Spaces:
Runtime error
Runtime error
LennardZuendorf
commited on
Commit
•
517fd4c
1
Parent(s):
1f063be
fix: final fix of attention
Browse files- explanation/attention.py +2 -2
- explanation/markup.py +1 -1
- main.py +4 -4
- model/mistral.py +2 -3
- utils/formatting.py +2 -3
- utils/modelling.py +7 -3
explanation/attention.py
CHANGED
@@ -37,8 +37,8 @@ def chat_explained(model, prompt):
|
|
37 |
attention_output = mdl.format_mistral_attention(attention_output)
|
38 |
averaged_attention = fmt.avg_attention(attention_output, model="mistral")
|
39 |
|
40 |
-
|
41 |
-
response_text = mistral.format_answer(
|
42 |
|
43 |
# otherwise use attention visualization for godel
|
44 |
else:
|
|
|
37 |
attention_output = mdl.format_mistral_attention(attention_output)
|
38 |
averaged_attention = fmt.avg_attention(attention_output, model="mistral")
|
39 |
|
40 |
+
output_text = fmt.format_output_text(output_text)
|
41 |
+
response_text = mistral.format_answer(output_text)
|
42 |
|
43 |
# otherwise use attention visualization for godel
|
44 |
else:
|
explanation/markup.py
CHANGED
@@ -10,7 +10,7 @@ from utils import formatting as fmt
|
|
10 |
|
11 |
# main function that assigns each text snipped a marked bucket
|
12 |
def markup_text(input_text: list, text_values: ndarray, variant: str):
|
13 |
-
print(f"Marking up text {input_text}
|
14 |
|
15 |
# naming of the 11 buckets
|
16 |
bucket_tags = ["-5", "-4", "-3", "-2", "-1", "0", "+1", "+2", "+3", "+4", "+5"]
|
|
|
10 |
|
11 |
# main function that assigns each text snipped a marked bucket
|
12 |
def markup_text(input_text: list, text_values: ndarray, variant: str):
|
13 |
+
print(f"Marking up text {input_text} for {variant}.")
|
14 |
|
15 |
# naming of the 11 buckets
|
16 |
bucket_tags = ["-5", "-4", "-3", "-2", "-1", "0", "+1", "+2", "+3", "+4", "+5"]
|
main.py
CHANGED
@@ -252,8 +252,8 @@ with gr.Blocks(
|
|
252 |
],
|
253 |
inputs=[
|
254 |
user_prompt,
|
255 |
-
system_prompt,
|
256 |
xai_selection,
|
|
|
257 |
model_selection,
|
258 |
knowledge_input,
|
259 |
],
|
@@ -266,6 +266,7 @@ with gr.Blocks(
|
|
266 |
examples=[
|
267 |
[
|
268 |
"Does money buy happiness?",
|
|
|
269 |
(
|
270 |
"Some studies have found a correlation between income"
|
271 |
" and happiness, but this relationship often has"
|
@@ -275,10 +276,10 @@ with gr.Blocks(
|
|
275 |
),
|
276 |
"",
|
277 |
"GODEL",
|
278 |
-
"SHAP",
|
279 |
],
|
280 |
[
|
281 |
"Does money buy happiness?",
|
|
|
282 |
(
|
283 |
"Some studies have found a correlation between income"
|
284 |
" and happiness, but this relationship often has"
|
@@ -288,14 +289,13 @@ with gr.Blocks(
|
|
288 |
),
|
289 |
"",
|
290 |
"GODEL",
|
291 |
-
"Attention",
|
292 |
],
|
293 |
[
|
294 |
"Does money buy happiness?",
|
|
|
295 |
"",
|
296 |
"",
|
297 |
"GODEL",
|
298 |
-
"Attention",
|
299 |
],
|
300 |
],
|
301 |
inputs=[
|
|
|
252 |
],
|
253 |
inputs=[
|
254 |
user_prompt,
|
|
|
255 |
xai_selection,
|
256 |
+
system_prompt,
|
257 |
model_selection,
|
258 |
knowledge_input,
|
259 |
],
|
|
|
266 |
examples=[
|
267 |
[
|
268 |
"Does money buy happiness?",
|
269 |
+
"SHAP",
|
270 |
(
|
271 |
"Some studies have found a correlation between income"
|
272 |
" and happiness, but this relationship often has"
|
|
|
276 |
),
|
277 |
"",
|
278 |
"GODEL",
|
|
|
279 |
],
|
280 |
[
|
281 |
"Does money buy happiness?",
|
282 |
+
"Attention",
|
283 |
(
|
284 |
"Some studies have found a correlation between income"
|
285 |
" and happiness, but this relationship often has"
|
|
|
289 |
),
|
290 |
"",
|
291 |
"GODEL",
|
|
|
292 |
],
|
293 |
[
|
294 |
"Does money buy happiness?",
|
295 |
+
"Attention",
|
296 |
"",
|
297 |
"",
|
298 |
"GODEL",
|
|
|
299 |
],
|
300 |
],
|
301 |
inputs=[
|
model/mistral.py
CHANGED
@@ -32,12 +32,11 @@ TOKENIZER = AutoTokenizer.from_pretrained("mistralai/Mistral-7B-Instruct-v0.2")
|
|
32 |
# default model config
|
33 |
CONFIG = GenerationConfig.from_pretrained("mistralai/Mistral-7B-Instruct-v0.2")
|
34 |
base_config_dict = {
|
35 |
-
"temperature":
|
36 |
-
"max_new_tokens":
|
37 |
"top_p": 0.9,
|
38 |
"repetition_penalty": 1.2,
|
39 |
"do_sample": True,
|
40 |
-
"seed": 42,
|
41 |
}
|
42 |
CONFIG.update(**base_config_dict)
|
43 |
|
|
|
32 |
# default model config
|
33 |
CONFIG = GenerationConfig.from_pretrained("mistralai/Mistral-7B-Instruct-v0.2")
|
34 |
base_config_dict = {
|
35 |
+
"temperature": 1,
|
36 |
+
"max_new_tokens": 100,
|
37 |
"top_p": 0.9,
|
38 |
"repetition_penalty": 1.2,
|
39 |
"do_sample": True,
|
|
|
40 |
}
|
41 |
CONFIG.update(**base_config_dict)
|
42 |
|
utils/formatting.py
CHANGED
@@ -92,15 +92,14 @@ def avg_attention(attention_values, model: str):
|
|
92 |
# check if model is godel
|
93 |
if model == "godel":
|
94 |
# get attention values for the input and output vectors
|
95 |
-
attention = attention_values.
|
96 |
-
return np.mean(attention, axis=
|
97 |
|
98 |
# extracting attention values for mistral
|
99 |
attention = attention_values.to(torch.device("cpu")).detach().numpy()
|
100 |
|
101 |
# removing the last dimension and transposing to get the correct shape
|
102 |
attention = attention[:, :, :, 0]
|
103 |
-
attention = attention.transpose()
|
104 |
|
105 |
# return the averaged attention values
|
106 |
return np.mean(attention, axis=1)
|
|
|
92 |
# check if model is godel
|
93 |
if model == "godel":
|
94 |
# get attention values for the input and output vectors
|
95 |
+
attention = attention_values.encoder_attentions[0][0].detach().numpy()
|
96 |
+
return np.mean(attention, axis=1)
|
97 |
|
98 |
# extracting attention values for mistral
|
99 |
attention = attention_values.to(torch.device("cpu")).detach().numpy()
|
100 |
|
101 |
# removing the last dimension and transposing to get the correct shape
|
102 |
attention = attention[:, :, :, 0]
|
|
|
103 |
|
104 |
# return the averaged attention values
|
105 |
return np.mean(attention, axis=1)
|
utils/modelling.py
CHANGED
@@ -100,11 +100,15 @@ def gpu_loading_config(max_memory: str = "15000MB"):
|
|
100 |
|
101 |
|
102 |
# formatting mistral attention values
|
103 |
-
# CREDIT: copied
|
104 |
# see https://github.com/jessevig/bertviz
|
105 |
-
def format_mistral_attention(attention_values):
|
|
|
|
|
106 |
squeezed = []
|
107 |
for layer_attention in attention_values:
|
108 |
layer_attention = layer_attention.squeeze(0)
|
|
|
|
|
109 |
squeezed.append(layer_attention)
|
110 |
-
return torch.stack(squeezed)
|
|
|
100 |
|
101 |
|
102 |
# formatting mistral attention values
|
103 |
+
# CREDIT: copied from BERTViz
|
104 |
# see https://github.com/jessevig/bertviz
|
105 |
+
def format_mistral_attention(attention_values, layers=None, heads=None):
|
106 |
+
if layers:
|
107 |
+
attention_values = [attention_values[layer_index] for layer_index in layers]
|
108 |
squeezed = []
|
109 |
for layer_attention in attention_values:
|
110 |
layer_attention = layer_attention.squeeze(0)
|
111 |
+
if heads:
|
112 |
+
layer_attention = layer_attention[heads]
|
113 |
squeezed.append(layer_attention)
|
114 |
+
return torch.stack(squeezed)
|