Add pipeline parallel support to TransformersModel (#12832)

Signed-off-by: Harry Mellor <[email protected]>
Signed-off-by: Isotr0py <[email protected]>
Co-authored-by: Isotr0py <[email protected]>

Author: hmellor

docs/source/models/supported_models.md

@@ -73,7 +73,7 @@ The Transformers fallback explicitly supports the following features:
 
 - <project:#quantization-index> (except GGUF)
 - <project:#lora-adapter>
-- <project:#distributed-serving> (pipeline parallel coming soon <gh-pr:12832>!)
+- <project:#distributed-serving> (requires `transformers>=4.49.0`)
 
 #### Remote code
 

tests/distributed/test_pipeline_parallel.py

@@ -175,6 +175,8 @@ def iter_params(self, model_id: str):
     "inceptionai/jais-13b-chat": PPTestSettings.fast(),
     "ai21labs/Jamba-tiny-dev": PPTestSettings.fast(),
     "meta-llama/Llama-3.2-1B-Instruct": PPTestSettings.detailed(),
+    # Tests TransformersModel
+    "ArthurZ/Ilama-3.2-1B": PPTestSettings.fast(),
     "openbmb/MiniCPM-2B-sft-bf16": PPTestSettings.fast(),
     "openbmb/MiniCPM3-4B": PPTestSettings.fast(),
     # Uses Llama
@@ -243,6 +245,7 @@ def iter_params(self, model_id: str):
     # [LANGUAGE GENERATION]
     "microsoft/Phi-3.5-MoE-instruct",
     "meta-llama/Llama-3.2-1B-Instruct",
+    # "ArthurZ/Ilama-3.2-1B", NOTE: Uncomment after #13905
     "ibm/PowerLM-3b",
     # [LANGUAGE EMBEDDING]
     "intfloat/e5-mistral-7b-instruct",