Spaces:
Runtime error
Runtime error
Andrea Seveso
commited on
Commit
·
ff797f9
1
Parent(s):
8d2a6ac
Remove precision
Browse files- app.py +0 -12
- src/display/utils.py +0 -16
- src/leaderboard/read_evals.py +8 -14
- src/submission/check_validity.py +3 -4
- src/submission/submit.py +3 -6
app.py
CHANGED
|
@@ -21,7 +21,6 @@ from src.display.utils import (
|
|
| 21 |
AutoEvalColumn,
|
| 22 |
ModelType,
|
| 23 |
fields,
|
| 24 |
-
Precision
|
| 25 |
)
|
| 26 |
from src.envs import API, EVAL_REQUESTS_PATH, EVAL_RESULTS_PATH, QUEUE_REPO, REPO_ID, RESULTS_REPO, TOKEN
|
| 27 |
from src.populate import get_evaluation_queue_df, get_leaderboard_df
|
|
@@ -77,8 +76,6 @@ def init_leaderboard(dataframe):
|
|
| 77 |
filter_columns=[
|
| 78 |
ColumnFilter(AutoEvalColumn.model_type.name,
|
| 79 |
type="checkboxgroup", label="Model types"),
|
| 80 |
-
ColumnFilter(AutoEvalColumn.precision.name,
|
| 81 |
-
type="checkboxgroup", label="Precision"),
|
| 82 |
ColumnFilter(
|
| 83 |
AutoEvalColumn.params.name,
|
| 84 |
type="slider",
|
|
@@ -167,14 +164,6 @@ with demo:
|
|
| 167 |
)
|
| 168 |
|
| 169 |
with gr.Column():
|
| 170 |
-
precision = gr.Dropdown(
|
| 171 |
-
choices=[i.value.name for i in Precision if i !=
|
| 172 |
-
Precision.Unknown],
|
| 173 |
-
label="Precision",
|
| 174 |
-
multiselect=False,
|
| 175 |
-
value="float16",
|
| 176 |
-
interactive=True,
|
| 177 |
-
)
|
| 178 |
base_model_name_textbox = gr.Textbox(
|
| 179 |
label="Base model (for delta or adapter weights)")
|
| 180 |
|
|
@@ -186,7 +175,6 @@ with demo:
|
|
| 186 |
model_name_textbox,
|
| 187 |
base_model_name_textbox,
|
| 188 |
revision_name_textbox,
|
| 189 |
-
precision,
|
| 190 |
model_type,
|
| 191 |
],
|
| 192 |
submission_result,
|
|
|
|
| 21 |
AutoEvalColumn,
|
| 22 |
ModelType,
|
| 23 |
fields,
|
|
|
|
| 24 |
)
|
| 25 |
from src.envs import API, EVAL_REQUESTS_PATH, EVAL_RESULTS_PATH, QUEUE_REPO, REPO_ID, RESULTS_REPO, TOKEN
|
| 26 |
from src.populate import get_evaluation_queue_df, get_leaderboard_df
|
|
|
|
| 76 |
filter_columns=[
|
| 77 |
ColumnFilter(AutoEvalColumn.model_type.name,
|
| 78 |
type="checkboxgroup", label="Model types"),
|
|
|
|
|
|
|
| 79 |
ColumnFilter(
|
| 80 |
AutoEvalColumn.params.name,
|
| 81 |
type="slider",
|
|
|
|
| 164 |
)
|
| 165 |
|
| 166 |
with gr.Column():
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 167 |
base_model_name_textbox = gr.Textbox(
|
| 168 |
label="Base model (for delta or adapter weights)")
|
| 169 |
|
|
|
|
| 175 |
model_name_textbox,
|
| 176 |
base_model_name_textbox,
|
| 177 |
revision_name_textbox,
|
|
|
|
| 178 |
model_type,
|
| 179 |
],
|
| 180 |
submission_result,
|
src/display/utils.py
CHANGED
|
@@ -37,8 +37,6 @@ for task in Tasks:
|
|
| 37 |
# Model information
|
| 38 |
auto_eval_column_dict.append(
|
| 39 |
["model_type", ColumnContent, ColumnContent("Type", "str", False)])
|
| 40 |
-
auto_eval_column_dict.append(
|
| 41 |
-
["precision", ColumnContent, ColumnContent("Precision", "str", False)])
|
| 42 |
auto_eval_column_dict.append(
|
| 43 |
["params", ColumnContent, ColumnContent("#Params (B)", "number", False)])
|
| 44 |
auto_eval_column_dict.append(
|
|
@@ -56,7 +54,6 @@ class EvalQueueColumn: # Queue column
|
|
| 56 |
model = ColumnContent("model", "markdown", True)
|
| 57 |
revision = ColumnContent("revision", "str", True)
|
| 58 |
private = ColumnContent("private", "bool", True)
|
| 59 |
-
precision = ColumnContent("precision", "str", True)
|
| 60 |
status = ColumnContent("status", "str", True)
|
| 61 |
|
| 62 |
# All the model information that we might need
|
|
@@ -86,19 +83,6 @@ class ModelType(Enum):
|
|
| 86 |
return ModelType.Unknown
|
| 87 |
|
| 88 |
|
| 89 |
-
class Precision(Enum):
|
| 90 |
-
float16 = ModelDetails("float16")
|
| 91 |
-
bfloat16 = ModelDetails("bfloat16")
|
| 92 |
-
Unknown = ModelDetails("?")
|
| 93 |
-
|
| 94 |
-
def from_str(precision):
|
| 95 |
-
if precision in ["torch.float16", "float16"]:
|
| 96 |
-
return Precision.float16
|
| 97 |
-
if precision in ["torch.bfloat16", "bfloat16"]:
|
| 98 |
-
return Precision.bfloat16
|
| 99 |
-
return Precision.Unknown
|
| 100 |
-
|
| 101 |
-
|
| 102 |
# Column selection
|
| 103 |
COLS = [c.name for c in fields(AutoEvalColumn) if not c.hidden]
|
| 104 |
|
|
|
|
| 37 |
# Model information
|
| 38 |
auto_eval_column_dict.append(
|
| 39 |
["model_type", ColumnContent, ColumnContent("Type", "str", False)])
|
|
|
|
|
|
|
| 40 |
auto_eval_column_dict.append(
|
| 41 |
["params", ColumnContent, ColumnContent("#Params (B)", "number", False)])
|
| 42 |
auto_eval_column_dict.append(
|
|
|
|
| 54 |
model = ColumnContent("model", "markdown", True)
|
| 55 |
revision = ColumnContent("revision", "str", True)
|
| 56 |
private = ColumnContent("private", "bool", True)
|
|
|
|
| 57 |
status = ColumnContent("status", "str", True)
|
| 58 |
|
| 59 |
# All the model information that we might need
|
|
|
|
| 83 |
return ModelType.Unknown
|
| 84 |
|
| 85 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 86 |
# Column selection
|
| 87 |
COLS = [c.name for c in fields(AutoEvalColumn) if not c.hidden]
|
| 88 |
|
src/leaderboard/read_evals.py
CHANGED
|
@@ -1,5 +1,5 @@
|
|
| 1 |
from src.submission.check_validity import is_model_on_hub
|
| 2 |
-
from src.display.utils import AutoEvalColumn, ModelType, Tasks
|
| 3 |
from src.display.formatting import make_clickable_model
|
| 4 |
import numpy as np
|
| 5 |
import dateutil
|
|
@@ -15,13 +15,12 @@ print("--- CONFIRMED: Running the modified version of read_evals.py ---")
|
|
| 15 |
class EvalResult:
|
| 16 |
"""Represents one full evaluation. Built from a combination of the result and request file for a given run.
|
| 17 |
"""
|
| 18 |
-
eval_name: str #
|
| 19 |
full_model: str # org/model (path on hub)
|
| 20 |
org: str
|
| 21 |
model: str
|
| 22 |
revision: str # commit hash, "" if main
|
| 23 |
results: dict
|
| 24 |
-
precision: Precision = Precision.Unknown
|
| 25 |
model_type: ModelType = ModelType.Unknown # Pretrained, fine tuned, ...
|
| 26 |
architecture: str = "Unknown"
|
| 27 |
likes: int = 0
|
|
@@ -37,9 +36,6 @@ class EvalResult:
|
|
| 37 |
|
| 38 |
config = data.get("config")
|
| 39 |
|
| 40 |
-
# Precision
|
| 41 |
-
precision = Precision.from_str(config.get("model_dtype"))
|
| 42 |
-
|
| 43 |
# Get model and org
|
| 44 |
org_and_model = config.get(
|
| 45 |
"model_name", config.get("model_args", None))
|
|
@@ -48,11 +44,11 @@ class EvalResult:
|
|
| 48 |
if len(org_and_model) == 1:
|
| 49 |
org = None
|
| 50 |
model = org_and_model[0]
|
| 51 |
-
result_key = f"{model}
|
| 52 |
else:
|
| 53 |
org = org_and_model[0]
|
| 54 |
model = org_and_model[1]
|
| 55 |
-
result_key = f"{org}
|
| 56 |
full_model = "/".join(org_and_model)
|
| 57 |
|
| 58 |
still_on_hub, _, model_config = is_model_on_hub(
|
|
@@ -79,7 +75,6 @@ class EvalResult:
|
|
| 79 |
org=org,
|
| 80 |
model=model,
|
| 81 |
results=results,
|
| 82 |
-
precision=precision,
|
| 83 |
revision=config.get("model_sha", ""),
|
| 84 |
still_on_hub=still_on_hub,
|
| 85 |
)
|
|
@@ -87,7 +82,7 @@ class EvalResult:
|
|
| 87 |
def update_with_request_file(self, requests_path):
|
| 88 |
"""Finds the relevant request file for the current model and updates info with it"""
|
| 89 |
request_file = get_request_file_for_model(
|
| 90 |
-
requests_path, self.full_model
|
| 91 |
|
| 92 |
try:
|
| 93 |
with open(request_file, "r") as f:
|
|
@@ -97,13 +92,12 @@ class EvalResult:
|
|
| 97 |
self.date = request.get("submitted_time", "")
|
| 98 |
except Exception:
|
| 99 |
print(
|
| 100 |
-
f"Could not find request file for {self.org}/{self.model}
|
| 101 |
|
| 102 |
def to_dict(self):
|
| 103 |
"""Converts the Eval Result to a dict compatible with our dataframe display"""
|
| 104 |
data_dict = {
|
| 105 |
"eval_name": self.eval_name, # not a column, just a save name,
|
| 106 |
-
AutoEvalColumn.precision.name: self.precision.value.name,
|
| 107 |
AutoEvalColumn.model_type.name: self.model_type.value.name,
|
| 108 |
AutoEvalColumn.model_type_symbol.name: self.model_type.value.symbol,
|
| 109 |
AutoEvalColumn.model.name: make_clickable_model(self.full_model),
|
|
@@ -118,7 +112,7 @@ class EvalResult:
|
|
| 118 |
return data_dict
|
| 119 |
|
| 120 |
|
| 121 |
-
def get_request_file_for_model(requests_path, model_name
|
| 122 |
"""Selects the correct request file for a given model. Only keeps runs tagged as FINISHED"""
|
| 123 |
request_files = os.path.join(
|
| 124 |
requests_path,
|
|
@@ -126,7 +120,7 @@ def get_request_file_for_model(requests_path, model_name, precision=None):
|
|
| 126 |
)
|
| 127 |
request_files = glob.glob(request_files)
|
| 128 |
|
| 129 |
-
# Select correct request file
|
| 130 |
request_file = ""
|
| 131 |
request_files = sorted(request_files, reverse=True)
|
| 132 |
for tmp_request_file in request_files:
|
|
|
|
| 1 |
from src.submission.check_validity import is_model_on_hub
|
| 2 |
+
from src.display.utils import AutoEvalColumn, ModelType, Tasks
|
| 3 |
from src.display.formatting import make_clickable_model
|
| 4 |
import numpy as np
|
| 5 |
import dateutil
|
|
|
|
| 15 |
class EvalResult:
|
| 16 |
"""Represents one full evaluation. Built from a combination of the result and request file for a given run.
|
| 17 |
"""
|
| 18 |
+
eval_name: str # org_model (uid)
|
| 19 |
full_model: str # org/model (path on hub)
|
| 20 |
org: str
|
| 21 |
model: str
|
| 22 |
revision: str # commit hash, "" if main
|
| 23 |
results: dict
|
|
|
|
| 24 |
model_type: ModelType = ModelType.Unknown # Pretrained, fine tuned, ...
|
| 25 |
architecture: str = "Unknown"
|
| 26 |
likes: int = 0
|
|
|
|
| 36 |
|
| 37 |
config = data.get("config")
|
| 38 |
|
|
|
|
|
|
|
|
|
|
| 39 |
# Get model and org
|
| 40 |
org_and_model = config.get(
|
| 41 |
"model_name", config.get("model_args", None))
|
|
|
|
| 44 |
if len(org_and_model) == 1:
|
| 45 |
org = None
|
| 46 |
model = org_and_model[0]
|
| 47 |
+
result_key = f"{model}"
|
| 48 |
else:
|
| 49 |
org = org_and_model[0]
|
| 50 |
model = org_and_model[1]
|
| 51 |
+
result_key = f"{org}"
|
| 52 |
full_model = "/".join(org_and_model)
|
| 53 |
|
| 54 |
still_on_hub, _, model_config = is_model_on_hub(
|
|
|
|
| 75 |
org=org,
|
| 76 |
model=model,
|
| 77 |
results=results,
|
|
|
|
| 78 |
revision=config.get("model_sha", ""),
|
| 79 |
still_on_hub=still_on_hub,
|
| 80 |
)
|
|
|
|
| 82 |
def update_with_request_file(self, requests_path):
|
| 83 |
"""Finds the relevant request file for the current model and updates info with it"""
|
| 84 |
request_file = get_request_file_for_model(
|
| 85 |
+
requests_path, self.full_model)
|
| 86 |
|
| 87 |
try:
|
| 88 |
with open(request_file, "r") as f:
|
|
|
|
| 92 |
self.date = request.get("submitted_time", "")
|
| 93 |
except Exception:
|
| 94 |
print(
|
| 95 |
+
f"Could not find request file for {self.org}/{self.model}")
|
| 96 |
|
| 97 |
def to_dict(self):
|
| 98 |
"""Converts the Eval Result to a dict compatible with our dataframe display"""
|
| 99 |
data_dict = {
|
| 100 |
"eval_name": self.eval_name, # not a column, just a save name,
|
|
|
|
| 101 |
AutoEvalColumn.model_type.name: self.model_type.value.name,
|
| 102 |
AutoEvalColumn.model_type_symbol.name: self.model_type.value.symbol,
|
| 103 |
AutoEvalColumn.model.name: make_clickable_model(self.full_model),
|
|
|
|
| 112 |
return data_dict
|
| 113 |
|
| 114 |
|
| 115 |
+
def get_request_file_for_model(requests_path, model_name):
|
| 116 |
"""Selects the correct request file for a given model. Only keeps runs tagged as FINISHED"""
|
| 117 |
request_files = os.path.join(
|
| 118 |
requests_path,
|
|
|
|
| 120 |
)
|
| 121 |
request_files = glob.glob(request_files)
|
| 122 |
|
| 123 |
+
# Select correct request file
|
| 124 |
request_file = ""
|
| 125 |
request_files = sorted(request_files, reverse=True)
|
| 126 |
for tmp_request_file in request_files:
|
src/submission/check_validity.py
CHANGED
|
@@ -55,15 +55,14 @@ def is_model_on_hub(model_name: str, revision: str, token: str = None, trust_rem
|
|
| 55 |
return False, "was not found on hub!", None
|
| 56 |
|
| 57 |
|
| 58 |
-
def get_model_size(model_info: ModelInfo
|
| 59 |
"""Gets the model size from the configuration, or the model name if the configuration does not contain the information."""
|
| 60 |
try:
|
| 61 |
model_size = round(model_info.safetensors["total"] / 1e9, 3)
|
| 62 |
except (AttributeError, TypeError):
|
| 63 |
return 0 # Unknown model sizes are indicated as 0, see NUMERIC_INTERVALS in app.py
|
| 64 |
|
| 65 |
-
size_factor = 8
|
| 66 |
-
precision == "GPTQ" or "gptq" in model_info.modelId.lower()) else 1
|
| 67 |
model_size = size_factor * model_size
|
| 68 |
return model_size
|
| 69 |
|
|
@@ -88,7 +87,7 @@ def already_submitted_models(requested_models_dir: str) -> set[str]:
|
|
| 88 |
with open(os.path.join(root, file), "r") as f:
|
| 89 |
info = json.load(f)
|
| 90 |
file_names.append(
|
| 91 |
-
f"{info['model']}_{info['revision']}
|
| 92 |
|
| 93 |
# Select organisation
|
| 94 |
if info["model"].count("/") == 0 or "submitted_time" not in info:
|
|
|
|
| 55 |
return False, "was not found on hub!", None
|
| 56 |
|
| 57 |
|
| 58 |
+
def get_model_size(model_info: ModelInfo):
|
| 59 |
"""Gets the model size from the configuration, or the model name if the configuration does not contain the information."""
|
| 60 |
try:
|
| 61 |
model_size = round(model_info.safetensors["total"] / 1e9, 3)
|
| 62 |
except (AttributeError, TypeError):
|
| 63 |
return 0 # Unknown model sizes are indicated as 0, see NUMERIC_INTERVALS in app.py
|
| 64 |
|
| 65 |
+
size_factor = 8 # Default size factor for float32 models
|
|
|
|
| 66 |
model_size = size_factor * model_size
|
| 67 |
return model_size
|
| 68 |
|
|
|
|
| 87 |
with open(os.path.join(root, file), "r") as f:
|
| 88 |
info = json.load(f)
|
| 89 |
file_names.append(
|
| 90 |
+
f"{info['model']}_{info['revision']}")
|
| 91 |
|
| 92 |
# Select organisation
|
| 93 |
if info["model"].count("/") == 0 or "submitted_time" not in info:
|
src/submission/submit.py
CHANGED
|
@@ -19,7 +19,6 @@ def add_new_eval(
|
|
| 19 |
model: str,
|
| 20 |
base_model: str,
|
| 21 |
revision: str,
|
| 22 |
-
precision: str,
|
| 23 |
model_type: str,
|
| 24 |
):
|
| 25 |
global REQUESTED_MODELS
|
|
@@ -34,7 +33,6 @@ def add_new_eval(
|
|
| 34 |
user_name = model.split("/")[0]
|
| 35 |
model_path = model.split("/")[1]
|
| 36 |
|
| 37 |
-
precision = precision.split(" ")[0]
|
| 38 |
current_time = datetime.now(timezone.utc).strftime("%Y-%m-%dT%H:%M:%SZ")
|
| 39 |
|
| 40 |
if model_type is None or model_type == "":
|
|
@@ -50,7 +48,7 @@ def add_new_eval(
|
|
| 50 |
except Exception:
|
| 51 |
return styled_error("Could not get your model information. Please fill it up properly.")
|
| 52 |
|
| 53 |
-
model_size = get_model_size(model_info=model_info
|
| 54 |
|
| 55 |
# Were the model card and license filled?
|
| 56 |
try:
|
|
@@ -69,7 +67,6 @@ def add_new_eval(
|
|
| 69 |
"model": model,
|
| 70 |
"base_model": base_model,
|
| 71 |
"revision": revision,
|
| 72 |
-
"precision": precision,
|
| 73 |
"status": "PENDING",
|
| 74 |
"submitted_time": current_time,
|
| 75 |
"model_type": model_type,
|
|
@@ -80,13 +77,13 @@ def add_new_eval(
|
|
| 80 |
}
|
| 81 |
|
| 82 |
# Check for duplicate submission
|
| 83 |
-
if f"{model}_{revision}
|
| 84 |
return styled_warning("This model has been already submitted.")
|
| 85 |
|
| 86 |
print("Creating eval file")
|
| 87 |
OUT_DIR = f"{EVAL_REQUESTS_PATH}/{user_name}"
|
| 88 |
os.makedirs(OUT_DIR, exist_ok=True)
|
| 89 |
-
out_path = f"{OUT_DIR}/{model_path}
|
| 90 |
|
| 91 |
with open(out_path, "w") as f:
|
| 92 |
f.write(json.dumps(eval_entry))
|
|
|
|
| 19 |
model: str,
|
| 20 |
base_model: str,
|
| 21 |
revision: str,
|
|
|
|
| 22 |
model_type: str,
|
| 23 |
):
|
| 24 |
global REQUESTED_MODELS
|
|
|
|
| 33 |
user_name = model.split("/")[0]
|
| 34 |
model_path = model.split("/")[1]
|
| 35 |
|
|
|
|
| 36 |
current_time = datetime.now(timezone.utc).strftime("%Y-%m-%dT%H:%M:%SZ")
|
| 37 |
|
| 38 |
if model_type is None or model_type == "":
|
|
|
|
| 48 |
except Exception:
|
| 49 |
return styled_error("Could not get your model information. Please fill it up properly.")
|
| 50 |
|
| 51 |
+
model_size = get_model_size(model_info=model_info)
|
| 52 |
|
| 53 |
# Were the model card and license filled?
|
| 54 |
try:
|
|
|
|
| 67 |
"model": model,
|
| 68 |
"base_model": base_model,
|
| 69 |
"revision": revision,
|
|
|
|
| 70 |
"status": "PENDING",
|
| 71 |
"submitted_time": current_time,
|
| 72 |
"model_type": model_type,
|
|
|
|
| 77 |
}
|
| 78 |
|
| 79 |
# Check for duplicate submission
|
| 80 |
+
if f"{model}_{revision}" in REQUESTED_MODELS:
|
| 81 |
return styled_warning("This model has been already submitted.")
|
| 82 |
|
| 83 |
print("Creating eval file")
|
| 84 |
OUT_DIR = f"{EVAL_REQUESTS_PATH}/{user_name}"
|
| 85 |
os.makedirs(OUT_DIR, exist_ok=True)
|
| 86 |
+
out_path = f"{OUT_DIR}/{model_path}_eval_request_False.json"
|
| 87 |
|
| 88 |
with open(out_path, "w") as f:
|
| 89 |
f.write(json.dumps(eval_entry))
|