Merge branch 'branch-25.02' of github.com:nv-morpheus/Morpheus into david-stubgen-2074

Author: dagardner-nv

ci/conda/recipes/morpheus-libs/meta.yaml

@@ -64,6 +64,7 @@ outputs:
         - cudf {{ rapids_version }}
         - cython 3.0.*
         - glog >=0.7.1,<0.8
+        - indicators=2.3
         - libcudf {{ rapids_version }}
         - librdkafka >=1.9.2,<1.10.0a0
         - mrc {{ minor_version }}

ci/conda/recipes/morpheus/meta.yaml

@@ -69,6 +69,7 @@ outputs:
         - cudf {{ rapids_version }}
         - cython 3.0.*
         - glog >=0.7.1,<0.8
+        - indicators=2.3
         - libcudf {{ rapids_version }}
         - librdkafka >=1.9.2,<1.10.0a0
         - mrc {{ minor_version }}

conda/environments/all_cuda-125_arch-x86_64.yaml

@@ -51,6 +51,7 @@ dependencies:
 - grpcio-status
 - gtest=1.14
 - gxx=12.1
+- httpx>=0.23,<0.28
 - huggingface_hub=0.20.2
 - include-what-you-use=0.20
 - indicators=2.3
@@ -136,6 +137,7 @@ dependencies:
   - faiss-cpu
   - google-search-results==2.4
   - langchain-nvidia-ai-endpoints==0.0.11
+  - langchain-openai==0.1.3
   - langchain==0.1.16
   - milvus==2.3.5
   - nemollm==0.3.5

conda/environments/examples_cuda-125_arch-x86_64.yaml

@@ -25,6 +25,7 @@ dependencies:
 - feedparser=6.0
 - grpcio
 - grpcio-status
+- httpx>=0.23,<0.28
 - huggingface_hub=0.20.2
 - jsonpatch>=1.33
 - kfp
@@ -73,6 +74,7 @@ dependencies:
   - faiss-cpu
   - google-search-results==2.4
   - langchain-nvidia-ai-endpoints==0.0.11
+  - langchain-openai==0.1.3
   - langchain==0.1.16
   - milvus==2.3.5
   - nemollm==0.3.5

dependencies.yaml

@@ -430,13 +430,14 @@ dependencies:
     common:
       - output_types: [requirements]
         packages:
-           - faiss-cpu
-           - google-search-results==2.4
-           - langchain==0.1.16
-           - langchain-nvidia-ai-endpoints==0.0.11
+           - &faiss-cpu faiss-cpu
+           - &google-search-results google-search-results==2.4
+           - &langchain langchain==0.1.16
+           - &langchain-nvidia-ai-endpoints langchain-nvidia-ai-endpoints==0.0.11
+           - &langchain-openai langchain-openai==0.1.3
            - milvus==2.3.5 # update to match pymilvus when available
            - pymilvus==2.3.6
-           - nemollm==0.3.5
+           - &nemollm nemollm==0.3.5
 
   example-dfp-prod:
     common:
@@ -487,6 +488,7 @@ dependencies:
           - &transformers transformers=4.36.2 # newer versions are incompatible with our pinned version of huggingface_hub
           - anyio>=3.7
           - arxiv=1.4
+          - httpx>=0.23,<0.28 # work-around for https://github.com/openai/openai-python/issues/1915
           - huggingface_hub=0.20.2 # work-around for https://github.com/UKPLab/sentence-transformers/issues/1762
           - jsonpatch>=1.33
           - newspaper3k=0.2
@@ -499,11 +501,12 @@ dependencies:
           - requests-toolbelt=1.0 # Transitive dep needed by nemollm, specified here to ensure we get a compatible version
           - pip
           - pip:
-            - langchain==0.1.16
-            - langchain-nvidia-ai-endpoints==0.0.11
-            - faiss-cpu
-            - google-search-results==2.4
-            - nemollm==0.3.5
+            - *faiss-cpu
+            - *google-search-results
+            - *langchain
+            - *langchain-nvidia-ai-endpoints
+            - *langchain-openai
+            - *nemollm
             - sentence-transformers==2.7 # using pip now instead of conda to avoid install of pytorch cpu
 
   model-training-tuning:

docker/Dockerfile

@@ -185,6 +185,7 @@ COPY . ./
 
 RUN --mount=type=cache,id=workspace_cache,target=/workspace/.cache,sharing=locked \
     --mount=type=cache,id=conda_pkgs,target=/opt/conda/pkgs,sharing=locked \
+    --mount=type=cache,id=pip_cache,target=/root/.cache/pip,sharing=locked \
     # Install git-lfs before running the build to avoid errors during conda build
     /opt/conda/bin/mamba install -y -n base -c conda-forge "git-lfs" &&\
     source activate base &&\
@@ -227,6 +228,7 @@ COPY ${MORPHEUS_ROOT_HOST}/conda/environments/dev_cuda-${CUDA_MAJOR_VER}${CUDA_M
 
 # Update the morpheus environment
 RUN --mount=type=cache,id=conda_pkgs,target=/opt/conda/pkgs,sharing=locked \
+    --mount=type=cache,id=pip_cache,target=/root/.cache/pip,sharing=locked \
     # Temp add channel_alias to get around conda 404 errors
     conda config --env --set channel_alias ${CONDA_CHANNEL_ALIAS} &&\
     /opt/conda/bin/conda env update --solver=libmamba -n morpheus --file conda/environments/dev.yaml &&\
@@ -263,6 +265,7 @@ COPY . ./
 
 RUN --mount=type=cache,id=workspace_cache,target=/workspace/.cache,sharing=locked \
     --mount=type=cache,id=conda_pkgs,target=/opt/conda/pkgs,sharing=locked \
+    --mount=type=cache,id=pip_cache,target=/root/.cache/pip,sharing=locked \
     # Install git-lfs before running the build to avoid errors during conda build
     /opt/conda/bin/mamba install -y -n base -c conda-forge "git-lfs" &&\
     source activate base &&\
@@ -285,6 +288,7 @@ COPY . ./
 RUN --mount=type=cache,id=workspace_cache,target=/workspace/.cache,sharing=locked \
     --mount=type=bind,from=conda_bld_morpheus,source=/opt/conda/conda-bld,target=/opt/conda/conda-bld \
     --mount=type=cache,id=conda_pkgs,target=/opt/conda/pkgs,sharing=locked \
+    --mount=type=cache,id=pip_cache,target=/root/.cache/pip,sharing=locked \
     source activate morpheus &&\
     CONDA_ALWAYS_YES=true /opt/conda/bin/mamba install -n morpheus \
         -c local \
@@ -314,6 +318,7 @@ COPY "${MORPHEUS_ROOT_HOST}/conda/environments/runtime_cuda-${CUDA_MAJOR_VER}${C
 # Mount Morpheus conda package build in `conda_bld_morpheus`
 RUN --mount=type=bind,from=conda_bld_morpheus,source=/opt/conda/conda-bld,target=/opt/conda/conda-bld \
     --mount=type=cache,id=conda_pkgs,target=/opt/conda/pkgs,sharing=locked \
+    --mount=type=cache,id=pip_cache,target=/root/.cache/pip,sharing=locked \
     # CVE-2018-20225 for the base pip, not the env one
     # conda will ignore the request to remove pip
     python -m pip uninstall -y pip && \

examples/llm/agents/common.py

@@ -18,7 +18,7 @@
 from langchain.agents import initialize_agent
 from langchain.agents import load_tools
 from langchain.agents.agent import AgentExecutor
-from langchain.llms.openai import OpenAI
+from langchain_openai import OpenAI
 
 from morpheus.config import Config
 from morpheus.pipeline.linear_pipeline import LinearPipeline

python/morpheus/morpheus/_lib/cudf_helpers.pyx

@@ -13,7 +13,10 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import itertools
+
 import cudf
+from cudf.core.column import ColumnBase
 from cudf.core.dtypes import StructDtype
 
 from libcpp.string cimport string
@@ -26,8 +29,6 @@ from pylibcudf.libcudf.table.table_view cimport table_view
 from pylibcudf.libcudf.types cimport size_type
 
 from cudf._lib.column cimport Column
-from cudf._lib.utils cimport data_from_unique_ptr
-from cudf._lib.utils cimport table_view_from_table
 
 ##### THE FOLLOWING CODE IS COPIED FROM CUDF AND SHOULD BE REMOVED WHEN UPDATING TO cudf>=24.12 #####
 # see https://github.com/rapidsai/cudf/pull/17193 for details
@@ -39,6 +40,7 @@ cimport pylibcudf.libcudf.copying as cpp_copying
 from pylibcudf.libcudf.column.column_view cimport column_view
 from libcpp.memory cimport make_unique, unique_ptr
 from pylibcudf.libcudf.scalar.scalar cimport scalar
+from pylibcudf cimport Table as plc_Table
 from cudf._lib.scalar cimport DeviceScalar
 
 # imports needed for from_column_view_with_fix
@@ -289,8 +291,35 @@ cdef public api:
         index_names = schema_infos[0:index_col_count] if index_col_count > 0 else None
         column_names = schema_infos[index_col_count:]
 
-        data, index = data_from_unique_ptr(move(table.tbl), column_names=column_names, index_names=index_names)
+        plc_table = plc_Table.from_libcudf(move(table.tbl))
 
+        if index_names is None:
+            index = None
+            data = {
+                col_name: ColumnBase.from_pylibcudf(col)
+                for col_name, col in zip(
+                    column_names, plc_table.columns()
+                )
+            }
+        else:
+            result_columns = [
+                ColumnBase.from_pylibcudf(col)
+                for col in plc_table.columns()
+            ]
+            index = cudf.Index._from_data(
+                dict(
+                    zip(
+                        index_names,
+                        result_columns[: len(index_names)],
+                    )
+                )
+            )
+            data = dict(
+                zip(
+                    column_names,
+                    result_columns[len(index_names) :],
+                )
+            )
         df = cudf.DataFrame._from_data(data, index)
 
         # Update the struct field names after the DataFrame is created
@@ -356,7 +385,13 @@ cdef public api:
 
         cdef vector[string] temp_col_names = get_column_names(table, True)
 
-        cdef table_view input_table_view = table_view_from_table(table, ignore_index=False)
+        cdef plc_Table plc_table = plc_Table(
+            [
+                col.to_pylibcudf(mode="read")
+                for col in itertools.chain(table.index._columns, table._columns)
+            ]
+        )
+        cdef table_view input_table_view = plc_table.view()
         cdef vector[string] index_names
         cdef vector[string] column_names
 

python/morpheus_llm/morpheus_llm/requirements_morpheus_llm.txt

@@ -4,6 +4,7 @@
 faiss-cpu
 google-search-results==2.4
 langchain-nvidia-ai-endpoints==0.0.11
+langchain-openai==0.1.3
 langchain==0.1.16
 milvus==2.3.5
 nemollm==0.3.5

tests/_utils/llm.py

@@ -89,13 +89,11 @@ def mk_mock_openai_response(messages: list[str]) -> mock.MagicMock:
     response = mock.MagicMock()
 
     response.choices = [_mk_mock_choice(message) for message in messages]
-    response.dict.return_value = {
-        "choices": [{
-            'message': {
-                'role': 'assistant', 'content': message
-            }
-        } for message in messages]
-    }
+
+    response_dict = {"choices": [{'message': {'role': 'assistant', 'content': message}} for message in messages]}
+
+    response.dict.return_value = response_dict
+    response.model_dump.return_value = response_dict
 
     return response