Update app.py
Browse files
app.py
CHANGED
@@ -67,55 +67,27 @@ def get_titan_embedding(bedrock_client, doc_name, text, attempt=0, cutoff=10000)
|
|
67 |
|
68 |
retries = 5
|
69 |
|
70 |
-
|
71 |
-
|
72 |
-
|
73 |
-
|
74 |
-
|
75 |
-
|
76 |
-
|
77 |
-
|
78 |
-
|
79 |
-
|
80 |
-
|
81 |
-
|
82 |
-
|
83 |
-
|
84 |
-
|
85 |
-
|
86 |
-
|
87 |
-
|
88 |
-
response_body = json.loads(response['body'].read())
|
89 |
|
90 |
|
91 |
-
# Handle a few common client exceptions
|
92 |
-
except botocore.exceptions.ClientError as error:
|
93 |
-
if error.response['Error']['Code'] == 'ThrottlingException':
|
94 |
-
if attempt + 1 == retries:
|
95 |
-
return None
|
96 |
-
|
97 |
-
delay = 2 ** (attempt + 1);
|
98 |
-
time.sleep(delay)
|
99 |
-
return get_titan_embedding(bedrock_client, doc_name, text, attempt=attempt + 1)
|
100 |
-
|
101 |
-
elif error.response['Error']['Code'] == 'ValidationException':
|
102 |
-
# get chunks of text length 20000 characters
|
103 |
-
text_chunks = [text[i:i+cutoff] for i in range(0, len(text), cutoff)]
|
104 |
-
embeddings = []
|
105 |
-
for chunk in text_chunks:
|
106 |
-
embeddings.append(get_titan_embedding(bedrock_client, doc_name, chunk))
|
107 |
-
|
108 |
-
# return the average of the embeddinngs
|
109 |
-
return np.mean(embeddings, axis=0)
|
110 |
-
|
111 |
-
else:
|
112 |
-
yield f"Unhandled Exception when processing {doc_name}! : {error.response['Error']['Code']}"
|
113 |
-
return None
|
114 |
|
115 |
-
# Catch-all for any other exceptions
|
116 |
-
except Exception as error:
|
117 |
-
yield f"Unhandled Exception when processing {doc_name}: {type(error).__name__}"
|
118 |
-
return None
|
119 |
|
120 |
return response_body.get('embedding')
|
121 |
|
@@ -129,6 +101,8 @@ def ask_ds(message, history):
|
|
129 |
|
130 |
# RAG
|
131 |
question_embedding = get_titan_embedding(bedrock_client, 'question', question)
|
|
|
|
|
132 |
|
133 |
similar_documents = []
|
134 |
for file, data in extractions.items():
|
|
|
67 |
|
68 |
retries = 5
|
69 |
|
70 |
+
model_id = 'amazon.titan-embed-text-v1'
|
71 |
+
accept = 'application/json'
|
72 |
+
content_type = 'application/json'
|
73 |
+
|
74 |
+
body = json.dumps({
|
75 |
+
"inputText": text,
|
76 |
+
})
|
77 |
+
|
78 |
+
# Invoke model
|
79 |
+
response = bedrock_client.invoke_model(
|
80 |
+
body=body,
|
81 |
+
modelId=model_id,
|
82 |
+
accept=accept,
|
83 |
+
contentType=content_type
|
84 |
+
)
|
85 |
+
|
86 |
+
# Print response
|
87 |
+
response_body = json.loads(response['body'].read())
|
|
|
88 |
|
89 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
90 |
|
|
|
|
|
|
|
|
|
91 |
|
92 |
return response_body.get('embedding')
|
93 |
|
|
|
101 |
|
102 |
# RAG
|
103 |
question_embedding = get_titan_embedding(bedrock_client, 'question', question)
|
104 |
+
|
105 |
+
yield f"question embedding: {question_embedding}"
|
106 |
|
107 |
similar_documents = []
|
108 |
for file, data in extractions.items():
|