Don't include bitsandbytes by default, update accelerate version

Author: njhill

Dockerfile

@@ -253,9 +253,7 @@ COPY --from=flash-att-v2-cache /usr/src/flash-attention-v2/build/lib.linux-x86_6
 # Install server
 COPY proto proto
 COPY server server
-RUN cd server && \
-    make gen-server && \
-    pip install ".[bnb, accelerate, onnx-gpu]" --no-cache-dir
+RUN cd server && make gen-server && pip install ".[accelerate, onnx-gpu]" --no-cache-dir
 
 # Patch codegen model changes into transformers 4.31
 RUN cp server/transformers_patch/modeling_codegen.py \

server/poetry.lock

@@ -8,14 +8,15 @@ authors = ["Olivier Dehaene <[email protected]>"]
 text-generation-server = 'text_generation_server.cli:app'
 
 [tool.poetry.dependencies]
-python = "^3.9"
+python = ">=3.9.0,<3.13"
 protobuf = "^4.24.3"
 grpcio = "^1.58.0"
 grpcio-reflection = "^1.58.0"
 loguru = "^0.7.0"
 typer = "^0.9.0"
-accelerate = { version = "0.20.3", optional = true }
+accelerate = { version = "0.23.0", optional = true }
 bitsandbytes = { version = "^0.41.0", optional = true }
+scipy = { version = "^1.11.2", optional = true }
 safetensors = "^0.3.3"
 sentencepiece = "^0.1.99"
 transformers = "4.33.1"
@@ -33,7 +34,7 @@ cryptography = ">=41.0.3"
 
 [tool.poetry.extras]
 accelerate = ["accelerate"]
-bnb = ["bitsandbytes"]
+bnb = ["bitsandbytes", "scipy"]  # BNB doesn't declare a dep on scipy https://github.com/TimDettmers/bitsandbytes/pull/525
 onnx = ["optimum", "onnxruntime", "onnx"]
 onnx-gpu = ["optimum", "onnxruntime-gpu", "onnx"]