README.md

metadata

language: en
license: apache-2.0

SQFT: Low-cost Model Adaptation in Low-precision Sparse Foundation Models

• Base Model: IntelLabs/sqft-phi-3-mini-4k-50-base-gptq
• Sparsity: 50%
• Quantization: INT4 (GPTQ)
• Finetune Method: SQFT
• Finetune data: 10K instruction-following math reasoning training dataset from LLM-Adapters (math_10k.json)
• Sub-Adapter: Heuristic

Evaluation

git clone https://github.com/IntelLabs/Hardware-Aware-Automated-Machine-Learning.git haaml && cd haaml/SQFT

BASE_MODEL_PATH=IntelLabs/sqft-phi-3-mini-4k-50-base-gptq
ADAPTER_MODEL_PATH=IntelLabs/sqft-phi-3-mini-4k-50-gptq-math-heu-adapter
OUTPUT_DIR=./results
python eval/evaluate_math.py --base_model_path ${BASE_MODEL_PATH} --adapter_model_path ${ADAPTER_MODEL_PATH} --output_dir ${OUTPUT_DIR}

Refer to our repo for the environment information to run this command.