rdiehlmartinez
commited on
Commit
·
2178182
1
Parent(s):
a378726
tested working version of weight subconfig
Browse files- pythia-training-metrics.py +12 -37
pythia-training-metrics.py
CHANGED
@@ -1,7 +1,6 @@
|
|
1 |
import datasets
|
2 |
import pickle
|
3 |
|
4 |
-
|
5 |
_DESCRIPTION = """\
|
6 |
Dataset for storing training metrics of pythia models
|
7 |
"""
|
@@ -12,10 +11,8 @@ class PythiaTrainingMetrics(datasets.GeneratorBasedBuilder):
|
|
12 |
"70m",
|
13 |
"160m",
|
14 |
"410m",
|
15 |
-
"1b",
|
16 |
"1.4b",
|
17 |
"2.8b",
|
18 |
-
"6.9b"
|
19 |
]
|
20 |
|
21 |
_GRADIENTS_DESCRIPTION = """\
|
@@ -56,36 +53,16 @@ class PythiaTrainingMetrics(datasets.GeneratorBasedBuilder):
|
|
56 |
description=_WEIGHTS_DESCRIPTION,
|
57 |
version="1.0.0",
|
58 |
),
|
59 |
-
|
60 |
-
name="all",
|
61 |
-
description="All the metrics",
|
62 |
-
version="1.0.0",
|
63 |
-
)
|
64 |
-
]
|
65 |
|
66 |
def _info(self):
|
67 |
"""
|
68 |
-
|
69 |
-
|
70 |
-
how do we do this if each feature is dependent on the model size?
|
71 |
"""
|
72 |
|
73 |
-
features_dict = {
|
74 |
-
"checkpoint_step": datasets.Value('int32'),
|
75 |
-
"layer_name": datasets.Value('string'),
|
76 |
-
}
|
77 |
-
|
78 |
-
if self.config.name in ["activations", "weights"]:
|
79 |
-
features_dict['data'] = datasets.Sequence(datasets.Value('float32'))
|
80 |
-
elif self.config_name in ["gradients", "gradients_mini"]:
|
81 |
-
features_dict['gradient_step'] = datasets.Value('int32')
|
82 |
-
features_dict['gradient'] = datasets.Sequence(datasets.Value('float32'))
|
83 |
-
|
84 |
-
features = datasets.Features(features_dict)
|
85 |
-
|
86 |
return datasets.DatasetInfo(
|
87 |
description=_DESCRIPTION,
|
88 |
-
features=features,
|
89 |
)
|
90 |
|
91 |
|
@@ -112,12 +89,12 @@ class PythiaTrainingMetrics(datasets.GeneratorBasedBuilder):
|
|
112 |
|
113 |
if self.config.name == "activations":
|
114 |
model_size_to_fp[model_size].append(f"{directory_path}/checkpoint_activations.pickle")
|
115 |
-
elif self.
|
116 |
model_size_to_fp[model_size].append(f"{directory_path}/checkpoint_weights.pickle")
|
117 |
-
elif self.
|
118 |
for gradient_step in get_gradient_step(checkpoint_step):
|
119 |
model_size_to_fp[model_size].append(f"{directory_path}/checkpoint_gradients_{gradient_step}.pickle")
|
120 |
-
elif self.
|
121 |
for gradient_step in get_gradient_step(checkpoint_step)[:2]:
|
122 |
model_size_to_fp[model_size].append(f"{directory_path}/checkpoint_gradients_mini_{gradient_step}.pickle")
|
123 |
else:
|
@@ -134,29 +111,27 @@ class PythiaTrainingMetrics(datasets.GeneratorBasedBuilder):
|
|
134 |
) for model_size_name, downloaded_fps in downloaded_files.items()
|
135 |
]
|
136 |
|
137 |
-
def _generate_examples(self, filepaths
|
138 |
|
139 |
# the filepaths should be a list of filepaths
|
140 |
if isinstance(filepaths, str):
|
141 |
filepaths = [filepaths]
|
142 |
-
|
143 |
global_idx = 0 # the unique identifier for the example
|
144 |
|
145 |
for filepath in filepaths:
|
146 |
-
with open(filepath,
|
147 |
data = pickle.load(f)
|
148 |
|
149 |
# extract checkpoint step from the filepath
|
150 |
-
checkpoint_step = int(filepath.split("/")[
|
151 |
|
152 |
if self.config.name in ["activations", "weights"]:
|
153 |
for layer_name, layer_data in data.items():
|
154 |
-
yield global_idx, {"checkpoint_step": checkpoint_step, "layer_name": layer_name, "data":
|
155 |
global_idx += 1
|
156 |
elif self.config.name in ["gradients", "gradients_mini"]:
|
157 |
-
|
158 |
gradient_step = int(filepath.split('/')[-1].split("_")[-1].split(".")[0])
|
159 |
-
|
160 |
for layer_name, layer_data in data.items():
|
161 |
-
yield global_idx, {"checkpoint_step": checkpoint_step, "layer_name": layer_name, "gradient_step": gradient_step, "
|
162 |
global_idx += 1
|
|
|
1 |
import datasets
|
2 |
import pickle
|
3 |
|
|
|
4 |
_DESCRIPTION = """\
|
5 |
Dataset for storing training metrics of pythia models
|
6 |
"""
|
|
|
11 |
"70m",
|
12 |
"160m",
|
13 |
"410m",
|
|
|
14 |
"1.4b",
|
15 |
"2.8b",
|
|
|
16 |
]
|
17 |
|
18 |
_GRADIENTS_DESCRIPTION = """\
|
|
|
53 |
description=_WEIGHTS_DESCRIPTION,
|
54 |
version="1.0.0",
|
55 |
),
|
56 |
+
]
|
|
|
|
|
|
|
|
|
|
|
57 |
|
58 |
def _info(self):
|
59 |
"""
|
60 |
+
NOTE: we might want to specify features, but since the featuers are different for each
|
61 |
+
model size it's annoying and kind of pointless since hf does it automatically
|
|
|
62 |
"""
|
63 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
64 |
return datasets.DatasetInfo(
|
65 |
description=_DESCRIPTION,
|
|
|
66 |
)
|
67 |
|
68 |
|
|
|
89 |
|
90 |
if self.config.name == "activations":
|
91 |
model_size_to_fp[model_size].append(f"{directory_path}/checkpoint_activations.pickle")
|
92 |
+
elif self.config.name == "weights":
|
93 |
model_size_to_fp[model_size].append(f"{directory_path}/checkpoint_weights.pickle")
|
94 |
+
elif self.config.name == "gradients":
|
95 |
for gradient_step in get_gradient_step(checkpoint_step):
|
96 |
model_size_to_fp[model_size].append(f"{directory_path}/checkpoint_gradients_{gradient_step}.pickle")
|
97 |
+
elif self.config.name == "gradients_mini":
|
98 |
for gradient_step in get_gradient_step(checkpoint_step)[:2]:
|
99 |
model_size_to_fp[model_size].append(f"{directory_path}/checkpoint_gradients_mini_{gradient_step}.pickle")
|
100 |
else:
|
|
|
111 |
) for model_size_name, downloaded_fps in downloaded_files.items()
|
112 |
]
|
113 |
|
114 |
+
def _generate_examples(self, filepaths):
|
115 |
|
116 |
# the filepaths should be a list of filepaths
|
117 |
if isinstance(filepaths, str):
|
118 |
filepaths = [filepaths]
|
119 |
+
|
120 |
global_idx = 0 # the unique identifier for the example
|
121 |
|
122 |
for filepath in filepaths:
|
123 |
+
with open(filepath, 'rb') as f:
|
124 |
data = pickle.load(f)
|
125 |
|
126 |
# extract checkpoint step from the filepath
|
127 |
+
checkpoint_step = int(filepath.split("/")[-2].split("_")[-1])
|
128 |
|
129 |
if self.config.name in ["activations", "weights"]:
|
130 |
for layer_name, layer_data in data.items():
|
131 |
+
yield global_idx, {"checkpoint_step": checkpoint_step, "layer_name": layer_name, "data": layer_data}
|
132 |
global_idx += 1
|
133 |
elif self.config.name in ["gradients", "gradients_mini"]:
|
|
|
134 |
gradient_step = int(filepath.split('/')[-1].split("_")[-1].split(".")[0])
|
|
|
135 |
for layer_name, layer_data in data.items():
|
136 |
+
yield global_idx, {"checkpoint_step": checkpoint_step, "layer_name": layer_name, "gradient_step": gradient_step, "data": layer_data}
|
137 |
global_idx += 1
|