Chris Alexiuk
commited on
Commit
·
3391cce
1
Parent(s):
643f5c3
Update app.py
Browse files
app.py
CHANGED
|
@@ -56,55 +56,12 @@ async def init():
|
|
| 56 |
# docsearch = await cl.make_async(Chroma.from_documents)(pdf_data, embeddings)
|
| 57 |
docsearch = Chroma.from_documents(pdf_data, embeddings)
|
| 58 |
|
| 59 |
-
# custom SageMaker Model
|
| 60 |
-
class Llama2SageMaker(LLM):
|
| 61 |
-
max_new_tokens: int = 256
|
| 62 |
-
top_p: float = 0.9
|
| 63 |
-
temperature: float = 0.1
|
| 64 |
-
|
| 65 |
-
@property
|
| 66 |
-
def _llm_type(self) -> str:
|
| 67 |
-
return "Llama2SageMaker"
|
| 68 |
-
|
| 69 |
-
def _call(
|
| 70 |
-
self,
|
| 71 |
-
prompt: str,
|
| 72 |
-
stop: Optional[List[str]] = None,
|
| 73 |
-
run_manager: Optional[CallbackManagerForLLMRun] = None,
|
| 74 |
-
) -> str:
|
| 75 |
-
if stop is not None:
|
| 76 |
-
raise ValueError("stop kwargs are not permitted.")
|
| 77 |
-
|
| 78 |
-
json_body = {
|
| 79 |
-
"inputs" : [
|
| 80 |
-
[{"role" : "user", "content" : prompt}]
|
| 81 |
-
],
|
| 82 |
-
"parameters" : {
|
| 83 |
-
"max_new_tokens" : self.max_new_tokens,
|
| 84 |
-
"top_p" : self.top_p,
|
| 85 |
-
"temperature" : self.temperature
|
| 86 |
-
}
|
| 87 |
-
}
|
| 88 |
-
|
| 89 |
-
response = requests.post(model_api_gateway, json=json_body)
|
| 90 |
-
|
| 91 |
-
return response.json()[0]["generation"]["content"]
|
| 92 |
-
|
| 93 |
-
@property
|
| 94 |
-
def _identifying_params(self) -> Mapping[str, Any]:
|
| 95 |
-
"""Get the identifying parameters."""
|
| 96 |
-
return {
|
| 97 |
-
"max_new_tokens" : self.max_new_tokens,
|
| 98 |
-
"top_p" : self.top_p,
|
| 99 |
-
"temperature" : self.temperature
|
| 100 |
-
}
|
| 101 |
-
|
| 102 |
-
# set our llm to the custom Llama2SageMaker endpoint model
|
| 103 |
-
llm = Llama2SageMaker()
|
| 104 |
-
|
| 105 |
# Create a chain that uses the Chroma vector store
|
| 106 |
chain = RetrievalQAWithSourcesChain.from_chain_type(
|
| 107 |
-
|
|
|
|
|
|
|
|
|
|
| 108 |
chain_type="stuff",
|
| 109 |
retriever=docsearch.as_retriever(),
|
| 110 |
return_source_documents=True,
|
|
|
|
| 56 |
# docsearch = await cl.make_async(Chroma.from_documents)(pdf_data, embeddings)
|
| 57 |
docsearch = Chroma.from_documents(pdf_data, embeddings)
|
| 58 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 59 |
# Create a chain that uses the Chroma vector store
|
| 60 |
chain = RetrievalQAWithSourcesChain.from_chain_type(
|
| 61 |
+
ChatOpenAI(
|
| 62 |
+
model="gpt-3.5-turbo",
|
| 63 |
+
temperature=0.0
|
| 64 |
+
),
|
| 65 |
chain_type="stuff",
|
| 66 |
retriever=docsearch.as_retriever(),
|
| 67 |
return_source_documents=True,
|