Latest changes to aadpt to upstream master

christian-pinto · christian-pinto · commit c7bef7048b1a · 2025-06-24T14:25:03.000Z
diff --git a/examples/offline_inference/prithvi_geospatial_mae.py b/examples/offline_inference/prithvi_geospatial_mae.py
@@ -143,7 +143,8 @@ def __init__(self):
         self.model = LLM(
             model=os.path.join(os.path.dirname(__file__), "./model"),
             skip_tokenizer_init=True,
-            dtype="float32",
+            dtype="float16",
+            enforce_eager=True
         )
 
     def run(self, input_data, location_coords):
diff --git a/vllm/v1/engine/core.py b/vllm/v1/engine/core.py
@@ -139,7 +139,7 @@ def _initialize_kv_caches(
             # is attention free.
             kv_cache_specs = []
             kv_cache_configs = [
-                    KVCacheConfig(num_blocks=0, tensors={}, kv_cache_groups=[])
+                    KVCacheConfig(num_blocks=0, kv_cache_tensors={}, kv_cache_groups=[])
                 ]
         else:
             # Get all kv cache needed by the model

Original file line number	Diff line number	Diff line change
`@@ -143,7 +143,8 @@ def __init__(self):`
`143`	`143`	`self.model = LLM(`
`144`	`144`	`model=os.path.join(os.path.dirname(__file__), "./model"),`
`145`	`145`	`skip_tokenizer_init=True,`
`146`		`- dtype="float32",`
	`146`	`+ dtype="float16",`
	`147`	`+ enforce_eager=True`
`147`	`148`	`)`
`148`	`149`
`149`	`150`	`def run(self, input_data, location_coords):`
Original file line number	Diff line number	Diff line change
`@@ -139,7 +139,7 @@ def _initialize_kv_caches(`
`139`	`139`	`# is attention free.`
`140`	`140`	`kv_cache_specs = []`
`141`	`141`	`kv_cache_configs = [`
`142`		`- KVCacheConfig(num_blocks=0, tensors={}, kv_cache_groups=[])`
	`142`	`+ KVCacheConfig(num_blocks=0, kv_cache_tensors={}, kv_cache_groups=[])`
`143`	`143`	`]`
`144`	`144`	`else:`
`145`	`145`	`# Get all kv cache needed by the model`