Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -48,8 +48,8 @@ async def lifespan(app: FastAPI):
|
|
| 48 |
|
| 49 |
# Load models into the shared dictionary
|
| 50 |
ml_models["embedder"] = HuggingFaceEmbeddings(model_name="BAAI/bge-base-en-v1.5")
|
| 51 |
-
|
| 52 |
-
cross_encoder_model = HuggingFaceCrossEncoder(model_name="BAAI/bge-reranker-large")
|
| 53 |
ml_models["reranker_compressor"] = CrossEncoderReranker(model=cross_encoder_model, top_n=5)
|
| 54 |
ml_models["llm"] = ChatGoogleGenerativeAI(model="gemini-1.5-flash", api_key=GOOGLE_API_KEY)
|
| 55 |
ml_models["prompt_template"] = PromptTemplate.from_template(
|
|
@@ -63,6 +63,10 @@ If not, the raw question is provided instead:
|
|
| 63 |
We retrieved the following policy clauses and rules relevant to this case:
|
| 64 |
{context}
|
| 65 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 66 |
### Task:
|
| 67 |
If the details about age , gender , procedure , location , policy duration are available, do all of the following:
|
| 68 |
1. Decide whether the procedure is covered.
|
|
@@ -70,7 +74,7 @@ If the details about age , gender , procedure , location , policy duration are a
|
|
| 70 |
3. Justify with the relevant clause.
|
| 71 |
and answer the query precisely as insurance agent.
|
| 72 |
|
| 73 |
-
Otherwise, answer the question concisely and clearly using the retrieved context.
|
| 74 |
|
| 75 |
### Output format:
|
| 76 |
If query involes age , gender , procedure , location , policy duration answer like below:
|
|
@@ -78,16 +82,14 @@ If query involes age , gender , procedure , location , policy duration answer li
|
|
| 78 |
"decision": "approved / rejected",
|
| 79 |
"amount": "INR amount or null",
|
| 80 |
"justification": "Refer to specific clause"
|
| 81 |
-
Make it
|
| 82 |
}}
|
| 83 |
Else:
|
| 84 |
{{
|
| 85 |
"response": "Concise natural language answer"
|
| 86 |
-
|
| 87 |
}}
|
| 88 |
-
|
| 89 |
-
##NOTE : Do not mention document id or its page number just mention clauses if applicable .
|
| 90 |
-
"""
|
| 91 |
)
|
| 92 |
print("✅ Models and prompt loaded successfully!")
|
| 93 |
except Exception as e:
|
|
|
|
| 48 |
|
| 49 |
# Load models into the shared dictionary
|
| 50 |
ml_models["embedder"] = HuggingFaceEmbeddings(model_name="BAAI/bge-base-en-v1.5")
|
| 51 |
+
cross_encoder_model = HuggingFaceCrossEncoder(model_name="BAAI/bge-reranker-base")
|
| 52 |
+
# cross_encoder_model = HuggingFaceCrossEncoder(model_name="BAAI/bge-reranker-large")
|
| 53 |
ml_models["reranker_compressor"] = CrossEncoderReranker(model=cross_encoder_model, top_n=5)
|
| 54 |
ml_models["llm"] = ChatGoogleGenerativeAI(model="gemini-1.5-flash", api_key=GOOGLE_API_KEY)
|
| 55 |
ml_models["prompt_template"] = PromptTemplate.from_template(
|
|
|
|
| 63 |
We retrieved the following policy clauses and rules relevant to this case:
|
| 64 |
{context}
|
| 65 |
|
| 66 |
+
Remember the below three points while answering or responsing the query :
|
| 67 |
+
1:Do not mention document id or its page number just ,mention clauses if applicable .
|
| 68 |
+
2:Do not make such statement in the response that you need more document or information to answer the query, just answer from retrieved data.
|
| 69 |
+
3:UNDERSTAND THE QUERY ASKED SMARTLY AND ANSWER IT.
|
| 70 |
### Task:
|
| 71 |
If the details about age , gender , procedure , location , policy duration are available, do all of the following:
|
| 72 |
1. Decide whether the procedure is covered.
|
|
|
|
| 74 |
3. Justify with the relevant clause.
|
| 75 |
and answer the query precisely as insurance agent.
|
| 76 |
|
| 77 |
+
Otherwise, answer the question concisely , relevently and clearly using the retrieved context.
|
| 78 |
|
| 79 |
### Output format:
|
| 80 |
If query involes age , gender , procedure , location , policy duration answer like below:
|
|
|
|
| 82 |
"decision": "approved / rejected",
|
| 83 |
"amount": "INR amount or null",
|
| 84 |
"justification": "Refer to specific clause"
|
| 85 |
+
Make it concise ,perfect, relevent and clear .
|
| 86 |
}}
|
| 87 |
Else:
|
| 88 |
{{
|
| 89 |
"response": "Concise natural language answer"
|
| 90 |
+
Make it concise ,perfect, relevent and clear .
|
| 91 |
}}
|
| 92 |
+
"""
|
|
|
|
|
|
|
| 93 |
)
|
| 94 |
print("✅ Models and prompt loaded successfully!")
|
| 95 |
except Exception as e:
|