Spaces:
Runtime error
Runtime error
Adding directions exploration.
Browse files
app.py
CHANGED
@@ -12,6 +12,7 @@ alt.data_transformers.disable_max_rows()
|
|
12 |
number_re = re.compile(r"\.[0-9]*\.")
|
13 |
|
14 |
STATE_DICT = {}
|
|
|
15 |
DATA = pd.DataFrame()
|
16 |
|
17 |
|
@@ -30,15 +31,20 @@ def scatter_plot_fn(group_name):
|
|
30 |
|
31 |
def find_choices(state_dict):
|
32 |
if not state_dict:
|
33 |
-
return []
|
34 |
global DATA
|
35 |
-
layered_tensors = [
|
|
|
|
|
36 |
choices = set()
|
37 |
data = []
|
|
|
38 |
for name in layered_tensors:
|
39 |
group_name = number_re.sub(".{N}.", name)
|
40 |
choices.add(group_name)
|
41 |
layer = int(number_re.search(name).group()[1:-1])
|
|
|
|
|
42 |
|
43 |
svdvals = torch.linalg.svdvals(state_dict[name])
|
44 |
svdvals /= svdvals.sum()
|
@@ -49,19 +55,64 @@ def find_choices(state_dict):
|
|
49 |
DATA["val"] = DATA["val"].astype("float")
|
50 |
DATA["layer"] = DATA["layer"].astype("category")
|
51 |
DATA["rank"] = DATA["rank"].astype("int32")
|
52 |
-
return choices
|
53 |
|
54 |
|
55 |
def weights_fn(model_id):
|
56 |
-
global STATE_DICT
|
57 |
try:
|
58 |
pipe = pipeline(model=model_id)
|
|
|
59 |
STATE_DICT = pipe.model.state_dict()
|
60 |
except Exception as e:
|
61 |
print(e)
|
62 |
STATE_DICT = {}
|
63 |
-
choices = find_choices(STATE_DICT)
|
64 |
-
return gr.Dropdown.update(choices=choices)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
65 |
|
66 |
|
67 |
with gr.Blocks() as scatter_plot:
|
@@ -69,10 +120,16 @@ with gr.Blocks() as scatter_plot:
|
|
69 |
with gr.Column():
|
70 |
model_id = gr.Textbox(label="model_id")
|
71 |
weights = gr.Dropdown(label="weights")
|
|
|
72 |
with gr.Column():
|
73 |
plot = gr.LinePlot(show_label=False).style(container=True)
|
74 |
-
|
|
|
|
|
75 |
weights.change(fn=scatter_plot_fn, inputs=weights, outputs=plot)
|
|
|
|
|
|
|
76 |
|
77 |
if __name__ == "__main__":
|
78 |
scatter_plot.launch()
|
|
|
12 |
number_re = re.compile(r"\.[0-9]*\.")
|
13 |
|
14 |
STATE_DICT = {}
|
15 |
+
PIPE = None
|
16 |
DATA = pd.DataFrame()
|
17 |
|
18 |
|
|
|
31 |
|
32 |
def find_choices(state_dict):
|
33 |
if not state_dict:
|
34 |
+
return [], []
|
35 |
global DATA
|
36 |
+
layered_tensors = [
|
37 |
+
k for k, v in state_dict.items() if number_re.findall(k) and len(v.shape) == 2
|
38 |
+
]
|
39 |
choices = set()
|
40 |
data = []
|
41 |
+
max_layer = 0
|
42 |
for name in layered_tensors:
|
43 |
group_name = number_re.sub(".{N}.", name)
|
44 |
choices.add(group_name)
|
45 |
layer = int(number_re.search(name).group()[1:-1])
|
46 |
+
if layer > max_layer:
|
47 |
+
max_layer = layer
|
48 |
|
49 |
svdvals = torch.linalg.svdvals(state_dict[name])
|
50 |
svdvals /= svdvals.sum()
|
|
|
55 |
DATA["val"] = DATA["val"].astype("float")
|
56 |
DATA["layer"] = DATA["layer"].astype("category")
|
57 |
DATA["rank"] = DATA["rank"].astype("int32")
|
58 |
+
return choices, list(range(max_layer + 1))
|
59 |
|
60 |
|
61 |
def weights_fn(model_id):
|
62 |
+
global STATE_DICT, PIPE
|
63 |
try:
|
64 |
pipe = pipeline(model=model_id)
|
65 |
+
PIPE = pipe
|
66 |
STATE_DICT = pipe.model.state_dict()
|
67 |
except Exception as e:
|
68 |
print(e)
|
69 |
STATE_DICT = {}
|
70 |
+
choices, layers = find_choices(STATE_DICT)
|
71 |
+
return [gr.Dropdown.update(choices=choices), gr.Dropdown.update(choices=layers)]
|
72 |
+
|
73 |
+
|
74 |
+
def layer_fn(weights, layer):
|
75 |
+
k = 5
|
76 |
+
directions = 10
|
77 |
+
|
78 |
+
embeddings = PIPE.model.get_input_embeddings().weight
|
79 |
+
weight_name = weights.replace("{N}", str(layer))
|
80 |
+
|
81 |
+
weight = STATE_DICT[weight_name]
|
82 |
+
|
83 |
+
U, S, Vh = torch.linalg.svd(weight)
|
84 |
+
|
85 |
+
D = U if U.shape[0] == embeddings.shape[0] else Vh
|
86 |
+
|
87 |
+
# words = D[:directions].matmul(embeddings.T).topk(k=k)
|
88 |
+
# words_t = D[:, :directions].T.matmul(embeddings.T).topk(k=k)
|
89 |
+
|
90 |
+
# Cosine similarity
|
91 |
+
words = (
|
92 |
+
(D[:directions] / D[:directions].norm(dim=0))
|
93 |
+
.matmul(embeddings.T / embeddings.T.norm(dim=0))
|
94 |
+
.topk(k=k)
|
95 |
+
)
|
96 |
+
words_t = (
|
97 |
+
(D[:, :directions].T / D[:, :directions].norm(dim=1))
|
98 |
+
.matmul(embeddings.T / embeddings.T.norm(dim=0))
|
99 |
+
.topk(k=k)
|
100 |
+
)
|
101 |
+
|
102 |
+
data = [[PIPE.tokenizer.decode(w) for w in indices] for indices in words.indices]
|
103 |
+
data = np.array(data)
|
104 |
+
data = pd.DataFrame(data)
|
105 |
+
|
106 |
+
data_t = [
|
107 |
+
[PIPE.tokenizer.decode(w) for w in indices] for indices in words_t.indices
|
108 |
+
]
|
109 |
+
data_t = np.array(data_t)
|
110 |
+
data_t = pd.DataFrame(data_t)
|
111 |
+
|
112 |
+
return (
|
113 |
+
gr.Dataframe.update(value=data, interactive=False),
|
114 |
+
gr.Dataframe.update(value=data_t, interactive=False),
|
115 |
+
)
|
116 |
|
117 |
|
118 |
with gr.Blocks() as scatter_plot:
|
|
|
120 |
with gr.Column():
|
121 |
model_id = gr.Textbox(label="model_id")
|
122 |
weights = gr.Dropdown(label="weights")
|
123 |
+
layer = gr.Dropdown(label="layer")
|
124 |
with gr.Column():
|
125 |
plot = gr.LinePlot(show_label=False).style(container=True)
|
126 |
+
directions = gr.Dataframe(interactive=False)
|
127 |
+
directions_t = gr.Dataframe(interactive=False)
|
128 |
+
model_id.change(weights_fn, inputs=model_id, outputs=[weights, layer])
|
129 |
weights.change(fn=scatter_plot_fn, inputs=weights, outputs=plot)
|
130 |
+
layer.change(
|
131 |
+
fn=layer_fn, inputs=[weights, layer], outputs=[directions, directions_t]
|
132 |
+
)
|
133 |
|
134 |
if __name__ == "__main__":
|
135 |
scatter_plot.launch()
|