Instructions to use GreenBitAI/DeepSeek-R1-0528-Qwen3-8B-layer-mix-bpw-4.0-mlx with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use GreenBitAI/DeepSeek-R1-0528-Qwen3-8B-layer-mix-bpw-4.0-mlx with MLX:
# Download the model from the Hub pip install huggingface_hub[hf_xet] huggingface-cli download --local-dir DeepSeek-R1-0528-Qwen3-8B-layer-mix-bpw-4.0-mlx GreenBitAI/DeepSeek-R1-0528-Qwen3-8B-layer-mix-bpw-4.0-mlx
- Notebooks
- Google Colab
- Kaggle
- Local Apps
- LM Studio
| { | |
| "measurement": { | |
| "model.layers.0": { | |
| "accuracy": 0.9141184650361538, | |
| "total_bits": 585302016.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "2": 64 | |
| }, | |
| "bits": [ | |
| 2 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "2": 64 | |
| }, | |
| "bits": [ | |
| 2 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "2": 64 | |
| }, | |
| "bits": [ | |
| 2 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.1": { | |
| "accuracy": 0.9733483912423253, | |
| "total_bits": 832045056.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.2": { | |
| "accuracy": 0.9746541411150247, | |
| "total_bits": 832045056.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.3": { | |
| "accuracy": 0.9737034396966919, | |
| "total_bits": 832045056.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.4": { | |
| "accuracy": 0.9715883855242282, | |
| "total_bits": 832045056.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.5": { | |
| "accuracy": 0.9714305921224877, | |
| "total_bits": 832045056.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.6": { | |
| "accuracy": 0.9588503530249, | |
| "total_bits": 832045056.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.7": { | |
| "accuracy": 0.9866362753673457, | |
| "total_bits": 786825216.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.8": { | |
| "accuracy": 0.9876651579979807, | |
| "total_bits": 786825216.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.9": { | |
| "accuracy": 0.9863171711331233, | |
| "total_bits": 786825216.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.10": { | |
| "accuracy": 0.9849221103359014, | |
| "total_bits": 786825216.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.11": { | |
| "accuracy": 0.9851903039962053, | |
| "total_bits": 786825216.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.12": { | |
| "accuracy": 0.9855111946817487, | |
| "total_bits": 786825216.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.13": { | |
| "accuracy": 0.9848670585197397, | |
| "total_bits": 786825216.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.14": { | |
| "accuracy": 0.9842190744238906, | |
| "total_bits": 786825216.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.15": { | |
| "accuracy": 0.9836641177535057, | |
| "total_bits": 786825216.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.16": { | |
| "accuracy": 0.9555211160331964, | |
| "total_bits": 593362944.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "2": 64 | |
| }, | |
| "bits": [ | |
| 2 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "2": 64 | |
| }, | |
| "bits": [ | |
| 2 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.17": { | |
| "accuracy": 0.9688245259458199, | |
| "total_bits": 593362944.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "2": 64 | |
| }, | |
| "bits": [ | |
| 2 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "2": 64 | |
| }, | |
| "bits": [ | |
| 2 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.18": { | |
| "accuracy": 0.9875572984456085, | |
| "total_bits": 786825216.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.19": { | |
| "accuracy": 0.9847902545006946, | |
| "total_bits": 786825216.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.20": { | |
| "accuracy": 0.9847954767756164, | |
| "total_bits": 786825216.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.21": { | |
| "accuracy": 0.9807497714064084, | |
| "total_bits": 786825216.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.22": { | |
| "accuracy": 0.9790314845740795, | |
| "total_bits": 786825216.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.23": { | |
| "accuracy": 0.97751333063934, | |
| "total_bits": 786825216.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.24": { | |
| "accuracy": 0.9759847359964624, | |
| "total_bits": 786825216.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.25": { | |
| "accuracy": 0.9775375220924616, | |
| "total_bits": 786825216.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.26": { | |
| "accuracy": 0.9779737029457465, | |
| "total_bits": 786825216.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.27": { | |
| "accuracy": 0.9816575867007487, | |
| "total_bits": 832045056.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.28": { | |
| "accuracy": 0.9733700336655602, | |
| "total_bits": 786825216.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.29": { | |
| "accuracy": 0.9744192563230172, | |
| "total_bits": 786825216.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.30": { | |
| "accuracy": 0.9782909244531766, | |
| "total_bits": 832045056.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.31": { | |
| "accuracy": 0.9784003929235041, | |
| "total_bits": 832045056.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.32": { | |
| "accuracy": 0.9765235031954944, | |
| "total_bits": 832045056.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.33": { | |
| "accuracy": 0.978374652331695, | |
| "total_bits": 832045056.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 32 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.34": { | |
| "accuracy": 0.9910124130547047, | |
| "total_bits": 585302016.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "2": 64 | |
| }, | |
| "bits": [ | |
| 2 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "2": 64 | |
| }, | |
| "bits": [ | |
| 2 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "2": 64 | |
| }, | |
| "bits": [ | |
| 2 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.35": { | |
| "accuracy": 0.990355659276247, | |
| "total_bits": 520814592.0, | |
| "o_proj": { | |
| "group_size": { | |
| "2": 64 | |
| }, | |
| "bits": [ | |
| 2 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "q_proj": { | |
| "group_size": { | |
| "2": 64 | |
| }, | |
| "bits": [ | |
| 2 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "2": 64 | |
| }, | |
| "bits": [ | |
| 2 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "2": 64 | |
| }, | |
| "bits": [ | |
| 2 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "2": 64 | |
| }, | |
| "bits": [ | |
| 2 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| } | |
| } | |
| } |