Pipelines working with granite (#154)

- New templates - Updated images - Updated langchain - Added missing image bucket env var
rh-aiservices-bu · Sep 30, 2024 · 057f210 · 057f210
1 parent 75990d1
commit 057f210
Show file tree

Hide file tree

Showing 15 changed files with 114 additions and 85 deletions.
diff --git a/lab-materials/03/06/llm_usage.py b/lab-materials/03/06/llm_usage.py
@@ -1,11 +1,9 @@
-from langchain.llms import VLLMOpenAI
-from langchain.chains.combine_documents.stuff import StuffDocumentsChain
-from langchain.chains import LLMChain
-from langchain.prompts import PromptTemplate
 from langchain.evaluation import load_evaluator
-from langchain.embeddings import HuggingFaceEmbeddings
+from langchain_huggingface import HuggingFaceEmbeddings
+from langchain.prompts import PromptTemplate
+from langchain_community.llms import VLLMOpenAI
 
-INFERENCE_SERVER_URL = "http://llm.ic-shared-llm.svc.cluster.local:8000"
+INFERENCE_SERVER_URL = "http://granite-7b-instruct-predictor.ic-shared-llm.svc.cluster.local:8080"
 MAX_NEW_TOKENS = 512
 TOP_P = 0.95
 TEMPERATURE = 0.01
@@ -15,7 +13,7 @@ def infer_with_template(input_text, template):
     llm = VLLMOpenAI(
         openai_api_key="EMPTY",
         openai_api_base= f"{INFERENCE_SERVER_URL}/v1",
-        model_name="mistralai/Mistral-7B-Instruct-v0.2",
+        model_name="granite-7b-instruct",
         max_tokens=MAX_NEW_TOKENS,
         top_p=TOP_P,
         temperature=TEMPERATURE,
@@ -24,11 +22,11 @@ def infer_with_template(input_text, template):
         verbose=False,
     )
 
-    PROMPT = PromptTemplate.from_template(template)
+    prompt = PromptTemplate(input_variables=["car_claim"], template=template)
 
-    llm_chain = LLMChain(llm=llm, prompt=PROMPT)
+    conversation = prompt | llm
 
-    return llm_chain.run(input_text)
+    return conversation.invoke(input=input_text)
 
 def similarity_metric(predicted_text, reference_text):
     embedding_model = HuggingFaceEmbeddings()

diff --git a/lab-materials/03/06/requirements.txt b/lab-materials/03/06/requirements.txt
@@ -1,4 +1,8 @@
-langchain==0.0.340
+langchain==0.3.1
+langchain-community==0.3.1
+langchain-core==0.3.6
+langchain-huggingface==0.1.0
+langchain-text-splitters==0.3.0
 text_generation==0.6.1
-sentence_transformers==2.2.2
-openai==1.13.3
+sentence-transformers==3.1.1
+openai==1.47.1
diff --git a/lab-materials/03/06/summary_template.txt b/lab-materials/03/06/summary_template.txt
@@ -1,5 +1,7 @@
 <s>[INST] <<SYS>>
 Summarize the below car insurance claim:
 <</SYS>>
+### TEXT:
 {car_claim}
+### ANSWER:
 [/INST]
diff --git a/lab-materials/03/06/test_responsetime.py b/lab-materials/03/06/test_responsetime.py
@@ -13,7 +13,9 @@ def test_responsetime():
     TEMPLATE = """<s>[INST] <<SYS>>
 Answer below truthfully and in less than 10 words:
 <</SYS>>
+### QUESTION:
 {silly_question}
+### ANSWER:
 [/INST]"""
 
     start = time.perf_counter()

diff --git a/lab-materials/03/06/test_security.py b/lab-materials/03/06/test_security.py
@@ -21,6 +21,6 @@ def test_security(endpoint, expected_id,):
         }, f)
 
 if __name__ == '__main__':
-    info_endpoint = "http://llm.ic-shared-llm.svc.cluster.local:8000" + "/v1/models"
+    info_endpoint = "http://granite-7b-instruct-predictor.ic-shared-llm.svc.cluster.local:8080" + "/v1/models"
     expected_id = "modelperm" #This is just for the demo, in a real scenario you would input the model id here
     test_security(info_endpoint, expected_id)
diff --git a/lab-materials/05/05-05/container/requirements.txt b/lab-materials/05/05-05/container/requirements.txt
@@ -1,8 +1,12 @@
-langchain==0.0.340
+langchain==0.3.1
+langchain-community==0.3.1
+langchain-core==0.3.6
+langchain-huggingface==0.1.0
+langchain-text-splitters==0.3.0
 text_generation==0.6.1
-sentence_transformers==2.2.2
+sentence-transformers==3.1.1
+openai==1.47.1
 opencv-python-headless==4.8.1.78
 python-dotenv==1.0.0
 psycopg2-binary==2.9.9
-boto3==1.26.165
-openai==1.13.3
+boto3==1.26.165
diff --git a/lab-materials/05/05-05/llm_usage.py b/lab-materials/05/05-05/llm_usage.py
@@ -1,13 +1,10 @@
-import os
-
-from langchain.llms import VLLMOpenAI
-from langchain.chains.combine_documents.stuff import StuffDocumentsChain
-from langchain.chains import LLMChain
-from langchain.prompts import PromptTemplate
 from langchain.evaluation import load_evaluator
-from langchain.embeddings import HuggingFaceEmbeddings
+from langchain_huggingface import HuggingFaceEmbeddings
+from langchain.prompts import PromptTemplate
+from langchain_community.llms import VLLMOpenAI
+
 
-INFERENCE_SERVER_URL = "http://llm.ic-shared-llm.svc.cluster.local:8000"
+INFERENCE_SERVER_URL = "http://granite-7b-instruct-predictor.ic-shared-llm.svc.cluster.local:8080"
 MAX_NEW_TOKENS = 512
 TOP_P = 0.95
 TEMPERATURE = 0.01
@@ -17,7 +14,7 @@ def infer_with_template(input_text, template):
     llm = VLLMOpenAI(
         openai_api_key="EMPTY",
         openai_api_base= f"{INFERENCE_SERVER_URL}/v1",
-        model_name="mistralai/Mistral-7B-Instruct-v0.2",
+        model_name="granite-7b-instruct",
         max_tokens=MAX_NEW_TOKENS,
         top_p=TOP_P,
         temperature=TEMPERATURE,
@@ -26,11 +23,11 @@ def infer_with_template(input_text, template):
         verbose=False,
     )
 
-    PROMPT = PromptTemplate.from_template(template)
+    prompt = PromptTemplate(input_variables=["car_claim"], template=template)
 
-    llm_chain = LLMChain(llm=llm, prompt=PROMPT)
+    conversation = prompt | llm
 
-    return llm_chain.run(input_text)
+    return conversation.invoke(input=input_text)
 
 def similarity_metric(predicted_text, reference_text):
     embedding_model = HuggingFaceEmbeddings()

diff --git a/lab-materials/05/05-05/pipeline_components.py b/lab-materials/05/05-05/pipeline_components.py
@@ -12,7 +12,7 @@
 @dsl.container_component
 def initialize():
     return dsl.ContainerSpec(
-        image='quay.io/rh-aiservices-bu/rhoai-lab-insurance-claim-processing-pipeline:1.1', 
+        image='quay.io/rh-aiservices-bu/rhoai-lab-insurance-claim-processing-pipeline:1.2', 
         command=[
             'sh',
             '-c', 
@@ -30,17 +30,17 @@ def initialize():
 @dsl.container_component
 def get_claims(claim_ids: int):
     return dsl.ContainerSpec(
-        image='quay.io/rh-aiservices-bu/rhoai-lab-insurance-claim-processing-pipeline:1.1', 
+        image='quay.io/rh-aiservices-bu/rhoai-lab-insurance-claim-processing-pipeline:1.2', 
         command=[
             'sh',
             '-c',
             '''claim_id="$0"
             export claim_id=$claim_id
             export POSTGRES_HOST=claimdb.$NAMESPACE.svc.cluster.local
+            export IMAGES_BUCKET=$NAMESPACE
             cd /shared-data
             cd parasol-insurance/lab-materials/05/05-05
             python get_claims.py
-            ls
             ''',
         ],
         args=[claim_ids]
@@ -49,11 +49,12 @@ def get_claims(claim_ids: int):
 @dsl.container_component
 def get_accident_time():
     return dsl.ContainerSpec(
-        image='quay.io/rh-aiservices-bu/rhoai-lab-insurance-claim-processing-pipeline:1.1', 
+        image='quay.io/rh-aiservices-bu/rhoai-lab-insurance-claim-processing-pipeline:1.2', 
         command=[
             'sh',
             '-c',
             '''export POSTGRES_HOST=claimdb.$NAMESPACE.svc.cluster.local
+            export IMAGES_BUCKET=$NAMESPACE
             cd /shared-data
             cd parasol-insurance/lab-materials/05/05-05
             python get_accident_time.py
@@ -65,11 +66,12 @@ def get_accident_time():
 @dsl.container_component
 def get_location():
     return dsl.ContainerSpec(
-        image='quay.io/rh-aiservices-bu/rhoai-lab-insurance-claim-processing-pipeline:1.1', 
+        image='quay.io/rh-aiservices-bu/rhoai-lab-insurance-claim-processing-pipeline:1.2', 
         command=[
             'sh',
             '-c',
             '''export POSTGRES_HOST=claimdb.$NAMESPACE.svc.cluster.local
+            export IMAGES_BUCKET=$NAMESPACE
             cd /shared-data
             cd parasol-insurance/lab-materials/05/05-05
             python get_location.py
@@ -81,11 +83,12 @@ def get_location():
 @dsl.container_component
 def get_sentiment():
     return dsl.ContainerSpec(
-        image='quay.io/rh-aiservices-bu/rhoai-lab-insurance-claim-processing-pipeline:1.1', 
+        image='quay.io/rh-aiservices-bu/rhoai-lab-insurance-claim-processing-pipeline:1.2', 
         command=[
             'sh',
             '-c',
             '''export POSTGRES_HOST=claimdb.$NAMESPACE.svc.cluster.local
+            export IMAGES_BUCKET=$NAMESPACE
             cd /shared-data
             cd parasol-insurance/lab-materials/05/05-05
             python get_sentiment.py
@@ -97,13 +100,14 @@ def get_sentiment():
 @dsl.container_component
 def detect_objects(detection_endpoint: str):
     return dsl.ContainerSpec(
-        image='quay.io/rh-aiservices-bu/rhoai-lab-insurance-claim-processing-pipeline:1.1', 
+        image='quay.io/rh-aiservices-bu/rhoai-lab-insurance-claim-processing-pipeline:1.2', 
         command=[
             'sh',
             '-c',
             '''detection_endpoint="$0"
             export detection_endpoint=$detection_endpoint
             export POSTGRES_HOST=claimdb.$NAMESPACE.svc.cluster.local
+            export IMAGES_BUCKET=$NAMESPACE
             cd /shared-data
             cd parasol-insurance/lab-materials/05/05-05
             python detect_objects.py
@@ -115,11 +119,12 @@ def detect_objects(detection_endpoint: str):
 @dsl.container_component
 def summarize_text():
     return dsl.ContainerSpec(
-        image='quay.io/rh-aiservices-bu/rhoai-lab-insurance-claim-processing-pipeline:1.1', 
+        image='quay.io/rh-aiservices-bu/rhoai-lab-insurance-claim-processing-pipeline:1.2', 
         command=[
             'sh',
             '-c',
             '''export POSTGRES_HOST=claimdb.$NAMESPACE.svc.cluster.local
+            export IMAGES_BUCKET=$NAMESPACE
             cd /shared-data
             cd parasol-insurance/lab-materials/05/05-05
             python summarize_text.py