A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 16, 02:00 UTC
[Bugfix] Fix _get_lora_device for HQQ marlin (#12090)
Signed-off-by: Varun Sundar Rabindranath <varun@neuralmagic.com> Co-authored-by: Varun Sundar Rabindranath <varun@neuralmagic.com>
Varun Sundar Rabindranath committed
Jan 16, 2025 at 01:29 UTC
ebd8c669efa54a218eb83735fd7ba40922f5f3ad
1 file changed
+3
vllm/lora/layers.py
+3
@@ -51,6 +51,9 @@ def _get_lora_device(base_layer: nn.Module) -> torch.device:
51
# marlin
52
elif hasattr(base_layer, "B"):
53
return base_layer.B.device
54
+ # HQQ marlin
55
+ elif hasattr(base_layer, "W_q"):
56
+ return base_layer.W_q.device
57
else:
58
raise ValueError(f"Unsupported base layer: {base_layer}")
59