[Core] Account for num_gpu_blocks_override in max_model_len checks (#41069)
Signed-off-by: Nick Hill <nickhill123@gmail.com>
This commit is contained in:
@@ -5874,7 +5874,7 @@ class GPUModelRunner(
|
||||
saved_override = self.cache_config.num_gpu_blocks_override
|
||||
self.cache_config.num_gpu_blocks_override = min_blocks
|
||||
minimal_config = get_kv_cache_config_from_groups(
|
||||
self.vllm_config, kv_cache_groups, available_memory=0, suppress_log=True
|
||||
self.vllm_config, kv_cache_groups, available_memory=0
|
||||
)
|
||||
self.cache_config.num_gpu_blocks_override = saved_override
|
||||
|
||||
|
||||
Reference in New Issue
Block a user