Spaces:
Running
on
CPU Upgrade
Running
on
CPU Upgrade
check-submit
#920
by
alozowski
HF staff
- opened
- src/submission/submit.py +14 -11
src/submission/submit.py
CHANGED
@@ -66,8 +66,10 @@ def add_new_eval(
|
|
66 |
weight_type: str,
|
67 |
model_type: str,
|
68 |
use_chat_template: bool,
|
69 |
-
profile: gr.OAuthProfile | None
|
70 |
-
|
|
|
|
|
71 |
# Login required
|
72 |
if profile is None:
|
73 |
return styled_error("Hub Login Required")
|
@@ -75,11 +77,10 @@ def add_new_eval(
|
|
75 |
# Name of the actual user who sent the request
|
76 |
username = profile.username
|
77 |
|
78 |
-
|
79 |
-
|
80 |
-
if
|
81 |
-
|
82 |
-
|
83 |
|
84 |
org_or_user = ""
|
85 |
model_path = model
|
@@ -96,7 +97,7 @@ def add_new_eval(
|
|
96 |
# Is the user rate limited?
|
97 |
if org_or_user != "":
|
98 |
user_can_submit, error_msg = user_submission_permission(
|
99 |
-
org_or_user,
|
100 |
)
|
101 |
if not user_can_submit:
|
102 |
return styled_error(error_msg)
|
@@ -112,15 +113,18 @@ def add_new_eval(
|
|
112 |
model_info = API.model_info(repo_id=model, revision=revision)
|
113 |
except Exception as e:
|
114 |
return styled_error("Could not get your model information. Please fill it up properly.")
|
|
|
|
|
|
|
|
|
115 |
|
116 |
# Check model size early
|
117 |
model_size = get_model_size(model_info=model_info, precision=precision)
|
118 |
-
|
119 |
# First check: Absolute size limit for float16 and bfloat16
|
120 |
if precision in ["float16", "bfloat16"] and model_size > 100:
|
121 |
return styled_error(f"Sadly, models larger than 100B parameters cannot be submitted in {precision} precision at this time. "
|
122 |
f"Your model size: {model_size:.2f}B parameters.")
|
123 |
-
|
124 |
# Second check: Precision-adjusted size limit for 8bit, 4bit, and GPTQ
|
125 |
if precision in ["8bit", "4bit", "GPTQ"]:
|
126 |
size_checker = ModelSizeChecker(model=model, precision=precision, model_size_in_b=model_size)
|
@@ -147,7 +151,6 @@ def add_new_eval(
|
|
147 |
architectures = getattr(model_config, "architectures", None)
|
148 |
if architectures:
|
149 |
architecture = ";".join(architectures)
|
150 |
-
|
151 |
# Were the model card and license filled?
|
152 |
try:
|
153 |
model_info.cardData["license"]
|
|
|
66 |
weight_type: str,
|
67 |
model_type: str,
|
68 |
use_chat_template: bool,
|
69 |
+
profile: gr.OAuthProfile | None,
|
70 |
+
requested_models: set[str] = None,
|
71 |
+
users_to_submission_dates: dict[str, list[str]] = None,
|
72 |
+
):
|
73 |
# Login required
|
74 |
if profile is None:
|
75 |
return styled_error("Hub Login Required")
|
|
|
77 |
# Name of the actual user who sent the request
|
78 |
username = profile.username
|
79 |
|
80 |
+
# Initialize the requested_models and users_to_submission_dates variables
|
81 |
+
# If the caller did not provide these values, fetch them from the EVAL_REQUESTS_PATH
|
82 |
+
if requested_models is None or users_to_submission_dates is None:
|
83 |
+
requested_models, users_to_submission_dates = already_submitted_models(EVAL_REQUESTS_PATH)
|
|
|
84 |
|
85 |
org_or_user = ""
|
86 |
model_path = model
|
|
|
97 |
# Is the user rate limited?
|
98 |
if org_or_user != "":
|
99 |
user_can_submit, error_msg = user_submission_permission(
|
100 |
+
org_or_user, users_to_submission_dates, RATE_LIMIT_PERIOD, RATE_LIMIT_QUOTA
|
101 |
)
|
102 |
if not user_can_submit:
|
103 |
return styled_error(error_msg)
|
|
|
113 |
model_info = API.model_info(repo_id=model, revision=revision)
|
114 |
except Exception as e:
|
115 |
return styled_error("Could not get your model information. Please fill it up properly.")
|
116 |
+
|
117 |
+
model_key = f"{model}_{model_info.sha}_{precision}"
|
118 |
+
if model_key in requested_models:
|
119 |
+
return styled_error(f"The model '{model}' with revision '{model_info.sha}' and precision '{precision}' has already been submitted.")
|
120 |
|
121 |
# Check model size early
|
122 |
model_size = get_model_size(model_info=model_info, precision=precision)
|
|
|
123 |
# First check: Absolute size limit for float16 and bfloat16
|
124 |
if precision in ["float16", "bfloat16"] and model_size > 100:
|
125 |
return styled_error(f"Sadly, models larger than 100B parameters cannot be submitted in {precision} precision at this time. "
|
126 |
f"Your model size: {model_size:.2f}B parameters.")
|
127 |
+
|
128 |
# Second check: Precision-adjusted size limit for 8bit, 4bit, and GPTQ
|
129 |
if precision in ["8bit", "4bit", "GPTQ"]:
|
130 |
size_checker = ModelSizeChecker(model=model, precision=precision, model_size_in_b=model_size)
|
|
|
151 |
architectures = getattr(model_config, "architectures", None)
|
152 |
if architectures:
|
153 |
architecture = ";".join(architectures)
|
|
|
154 |
# Were the model card and license filled?
|
155 |
try:
|
156 |
model_info.cardData["license"]
|