From 018e2dd7b2a607b22c4abf6e2247f6a314664324 Mon Sep 17 00:00:00 2001 From: sekyondaMeta <127536312+sekyondaMeta@users.noreply.github.com> Date: Tue, 21 Jan 2025 10:51:31 -0500 Subject: [PATCH] Update rpc_tutorial.rst Fixing broken links in RPC Tutorial --- intermediate_source/rpc_tutorial.rst | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/intermediate_source/rpc_tutorial.rst b/intermediate_source/rpc_tutorial.rst index 835e6f0649f..dd8af47e62a 100644 --- a/intermediate_source/rpc_tutorial.rst +++ b/intermediate_source/rpc_tutorial.rst @@ -59,7 +59,7 @@ Distributed Reinforcement Learning using RPC and RRef ----------------------------------------------------- This section describes steps to build a toy distributed reinforcement learning -model using RPC to solve CartPole-v1 from `OpenAI Gym `__. +model using RPC to solve CartPole-v1 from `OpenAI Gym `__. The policy code is mostly borrowed from the existing single-thread `example `__ as shown below. We will skip details of the ``Policy`` design, and focus on RPC @@ -156,7 +156,7 @@ send commands. Applications don't need to worry about the lifetime of ``RRefs``. The owner of each ``RRef`` maintains a reference counting map to track its lifetime, and guarantees the remote data object will not be deleted as long as there is any live user of that ``RRef``. Please refer to the ``RRef`` -`design doc `__ for details. +`design doc `__ for details. .. code:: python @@ -531,7 +531,7 @@ the given arguments (i.e., ``lr=0.05``). In the training loop, it first creates a distributed autograd context, which will help the distributed autograd engine to find gradients and involved RPC send/recv functions. The design details of the distributed autograd engine can -be found in its `design note `__. +be found in its `design note `__. Then, it kicks off the forward pass as if it is a local model, and run the distributed backward pass. For the distributed backward, you only need to specify a list of roots, in this case, it is the loss ``Tensor``.