[Fix] Fix Inference Example, Tests, and Requirements (#5688)

* clean requirements

* modify example inference struct

* add test ci scripts

* mark test_infer as submodule

* rm deprecated cls & deps

* import of HAS_FLASH_ATTN

* prune inference tests to be run

* prune triton kernel tests

* increment pytest timeout mins

* revert import path in openmoe

examples/inference/llama/benchmark_llama3.py

@@ -182,7 +182,7 @@ def benchmark_inference(args):
 
 
 def inference(rank, world_size, port, args):
-    colossalai.launch(config={}, rank=rank, world_size=world_size, host="localhost", port=port, backend="nccl")
+    colossalai.launch(rank=rank, world_size=world_size, host="localhost", port=port, backend="nccl")
     benchmark_inference(args)
 
 

examples/inference/llama/llama_generation.py

@@ -17,7 +17,7 @@ def infer(args):
     # ==============================
     # Launch colossalai, setup distributed environment
     # ==============================
-    colossalai.launch_from_torch(config={})
+    colossalai.launch_from_torch()
     coordinator = DistCoordinator()
 
     # ==============================
@@ -59,7 +59,7 @@ def infer(args):
     coordinator.print_on_master(out[0])
 
 
-# colossalai run --nproc_per_node 1 llama_gen.py -m MODEL_PATH
+# colossalai run --nproc_per_node 1 llama_generation.py -m MODEL_PATH
 if __name__ == "__main__":
     # ==============================
     # Parse Arguments

examples/inference/llama/test_ci.sh

@@ -0,0 +1,4 @@
+#!/bin/bash
+echo "Skip the test (this test is slow)"
+
+# bash ./run_benchmark.sh

examples/language/openmoe/model/modeling_openmoe.py

@@ -35,7 +35,7 @@ from transformers.utils import (
     replace_return_docstrings,
 )
 
-from colossalai.kernel.extensions.pybind.flash_attention import HAS_FLASH_ATTN
+from colossalai.kernel.extensions.flash_attention import HAS_FLASH_ATTN
 from colossalai.kernel.triton.llama_act_combine_kernel import HAS_TRITON
 from colossalai.moe.layers import SparseMLP
 from colossalai.moe.manager import MOE_MANAGER