Spaces:
Running
on
Zero
Running
on
Zero
Commit
·
0b155f0
1
Parent(s):
0adad70
fix again
Browse files
app.py
CHANGED
|
@@ -76,11 +76,10 @@ def infer(code):
|
|
| 76 |
# ejs: Yeah, this var_name thing is really bizarre. But look at https://github.com/lt-asset/resym/blob/main/training_src/fielddecoder_inf.py
|
| 77 |
prompt = f"What are the original name and data types of variables {varstring}?\n```{code}\n```{var_name}"
|
| 78 |
|
| 79 |
-
prompt
|
| 80 |
-
print(prompt)
|
| 81 |
|
| 82 |
input_ids = tokenizer.encode(prompt, return_tensors="pt").cuda()[:, : 8192 - 1024]
|
| 83 |
-
|
| 84 |
input_ids=input_ids,
|
| 85 |
max_new_tokens=1024,
|
| 86 |
num_beams=4,
|
|
@@ -90,14 +89,30 @@ def infer(code):
|
|
| 90 |
pad_token_id=0,
|
| 91 |
eos_token_id=0,
|
| 92 |
)[0]
|
| 93 |
-
|
| 94 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 95 |
skip_special_tokens=True,
|
| 96 |
clean_up_tokenization_spaces=True,
|
| 97 |
)
|
| 98 |
|
| 99 |
-
|
| 100 |
-
|
|
|
|
| 101 |
|
| 102 |
|
| 103 |
demo = gr.Interface(
|
|
@@ -106,6 +121,7 @@ demo = gr.Interface(
|
|
| 106 |
gr.Textbox(lines=10, value=example),
|
| 107 |
],
|
| 108 |
outputs=[gr.Text(label="Var Decoder Output"),
|
|
|
|
| 109 |
gr.Text(label="Generated Variable List")],
|
| 110 |
description=description
|
| 111 |
)
|
|
|
|
| 76 |
# ejs: Yeah, this var_name thing is really bizarre. But look at https://github.com/lt-asset/resym/blob/main/training_src/fielddecoder_inf.py
|
| 77 |
prompt = f"What are the original name and data types of variables {varstring}?\n```{code}\n```{var_name}"
|
| 78 |
|
| 79 |
+
print(f"Prompt:\n{prompt}")
|
|
|
|
| 80 |
|
| 81 |
input_ids = tokenizer.encode(prompt, return_tensors="pt").cuda()[:, : 8192 - 1024]
|
| 82 |
+
var_output = vardecoder_model.generate(
|
| 83 |
input_ids=input_ids,
|
| 84 |
max_new_tokens=1024,
|
| 85 |
num_beams=4,
|
|
|
|
| 89 |
pad_token_id=0,
|
| 90 |
eos_token_id=0,
|
| 91 |
)[0]
|
| 92 |
+
var_output = tokenizer.decode(
|
| 93 |
+
var_output[input_ids.size(1) :],
|
| 94 |
+
skip_special_tokens=True,
|
| 95 |
+
clean_up_tokenization_spaces=True,
|
| 96 |
+
)
|
| 97 |
+
field_output = fielddecoder_model.generate(
|
| 98 |
+
input_ids=input_ids,
|
| 99 |
+
max_new_tokens=1024,
|
| 100 |
+
num_beams=4,
|
| 101 |
+
num_return_sequences=1,
|
| 102 |
+
do_sample=False,
|
| 103 |
+
early_stopping=False,
|
| 104 |
+
pad_token_id=0,
|
| 105 |
+
eos_token_id=0,
|
| 106 |
+
)[0]
|
| 107 |
+
field_output = tokenizer.decode(
|
| 108 |
+
field_output[input_ids.size(1) :],
|
| 109 |
skip_special_tokens=True,
|
| 110 |
clean_up_tokenization_spaces=True,
|
| 111 |
)
|
| 112 |
|
| 113 |
+
var_output = var_name + ":" + var_output
|
| 114 |
+
field_output = var_name + ":" + field_output
|
| 115 |
+
return var_output, field_output, varstring
|
| 116 |
|
| 117 |
|
| 118 |
demo = gr.Interface(
|
|
|
|
| 121 |
gr.Textbox(lines=10, value=example),
|
| 122 |
],
|
| 123 |
outputs=[gr.Text(label="Var Decoder Output"),
|
| 124 |
+
gr.Text(label="Field Decoder Output"),
|
| 125 |
gr.Text(label="Generated Variable List")],
|
| 126 |
description=description
|
| 127 |
)
|