Second attempt at compatible models
Browse files- config.json +1 -1
- decoder_model_merged.onnx +1 -1
- decoder_model_merged_quantized.onnx +1 -1
- generation_config.json +1 -1
- quantize_config.json +37 -37
config.json
CHANGED
|
@@ -43,7 +43,7 @@
|
|
| 43 |
"summary_proj_to_labels": true,
|
| 44 |
"summary_type": "cls_index",
|
| 45 |
"summary_use_proj": true,
|
| 46 |
-
"transformers_version": "4.
|
| 47 |
"use_cache": true,
|
| 48 |
"vocab_size": 50257,
|
| 49 |
"window_size": 256
|
|
|
|
| 43 |
"summary_proj_to_labels": true,
|
| 44 |
"summary_type": "cls_index",
|
| 45 |
"summary_use_proj": true,
|
| 46 |
+
"transformers_version": "4.32.0.dev0",
|
| 47 |
"use_cache": true,
|
| 48 |
"vocab_size": 50257,
|
| 49 |
"window_size": 256
|
decoder_model_merged.onnx
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 45031368
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:40d4f160d12634d87968bd8b31d162a561a3a333ad86306b1255292709bbf239
|
| 3 |
size 45031368
|
decoder_model_merged_quantized.onnx
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 24681847
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:cdacdf8b589e5654f95b352770e9795a791c2c17832e9a48a0e3b6823bbcd4d9
|
| 3 |
size 24681847
|
generation_config.json
CHANGED
|
@@ -2,5 +2,5 @@
|
|
| 2 |
"_from_model_config": true,
|
| 3 |
"bos_token_id": 50256,
|
| 4 |
"eos_token_id": 50256,
|
| 5 |
-
"transformers_version": "4.
|
| 6 |
}
|
|
|
|
| 2 |
"_from_model_config": true,
|
| 3 |
"bos_token_id": 50256,
|
| 4 |
"eos_token_id": 50256,
|
| 5 |
+
"transformers_version": "4.32.0.dev0"
|
| 6 |
}
|
quantize_config.json
CHANGED
|
@@ -5,82 +5,82 @@
|
|
| 5 |
"decoder_with_past_model": {
|
| 6 |
"op_types": [
|
| 7 |
"MatMul",
|
|
|
|
| 8 |
"Sub",
|
| 9 |
-
"
|
| 10 |
-
"
|
| 11 |
-
"Shape",
|
| 12 |
"Concat",
|
|
|
|
| 13 |
"Range",
|
| 14 |
-
"Unsqueeze",
|
| 15 |
"Mul",
|
| 16 |
-
"
|
| 17 |
"Transpose",
|
| 18 |
-
"Pow",
|
| 19 |
"Reshape",
|
| 20 |
-
"Constant",
|
| 21 |
-
"Div",
|
| 22 |
-
"Cast",
|
| 23 |
-
"Softmax",
|
| 24 |
"Slice",
|
|
|
|
| 25 |
"Add",
|
|
|
|
| 26 |
"Sqrt",
|
| 27 |
-
"
|
| 28 |
-
"
|
|
|
|
|
|
|
|
|
|
| 29 |
],
|
| 30 |
"weight_type": "QInt8"
|
| 31 |
},
|
| 32 |
"decoder_model_merged": {
|
| 33 |
"op_types": [
|
| 34 |
"MatMul",
|
|
|
|
| 35 |
"Sub",
|
| 36 |
-
"
|
| 37 |
-
"
|
| 38 |
-
"Shape",
|
| 39 |
"Concat",
|
|
|
|
| 40 |
"Range",
|
| 41 |
-
"Unsqueeze",
|
| 42 |
"Mul",
|
| 43 |
-
"
|
| 44 |
"Transpose",
|
| 45 |
-
"Pow",
|
| 46 |
"Reshape",
|
| 47 |
-
"Constant",
|
| 48 |
-
"Div",
|
| 49 |
-
"If",
|
| 50 |
-
"Cast",
|
| 51 |
-
"Softmax",
|
| 52 |
"Slice",
|
|
|
|
| 53 |
"Add",
|
|
|
|
| 54 |
"Sqrt",
|
| 55 |
-
"
|
| 56 |
-
"
|
|
|
|
|
|
|
|
|
|
|
|
|
| 57 |
],
|
| 58 |
"weight_type": "QInt8"
|
| 59 |
},
|
| 60 |
"decoder_model": {
|
| 61 |
"op_types": [
|
| 62 |
"MatMul",
|
|
|
|
| 63 |
"Sub",
|
| 64 |
-
"
|
| 65 |
-
"
|
| 66 |
-
"Shape",
|
| 67 |
"Concat",
|
|
|
|
| 68 |
"Range",
|
| 69 |
-
"Unsqueeze",
|
| 70 |
"Mul",
|
| 71 |
-
"
|
| 72 |
"Transpose",
|
| 73 |
-
"Pow",
|
| 74 |
"Reshape",
|
| 75 |
-
"Constant",
|
| 76 |
-
"Div",
|
| 77 |
-
"Cast",
|
| 78 |
-
"Softmax",
|
| 79 |
"Slice",
|
|
|
|
| 80 |
"Add",
|
|
|
|
| 81 |
"Sqrt",
|
| 82 |
-
"
|
| 83 |
-
"
|
|
|
|
|
|
|
|
|
|
| 84 |
],
|
| 85 |
"weight_type": "QInt8"
|
| 86 |
}
|
|
|
|
| 5 |
"decoder_with_past_model": {
|
| 6 |
"op_types": [
|
| 7 |
"MatMul",
|
| 8 |
+
"Pow",
|
| 9 |
"Sub",
|
| 10 |
+
"Unsqueeze",
|
| 11 |
+
"Constant",
|
|
|
|
| 12 |
"Concat",
|
| 13 |
+
"Where",
|
| 14 |
"Range",
|
|
|
|
| 15 |
"Mul",
|
| 16 |
+
"ReduceMean",
|
| 17 |
"Transpose",
|
|
|
|
| 18 |
"Reshape",
|
|
|
|
|
|
|
|
|
|
|
|
|
| 19 |
"Slice",
|
| 20 |
+
"Shape",
|
| 21 |
"Add",
|
| 22 |
+
"Tanh",
|
| 23 |
"Sqrt",
|
| 24 |
+
"Cast",
|
| 25 |
+
"Softmax",
|
| 26 |
+
"Squeeze",
|
| 27 |
+
"Gather",
|
| 28 |
+
"Div"
|
| 29 |
],
|
| 30 |
"weight_type": "QInt8"
|
| 31 |
},
|
| 32 |
"decoder_model_merged": {
|
| 33 |
"op_types": [
|
| 34 |
"MatMul",
|
| 35 |
+
"Pow",
|
| 36 |
"Sub",
|
| 37 |
+
"Unsqueeze",
|
| 38 |
+
"Constant",
|
|
|
|
| 39 |
"Concat",
|
| 40 |
+
"Where",
|
| 41 |
"Range",
|
|
|
|
| 42 |
"Mul",
|
| 43 |
+
"ReduceMean",
|
| 44 |
"Transpose",
|
|
|
|
| 45 |
"Reshape",
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 46 |
"Slice",
|
| 47 |
+
"Shape",
|
| 48 |
"Add",
|
| 49 |
+
"Tanh",
|
| 50 |
"Sqrt",
|
| 51 |
+
"Cast",
|
| 52 |
+
"Softmax",
|
| 53 |
+
"If",
|
| 54 |
+
"Squeeze",
|
| 55 |
+
"Gather",
|
| 56 |
+
"Div"
|
| 57 |
],
|
| 58 |
"weight_type": "QInt8"
|
| 59 |
},
|
| 60 |
"decoder_model": {
|
| 61 |
"op_types": [
|
| 62 |
"MatMul",
|
| 63 |
+
"Pow",
|
| 64 |
"Sub",
|
| 65 |
+
"Unsqueeze",
|
| 66 |
+
"Constant",
|
|
|
|
| 67 |
"Concat",
|
| 68 |
+
"Where",
|
| 69 |
"Range",
|
|
|
|
| 70 |
"Mul",
|
| 71 |
+
"ReduceMean",
|
| 72 |
"Transpose",
|
|
|
|
| 73 |
"Reshape",
|
|
|
|
|
|
|
|
|
|
|
|
|
| 74 |
"Slice",
|
| 75 |
+
"Shape",
|
| 76 |
"Add",
|
| 77 |
+
"Tanh",
|
| 78 |
"Sqrt",
|
| 79 |
+
"Cast",
|
| 80 |
+
"Softmax",
|
| 81 |
+
"Squeeze",
|
| 82 |
+
"Gather",
|
| 83 |
+
"Div"
|
| 84 |
],
|
| 85 |
"weight_type": "QInt8"
|
| 86 |
}
|