Remove explicit Torchvision dep and cleanup docs

Author: jmanning-stackav

README.md

@@ -32,6 +32,8 @@ Each operation is complete with a PyTorch-only reference implementation (and som
   - GEMM
     - Mixed-precision
     - Scaled
+- Vision
+  - Non-Max Suppression (NMS)
 - vLLM
   - KV cache operations
     - Copy blocks
@@ -99,6 +101,7 @@ We were inspired by and leverage components of the following libraries:
 
 - [bitsandbytes](https://github.com/bitsandbytes-foundation/bitsandbytes)
 - [GemLite](https://github.com/mobiusml/gemlite)
+- [Torchvision](https://github.com/pytorch/vision)
 - [vLLM](https://github.com/vllm-project/vllm)
 
 ## License

docs/getting_started/developer_environment.md

@@ -95,3 +95,14 @@ In order to use them, you can install vLLM (`pip install vllm`) and set the envi
 pip install vllm==0.9.1
 CONCH_ENABLE_VLLM=1 python benchmarks/paged_attention_benchmark.py
 ```
+
+### Optional: Benchmarking against Torchvision
+
+Some unit tests/benchmarks allow comparison to CUDA implementations of operations from Torchvision (rather than PyTorch-reference implementations).
+In order to use them, you can install Torchvision (`pip install torchvision`) and set the environment variable `CONCH_ENABLE_TORCHVISION=1`.
+
+```bash
+# Note: add `--extra-index-url https://download.pytorch.org/whl/rocm6.2.4` on ROCm
+pip install torchvision==0.22
+CONCH_ENABLE_TORCHVISION=1 python benchmarks/nms_benchmark.py
+```

setup.py

@@ -6,7 +6,6 @@
 from setuptools import setup
 
 _TORCH_VERSION: Final = "2.7.0"
-_TORCHVISION_VERSION: Final = "0.22"
 _TRITON_VERSION: Final = "3.3.0"
 
 _REQUIREMENTS: Final = [
@@ -17,23 +16,20 @@
 # --extra-index-url https://download.pytorch.org/whl/cpu
 _DEFAULT_PLATFORM_REQUIREMENTS: Final = [
     f"torch=={_TORCH_VERSION}",
-    f"torchvision=={_TORCHVISION_VERSION}",
     f"triton>={_TRITON_VERSION}",
 ]
 
 # For ROCm:
 # --extra-index-url https://download.pytorch.org/whl/rocm6.2.4
 _ROCM_PLATFORM_REQUIREMENTS: Final = [
     f"torch=={_TORCH_VERSION}+rocm6.2.4",
-    f"torchvision=={_TORCHVISION_VERSION}+rocm6.2.4",
     f"pytorch-triton-rocm>={_TRITON_VERSION}",
 ]
 
 # For XPU:
 # --extra-index-url https://download.pytorch.org/whl/xpu
 _XPU_PLATFORM_REQUIREMENTS: Final = [
     f"torch=={_TORCH_VERSION}+xpu",
-    f"torchvision=={_TORCHVISION_VERSION}+xpu",
     f"pytorch-triton-xpu>={_TRITON_VERSION}",
 ]