update ut/doc for glm/codegen (microsoft#4057)

* update ut/doc for glm/codegen

* formatting/spacing on docs

* re-order/alphabetize the models

---------

Co-authored-by: Logan Adams <114770087+loadams@users.noreply.github.com>
Co-authored-by: Logan Adams <loadams@microsoft.com>

docs/_tutorials/automatic-tensor-parallelism.md

@@ -123,11 +123,14 @@ The following model families have been successfully tested with automatic tensor
 - albert
 - bert
 - bigbird_pegasus
+- bloom
 - camembert
+- codegen
 - deberta_v2
 - electra
 - ernie
 - esm
+- glm
 - gpt-j
 - gpt-neo
 - gpt-neox
@@ -136,6 +139,7 @@ The following model families have been successfully tested with automatic tensor
 - llama
 - m2m_100
 - marian
+- mpt
 - mvp
 - nezha
 - openai
@@ -151,14 +155,11 @@ The following model families have been successfully tested with automatic tensor
 - xglm
 - xlm_roberta
 - yoso
-- bloom
-- mpt
 
 # Unsupported Models
 
 The following models are not currently supported with automatic tensor parallelism. They may still be compatible with other DeepSpeed features (e.g., kernel injection for Bloom):
 
-- codegen
 - deberta
 - flaubert
 - fsmt

tests/unit/inference/test_inference.py

@@ -478,12 +478,8 @@ def test(
 @pytest.mark.seq_inference
 @pytest.mark.parametrize(
     "model_w_task",
-    [
-        ("Helsinki-NLP/opus-mt-en-de", "translation"),
-    ],
-    ids=[
-        "marian",
-    ],
+    [("Helsinki-NLP/opus-mt-en-de", "translation"), ("Salesforce/codegen-350M-mono", "text-generation")],
+    ids=["marian", "codegen"],  #codegen has fusedqkv weight.
 )
 @pytest.mark.parametrize("dtype", [torch.float16], ids=["fp16"])
 class TestAutoTensorParallelism(DistributedTest):