The most powerful and modular diffusion model GUI, api and backend with a graph/nodes interface. The fastest local inference engine in the world.
mm: wrap the raw stream in context manager (#10958)
rattus0ff045789246
diff --git a/comfy/model_management.py b/comfy/model_management.pyindex d8ce80010..aeddbaefe 100644--- a/comfy/model_management.py+++ b/comfy/model_management.py@@ -1055,7 +1055,9 @@ def get_offload_stream(device):elif is_device_cuda(device):ss = []for k in range(NUM_STREAMS):- ss.append(torch.cuda.Stream(device=device, priority=0))+ s1 = torch.cuda.Stream(device=device, priority=0)+ s1.as_context = torch.cuda.stream+ ss.append(s1)STREAMS[device] = sss = ss[stream_counter]stream_counters[device] = stream_counter@@ -1063,7 +1065,9 @@ def get_offload_stream(device):elif is_device_xpu(device):ss = []for k in range(NUM_STREAMS):- ss.append(torch.xpu.Stream(device=device, priority=0))+ s1 = torch.xpu.Stream(device=device, priority=0)+ s1.as_context = torch.xpu.stream+ ss.append(s1)STREAMS[device] = sss = ss[stream_counter]stream_counters[device] = stream_counter@@ -1081,12 +1085,19 @@ def cast_to(weight, dtype=None, device=None, non_blocking=False, copy=False, strif dtype is None or weight.dtype == dtype:return weightif stream is not None:- with stream:+ wf_context = stream+ if hasattr(wf_context, "as_context"):+ wf_context = wf_context.as_context(stream)+ with wf_context:return weight.to(dtype=dtype, copy=copy)return weight.to(dtype=dtype, copy=copy)+if stream is not None:- with stream:+ wf_context = stream+ if hasattr(wf_context, "as_context"):+ wf_context = wf_context.as_context(stream)+ with wf_context:r = torch.empty_like(weight, dtype=dtype, device=device)r.copy_(weight, non_blocking=non_blocking)else:diff --git a/comfy/ops.py b/comfy/ops.pyindex a0ff4e8f1..61a2f0754 100644--- a/comfy/ops.py+++ b/comfy/ops.py@@ -95,6 +95,8 @@ def cast_bias_weight(s, input=None, dtype=None, device=None, bias_dtype=None, ofif offload_stream is not None:wf_context = offload_stream+ if hasattr(wf_context, "as_context"):+ wf_context = wf_context.as_context(offload_stream)else:wf_context = contextlib.nullcontext()