Spaces:
Running
on
T4
Running
on
T4
model changed to Haiku 3.5 for query re=write
Browse files
RAG/rag_DocumentSearcher.py
CHANGED
@@ -325,7 +325,7 @@ def query_(awsauth,inputs, session_id,search_types):
|
|
325 |
if(hit["_source"]["image"]!="None"):
|
326 |
with open(parent_dirname+'/figures/'+st.session_state.input_index+"/"+hit["_source"]["raw_element_type"].split("_")[1].replace(".jpg","")+"-resized.jpg", "rb") as read_img:
|
327 |
input_encoded = base64.b64encode(read_img.read()).decode("utf8")
|
328 |
-
context.append(str(id+1) + " : Reference from a image :" + invoke_models.generate_image_captions_llm(input_encoded,question))
|
329 |
else:
|
330 |
context.append(str(id+1) + " : Reference from a text chunk :" + hit["_source"]["processed_element"])
|
331 |
|
@@ -349,7 +349,7 @@ def query_(awsauth,inputs, session_id,search_types):
|
|
349 |
|
350 |
llm_prompt = prompt_template.format(context="\n".join(total_context[0:3]),question=question)
|
351 |
print("started LLM prompt: "+st.session_state.input_index)
|
352 |
-
output = invoke_models.invoke_llm_model( "\n\nHuman: {input}\n\nAssistant:".format(input=llm_prompt) ,False)
|
353 |
print("Finished LLM prompt: "+st.session_state.input_index)
|
354 |
if(len(images_2)==0):
|
355 |
images_2 = images
|
|
|
325 |
if(hit["_source"]["image"]!="None"):
|
326 |
with open(parent_dirname+'/figures/'+st.session_state.input_index+"/"+hit["_source"]["raw_element_type"].split("_")[1].replace(".jpg","")+"-resized.jpg", "rb") as read_img:
|
327 |
input_encoded = base64.b64encode(read_img.read()).decode("utf8")
|
328 |
+
context.append(str(id+1) + " : Reference from a image :" + invoke_models.generate_image_captions_llm(input_encoded,question,"anthropic.claude-3-haiku-20240307-v1:0"))
|
329 |
else:
|
330 |
context.append(str(id+1) + " : Reference from a text chunk :" + hit["_source"]["processed_element"])
|
331 |
|
|
|
349 |
|
350 |
llm_prompt = prompt_template.format(context="\n".join(total_context[0:3]),question=question)
|
351 |
print("started LLM prompt: "+st.session_state.input_index)
|
352 |
+
output = invoke_models.invoke_llm_model( "\n\nHuman: {input}\n\nAssistant:".format(input=llm_prompt) ,False,"anthropic.claude-3-haiku-20240307-v1:0")
|
353 |
print("Finished LLM prompt: "+st.session_state.input_index)
|
354 |
if(len(images_2)==0):
|
355 |
images_2 = images
|
utilities/invoke_models.py
CHANGED
@@ -58,10 +58,12 @@ def invoke_model_mm(text,img):
|
|
58 |
#print(response_body)
|
59 |
return response_body.get("embedding")
|
60 |
|
61 |
-
def invoke_llm_model(input,is_stream):
|
|
|
|
|
62 |
if(is_stream == False):
|
63 |
response = bedrock_runtime_client.invoke_model(
|
64 |
-
modelId=
|
65 |
contentType = "application/json",
|
66 |
accept = "application/json",
|
67 |
performanceConfigLatency='optimized',
|
@@ -126,11 +128,12 @@ def read_from_table(file,question):
|
|
126 |
agent_res = agent.invoke(question)['output']
|
127 |
return agent_res
|
128 |
|
129 |
-
def generate_image_captions_llm(base64_string,question):
|
130 |
-
|
|
|
131 |
|
132 |
response = bedrock_runtime_client.invoke_model(
|
133 |
-
modelId=
|
134 |
contentType = "application/json",
|
135 |
accept = "application/json",
|
136 |
|
|
|
58 |
#print(response_body)
|
59 |
return response_body.get("embedding")
|
60 |
|
61 |
+
def invoke_llm_model(input,is_stream,model_id):
|
62 |
+
if(model_id is None):
|
63 |
+
model_id = "us.anthropic.claude-3-5-haiku-20241022-v1:0"
|
64 |
if(is_stream == False):
|
65 |
response = bedrock_runtime_client.invoke_model(
|
66 |
+
modelId= model_id,#"us.anthropic.claude-3-5-haiku-20241022-v1:0",#"anthropic.claude-3-5-haiku-20241022-v1:0",#"anthropic.claude-3-5-sonnet-20240620-v1:0",,
|
67 |
contentType = "application/json",
|
68 |
accept = "application/json",
|
69 |
performanceConfigLatency='optimized',
|
|
|
128 |
agent_res = agent.invoke(question)['output']
|
129 |
return agent_res
|
130 |
|
131 |
+
def generate_image_captions_llm(base64_string,question,model_id):
|
132 |
+
if(model_id is None):
|
133 |
+
model_id = "us.anthropic.claude-3-5-haiku-20241022-v1:0"
|
134 |
|
135 |
response = bedrock_runtime_client.invoke_model(
|
136 |
+
modelId= model_id,
|
137 |
contentType = "application/json",
|
138 |
accept = "application/json",
|
139 |
|