Update README.md

Signed-off-by: jinjieyuan <[email protected]>

README.md

@@ -3,26 +3,27 @@ language: en
 license: apache-2.0
 ---
 
-# Shears Model Card: Shears-llama-13b-50-math-heuristic
+# Shears Model Card: shears-llama-13b-50-math-heuristic
 
-Fine tuned model on LLaMA-13B with some math reasoning datasets using Shears.
+The heuristic subnetwork discovered from the [super-network](https://huggingface.co/IntelLabs/shears-llama-13b-50-math-super) fine-tuned on LLaMA-13B with some math reasoning datasets using Shears.
 
 ## Model Details
 
 ### Information
 
-- **Model name:** Shears-llama-13b-50-math-heuristic
+- **Model name:** shears-llama-13b-50-math-heuristic
 - **Base model:** [LLaMA-13b](https://huggingface.co/yahma/llama-13b-hf)
 - **Sparsity:** 50%
 - **Domain:** Math
 - **Subnetwork version:** Heuristic
+- **NNCF Configuration:** [nncf_shears_llama_13b_sparsity50.json](https://github.com/IntelLabs/Hardware-Aware-Automated-Machine-Learning/tree/main/Shears/nncf_config/unified_math/nncf_shears_llama_13b_sparsity50.json)
 
 ### Adapter Configuration
 
 - **LoRA rank:** 32 (24 in the heuristic subnetwork)
 - **LoRA alpha:** 64
 - **LoRA target modules:** q_proj, k_proj, v_proj, up_proj, down_proj
-- **LoRA rank search space:** [32, 24, 16]
+- **LoRA rank search space:** [32, 24, 16] (for each LoRA module)
 
 ### Training Hyperparameters
 
@@ -40,6 +41,12 @@ Unified math reasoning dataset: [math_10k.json](https://github.com/AGI-Edgerunne
 
 ## How to use
 
+Use our modified PEFT library (apply [patch](https://github.com/IntelLabs/Hardware-Aware-Automated-Machine-Learning/tree/main/Shears/patches/peft-modifications-for-shears-inference-usage.patch)):
+```bash
+git clone https://github.com/huggingface/peft.git
+pushd peft && git checkout v0.5.0 && git apply --ignore-space-change --ignore-whitespace peft-modifications-for-shears-inference-usage.patch && pip install -e . && popd
+```
+
 ```python
 import torch
 from peft import PeftModel