Fix ZeroGPU compat: torch 2.6.0 -> 2.8.0, update causal_conv1d wheel

Dockerfile

@@ -3,8 +3,9 @@
 # Mirrors the HF Space (ZeroGPU) stack so we can validate the inference path
 # locally without the host's CUDA-13/Python-3.12 incompatibilities:
 #   - Ubuntu 22.04 ships Python 3.10 (= Space's Python)
-#   - cu12.4 runtime libs (pre-built torch 2.6.0 cu12 wheels target this)
-#   - causal_conv1d wheel for cu12torch2.6 / cp310 / cxx11abi=FALSE
+#   - cu12.8 runtime libs (pre-built torch 2.8.0 cu12 wheels target this)
+#   - causal_conv1d wheel for cu12torch2.8 / cp310 / cxx11abi=TRUE
+#     (torch >=2.8 ships with cxx11abi=TRUE by default)
 #
 # Build:
 #   docker build -t qwen3.5-ocr-jp-2b-local spaces/Qwen3.5-ocr-jp-2b/
@@ -15,7 +16,7 @@
 #     qwen3.5-ocr-jp-2b-local
 # Open http://localhost:7860
 
-FROM nvidia/cuda:12.4.1-devel-ubuntu22.04
+FROM nvidia/cuda:12.8.1-devel-ubuntu22.04
 
 # `devel` (not `runtime`) is required: triton JIT-compiles its CUDA driver
 # bindings (`backends/nvidia/driver.c` → `cuda_utils.so`) at first invocation,
@@ -39,18 +40,18 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
 
 WORKDIR /app
 
-# Pin torch+vision from the cu12.4 index so the wheel ABI matches our cuda image.
+# Pin torch+vision from the cu12.8 index so the wheel ABI matches our cuda image.
 RUN pip install --no-cache-dir \
-        --extra-index-url https://download.pytorch.org/whl/cu124 \
-        torch==2.6.0 torchvision==0.21.0
+        --extra-index-url https://download.pytorch.org/whl/cu128 \
+        torch==2.8.0 torchvision==0.23.0
 
 COPY requirements.txt .
 RUN pip install --no-cache-dir -r requirements.txt
 
-# Pre-install causal_conv1d for fla fast path (Python 3.10 / cu12 / torch 2.6 / cxx11abi=FALSE).
+# Pre-install causal_conv1d for fla fast path (Python 3.10 / cu12 / torch 2.8 / cxx11abi=TRUE).
 # app.py also tries to install at runtime; this pre-install just makes startup faster.
 RUN pip install --no-cache-dir --no-deps \
-        https://github.com/Dao-AILab/causal-conv1d/releases/download/v1.5.0.post8/causal_conv1d-1.5.0.post8+cu12torch2.6cxx11abiFALSE-cp310-cp310-linux_x86_64.whl
+        https://github.com/Dao-AILab/causal-conv1d/releases/download/v1.6.2.post1/causal_conv1d-1.6.2.post1+cu12torch2.8cxx11abiTRUE-cp310-cp310-linux_x86_64.whl
 
 COPY app.py .