Update app.py
Browse files
app.py
CHANGED
@@ -65,6 +65,110 @@ def dspy_generate_agent_prompts(prompt):
|
|
65 |
|
66 |
return agent_prompts
|
67 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
68 |
# Define the main function to be used with Gradio
|
69 |
def generate_outputs(user_prompt):
|
70 |
# 1. Process prompt with langchain (replace with your actual implementation)
|
@@ -79,11 +183,12 @@ def generate_outputs(user_prompt):
|
|
79 |
# 4. Generate prompts for agents using DSPy
|
80 |
agent_prompts = dspy_generate_agent_prompts(processed_prompt)
|
81 |
|
82 |
-
# 5. Use the chosen LLM for two of the prompts
|
83 |
-
output_1 = llm(agent_prompts[0], max_length=100)[0][
|
84 |
-
output_2 = llm(agent_prompts[1], max_length=100)[0][
|
|
|
85 |
|
86 |
-
# 6. Produce outputs with Langchain or DSPy (
|
87 |
report, recommendations, visualization = produce_outputs(combined_data)
|
88 |
|
89 |
return report, recommendations, visualization
|
|
|
65 |
|
66 |
return agent_prompts
|
67 |
|
68 |
+
def query_vectara(text):
|
69 |
+
user_message = text
|
70 |
+
|
71 |
+
# Read authentication parameters from the .env file
|
72 |
+
customer_id = os.getenv('CUSTOMER_ID')
|
73 |
+
corpus_id = os.getenv('CORPUS_ID')
|
74 |
+
api_key = os.getenv('API_KEY')
|
75 |
+
|
76 |
+
# Define the headers
|
77 |
+
api_key_header = {
|
78 |
+
"customer-id": customer_id,
|
79 |
+
"x-api-key": api_key
|
80 |
+
}
|
81 |
+
|
82 |
+
# Define the request body in the structure provided in the example
|
83 |
+
request_body = {
|
84 |
+
"query": [
|
85 |
+
{
|
86 |
+
"query": user_message,
|
87 |
+
"queryContext": "",
|
88 |
+
"start": 1,
|
89 |
+
"numResults": 25,
|
90 |
+
"contextConfig": {
|
91 |
+
"charsBefore": 0,
|
92 |
+
"charsAfter": 0,
|
93 |
+
"sentencesBefore": 2,
|
94 |
+
"sentencesAfter": 2,
|
95 |
+
"startTag": "%START_SNIPPET%",
|
96 |
+
"endTag": "%END_SNIPPET%",
|
97 |
+
},
|
98 |
+
"rerankingConfig": {
|
99 |
+
"rerankerId": 272725718,
|
100 |
+
"mmrConfig": {
|
101 |
+
"diversityBias": 0.35
|
102 |
+
}
|
103 |
+
},
|
104 |
+
"corpusKey": [
|
105 |
+
{
|
106 |
+
"customerId": customer_id,
|
107 |
+
"corpusId": corpus_id,
|
108 |
+
"semantics": 0,
|
109 |
+
"metadataFilter": "",
|
110 |
+
"lexicalInterpolationConfig": {
|
111 |
+
"lambda": 0
|
112 |
+
},
|
113 |
+
"dim": []
|
114 |
+
}
|
115 |
+
],
|
116 |
+
"summary": [
|
117 |
+
{
|
118 |
+
"maxSummarizedResults": 5,
|
119 |
+
"responseLang": "auto",
|
120 |
+
"summarizerPromptName": "vectara-summary-ext-v1.2.0"
|
121 |
+
}
|
122 |
+
]
|
123 |
+
}
|
124 |
+
]
|
125 |
+
}
|
126 |
+
|
127 |
+
# Make the API request using Gradio
|
128 |
+
response = requests.post(
|
129 |
+
"https://api.vectara.io/v1/query",
|
130 |
+
json=request_body, # Use json to automatically serialize the request body
|
131 |
+
verify=True,
|
132 |
+
headers=api_key_header
|
133 |
+
)
|
134 |
+
|
135 |
+
if response.status_code == 200:
|
136 |
+
query_data = response.json()
|
137 |
+
if query_data:
|
138 |
+
sources_info = []
|
139 |
+
|
140 |
+
# Extract the summary.
|
141 |
+
summary = query_data['responseSet'][0]['summary'][0]['text']
|
142 |
+
|
143 |
+
# Iterate over all response sets
|
144 |
+
for response_set in query_data.get('responseSet', []):
|
145 |
+
# Extract sources
|
146 |
+
# Limit to top 5 sources.
|
147 |
+
for source in response_set.get('response', [])[:5]:
|
148 |
+
source_metadata = source.get('metadata', [])
|
149 |
+
source_info = {}
|
150 |
+
|
151 |
+
for metadata in source_metadata:
|
152 |
+
metadata_name = metadata.get('name', '')
|
153 |
+
metadata_value = metadata.get('value', '')
|
154 |
+
|
155 |
+
if metadata_name == 'title':
|
156 |
+
source_info['title'] = metadata_value
|
157 |
+
elif metadata_name == 'author':
|
158 |
+
source_info['author'] = metadata_value
|
159 |
+
elif metadata_name == 'pageNumber':
|
160 |
+
source_info['page number'] = metadata_value
|
161 |
+
|
162 |
+
if source_info:
|
163 |
+
sources_info.append(source_info)
|
164 |
+
|
165 |
+
result = {"summary": summary, "sources": sources_info}
|
166 |
+
return f"{json.dumps(result, indent=2)}"
|
167 |
+
else:
|
168 |
+
return "No data found in the response."
|
169 |
+
else:
|
170 |
+
return f"Error: {response.status_code}"
|
171 |
+
|
172 |
# Define the main function to be used with Gradio
|
173 |
def generate_outputs(user_prompt):
|
174 |
# 1. Process prompt with langchain (replace with your actual implementation)
|
|
|
183 |
# 4. Generate prompts for agents using DSPy
|
184 |
agent_prompts = dspy_generate_agent_prompts(processed_prompt)
|
185 |
|
186 |
+
# 5. Use the chosen LLM for two of the prompts and vectara tool use for the third agent
|
187 |
+
output_1 = llm(agent_prompts[0], max_length=100)[0][combined_data]
|
188 |
+
output_2 = llm(agent_prompts[1], max_length=100)[0][combined_data]
|
189 |
+
output_3 = query_vectara(prompt)
|
190 |
|
191 |
+
# 6. Produce outputs with Langchain or DSPy (stand in section)
|
192 |
report, recommendations, visualization = produce_outputs(combined_data)
|
193 |
|
194 |
return report, recommendations, visualization
|