fix(python): move accelerate and GPU-specific libs to build-type (#3194)

Some of the dependencies in `requirements.txt`, even if generic, pulls
down the line CUDA libraries.

This changes moves mostly all GPU-specific libs to the build-type, and
tries a safer approach. In `requirements.txt` now are listed only
"first-level" dependencies, for instance, grpc, but libs-dependencies
are moved down to the respective build-type `requirements.txt` to avoid
any mixin.

This should fix #2737 and #1592.

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>

backend/python/vllm/requirements-after.txt

@@ -0,0 +1 @@
+flash-attn

backend/python/vllm/requirements-cpu.txt

@@ -0,0 +1,4 @@
+accelerate
+torch
+transformers
+vllm

backend/python/vllm/requirements-cublas11.txt

@@ -1,3 +1,5 @@
 --extra-index-url https://download.pytorch.org/whl/cu118
+accelerate
 torch
-flash-attn
+transformers
+vllm

backend/python/vllm/requirements-cublas12.txt

@@ -1,2 +1,4 @@
+accelerate
 torch
-flash-attn
+transformers
+vllm

backend/python/vllm/requirements-hipblas.txt

@@ -1,2 +1,5 @@
 --extra-index-url https://download.pytorch.org/whl/rocm6.0
-torch
+accelerate
+torch
+transformers
+vllm

backend/python/vllm/requirements-intel.txt

@@ -1,5 +1,8 @@
 --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/
 intel-extension-for-pytorch
+accelerate
 torch
+transformers
 optimum[openvino]
-setuptools==70.3.0 # https://github.com/mudler/LocalAI/issues/2406
+setuptools==70.3.0 # https://github.com/mudler/LocalAI/issues/2406
+vllm

backend/python/vllm/requirements.txt

@@ -1,7 +1,4 @@
-accelerate
-vllm
 grpcio==1.65.4
 protobuf
 certifi
-transformers
 setuptools