833d9a0a9eb77fcd9874
-size 301850608
+oid sha256:9623536c11cf88eff7a7ba8cc96e0a7ed709c8287e5af5d9c98bbdf3c92d76d6
+size 195668862

onnx/decoder_with_past_model_quantized.onnx

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:835b6873332400f93027d88c8a38dd089f9489a34c78954467aa7bb16894e927
-size 76595926
+oid sha256:9d57a5d2b6c2e247c2e26818a937fdd9338ac90a961bcc6908eff422b45cf6ef
+size 50087725

onnx/encoder_model.onnx

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a24ce0db1cb944fe7b16e410463e83572500a7301a451b9d30f3e4695d850848
-size 82461621
+oid sha256:3af4d8b7515f01f1c313afe0d691768ff084169b8e1ec1cb4d975ce2ac465f6e
+size 82474863

onnx/encoder_model_quantized.onnx

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7dc3dace4e873e8d62b34fa0524b9d26198eebb06b6282785716cbd53c853026
-size 23183306
+oid sha256:d0d4e59e2842617b39787cece73d7e8f76f99b1697d3386c0e682eca2269f4a1
+size 23200856

quantize_config.json

@@ -0,0 +1,115 @@
+{
+    "per_channel": false,
+    "reduce_range": false,
+    "per_model_config": {
+        "decoder_model": {
+            "op_types": [
+                "Softmax",
+                "Div",
+                "ConstantOfShape",
+                "Reshape",
+                "Squeeze",
+                "Unsqueeze",
+                "Add",
+                "Mul",
+                "Sub",
+                "Pow",
+                "Erf",
+                "Cast",
+                "Less",
+                "Where",
+                "MatMul",
+                "Gather",
+                "Equal",
+                "Concat",
+                "Sqrt",
+                "Expand",
+                "Range",
+                "ReduceMean",
+                "Shape",
+                "Constant",
+                "Transpose",
+                "Slice"
+            ],
+            "weight_type": "QInt8"
+        },
+        "decoder_model_merged": {
+            "op_types": [
+                "Softmax",
+                "Div",
+                "ConstantOfShape",
+                "Reshape",
+                "Squeeze",
+                "Unsqueeze",
+                "Add",
+                "Mul",
+                "Sub",
+                "Pow",
+                "Erf",
+                "Cast",
+                "Less",
+                "Where",
+                "MatMul",
+                "Gather",
+                "Equal",
+                "Concat",
+                "Sqrt",
+                "Expand",
+                "Range",
+                "ReduceMean",
+                "If",
+                "Shape",
+                "Constant",
+                "Transpose",
+                "Slice"
+            ],
+            "weight_type": "QInt8"
+        },
+        "decoder_with_past_model": {
+            "op_types": [
+                "Mul",
+                "Softmax",
+                "ReduceMean",
+                "Sub",
+                "Sqrt",
+                "MatMul",
+                "Shape",
+                "Gather",
+                "Pow",
+                "Div",
+                "Erf",
+                "Constant",
+                "Reshape",
+                "Concat",
+                "Unsqueeze",
+                "Transpose",
+                "Slice",
+                "Add"
+            ],
+            "weight_type": "QInt8"
+        },
+        "encoder_model": {
+            "op_types": [
+                "Mul",
+                "Softmax",
+                "ReduceMean",
+                "Sub",
+                "MatMul",
+                "Unsqueeze",
+                "Div",
+                "Pow",
+                "Shape",
+                "Conv",
+                "Erf",
+                "Constant",
+                "Gather",
+                "Concat",
+                "Reshape",
+                "Sqrt",
+                "Transpose",
+                "Add"
+            ],
+            "weight_type": "QUInt8"
+        }
+    }
+}