CI fix (#186)

vllm-project/vllm#24795 and
vllm-project/vllm#24615 and
vllm-project/vllm#24078

---------

Signed-off-by: Agata Dobrzyniewicz <[email protected]>

Author: adobrzyn

tests/unit_tests/worker/test_hpu_input_batch.py

@@ -15,6 +15,7 @@
 from vllm.v1.pool.metadata import PoolingMetadata
 from vllm.v1.sample.logits_processor import LogitsProcessors
 from vllm.v1.sample.metadata import SamplingMetadata
+from vllm.v1.utils import CpuGpuBuffer
 from vllm.v1.worker.block_table import BlockTable, MultiGroupBlockTable
 from vllm_gaudi.v1.worker.hpu_input_batch import InputBatch, CachedRequestState
 
@@ -37,7 +38,7 @@ def _compare_objs(obj1, obj2, skip: Sequence = ("logitsprocs", "batch_update_bui
 
         is_same = False
         if isinstance(a, torch.Tensor):
-            if (a.numel() == 0 or b.numel() == 0):
+            if a.numel() == 0 or b.numel() == 0:
                 is_same = (a.numel() == 0 and b.numel() == 0)
             elif torch.allclose(a, b):
                 is_same = True
@@ -53,6 +54,8 @@ def _compare_objs(obj1, obj2, skip: Sequence = ("logitsprocs", "batch_update_bui
             is_same = True  # if we make it here must be same
         elif a == b:
             is_same = True
+        elif isinstance(a, CpuGpuBuffer):
+            is_same = np.allclose(a.np, b.np) and torch.allclose(a.gpu, b.gpu)
         assert is_same, f"Attribute {attr_name} is different"\
             f" in {obj1} and {obj2}: {a} != {b}"
 

tests/unit_tests/worker/test_hpu_model_runner.py

@@ -157,7 +157,7 @@ def _is_req_state_block_table_match(model_runner, req_id: str) -> bool:
     if block_table.num_blocks_per_row[req_index] != len(req_state.block_ids[0]):
         return False
     num_blocks = block_table.num_blocks_per_row[req_index]
-    return (block_table.block_table_np[req_index, :num_blocks] == req_state.block_ids[0]).all()
+    return (block_table.block_table.np[req_index, :num_blocks] == req_state.block_ids[0]).all()
 
 
 def test_update_states_new_request(model_runner, dist_init):

vllm_gaudi/v1/worker/hpu_model_runner.py

@@ -43,7 +43,7 @@
 from vllm.model_executor.layers.rotary_embedding import MRotaryEmbedding
 from vllm.multimodal.inputs import PlaceholderRange
 from vllm.sampling_params import SamplingType
-from vllm.transformers_utils.tokenizer_group import init_tokenizer_from_configs
+from vllm.transformers_utils.tokenizer import init_tokenizer_from_configs
 from vllm.utils import (STR_DTYPE_TO_TORCH_DTYPE, LayerBlockType, cdiv, is_pin_memory_available, LazyLoader)
 from vllm_gaudi.utils import (HPUCompileConfig, is_fake_hpu, async_h2d_copy)
 from vllm_gaudi.v1.attention.backends.hpu_attn import HPUAttentionMetadataV1
@@ -728,9 +728,7 @@ def __init__(
             logger.info("Bucketing is OFF.")
         self._PAD_SLOT_ID = -1
         self._PAD_BLOCK_ID = -1
-        self._tokenizer = init_tokenizer_from_configs(model_config=vllm_config.model_config,
-                                                      scheduler_config=vllm_config.scheduler_config,
-                                                      lora_config=vllm_config.lora_config).tokenizer
+        self._tokenizer = init_tokenizer_from_configs(model_config=vllm_config.model_config)
 
         # TODO(madamczyk-intel): add a knob for that
         # TODO(madamczyk-intel): debug why increasing it lowers acc