make it lighten (#2577)

### What problem does this PR solve?

#2295

### Type of change

- [x] Refactoring

Dockerfile.scratch

 RUN  sed -i 's|http://archive.ubuntu.com|https://mirrors.tuna.tsinghua.edu.cn|g' /etc/apt/sources.list.d/ubuntu.sources
 
 RUN --mount=type=cache,target=/var/cache/apt,sharing=locked \
-    apt update && apt install -y curl libpython3-dev nginx openmpi-bin openmpi-common libopenmpi-dev libglib2.0-0 libglx-mesa0 \
+    apt update && apt install -y curl libpython3-dev nginx libglib2.0-0 libglx-mesa0 \
     && rm -rf /var/lib/apt/lists/* \
     && curl -sSL https://install.python-poetry.org | python3 -
 
 WORKDIR /ragflow
 
 RUN --mount=type=cache,target=/var/cache/apt,sharing=locked \
-    apt update && apt install -y nodejs npm cargo && \
+    apt update && apt install -y nodejs npm && \
     rm -rf /var/lib/apt/lists/*
 
 COPY web web
 # install dependencies from poetry.lock file
 COPY pyproject.toml poetry.toml poetry.lock ./
 RUN --mount=type=cache,target=/root/.cache/pypoetry,sharing=locked \
-    /root/.local/bin/poetry install --sync --no-root
+    /root/.local/bin/poetry lock
+
+RUN --mount=type=cache,target=/root/.cache/pypoetry,sharing=locked \
+    /root/.local/bin/poetry install --sync --no-cache --no-root
 
 # production stage
 FROM base AS production
 # Download nltk data
 RUN python3 -m nltk.downloader wordnet punkt punkt_tab
 
-# Copy models downloaded via download_deps.sh
-COPY det.onnx layout.laws.onnx layout.manual.onnx layout.onnx layout.paper.onnx ocr.res rec.onnx tsr.onnx updown_concat_xgb.model /ragflow/rag/res/deepdoc/
 
 ENV PYTHONPATH=/ragflow/
 

pyproject.toml

 tiktoken = "0.6.0"
 torch = "2.3.0"
 transformers = "4.38.1"
-umap = "0.1.1"
+umap_learn = "0.5.6"
 vertexai = "1.64.0"
 volcengine = "1.0.146"
 voyageai = "0.2.3"

rag/llm/embedding_model.py

             threads: Optional[int] = None,
             **kwargs,
     ):
-        from fastembed import TextEmbedding
-        if not FastEmbed._model:
+        if not LIGHTEN and not FastEmbed._model:
+            from fastembed import TextEmbedding
             self._model = TextEmbedding(model_name, cache_dir, threads, **kwargs)
 
     def encode(self, texts: list, batch_size=32):
     _client = None
 
     def __init__(self, key=None, model_name="maidalun1020/bce-embedding-base_v1", **kwargs):
-        from BCEmbedding import EmbeddingModel as qanthing
-        if not YoudaoEmbed._client:
+        if not LIGHTEN and not YoudaoEmbed._client:
+            from BCEmbedding import EmbeddingModel as qanthing
             try:
                 print("LOADING BCE...")
                 YoudaoEmbed._client = qanthing(model_name_or_path=os.path.join(

rag/llm/rerank_model.py

     _model_lock = threading.Lock()
 
     def __init__(self, key=None, model_name="maidalun1020/bce-reranker-base_v1", **kwargs):
-        from BCEmbedding import RerankerModel
-        if not YoudaoRerank._model:
+        if not LIGHTEN and not YoudaoRerank._model:
+            from BCEmbedding import RerankerModel
             with YoudaoRerank._model_lock:
                 if not YoudaoRerank._model:
                     try: