Upload README.md with huggingface_hub

README.md

@@ -0,0 +1,47 @@
+---
+base_model: Qwen/Qwen3.6-27B
+tags:
+- compressed-tensors
+- 4-bit
+- dense
+- deltanet
+- thinking
+- vision
+- multimodal
+- rdna4
+- rocm
+- sglang
+- quantized
+license: apache-2.0
+---
+
+# Qwen3.6-27B AWQ 4-bit (compressed-tensors)
+
+Compressed-tensors output of GPTQ calibration of [Qwen3.6-27B](https://huggingface.co/Qwen/Qwen3.6-27B) with thinking + vision preserved.
+
+**Most users want [mattbucci/Qwen3.6-27B-AWQ](https://huggingface.co/mattbucci/Qwen3.6-27B-AWQ) instead** — same calibration weights repacked into native AWQ format. On ROCm/SGLang the native variant runs through SGLang's fused Triton AWQ GEMM kernel; the CT variant goes through the slower ROCm compressed-tensors path.
+
+This CT variant exists for: cross-engine compatibility, inspection of raw GPTQ output, and downstream re-conversion.
+
+## Model Details
+
+|  |  |
+|---|---|
+| **Base model** | [Qwen/Qwen3.6-27B](https://huggingface.co/Qwen/Qwen3.6-27B) |
+| **Architecture** | Qwen3.5 dense+DeltaNet hybrid + vision tower |
+| **Parameters** | 27B |
+| **Format** | compressed-tensors `pack-quantized` (W4A16, group_size=128) |
+| **Calibration** | GPTQ via llmcompressor, 256 samples × 1024 tokens, `thinking_vision` recipe |
+
+For ignore list and benchmark numbers see the [native variant's README](https://huggingface.co/mattbucci/Qwen3.6-27B-AWQ).
+
+## Convert to native AWQ
+
+```bash
+git clone https://github.com/mattbucci/2x-R9700-RDNA4-GFX1201-sglang-inference
+python scripts/quantize/convert_moe_ct_to_awq.py <local_path_to_this_repo> <output_dir> --group-size 128
+```
+
+## Hardware origin
+
+Calibrated on 2× AMD Radeon AI PRO R9700 (gfx1201, RDNA4) with ROCm 7.2 + SGLang v0.5.10 + RDNA4 patches.