---
license: apache-2.0
language:
- en
metrics:
- bleu
- rouge
tags:
- causal-lm
- code
- cypher
- graph
- neo4j
inference: false
widget:
- text: >-
    Show me the people who have Python and Cloud skills and have been in the
    company for at least 3 years.
  example_title: Example 1
- text: What is the IMDb rating of Pulp Fiction?
  example_title: Example 2
- text: >-
    Display the first 3 users followed by 'Neo4j' who have more than 10000
    followers.
  example_title: Example 3
base_model:
- stabilityai/stable-code-instruct-3b
base_model_relation: finetune
---

## Model Description

A specialized 3B parameters model beating SOTA models such as GPT4-o at generating CYPHER.
It's a finetune of https://huggingface.co./stabilityai/stable-code-instruct-3b trained on https://github.com/neo4j-labs/text2cypher/tree/main/datasets/synthetic_opus_demodbs to generate CYPHER queries from text to query GraphDB such as neo4j.

## Usage

### Safetensors (recommended)

```python
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained("lakkeo/stable-cypher-instruct-3b", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("lakkeo/stable-cypher-instruct-3b", torch_dtype=torch.bfloat16, trust_remote_code=True)

messages = [
    {
        "role": "user",
        "content": "Show me the people who have Python and Cloud skills and have been in the company for at least 3 years."
    }
]

prompt = tokenizer.apply_chat_template(messages, add_generation_prompt=True, tokenize=False)

inputs = tokenizer([prompt], return_tensors="pt").to(model.device)

tokens = model.generate(
        **inputs,
        max_new_tokens=128,
        do_sample=True,
        top_p=0.9,
        temperature=0.2,
        pad_token_id=tokenizer.eos_token_id,
    )

outputs = tokenizer.batch_decode(tokens[:, inputs.input_ids.shape[-1]:], skip_special_tokens=False)[0]
```

### GGUF

```python
from llama_cpp import Llama

# Load the GGUF model
print("Loading model...")
model = Llama(
    model_path=r"C:\Users\John\stable-cypher-instruct-3b.Q4_K_M.gguf",
    n_ctx=512,
    n_batch=512,
    n_gpu_layers=-1,  # Use all available GPU layers
    max_tokens=128,
    top_p=0.9,
    temperature=0.2,
    verbose=False 
)

# Define your question
question = "Show me the people who have Python and Cloud skills and have been in the company for at least 3 years."

# Create the full prompt (simulating the apply_chat_template function)
full_prompt = f"<|im_start|>system\nCreate a Cypher statement to answer the following question:<|im_end|>\n<|im_start|>user\n{question}<|im_end|>\n<|im_start|>assistant\n"

# Generate response
print("Generating response...")
response = model(
    full_prompt,
    max_tokens=128,
    stop=["<|im_end|>", "<|im_start|>"],
    echo=False
)

# Extract and print the generated response
answer = response['choices'][0]['text'].strip()
print("\nQuestion:", question)
print("\nGenerated Cypher statement:")
print(answer)
```

## Performance

| Metric       | stable-code-instruct-3b |   gpt4-o   | stable-cypher-instruct-3b |
| :----------: | :---------------------: | :--------: | :-----------------------: |
| BLEU-4       |         19.07           |   32.35    |         **88.63**         |
| ROUGE-1      |         39.49           |   69.17    |         **95.09**         |
| ROUGE-2      |         24.82           |   46.97    |         **90.71**         |
| ROUGE-L      |         29.63           |   65.24    |         **91.51**         |
| Jaro-Winkler |         52.21           |   86.38    |         **95.69**         |
| Jaccard      |         25.55           |   72.80    |         **90.78**         |
| Pass@1       |         0.00            |   0.00     |         **51.80**         |

### Example
![image/png](https://cdn-uploads.huggingface.co/production/uploads/6504bb76423b46492e7f38c7/pweL4qgmFaknLBYp-CGHm.png)

### Eval params
![image/png](https://cdn-uploads.huggingface.co/production/uploads/6504bb76423b46492e7f38c7/AT80-09XrHNz-dJs9TH3M.png)

## Reproducability

This is the config file from Llama Factory :

```json
{
  "top.model_name": "Custom",
  "top.finetuning_type": "lora",
  "top.adapter_path": [],
  "top.quantization_bit": "none",
  "top.template": "default",
  "top.rope_scaling": "none",
  "top.booster": "none",
  "train.training_stage": "Supervised Fine-Tuning",
  "train.dataset_dir": "data",
  "train.dataset": [
    "cypher_opus"
  ],
  "train.learning_rate": "2e-4",
  "train.num_train_epochs": "5.0",
  "train.max_grad_norm": "1.0",
  "train.max_samples": "5000",
  "train.compute_type": "fp16",
  "train.cutoff_len": 256,
  "train.batch_size": 16,
  "train.gradient_accumulation_steps": 2,
  "train.val_size": 0.1,
  "train.lr_scheduler_type": "cosine",
  "train.logging_steps": 10,
  "train.save_steps": 100,
  "train.warmup_steps": 20,
  "train.neftune_alpha": 0,
  "train.optim": "adamw_torch",
  "train.resize_vocab": false,
  "train.packing": false,
  "train.upcast_layernorm": false,
  "train.use_llama_pro": false,
  "train.shift_attn": false,
  "train.report_to": false,
  "train.num_layer_trainable": 3,
  "train.name_module_trainable": "all",
  "train.lora_rank": 64,
  "train.lora_alpha": 64,
  "train.lora_dropout": 0.1,
  "train.loraplus_lr_ratio": 0,
  "train.create_new_adapter": false,
  "train.use_rslora": false,
  "train.use_dora": true,
  "train.lora_target": "",
  "train.additional_target": "",
  "train.dpo_beta": 0.1,
  "train.dpo_ftx": 0,
  "train.orpo_beta": 0.1,
  "train.reward_model": null,
  "train.use_galore": false,
  "train.galore_rank": 16,
  "train.galore_update_interval": 200,
  "train.galore_scale": 0.25,
  "train.galore_target": "all"
}
```

I used llama.cpp to merge the LoRa and generate the quants.

The progress achieved from the base model is significant but you will still need to finetune on your company's syntax and entities.
I've been tickering with the training parameters for a few batches of training but there is room for improvements.
I'm open to the idea of making a full tutorial if there is enough interest in this project.