Spaces:
Running
on
Zero
Running
on
Zero
Commit
·
d5e44b4
1
Parent(s):
803e1b0
Name constants
Browse files
app.py
CHANGED
|
@@ -11,6 +11,10 @@ import huggingface_hub
|
|
| 11 |
|
| 12 |
import prep_decompiled
|
| 13 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 14 |
hf_key = os.environ["HF_TOKEN"]
|
| 15 |
huggingface_hub.login(token=hf_key)
|
| 16 |
|
|
@@ -85,11 +89,11 @@ def infer(code):
|
|
| 85 |
print(f"Prompt:\n{repr(var_prompt)}")
|
| 86 |
|
| 87 |
var_input_ids = tokenizer.encode(var_prompt, return_tensors="pt").cuda()[
|
| 88 |
-
:, :
|
| 89 |
]
|
| 90 |
var_output = vardecoder_model.generate(
|
| 91 |
input_ids=var_input_ids,
|
| 92 |
-
max_new_tokens=
|
| 93 |
num_beams=4,
|
| 94 |
num_return_sequences=1,
|
| 95 |
do_sample=False,
|
|
@@ -112,12 +116,12 @@ def infer(code):
|
|
| 112 |
field_output = "Failed to parse fields" if field_prompt_result is None else "No fields"
|
| 113 |
else:
|
| 114 |
field_input_ids = tokenizer.encode(field_prompt_result, return_tensors="pt").cuda()[
|
| 115 |
-
:, :
|
| 116 |
]
|
| 117 |
|
| 118 |
field_output = fielddecoder_model.generate(
|
| 119 |
input_ids=field_input_ids,
|
| 120 |
-
max_new_tokens=
|
| 121 |
num_beams=4,
|
| 122 |
num_return_sequences=1,
|
| 123 |
do_sample=False,
|
|
|
|
| 11 |
|
| 12 |
import prep_decompiled
|
| 13 |
|
| 14 |
+
# Model configuration constants
|
| 15 |
+
MAX_CONTEXT_LENGTH = 8192
|
| 16 |
+
MAX_NEW_TOKENS = 1024
|
| 17 |
+
|
| 18 |
hf_key = os.environ["HF_TOKEN"]
|
| 19 |
huggingface_hub.login(token=hf_key)
|
| 20 |
|
|
|
|
| 89 |
print(f"Prompt:\n{repr(var_prompt)}")
|
| 90 |
|
| 91 |
var_input_ids = tokenizer.encode(var_prompt, return_tensors="pt").cuda()[
|
| 92 |
+
:, : MAX_CONTEXT_LENGTH - MAX_NEW_TOKENS
|
| 93 |
]
|
| 94 |
var_output = vardecoder_model.generate(
|
| 95 |
input_ids=var_input_ids,
|
| 96 |
+
max_new_tokens=MAX_NEW_TOKENS,
|
| 97 |
num_beams=4,
|
| 98 |
num_return_sequences=1,
|
| 99 |
do_sample=False,
|
|
|
|
| 116 |
field_output = "Failed to parse fields" if field_prompt_result is None else "No fields"
|
| 117 |
else:
|
| 118 |
field_input_ids = tokenizer.encode(field_prompt_result, return_tensors="pt").cuda()[
|
| 119 |
+
:, : MAX_CONTEXT_LENGTH - MAX_NEW_TOKENS
|
| 120 |
]
|
| 121 |
|
| 122 |
field_output = fielddecoder_model.generate(
|
| 123 |
input_ids=field_input_ids,
|
| 124 |
+
max_new_tokens=MAX_NEW_TOKENS,
|
| 125 |
num_beams=4,
|
| 126 |
num_return_sequences=1,
|
| 127 |
do_sample=False,
|