neo4j-labs
diff --git a/‎finetuning/unsloth-llama3/README.md
Lines changed: 68 additions & 2 deletions b/‎finetuning/unsloth-llama3/README.md
Lines changed: 68 additions & 2 deletions
@@ -1,7 +1,73 @@
 # Finetuning Llama3 using Unsloth
 
-This notebook will guide you step by step on how to train LLMs for text2cypher purposes (in this case, llama3, but you could use other models too).
+We have two notebooks here:
+
+## Using simple prompt template
+
+* Filename: `llama3_text2cypher_simple.ipynb`
+* Contributed by: [Geraldus Wilsen](https://github.com/projectwilsen/)
+* Dataset: synthetic_gpt4turbo_demodbs
+* Originally published: https://github.com/projectwilsen/KnowledgeGraphLLM
+
+This notebook uses simple prompt completion template to finetune Llama3 to construct Cypher statements on a single database (recommendations).
 
 For more information, you could watch this video tutorial: https://www.youtube.com/watch?v=7VU-xWJ39ng
 
-Explore other use case and tutorials: https://github.com/projectwilsen/KnowledgeGraphLLM
+## Using chat prompt template
+
+* Filename: `llama3_text2cypher_chat.ipynb`
+* Contributed by: [Tomaz Bratanic](https://github.com/tomasonjo)
+* Dataset: synthetic_gpt4o_demodbs
+* HuggingFace model: https://huggingface.co/tomasonjo/text2cypher-demo-16bit
+* Ollama model: https://ollama.com/tomasonjo/llama3-text2cypher-demo
+
+This notebook uses chat prompt template (system,user,assistant) to finetune Llama3 to construct Cypher statements on 16 different graph databases available on demo server.
+
+You can load use it in LangChain using the following code.
+First load the model using Ollama and install dependencies.
+
+```bash
+ollama pull tomasonjo/llama3-text2cypher-demo
+pip install langchain langchain-community neo4j
+```
+
+Now you can use the following code to generate Cypher statements with LangChain:
+
+```python
+from langchain_community.graphs import Neo4jGraph
+from langchain_community.chat_models import ChatOllama
+from langchain_core.prompts import ChatPromptTemplate
+
+DEMO_URL = "neo4j+s://demo.neo4jlabs.com"
+DATABASE = "recommendations"
+
+graph = Neo4jGraph(
+    url=DEMO_URL,
+    database=DATABASE,
+    username=DATABASE,
+    password=DATABASE,
+    enhanced_schema=True,
+    sanitize=True,
+)
+llm = ChatOllama(model="tomasonjo/llama3-text2cypher-demo")
+prompt = ChatPromptTemplate.from_messages(
+    [
+        (
+            "system",
+            "Given an input question, convert it to a Cypher query. No pre-amble.",
+        ),
+        (
+            "human",
+            (
+                "Based on the Neo4j graph schema below, write a Cypher query that would answer the user's question: "
+                "\n{schema} \nQuestion: {question} \nCypher query:"
+            ),
+        ),
+    ]
+)
+chain = prompt | llm
+
+question = "How many movies did Tom Hanks play in?"
+response = chain.invoke({"question": question, "schema": graph.schema})
+print(response.content)
+```