we got it to BUILD

python/vllm_engine/pyproject.toml

@@ -12,14 +12,13 @@ dependencies = [
 ]
 
 [tool.uv]
-environments = [
-    "sys_platform == 'linux' and platform_machine == 'aarch64'"
-]
+environments = ["sys_platform == 'linux' and platform_machine == 'aarch64'"]
 
 
 [tool.uv.sources]
 vllm = { git = "https://github.com/hmellor/vllm.git", rev = "b99bedc737166ae5ca98cb9e3534b96e0c8c69aa" }
-torch = [{ index = "pytorch-cu130", marker = "platform_machine == 'aarch64'" },
+torch = [
+  { index = "pytorch-cu130", marker = "platform_machine == 'aarch64'" },
   { index = "pytorch-cpu", marker = "platform_machine == 'x86_64'" },
 ]
 

python/vllm_engine/src/vllm_engine/growable_cache.py

@@ -339,9 +339,8 @@ def _patch_get_computed_blocks() -> None:
         ):
             return original(self, request)
 
-        from vllm.utils.math_utils import cdiv  # type: ignore[reportMissingImports]
-
         from exo.worker.engines.vllm.vllm_generator import _build_layer_groups
+        from vllm.utils.math_utils import cdiv  # type: ignore[reportMissingImports]
 
         num_groups = len(self.kv_cache_config.kv_cache_groups)
         null_block = self.block_pool.null_block

python/vllm_engine/src/vllm_engine/prompt_format.py

@@ -1,9 +1,9 @@
+from exo.shared.types.common import ModelId
+from exo.shared.types.text_generation import TextGenerationTaskParams
 from mlx_lm.tokenizer_utils import TokenizerWrapper
 from vllm.sampling_params import SamplingParams
 from vllm.v1.engine.llm_engine import LLMEngine
 
-from exo.shared.types.common import ModelId
-from exo.shared.types.text_generation import TextGenerationTaskParams
 from exo.worker.engines.mlx.utils_mlx import (
     apply_chat_template,
     get_eos_token_ids_for_model,

python/vllm_engine/src/vllm_engine/vllm_generator.py

@@ -8,11 +8,6 @@ from collections.abc import Callable, Generator
 from dataclasses import dataclass, field
 
 import torch
-from vllm.engine.arg_utils import EngineArgs
-from vllm.sampling_params import SamplingParams
-from vllm.v1.engine.llm_engine import LLMEngine
-from vllm.v1.kv_cache_interface import KVCacheConfig
-
 from exo.shared.types.api import (
     CompletionTokensDetails,
     GenerationStats,
@@ -24,8 +19,6 @@ from exo.shared.types.memory import Memory
 from exo.shared.types.tasks import TaskId
 from exo.shared.types.text_generation import TextGenerationTaskParams
 from exo.shared.types.worker.runner_response import GenerationResponse
-from exo.worker.engines.mlx.cache import KVPrefixCache
-from exo.worker.engines.mlx.utils_mlx import get_eos_token_ids_for_model
 from exo.worker.engines.vllm.growable_cache import (
     get_model_runner,
     patch_vllm,
@@ -36,6 +29,13 @@ from exo.worker.engines.vllm.prompt_format import (
     format_vllm_prompt,
     make_vllm_sampling_params,
 )
+from vllm.engine.arg_utils import EngineArgs
+from vllm.sampling_params import SamplingParams
+from vllm.v1.engine.llm_engine import LLMEngine
+from vllm.v1.kv_cache_interface import KVCacheConfig
+
+from exo.worker.engines.mlx.cache import KVPrefixCache
+from exo.worker.engines.mlx.utils_mlx import get_eos_token_ids_for_model
 from exo.worker.runner.bootstrap import logger
 from exo.worker.runner.llm_inference.tool_parsers import ToolParser, infer_tool_parser