24b1051 Week04 Track1 20% Submission02
Public compressed checkpoint for the CS6013 Efficient AI Week04 Track 1 submission.
- Base model: Qwen/Qwen3-4B-Instruct-2507
- Base revision: cdbee75f17c01a7cc42f958dc650907174af0554
- Hugging Face repository: https://huggingface.co/ParthivMEV/24b1051-Week04-Compression-20-Submission02
- Method: packed groupwise NF3 with selective packed W4, FP16 scales, group size 128
- BF16 baseline weights: 8044982000 bytes
- Compressed weights: 1596568736 bytes
- Ratio: 19.84552279%
- Reconstruction code: https://github.com/Parthiv-MEV/CS6013/tree/main/24b1051/Week04/Compression_20/Submission02
Use the submission's dequantize_to_bf16.py to reconstruct a standard BF16 Hugging Face checkpoint.
- Downloads last month
- 3
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support
Model tree for ParthivMEV/24b1051-Week04-Compression-20-Submission02
Base model
Qwen/Qwen3-4B-Instruct-2507