Merge branch 'main' into test-ci

Author: charlesbluca

.github/workflows/ci-pull-request.yml

@@ -1,11 +1,6 @@
 name: Pull Request Validation
 
 on:
-  pull_request:
-    types:
-      - opened
-      - synchronize
-      - reopened
   pull_request_target:
     types:
       - opened

.github/workflows/reusable-integration-test.yml

@@ -72,6 +72,7 @@ jobs:
           uv venv .venv
           source .venv/bin/activate
           uv pip install -e ./src -e ./api -e ./client
+          # block pulling in breaking changes in pypdfium2 v5
           uv pip install \
             opencv-python \
             llama-index-embeddings-nvidia \
@@ -80,7 +81,7 @@ jobs:
             'milvus-lite==2.4.12' \
             'nvidia-riva-client==2.20.0' \
             'unstructured-client>=0.25.9' \
-            'pypdfium2<5' \ # block pulling in breaking changes in pypdfium2 v5
+            'pypdfium2<5' \
             tritonclient \
             markitdown \
             glom
@@ -109,4 +110,4 @@ jobs:
           source .venv/bin/activate
           echo 'Running integration tests...'
           unset LD_LIBRARY_PATH
-          python -m pytest -rsx tests/integration
+          python -m pytest -rsx -m integration tests/integration

nemo_retriever/src/nemo_retriever/model/local/llama_nemotron_embed_1b_v2_embedder.py

@@ -45,7 +45,7 @@ def __post_init__(self) -> None:
         MODEL_ID = self.model_id or "nvidia/llama-3.2-nv-embedqa-1b-v2"
         dev = torch.device(self.device or ("cuda" if torch.cuda.is_available() else "cpu"))
         hf_cache_dir = self.hf_cache_dir or str(Path.home() / ".cache" / "huggingface")
-        self._tokenizer = AutoTokenizer.from_pretrained(MODEL_ID, cache_dir=hf_cache_dir)
+        self._tokenizer = AutoTokenizer.from_pretrained(MODEL_ID, trust_remote_code=True, cache_dir=hf_cache_dir)
         self._model = AutoModel.from_pretrained(MODEL_ID, trust_remote_code=True, cache_dir=hf_cache_dir)
         self._model = self._model.to(dev)
         self._model.eval()