Commit
β’
6e11c96
1
Parent(s):
cf9b005
update
Browse files
.DS_Store
ADDED
Binary file (6.15 kB). View file
|
|
generation_config.json
DELETED
@@ -1,7 +0,0 @@
|
|
1 |
-
{
|
2 |
-
"_from_model_config": true,
|
3 |
-
"decoder_start_token_id": 0,
|
4 |
-
"eos_token_id": 1,
|
5 |
-
"pad_token_id": 0,
|
6 |
-
"transformers_version": "4.27.0.dev0"
|
7 |
-
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
onnx/decoder_model_merged_quantized.onnx
DELETED
@@ -1,3 +0,0 @@
|
|
1 |
-
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:fe9751468ddf78017d5bbb317d1162049b57bc8b11321dcf3756907835245e42
|
3 |
-
size 20201510
|
|
|
|
|
|
|
|
onnx/{encoder_model_quantized.onnx β encoder_model_quant.onnx}
RENAMED
File without changes
|
onnx/{init_decoder_quantized.onnx β init_decoder_quant.onnx}
RENAMED
File without changes
|
scripts/gen.py
ADDED
@@ -0,0 +1,51 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from transformers import T5Tokenizer, T5ForConditionalGeneration, T5Config
|
2 |
+
import torch
|
3 |
+
from transformers.onnx import OnnxConfig, export
|
4 |
+
from pathlib import Path
|
5 |
+
|
6 |
+
# Load the T5-efficient-tiny model and tokenizer
|
7 |
+
model_name = "google/t5-efficient-tiny"
|
8 |
+
model = T5ForConditionalGeneration.from_pretrained(model_name)
|
9 |
+
tokenizer = T5Tokenizer.from_pretrained(model_name)
|
10 |
+
config = T5Config.from_pretrained(model_name)
|
11 |
+
|
12 |
+
# Prepare a sample input
|
13 |
+
text = "Translate English to French: The house is wonderful."
|
14 |
+
inputs = tokenizer(text, return_tensors="pt")
|
15 |
+
|
16 |
+
# Define the model configuration for ONNX
|
17 |
+
class T5OnnxConfig(OnnxConfig):
|
18 |
+
@property
|
19 |
+
def inputs(self):
|
20 |
+
return {
|
21 |
+
"input_ids": {
|
22 |
+
"shape": [self.batch_size, self.sequence_length],
|
23 |
+
"dtype": torch.int64,
|
24 |
+
},
|
25 |
+
"attention_mask": {
|
26 |
+
"shape": [self.batch_size, self.sequence_length],
|
27 |
+
"dtype": torch.int64,
|
28 |
+
},
|
29 |
+
}
|
30 |
+
|
31 |
+
@property
|
32 |
+
def outputs(self):
|
33 |
+
return {
|
34 |
+
"logits": {
|
35 |
+
"shape": [self.batch_size, self.sequence_length, self.config.vocab_size],
|
36 |
+
"dtype": torch.float32,
|
37 |
+
},
|
38 |
+
}
|
39 |
+
|
40 |
+
onnx_config = T5OnnxConfig(config, 1, 128)
|
41 |
+
|
42 |
+
# Export the model to ONNX format
|
43 |
+
output_path = Path("t5-efficient-tiny.onnx")
|
44 |
+
export(
|
45 |
+
preprocessor=tokenizer,
|
46 |
+
model=model,
|
47 |
+
config=onnx_config,
|
48 |
+
output=output_path
|
49 |
+
)
|
50 |
+
|
51 |
+
print("Model has been successfully exported to ONNX format.")
|
quantifiy.py β scripts/quantifiy.py
RENAMED
File without changes
|