passthru rpc_servers params

wip

Author: chraac

llama_cpp/llama.py

@@ -70,6 +70,7 @@ def __init__(
         split_mode: int = llama_cpp.LLAMA_SPLIT_MODE_LAYER,
         main_gpu: int = 0,
         tensor_split: Optional[List[float]] = None,
+        rpc_servers: Optional[str] = None,
         vocab_only: bool = False,
         use_mmap: bool = True,
         use_mlock: bool = False,
@@ -148,6 +149,7 @@ def __init__(
             split_mode: How to split the model across GPUs. See llama_cpp.LLAMA_SPLIT_* for options.
             main_gpu: main_gpu interpretation depends on split_mode: LLAMA_SPLIT_NONE: the GPU that is used for the entire model. LLAMA_SPLIT_ROW: the GPU that is used for small tensors and intermediate results. LLAMA_SPLIT_LAYER: ignored
             tensor_split: How split tensors should be distributed across GPUs. If None, the model is not split.
+            rpc_servers: Comma separated list of RPC servers to use for offloading
             vocab_only: Only load the vocabulary no weights.
             use_mmap: Use mmap if possible.
             use_mlock: Force the system to keep the model in RAM.
@@ -219,6 +221,7 @@ def __init__(
         )  # 0x7FFFFFFF is INT32 max, will be auto set to all layers
         self.model_params.split_mode = split_mode
         self.model_params.main_gpu = main_gpu
+        self.model_params.rpc_servers = rpc_servers
         self.tensor_split = tensor_split
         self._c_tensor_split = None
         if self.tensor_split is not None: