How to use from
SGLangUse Docker images
docker run --gpus all \
--shm-size 32g \
-p 30000:30000 \
-v ~/.cache/huggingface:/root/.cache/huggingface \
--env "HF_TOKEN=<secret>" \
--ipc=host \
lmsysorg/sglang:latest \
python3 -m sglang.launch_server \
--model-path "ai2sql/ai2sql_llama-2-7b" \
--host 0.0.0.0 \
--port 30000# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
-H "Content-Type: application/json" \
--data '{
"model": "ai2sql/ai2sql_llama-2-7b",
"prompt": "Once upon a time,",
"max_tokens": 512,
"temperature": 0.5
}'Quick Links
AI2sql
AI2sql is a state-of-the-art LLM for converting natural language questions to SQL queries.
Model Card: Fine-tuning Llama 2 for AI2SQL Query Generation
This model card outlines the fine-tuning of the Llama 2 model to generate SQL queries for AI2SQL tasks.
Model Details
- Original Model: NousResearch/Llama-2-7b-chat-hf
- Model Type: Large Language Model
- Fine-tuning Task: AI2SQL (SQL Query Generation)
- Fine-tuned Model Name: llama-2-7b-miniguanaco
Implementation
- Environment Requirement: GPU-supported platform with minimum 20GB RAM.
- Dependencies: accelerate==0.21.0, peft==0.4.0, bitsandbytes==0.40.2, transformers==4.31.0, trl==0.4.7
- GPU Specification: T4 or equivalent (as of 24 Aug 2023)
Training Details
- Dataset: WikiSQL
- Method: Supervised Fine-Tuning (SFT)
- Epochs: 1
- Batch Size: 4 per GPU
- Optimization: AdamW with cosine learning rate schedule
- Learning Rate: 2e-4
- Special Features:
- LoRA for efficient parameter adjustment.
- 4-bit precision model loading with BitsAndBytes.
- Gradient checkpointing and clipping.
Performance Metrics
- Accuracy: 85% (on a held-out test set from WikiSQL)
- Query Generation Time: Average of 0.5 seconds per query
- Resource Efficiency: Demonstrates 30% reduced memory usage compared to the base model
Usage and Applications
TBD
Note: The performance metrics provided here are hypothetical and for illustrative purposes only. Actual performance would depend on various factors, including the specifics of the dataset and training regimen.
- Downloads last month
- 10
Install from pip and serve model
# Install SGLang from pip: pip install sglang# Start the SGLang server: python3 -m sglang.launch_server \ --model-path "ai2sql/ai2sql_llama-2-7b" \ --host 0.0.0.0 \ --port 30000# Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "ai2sql/ai2sql_llama-2-7b", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'