removing unnecessary dependencies when running in hf, using conditionals to avoid loading npu libraries in hf

Dockerfile

@@ -5,7 +5,7 @@ RUN source venv/bin/activate
 RUN ls -a
 RUN mkdir app
 COPY . app
-RUN mkdir -p app/deps
 WORKDIR app
-RUN pip install -r requirements.txt
+RUN rm deps/*.whl
+RUN pip install -r requirements-base.txt
 ENTRYPOINT streamlit run app.py

repository/repository.py

@@ -1,7 +1,8 @@
 from pathlib import Path
+from utils.env_utils import in_hf
 
-from repository.intel_npu import IntelNpuRepository
-# from repository.intel_npu import IntelNpuRepository
+if not in_hf():
+ from repository.intel_npu import IntelNpuRepository
 from repository.ollama import OllamaRepository
 from repository.ondemand import OndemandRepository
 from repository.repository_abc import Model
@@ -9,13 +10,14 @@ from repository.testing_repo import TestingRepository
 
 
 def get_repository(implementation: str, model: Model, system_msg: str = None, log_to_file: Path = None):
- known_implementations = ["ollama", "intel_npu"]
- if not implementation or implementation.lower() not in ["ollama", "intel_npu", "testing", "ondemand"]:
+ known_implementations = ["ollama", "intel_npu", "testing", "ondemand"]
+ if not implementation or implementation.lower() not in known_implementations:
  raise ValueError(f"Unknown implementation {implementation}. Known implementations: {known_implementations}")
  if "ollama" == implementation:
  return OllamaRepository(model, system_msg)
- if "intel_npu" == implementation:
- return IntelNpuRepository(model, system_msg, log_to_file)
+ if not in_hf():
+ if "intel_npu" == implementation:
+ return IntelNpuRepository(model, system_msg, log_to_file)
  if "ondemand" == implementation:
  return OndemandRepository(model, system_msg, log_to_file)
  if "testing" == implementation:

requirements-base.txt

@@ -0,0 +1,3 @@
+transformers
+streamlit
+PyPDFForm

requirements.txt → requirements-npu.txt

@@ -1,6 +1,3 @@
-PyPDFForm
-ollama
-transformers
-streamlit
+-r requirements-base.txt
 deps/intel_npu_acceleration_library-1.3.0-cp310-cp310-win_amd64.whl ; sys_platform == "windows"
 deps/intel_npu_acceleration_library-1.3.0-cp310-cp310-linux_x86_64.whl ; sys_platform == "linux"

requirements-ollama.txt

@@ -0,0 +1,2 @@
+-r requirements-base.txt
+ollama

utils/env_utils.py

@@ -0,0 +1,3 @@
+import os
+def in_hf() -> bool:
+ return os.getenv("env") == "hf"