install flash-linear-attention (#3466)

* install flash-linear-attention

* handle prequant weights for fsdp2 and ensure loss is not zero

* fix type for cu_seqlen, uninstall causal_conv1d

* chore: lint

* uv pip uninstall doesn't need confirmation

cicd/Dockerfile-uv.jinja

@@ -33,6 +33,7 @@ RUN if [ "$NIGHTLY_BUILD" = "true" ] ; then \
 
 RUN uv pip install packaging==26.0 setuptools==75.8.0
 RUN uv pip install torchvision
+RUN uv pip uninstall causal_conv1d
 RUN if [ "$AXOLOTL_EXTRAS" != "" ] ; then \
         uv pip install --no-build-isolation -e .[deepspeed,flash-attn,ring-flash-attn,optimizers,ray,$AXOLOTL_EXTRAS] $AXOLOTL_ARGS; \
     else \

cicd/Dockerfile.jinja

@@ -33,6 +33,7 @@ RUN if [ "$NIGHTLY_BUILD" = "true" ] ; then \
     fi
 
 RUN pip install packaging==26.0 setuptools==75.8.0 psutil
+RUN pip uninstall -y causal_conv1d
 RUN if [ "$AXOLOTL_EXTRAS" != "" ] ; then \
         pip install --no-build-isolation -e .[deepspeed,flash-attn,ring-flash-attn,optimizers,ray,$AXOLOTL_EXTRAS] $AXOLOTL_ARGS; \
     else \