modify patch and workflow (#163)

Author: NaganooMei

.github/workflows/unifiedcache_test.yml

@@ -51,7 +51,7 @@ jobs:
       - name: Print debug info
         run: |
           echo "GITHUB_WORKSPACE is $GITHUB_WORKSPACE"
-          ls -l $GITHUB_WORKSPACE/unifiedcache/patch
+          ls -l $GITHUB_WORKSPACE/ucm/patch
 
       - name: Debug repo layout
         run: |
@@ -60,7 +60,7 @@ jobs:
           ls -l "$GITHUB_WORKSPACE"
       - name: Apply patch
         run: |
-          git -C /vllm-workspace/vllm apply $GITHUB_WORKSPACE/unifiedcache/patch/0.9.2/vllm-adapt.patch
+          git -C /vllm-workspace/vllm apply $GITHUB_WORKSPACE/ucm/patch/0.9.2/vllm-adapt.patch
       
 
       - name: Install unified-cache-management

ucm/patch/0.9.2/vllm-adapt-sparse.patch

@@ -21,7 +21,7 @@ index f0ad68b16..847c97371 100644
  from vllm.platforms import _Backend, current_platform
  from vllm.utils import direct_register_custom_op
  from vllm.v1.attention.backends.utils import validate_kv_sharing_target
-+from unifiedcache.integration.vllm.ucm_sparse.state import get_ucm_sparse, has_ucm_sparse
++from ucm.integration.vllm.ucm_sparse.state import get_ucm_sparse, has_ucm_sparse
 
 
  class Attention(nn.Module):
@@ -114,8 +114,8 @@ index 6937455e7..3a44db442 100644
  from vllm.v1.kv_cache_interface import KVCacheConfig
  from vllm.v1.metrics.stats import PrefixCacheStats
  from vllm.v1.request import Request, RequestStatus
-+from unifiedcache.integration.vllm.ucm_sparse.state import get_ucm_sparse, has_ucm_sparse
-+from unifiedcache.integration.vllm.ucm_sparse.base import INVALID_SLOT
++from ucm.integration.vllm.ucm_sparse.state import get_ucm_sparse, has_ucm_sparse
++from ucm.integration.vllm.ucm_sparse.base import INVALID_SLOT
 
  logger = init_logger(__name__)
 
@@ -179,8 +179,8 @@ index 22c0ad8d6..c5c39a2b8 100644
  from vllm.v1.spec_decode.metrics import SpecDecodingStats
  from vllm.v1.structured_output import StructuredOutputManager
 
-+from unifiedcache.integration.vllm.ucm_sparse.state import ensure_ucm_sparse_initialized, get_ucm_sparse, has_ucm_sparse
-+from unifiedcache.integration.vllm.ucm_sparse.base import UcmSparseBase, UcmSparseRole, INVALID_SLOT
++from ucm.integration.vllm.ucm_sparse.state import ensure_ucm_sparse_initialized, get_ucm_sparse, has_ucm_sparse
++from ucm.integration.vllm.ucm_sparse.base import UcmSparseBase, UcmSparseRole, INVALID_SLOT
 +
  logger = init_logger(__name__)
 
@@ -313,8 +313,8 @@ index 14278bb6a..84a597b0e 100644
  from .utils import (gather_mm_placeholders, initialize_kv_cache_for_kv_sharing,
                      sanity_check_mm_encoder_outputs, scatter_mm_placeholders)
 
-+from unifiedcache.integration.vllm.ucm_sparse.state import get_ucm_sparse, has_ucm_sparse
-+from unifiedcache.integration.vllm.ucm_sparse.base import UcmSparseMetadata, INVALID_SLOT
++from ucm.integration.vllm.ucm_sparse.state import get_ucm_sparse, has_ucm_sparse
++from ucm.integration.vllm.ucm_sparse.base import UcmSparseMetadata, INVALID_SLOT
 +
  if TYPE_CHECKING:
      import xgrammar as xgr
@@ -482,7 +482,7 @@ index 7117f60b5..c239e1f02 100644
  from vllm.v1.utils import report_usage_stats
  from vllm.v1.worker.gpu_model_runner import GPUModelRunner
  from vllm.v1.worker.worker_base import WorkerBase
-+from unifiedcache.integration.vllm.ucm_sparse.state import ensure_ucm_sparse_initialized
++from ucm.integration.vllm.ucm_sparse.state import ensure_ucm_sparse_initialized
 
  logger = init_logger(__name__)
 

ucm/patch/0.9.2/vllm-ascend-adapt-sparse.patch

@@ -17,7 +17,7 @@ index 915feb7..0600f35 100644
  from vllm_ascend.utils import (ACL_FORMAT_FRACTAL_NZ, aligned_16, is_310p,
                                 nd_to_nz_2d, nd_to_nz_spec)
 
-+from unifiedcache.integration.vllm.ucm_sparse.state import get_ucm_sparse, has_ucm_sparse
++from ucm.integration.vllm.ucm_sparse.state import get_ucm_sparse, has_ucm_sparse
 +
 
  class AscendAttentionBackend(AttentionBackend):
@@ -91,8 +91,8 @@ index f9cca93..31766b2 100644
  from vllm_ascend.worker.mtp_proposer_v1 import MtpProposer
  from vllm_ascend.worker.npu_input_batch import CachedRequestState, InputBatch
 
-+from unifiedcache.integration.vllm.ucm_sparse.state import get_ucm_sparse, has_ucm_sparse
-+from unifiedcache.integration.vllm.ucm_sparse.base import UcmSparseMetadata, INVALID_SLOT
++from ucm.integration.vllm.ucm_sparse.state import get_ucm_sparse, has_ucm_sparse
++from ucm.integration.vllm.ucm_sparse.base import UcmSparseMetadata, INVALID_SLOT
 +
  if TYPE_CHECKING:
      import xgrammar as xgr  # type: ignore[import-untyped]
@@ -250,7 +250,7 @@ index df03d50..6ea5bf3 100644
                                 read_kv_cache_bytes_from_file,
                                 sleep_mode_enabled, try_register_lib)
  from vllm_ascend.worker.model_runner_v1 import NPUModelRunner
-+from unifiedcache.integration.vllm.ucm_sparse.state import ensure_ucm_sparse_initialized
++from ucm.integration.vllm.ucm_sparse.state import ensure_ucm_sparse_initialized
 
 
  class NPUWorker(WorkerBase):