Spaces:
Running
Running
Fix sleep bug
Browse files
app.py
CHANGED
@@ -166,6 +166,7 @@ def grade(file_obj, progress=gr.Progress()):
|
|
166 |
grade_sample_run_complete = False
|
167 |
temperature = 0.0
|
168 |
|
|
|
169 |
while not grade_sample_run_complete:
|
170 |
try:
|
171 |
response = openai.ChatCompletion.create(
|
@@ -206,8 +207,15 @@ def grade(file_obj, progress=gr.Progress()):
|
|
206 |
grade_sample_run_complete = True
|
207 |
except:
|
208 |
# gpt4 may have token rate limit
|
|
|
|
|
|
|
|
|
|
|
|
|
209 |
print("sleep 30s")
|
210 |
time.sleep(30)
|
|
|
211 |
|
212 |
if len(sample_grade['model']) >= j + 1:
|
213 |
sample_grade['model'][j] = response['model']
|
@@ -298,7 +306,7 @@ markdown = """
|
|
298 |
|
299 |
In this demo, we offer MM-Vet LLM-based (GPT-4) evaluator to grade open-ended outputs from your models.
|
300 |
|
301 |
-
Plese upload your json file of your model results containing
|
302 |
|
303 |
The grading may last 5 minutes. Sine we only support 1 queue, the grading time may be longer when you need to wait for other users' grading to finish.
|
304 |
|
|
|
166 |
grade_sample_run_complete = False
|
167 |
temperature = 0.0
|
168 |
|
169 |
+
num_sleep = 0
|
170 |
while not grade_sample_run_complete:
|
171 |
try:
|
172 |
response = openai.ChatCompletion.create(
|
|
|
207 |
grade_sample_run_complete = True
|
208 |
except:
|
209 |
# gpt4 may have token rate limit
|
210 |
+
num_sleep += 1
|
211 |
+
if num_sleep > 2:
|
212 |
+
score = 0.0
|
213 |
+
grade_sample_run_complete = True
|
214 |
+
num_sleep = 0
|
215 |
+
continue
|
216 |
print("sleep 30s")
|
217 |
time.sleep(30)
|
218 |
+
|
219 |
|
220 |
if len(sample_grade['model']) >= j + 1:
|
221 |
sample_grade['model'][j] = response['model']
|
|
|
306 |
|
307 |
In this demo, we offer MM-Vet LLM-based (GPT-4) evaluator to grade open-ended outputs from your models.
|
308 |
|
309 |
+
Plese upload your json file of your model results containing `{v1_0: ..., v1_1: ..., }`like [this json file](https://raw.githubusercontent.com/yuweihao/MM-Vet/main/results/llava_llama2_13b_chat.json).
|
310 |
|
311 |
The grading may last 5 minutes. Sine we only support 1 queue, the grading time may be longer when you need to wait for other users' grading to finish.
|
312 |
|