adding more pages, links, etc.

deepspeedai · Dec 4, 2023 · fa5f948 · fa5f948
1 parent 5e2289a
commit fa5f948
Show file tree

Hide file tree

Showing 4 changed files with 18 additions and 0 deletions.
diff --git a/docs/source/deployment.rst b/docs/source/deployment.rst
@@ -102,6 +102,8 @@ override the default ``max_length``:
     response_long = client(prompt, max_length=1024)
     response_short = client(prompt, max_length=128)
 
+.. _deployment_model_parallelism:
+
 Model Parallelism
 -----------------
 
@@ -113,6 +115,8 @@ argument to :func:`mii.serve`:
 
     client = mii.serve("mistralai/Mistral-7B-v0.1", tensor_parallel=2)
 
+.. _deployment_model_replicas:
+
 Model Replicas
 --------------
 

diff --git a/docs/source/parallelism.rst b/docs/source/parallelism.rst
@@ -1,2 +1,10 @@
 Model parallelism
 =================
+
+DeepSpeed-MII supports model parallelism via tensor parallelism for splitting models across multiple GPUs.
+
+For model parallelism with :doc:`pipeline`, please see :ref:`Pipeline Model
+Parallelism <pipeline_model_parallelism>`.
+
+For model parallelism with :doc:`deployment`, please see :ref:`Persistent
+Deployment Model Parallelism <deployment_model_parallelism>`.
diff --git a/docs/source/pipeline.rst b/docs/source/pipeline.rst
@@ -70,6 +70,8 @@ pipeline. For example, you can control per-prompt generation length:
     response_long = pipeline(prompt, max_length=1024)
     response_short = pipeline(prompt, max_length=128)
 
+.. _pipeline_model_parallelism:
+
 Model Parallelism
 -----------------
 

diff --git a/docs/source/replicas.rst b/docs/source/replicas.rst
@@ -1,2 +1,6 @@
 Model Replicas
 ==============
+
+DeepSpeed-MII supports creating multiple replicas of a model with
+:doc:`deployment`. Please see :ref:`Persistent Deployment Model Replicas
+<deployment_model_replicas>`.