Fixes #1348

docs/README_MACOS.md

@@ -24,24 +24,24 @@ Supports CPU and MPS (Metal M1/M2).
     python -m pip install --upgrade setuptools
 
     # Install Torch:
-    pip install -r requirements.txt --extra-index https://download.pytorch.org/whl/cpu
+    pip install -r requirements.txt --extra-index https://download.pytorch.org/whl/cpu -c reqs_optional/reqs_constraints.txt
     ```
 * Install document question-answer dependencies:
     ```bash
     # Required for Doc Q/A: LangChain:
-    pip install -r reqs_optional/requirements_optional_langchain.txt
+    pip install -r reqs_optional/requirements_optional_langchain.txt -c reqs_optional/reqs_constraints.txt
     # Required for CPU: LLaMa/GPT4All:
     pip uninstall -y llama-cpp-python llama-cpp-python-cuda
-    pip install -r reqs_optional/requirements_optional_gpt4all.txt
-    pip install librosa
-    pip install llama-cpp-python
+    pip install -r reqs_optional/requirements_optional_gpt4all.txt -c reqs_optional/reqs_constraints.txt
+    pip install librosa -c reqs_optional/reqs_constraints.txt
+    pip install llama-cpp-python -c reqs_optional/reqs_constraints.txt
     # Optional: PyMuPDF/ArXiv:
-    pip install -r reqs_optional/requirements_optional_langchain.gpllike.txt
+    pip install -r reqs_optional/requirements_optional_langchain.gpllike.txt -c reqs_optional/reqs_constraints.txt
     # Optional: Selenium/PlayWright:
-    pip install -r reqs_optional/requirements_optional_langchain.urls.txt
+    pip install -r reqs_optional/requirements_optional_langchain.urls.txt -c reqs_optional/reqs_constraints.txt
     # Optional: DocTR OCR:
     conda install weasyprint pygobject -c conda-forge -y
-    pip install -r reqs_optional/requirements_optional_doctr.txt                     
+    pip install -r reqs_optional/requirements_optional_doctr.txt -c reqs_optional/reqs_constraints.txt
     # Optional: for supporting unstructured package
     python -m nltk.downloader all
   ```
@@ -109,14 +109,14 @@ and run `sh run.sh` from the terminal placed in the parent folder of `run.sh`
     ```
     during pip install.  If so, set your archflags during pip install. E.g.
     ```bash
-    ARCHFLAGS="-arch x86_64" pip install -r requirements.txt
+    ARCHFLAGS="-arch x86_64" pip install -r requirements.txt -c reqs_optional/reqs_constraints.txt
     ```
 * Metal M1/M2 Only
   * By default requirements_optional_gpt4all.txt should install correct llama_cpp_python packages for GGUF.  See [https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases](https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases) or [https://github.com/abetlen/llama-cpp-python/releases](https://github.com/abetlen/llama-cpp-python/releases) for other releases if you encounter any issues.
   * If any issues, then compile:
       ```bash
       pip uninstall llama-cpp-python -y
-      CMAKE_ARGS="-DLLAMA_METAL=on" FORCE_CMAKE=1 pip install -U llama-cpp-python==0.2.26 --no-cache-dir
+      CMAKE_ARGS="-DLLAMA_METAL=on" FORCE_CMAKE=1 pip install -U llama-cpp-python==0.2.26 --no-cache-dir -c reqs_optional/reqs_constraints.txt
       ```
 
 * If you encounter an error while building a wheel during the `pip install` process, you may need to install a C++ compiler on your computer.

docs/linux_install.sh

@@ -16,78 +16,78 @@ conda install weasyprint pygobject -c conda-forge -y
 pip uninstall -y pandoc pypandoc pypandoc-binary flash-attn
 # broad support, but no training-time or data creation dependencies
 
-  pip install -r requirements.txt
+pip install -r requirements.txt -c reqs_optional/reqs_constraints.txt
 #* Optional: Install document question-answer dependencies:
 # May be required for jq package:
 sudo apt-get -y install autoconf libtool
 # Required for Doc Q/A: LangChain:
-pip install -r reqs_optional/requirements_optional_langchain.txt
+pip install -r reqs_optional/requirements_optional_langchain.txt -c reqs_optional/reqs_constraints.txt
 # Required for CPU: LLaMa/GPT4All:
-pip install -r reqs_optional/requirements_optional_gpt4all.txt
+pip install -r reqs_optional/requirements_optional_gpt4all.txt -c reqs_optional/reqs_constraints.txt
 # Optional: PyMuPDF/ArXiv:
 #   Note!! that pymupdf is AGPL, requiring any source code be made available, but it's like GPL and too strong a constraint for general commercial use.
 if [ "${GPLOK}" -eq "1" ]
 then
-    pip install -r reqs_optional/requirements_optional_langchain.gpllike.txt
+    pip install -r reqs_optional/requirements_optional_langchain.gpllike.txt -c reqs_optional/reqs_constraints.txt
 fi
 # Optional: FAISS
-pip install -r reqs_optional/requirements_optional_faiss.txt
+pip install -r reqs_optional/requirements_optional_faiss.txt -c reqs_optional/reqs_constraints.txt
 # Optional: Selenium/PlayWright:
-pip install -r reqs_optional/requirements_optional_langchain.urls.txt
+pip install -r reqs_optional/requirements_optional_langchain.urls.txt -c reqs_optional/reqs_constraints.txt
 
 # Optional: support docx, pptx, ArXiv, etc. required by some python packages
 sudo apt-get install -y libmagic-dev poppler-utils tesseract-ocr libtesseract-dev libreoffice
 
 # Optional: For DocTR
-pip install -r reqs_optional/requirements_optional_doctr.txt
+pip install -r reqs_optional/requirements_optional_doctr.txt -c reqs_optional/reqs_constraints.txt
 # For DocTR: go back to older onnx so Tesseract OCR still works
-pip install onnxruntime==1.15.0
+pip install onnxruntime==1.15.0 -c reqs_optional/reqs_constraints.txt
 # GPU only:
-pip install onnxruntime-gpu==1.15.0
+pip install onnxruntime-gpu==1.15.0 -c reqs_optional/reqs_constraints.txt
 
 # Optional: for supporting unstructured package
 for i in 1 2 3 4; do python -m nltk.downloader all && break || sleep 1; done  # retry as frequently fails with github downloading issues
 # Optional: Required for PlayWright
 playwright install --with-deps
 # Audio transcription from Youtube videos and local mp3 files:
-pip install pydub==0.25.1 librosa==0.10.1 ffmpeg==1.4 yt_dlp==2023.10.13 wavio==0.0.8
+pip install pydub==0.25.1 librosa==0.10.1 ffmpeg==1.4 yt_dlp==2023.10.13 wavio==0.0.8 -c reqs_optional/reqs_constraints.txt
 # Audio speed-up and slowdown (best quality), if not installed can only speed-up with lower quality
 sudo apt-get install -y rubberband-cli
-pip install pyrubberband==0.3.0
+pip install pyrubberband==0.3.0 -c reqs_optional/reqs_constraints.txt
 # https://stackoverflow.com/questions/75813603/python-working-with-sound-librosa-and-pyrubberband-conflict
 pip uninstall -y pysoundfile soundfile
-pip install soundfile==0.12.1
+pip install soundfile==0.12.1 -c reqs_optional/reqs_constraints.txt
 # Optional: Only for testing for now
-pip install playsound==1.3.0
+pip install playsound==1.3.0 -c reqs_optional/reqs_constraints.txt
 # STT from microphone (may not be required if ffmpeg installed above)
 sudo apt-get install ffmpeg
 # for any TTS:
-pip install torchaudio soundfile==0.12.1
+pip install torchaudio soundfile==0.12.1 -c reqs_optional/reqs_constraints.txt
 # GPU Only: for Coqui XTTS (ensure CUDA_HOME set and consistent with added postfix for extra-index):
 # relaxed versions to avoid conflicts
-pip install TTS deepspeed noisereduce emoji ffmpeg-python==0.2.0 trainer pysbd coqpit
+pip install TTS deepspeed noisereduce emoji ffmpeg-python==0.2.0 trainer pysbd coqpit -c reqs_optional/reqs_constraints.txt
 # for Coqui XTTS language helpers (specific versions probably not required)
-pip install cutlet==0.3.0 langid==1.1.6 g2pkk==0.1.2 jamo==0.4.1 gruut[de,es,fr]==2.2.3 jieba==0.42.1
+pip install cutlet==0.3.0 langid==1.1.6 g2pkk==0.1.2 jamo==0.4.1 gruut[de,es,fr]==2.2.3 jieba==0.42.1 -c reqs_optional/reqs_constraints.txt
 # For faster whisper:
-pip install git+https://github.com/SYSTRAN/faster-whisper.git
+pip install git+https://github.com/SYSTRAN/faster-whisper.git -c reqs_optional/reqs_constraints.txt
 # needed for librosa/soundfile to work, but violates TTS, but that's probably just too strict as we have seen before)
-pip install numpy==1.23.0 --no-deps --upgrade
+pip install numpy==1.23.0 --no-deps --upgrade -c reqs_optional/reqs_constraints.txt
 # TTS or other deps load old librosa, fix:
-pip install librosa==0.10.1 --no-deps --upgrade
+pip install librosa==0.10.1 --no-deps --upgrade -c reqs_optional/reqs_constraints.txt
 #* STT and TTS Notes:
 #  * STT: Ensure microphone is on and in browser go to http://localhost:7860 instead of http://0.0.0.0:7860 for microphone to be possible to allow in browser.
 #  * TTS: For XTT models, ensure `CUDA_HOME` is set correctly, because deepspeed compiles at runtime using torch and nvcc.  Those must match CUDA version.  E.g. if used `--extra-index https://download.pytorch.org/whl/cu118`, then must have ENV `CUDA_HOME=/usr/local/cuda-11.7` or ENV from conda must be that version.  Since conda only has up to cuda 11.7 for dev toolkit, but H100+ need cuda 11.8, for those cases one should download the toolkit from NVIDIA.
 
 # Vision/Image packages
-pip install fiftyone
-pip install pytube
-pip install diffusers==0.24.0
+pip install fiftyone -c reqs_optional/reqs_constraints.txt
+pip install pytube -c reqs_optional/reqs_constraints.txt
+pip install diffusers==0.24.0 -c reqs_optional/reqs_constraints.txt
 
 #* HNSW issue:
 #    In some cases old chroma migration package will install old hnswlib and that may cause issues when making a database, then do:
 pip uninstall -y hnswlib chroma-hnswlib
 # restore correct version
-pip install chroma-hnswlib==0.7.3 --upgrade
+pip install chroma-hnswlib==0.7.3 --upgrade -c reqs_optional/reqs_constraints.txt
 #* Selenium needs to have chrome installed, e.g. on Ubuntu:
 sudo apt install -y unzip xvfb libxi6 libgconf-2-4 libu2f-udev
 sudo apt install -y default-jdk
@@ -109,24 +109,24 @@ sudo mv chromedriver /usr/bin/chromedriver
 sudo chown root:root /usr/bin/chromedriver
 sudo chmod +x /usr/bin/chromedriver
 #* GPU Optional: For AutoGPTQ support on x86_64 linux
-pip uninstall -y auto-gptq ; pip install auto-gptq==0.6.0 --extra-index-url https://huggingface.github.io/autogptq-index/whl/cu118/
+pip uninstall -y auto-gptq ; pip install auto-gptq==0.6.0 --extra-index-url https://huggingface.github.io/autogptq-index/whl/cu118/ -c reqs_optional/reqs_constraints.txt
 # in-transformers support of AutoGPTQ, requires also auto-gptq above to be installed since used internally by transformers/optimum
-pip install optimum==1.16.1
+pip install optimum==1.16.1 -c reqs_optional/reqs_constraints.txt
 #    See [AutoGPTQ](README_GPU.md#autogptq) about running AutoGPT models.
 #* GPU Optional: For AutoAWQ support on x86_64 linux
-pip uninstall -y autoawq ; pip install https://github.com/casper-hansen/AutoAWQ/releases/download/v0.1.8/autoawq-0.1.8+cu118-cp310-cp310-linux_x86_64.whl
+pip uninstall -y autoawq ; pip install https://github.com/casper-hansen/AutoAWQ/releases/download/v0.1.8/autoawq-0.1.8+cu118-cp310-cp310-linux_x86_64.whl -c reqs_optional/reqs_constraints.txt
 # fix version since don't need lm-eval to have its version of 1.5.0
-pip install sacrebleu==2.3.1 --upgrade
+pip install sacrebleu==2.3.1 --upgrade -c reqs_optional/reqs_constraints.txt
 #    If this has issues, you need to build:
 if [ 1 -eq 0 ]
 then
     pip uninstall -y autoawq
     git clone https://github.com/casper-hansen/AutoAWQ
     cd AutoAWQ
-    pip install .
+    pip install . -c reqs_optional/reqs_constraints.txt
 fi
 #* GPU Optional: For exllama support on x86_64 linux
-pip uninstall -y exllama ; pip install https://github.com/jllllll/exllama/releases/download/0.0.18/exllama-0.0.18+cu118-cp310-cp310-linux_x86_64.whl --no-cache-dir
+pip uninstall -y exllama ; pip install https://github.com/jllllll/exllama/releases/download/0.0.18/exllama-0.0.18+cu118-cp310-cp310-linux_x86_64.whl --no-cache-dir -c reqs_optional/reqs_constraints.txt
 #    See [exllama](README_GPU.md#exllama) about running exllama models.
 
 #  * If any issues with llama_cpp_python, then must compile llama-cpp-python with CUDA support:
@@ -136,7 +136,7 @@ then
     export LLAMA_CUBLAS=1
     export CMAKE_ARGS=-DLLAMA_CUBLAS=on
     export FORCE_CMAKE=1
-    CMAKE_ARGS="-DLLAMA_CUBLAS=on" FORCE_CMAKE=1 pip install llama-cpp-python==0.2.26 --no-cache-dir --verbose
+    CMAKE_ARGS="-DLLAMA_CUBLAS=on" FORCE_CMAKE=1 pip install llama-cpp-python==0.2.26 --no-cache-dir --verbose -c reqs_optional/reqs_constraints.txt
 fi
 #  * By default, we set `n_gpu_layers` to large value, so llama.cpp offloads all layers for maximum GPU performance.  You can control this by passing `--llamacpp_dict="{'n_gpu_layers':20}"` for value 20, or setting in UI.  For highest performance, offload *all* layers.
 #    That is, one gets maximum performance if one sees in startup of h2oGPT all layers offloaded:
@@ -153,14 +153,14 @@ fi
 if [[ -v CUDA_HOME ]];
 then
     pip install --upgrade pip
-    pip install flash-attn==2.4.2 --no-build-isolation --no-cache-dir
+    pip install flash-attn==2.4.2 --no-build-isolation --no-cache-dir -c reqs_optional/reqs_constraints.txt
 fi
 #* Control Core Count for chroma < 0.4 using chromamigdb package:
 #    * Duckdb used by Chroma < 0.4 uses DuckDB 0.8.1 that has no control over number of threads per database, `import duckdb` leads to all virtual cores as threads and each db consumes another number of threads equal to virtual cores.  To prevent this, one can rebuild duckdb using [this modification](https://github.com/h2oai/duckdb/commit/dcd8c1ffc53dd020623630efb99ba6a3a4cbc5ad) or one can try to use the prebuild wheel for x86_64 built on Ubuntu 20.
 pip uninstall -y pyduckdb duckdb
-pip install https://h2o-release.s3.amazonaws.com/h2ogpt/duckdb-0.8.2.dev4025%2Bg9698e9e6a8.d20230907-cp310-cp310-linux_x86_64.whl --no-cache-dir --force-reinstall --no-deps
+pip install https://h2o-release.s3.amazonaws.com/h2ogpt/duckdb-0.8.2.dev4025%2Bg9698e9e6a8.d20230907-cp310-cp310-linux_x86_64.whl --no-cache-dir --force-reinstall --no-deps -c reqs_optional/reqs_constraints.txt
 #* SERP for search:
-pip install -r reqs_optional/requirements_optional_agents.txt
+pip install -r reqs_optional/requirements_optional_agents.txt -c reqs_optional/reqs_constraints.txt
 #  For more info see [SERP Docs](README_SerpAPI.md).
 #* Deal with not-thread-safe things in LangChain:
 sp=`python3.10 -c 'import site; print(site.getsitepackages()[0])'`
@@ -181,7 +181,7 @@ sed -i 's/while True:/while True:\n            time.sleep(0.001)\n/g' $sp/gradio
 
 #* PDF View support
 # only if using gradio4
-#pip install https://h2o-release.s3.amazonaws.com/h2ogpt/gradio_pdf-0.0.3-py3-none-any.whl
+#pip install https://h2o-release.s3.amazonaws.com/h2ogpt/gradio_pdf-0.0.3-py3-none-any.whl -c reqs_optional/reqs_constraints.txt
 
 
 ### Compile Install Issues