text2cypher-smollm2/README.md

library_name, base_model, tags, language, license, datasets

library_name	base_model	tags	language	license	datasets
transformers	HuggingFaceTB/SmolLM2-135M-Instruct	text2cypher cypher graph fine-tuned	en	apache-2.0	RomanTeucher/text2cypher-curated

Text2Cypher — SmolLM2-135M Fine-tuned

A fine-tuned version of SmolLM2-135M-Instruct that generates Cypher queries from natural language questions and a graph schema.

Model Details

• Base model: HuggingFaceTB/SmolLM2-135M-Instruct
• Model type: Causal Language Model
• Language: English
• License: Apache 2.0
• Finetuned by: Anugya Sahu

Training Data

• Dataset: RomanTeucher/text2cypher-curated
• 1000 training samples, 75 validation, 50 test
• Each sample contains a graph schema, a natural language question, and a target Cypher query

How to Use

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("Anugya/text2cypher-smollm2")
tokenizer = AutoTokenizer.from_pretrained("Anugya/text2cypher-smollm2")
tokenizer.pad_token = tokenizer.eos_token

schema = "Movie {title, year}, Person {name}, (Person)-[:DIRECTED]->(Movie)"
question = "Which movies did Christopher Nolan direct before 2010?"

prompt = f"""### Schema:
{schema}

### Question:
{question}

### Cypher:"""

inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=128, do_sample=False)
generated = outputs[0][inputs["input_ids"].shape[1]:]
print(tokenizer.decode(generated, skip_special_tokens=True))

Training Details

• Full fine-tune — all weights updated, no LoRA
• Epochs: 3
• Learning rate: 2e-4
• Batch size: 4
• Max token length: 256
• Hardware: CPU (Apple M-series)
• Precision: float32

Evaluation

Evaluated on 50 test samples using:

• Exact Match — strict comparison after lowercasing and stripping
• Token F1 — token overlap between prediction and ground truth

Limitations

• 135M parameter model — generates Cypher that looks right but often isn't
• No query execution validation against a real Neo4j database
• May struggle with complex schemas or multi-hop queries
• Trained on CPU with limited epochs — larger training would improve results