Upload README.md with huggingface_hub
Browse files
README.md
CHANGED
|
@@ -1,3 +1,20 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
# Research Test: Qwen3-Coder-Next-REAP-AWQ
|
| 2 |
|
| 3 |
> Expert-pruned and AWQ-quantized Qwen3-Coder-Next using the REAP (Robust Efficient Architecture Pruning) pipeline. 20% of MoE experts removed via diverse-calibration saliency analysis, then quantized to W4A16 for efficient inference on consumer GPUs.
|
|
|
|
| 1 |
+
---
|
| 2 |
+
language:
|
| 3 |
+
- en
|
| 4 |
+
license: other
|
| 5 |
+
tags:
|
| 6 |
+
- moe
|
| 7 |
+
- pruning
|
| 8 |
+
- awq
|
| 9 |
+
- quantized
|
| 10 |
+
- qwen3
|
| 11 |
+
- reap
|
| 12 |
+
- expert-pruning
|
| 13 |
+
base_model: Qwen/Qwen3-Coder-Next
|
| 14 |
+
pipeline_tag: text-generation
|
| 15 |
+
library_name: transformers
|
| 16 |
+
---
|
| 17 |
+
|
| 18 |
# Research Test: Qwen3-Coder-Next-REAP-AWQ
|
| 19 |
|
| 20 |
> Expert-pruned and AWQ-quantized Qwen3-Coder-Next using the REAP (Robust Efficient Architecture Pruning) pipeline. 20% of MoE experts removed via diverse-calibration saliency analysis, then quantized to W4A16 for efficient inference on consumer GPUs.
|