Instructions to use Graph-COM/KVEraser with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- PEFT
How to use Graph-COM/KVEraser with PEFT:
from peft import PeftModel from transformers import AutoModel base_model = AutoModel.from_pretrained("Qwen/Qwen3-8B") model = PeftModel.from_pretrained(base_model, "Graph-COM/KVEraser") - Notebooks
- Google Colab
- Kaggle
Download training_config.json from Graph-COM/KVEraser: direct link, hf CLI and curl.
- Browser
- Download file 463 Bytes
-
https://huggingface.co/Graph-COM/KVEraser/resolve/main/training_config.json
- Command line
-
hf download hf://Graph-COM/KVEraser/training_config.json
-
curl -L -o training_config.json https://huggingface.co/Graph-COM/KVEraser/resolve/main/training_config.json
463 Bytes
| { | |
| "base_model": "Qwen/Qwen3-8B", | |
| "base_revision": "b968826d9c46dd6066d109eabc6255188de91218", | |
| "data_format": "qwen3_8b/v1", | |
| "optimizer": "AdamW", | |
| "seed": 42, | |
| "epochs": 1, | |
| "lr": 1e-05, | |
| "weight_decay": 0.01, | |
| "max_grad_norm": 1.0, | |
| "pretrain_train_samples": 72000, | |
| "pretrain_validation_samples": 8000, | |
| "pretrain_grad_accum_steps": 8, | |
| "finetune_train_samples": 11491, | |
| "finetune_validation_samples": 1642, | |
| "finetune_grad_accum_steps": 1 | |
| } | |