metadata
license: apache-2.0
base_model:
- Pinkstackorg/Qwen3-Coder-pruned-20B-A3B
This is a 4bit bnb quant of the original model.
This is a layer-wise pruned variant of cerebras/Qwen3-Coder-REAP-25B-A3B resulting in a ~20B model with ~3B active parameters. It has not been fine-tuned yet.
Prune info: Original model: 48 layers New model: 38 Layers
result: Similar model, it MUST be fine-tuned before usage as current performance is non-ideal. While it can absolutely be used as is, fine-tuning is needed.