Upload folder using huggingface_hub

config.json

@@ -50,6 +50,7 @@
   "max_length": 448,
   "max_source_positions": 1500,
   "max_target_positions": 448,
+  "median_filter_width": 7,
   "model_type": "whisper",
   "num_hidden_layers": 4,
   "num_mel_bins": 80,
@@ -145,7 +146,7 @@
     50361,
     50362
   ],
-  "transformers_version": "4.30.2",
+  "transformers_version": "4.33.0.dev0",
   "use_cache": true,
   "use_weighted_layer_sum": false,
   "vocab_size": 51865

generation_config.json

@@ -1,5 +1,30 @@
 {
-  "alignment_heads": [[2, 2], [3, 0], [3, 2], [3, 3], [3, 4], [3, 5]],
+  "alignment_heads": [
+    [
+      2,
+      2
+    ],
+    [
+      3,
+      0
+    ],
+    [
+      3,
+      2
+    ],
+    [
+      3,
+      3
+    ],
+    [
+      3,
+      4
+    ],
+    [
+      3,
+      5
+    ]
+  ],
   "begin_suppress_tokens": [
     220,
     50257
@@ -218,5 +243,5 @@
     "transcribe": 50359,
     "translate": 50358
   },
-  "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 sha256:28fc8dfa729d0fba4dfbd09e2c9f95a70eaa1a657f15a9fb0e6d00d1228ca51f
-size 198049530
+oid sha256:d7668e6e5d90c757759d2a29c918f5f24f6eef073fc5a17c5b86c1a68a17555f
+size 118406656

onnx/decoder_model_merged.onnx

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3a3f5dfc9cf89e4910a1be6574ed519181eefc743fd125ddcd4a97632fd44e94
-size 198197535
+oid sha256:b81258b0e8acdbedce92eb901d18f1c189b2d6f3d8668261d65117d423f9a5b9
+size 118579599

onnx/decoder_model_merged_quantized.onnx

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3630e9e28169f8664cdaed6d2b0f00abae079bae7e4336c249308022635721b4
-size 50584327
+oid sha256:6c0c125986b007d2e3734bec84c18bda0152071b90b87fadac6d7764499927a0
+size 30727765

onnx/decoder_model_quantized.onnx

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6ae32f7a6c43d40740199b20c1720539f3d6dcd73d1bc0ff0c9d22f4386bdc37
-size 50348480
+oid sha256:d1ac41063dd0a01d278dca49214ff8ce1b930cc9e34e2881c56a79a428650534
+size 30460101

onnx/decoder_with_past_model.onnx

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3df7e2682d1b77dc3706a9c444335ccf4e441fab2e64562d2e4934f70ecfaba2
-size 193295315
+oid sha256:25efcc2a9319b897bc0c3b6b2cd1f8ab8a25ebff6da4be828e4766f4a597d5f6
+size 113654551

onnx/decoder_with_past_model_quantized.onnx

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ac92773054aca908fca90d7e84a8c97a995526e9fe6292cf00c6e77ff7d223dd
-size 49114568
+oid sha256:26d8d7fa7829644b22762c965351893c3fb992501ed1776d54f126530eafabf8
+size 29229119

onnx/encoder_model.onnx

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4164766243b81a4bba7fc51bb0c37539d15b11e2d39820d15f65bdea39942feb
-size 32900723
+oid sha256:39e81b6c86a5b2b4beda1bb3145486a769d594801f780a66cad1ae72c7ad2c5e
+size 32909539

onnx/encoder_model_quantized.onnx

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ca9d7bb2836193704b7e2435e3bbadbed985ac3a79ab7406b244b8865ab1a5c0
-size 10113248
+oid sha256:fd9d995b9dcb0520f0dbf6cf68651af639fc385f594d9d876e69ca2802dc438e
+size 10124910

quantize_config.json

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