Removed vllm.hpu directory and changed relevant imports

.github/workflows/mypy.yaml

@@ -50,6 +50,5 @@ jobs:
         mypy vllm/transformers_utils --config-file pyproject.toml
         mypy vllm/usage --config-file pyproject.toml
         mypy vllm/worker --config-file pyproject.toml
-        mypy vllm/hpu --config-file pyproject.toml
 
 

format.sh

@@ -113,7 +113,6 @@ mypy vllm/spec_decode --config-file pyproject.toml
 mypy vllm/transformers_utils --config-file pyproject.toml
 mypy vllm/usage --config-file pyproject.toml
 mypy vllm/worker --config-file pyproject.toml
-mypy vllm/hpu --config-file pyproject.toml
 
 
 # If git diff returns a file that is in the skip list, the file may be checked anyway:

requirements-hpu.txt

@@ -6,3 +6,4 @@ ray == 2.32.0
 triton
 pandas
 tabulate
+vllm-hpu-extension @ git+https://github.com/HabanaAI/vllm-hpu-extension.git@30ee2d1

tests/lora/test_lora_hpu.py

@@ -1,7 +1,7 @@
 import pytest
 import torch
+from vllm_hpu_extension.ops import LoraMask
 
-from vllm.hpu.ops import LoraMask
 from vllm.lora.layers import _apply_lora, _apply_lora_packed_nslice
 
 from .utils import DummyLoRAManager

vllm/attention/backends/habana_attn.py

@@ -7,14 +7,14 @@
 from typing import Any, Dict, List, Optional, Tuple, Type
 
 import torch
+import vllm_hpu_extension.ops as ops
+from vllm_hpu_extension import cache_ops
+from vllm_hpu_extension.utils import Matmul, Softmax, VLLMKVCache
 
-import vllm.hpu.ops as ops
 from vllm.attention.backends.abstract import (AttentionBackend, AttentionImpl,
                                               AttentionMetadata, AttentionType)
 from vllm.attention.ops.habana_paged_attn import (HabanaPagedAttention,
                                                   HabanaPagedAttentionMetadata)
-from vllm.hpu import cache_ops
-from vllm.hpu.utils import Matmul, Softmax, VLLMKVCache
 from vllm.logger import init_logger
 
 logger = init_logger(__name__)

vllm/attention/ops/habana_paged_attn.py

@@ -6,8 +6,7 @@
 from typing import Dict, List, Optional, Tuple
 
 import torch
-
-from vllm.hpu import cache_ops, ops
+from vllm_hpu_extension import cache_ops, ops
 
 # Should be the same as PARTITION_SIZE in `paged_attention_v2_launcher`.
 _PARTITION_SIZE = 512