Release/v1.22.0 tmp

.github/workflows/unit-tests.yml

@@ -14,7 +14,9 @@ jobs:
           python-version: "3.11"
 
       - name: Install package and test dependencies
-        run: pip install -e ".[test]"
+        run: |
+          pip install -e ".[test]"
+          pip install onnx_ir
 
       - name: Run unit tests
         env:

.pre-commit-config.yaml

@@ -6,7 +6,7 @@ repos:
       # Run the linter.
       - id: ruff
         types_or: [ python, pyi, jupyter ]
-        args: [ --fix ]
+        args: [ --fix, --ignore, F ]
       # Run the formatter.
       - id: ruff-format
         types_or: [ python, pyi, jupyter ]

QEfficient/__init__.py

@@ -17,7 +17,25 @@
 # ----------------------------------------------------------------------------- #
 # Placeholder for all non-transformer models registered in QEfficient
 import warnings  # noqa: I001
+import transformers
+import transformers.utils as transformers_utils
 
+try:
+    from transformers import HybridCache as _TransformersHybridCache  # noqa: F401
+except ImportError:
+    from transformers.cache_utils import DynamicCache
+
+    class HybridCache(DynamicCache):
+        pass
+
+    class HybridChunkedCache(HybridCache):
+        pass
+
+    transformers.HybridCache = HybridCache
+    transformers.HybridChunkedCache = HybridChunkedCache
+
+if not hasattr(transformers_utils, "FLAX_WEIGHTS_NAME"):
+    transformers_utils.FLAX_WEIGHTS_NAME = "flax_model.msgpack"
 import QEfficient.utils.model_registery  # noqa: F401
 from QEfficient.base import (
     QEFFAutoModel,
@@ -29,16 +47,27 @@
     QEFFCommonLoader,
 )
 from QEfficient.compile.compile_helper import compile
-from QEfficient.diffusers.pipelines.flux.pipeline_flux import QEffFluxPipeline
-from QEfficient.diffusers.pipelines.wan.pipeline_wan import QEffWanPipeline
-from QEfficient.diffusers.pipelines.wan.pipeline_wan_i2v import QEffWanImageToVideoPipeline
 from QEfficient.exporter.export_hf_to_cloud_ai_100 import qualcomm_efficient_converter
 from QEfficient.generation.text_generation_inference import cloud_ai_100_exec_kv
 from QEfficient.peft import QEffAutoPeftModelForCausalLM
 from QEfficient.transformers.transform import transform
 from QEfficient.utils import custom_format_warning
 from QEfficient.utils.logging_utils import logger
 
+try:
+    from QEfficient.diffusers.pipelines.flux.pipeline_flux import QEffFluxPipeline
+    from QEfficient.diffusers.pipelines.wan.pipeline_wan import QEffWanPipeline
+    from QEfficient.diffusers.pipelines.wan.pipeline_wan_i2v import QEffWanImageToVideoPipeline
+except Exception:
+    QEffFluxPipeline = None
+    QEffWanPipeline = None
+    QEffWanImageToVideoPipeline = None
+
+try:
+    from QEfficient.peft import QEffAutoPeftModelForCausalLM
+except Exception:
+    QEffAutoPeftModelForCausalLM = None
+
 # custom warning for the better logging experience
 warnings.formatwarning = custom_format_warning
 
@@ -58,11 +87,15 @@
     "QEFFAutoModelForSequenceClassification",
     "QEFFAutoModelForSpeechSeq2Seq",
     "QEFFCommonLoader",
-    "QEffFluxPipeline",
-    "QEffWanPipeline",
-    "QEffWanImageToVideoPipeline",
 ]
 
+if QEffFluxPipeline is not None:
+    __all__.append("QEffFluxPipeline")
+if QEffWanPipeline is not None:
+    __all__.append("QEffWanPipeline")
+if QEffWanImageToVideoPipeline is not None:
+    __all__.append("QEffWanImageToVideoPipeline")
+
 
 # Conditionally import QAIC-related modules if the SDK is installed
 __version__ = "1.22.0.dev0"