Release: v0.13.2

* [Device map] nn.Parameter don't have children

* Update src/accelerate/utils/modeling.py

Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

setup.py

@@ -32,7 +32,7 @@ extras["sagemaker"] = [
 
 setup(
     name="accelerate",
-    version="0.13.0",
+    version="0.13.2",
     description="Accelerate",
     long_description=open("README.md", "r", encoding="utf-8").read(),
     long_description_content_type="text/markdown",

src/accelerate/__init__.py

@@ -2,7 +2,7 @@
 # There's no way to ignore "F401 '...' imported but unused" warnings in this
 # module, but to preserve other warnings. So, don't check this module at all.
 
-__version__ = "0.13.0"
+__version__ = "0.13.2"
 
 from .accelerator import Accelerator
 from .big_modeling import cpu_offload, disk_offload, dispatch_model, init_empty_weights, load_checkpoint_and_dispatch

src/accelerate/commands/config/cluster.py

@@ -35,6 +35,7 @@ def get_cluster_input():
 
     machine_rank = 0
     num_machines = 1
+    num_processes = 1
     gpu_ids = None
     main_process_ip = None
     main_process_port = None
@@ -294,12 +295,6 @@ def get_cluster_input():
             default=1,
             error_message="Please enter an integer.",
         )
-
-    if distributed_type in [DistributedType.MULTI_GPU, DistributedType.NO] and not use_cpu:
-        gpu_ids = _ask_field(
-            "What GPU(s) (by id) should be used for training on this machine as a comma-seperated list? [all]:",
-            default="all",
-        )
     elif distributed_type in [DistributedType.FSDP, DistributedType.DEEPSPEED]:
         num_processes = _ask_field(
             "How many GPU(s) should be used for distributed training? [1]:",
@@ -310,6 +305,12 @@ def get_cluster_input():
     else:
         num_processes = 1
 
+    if distributed_type in [DistributedType.MULTI_GPU, DistributedType.NO] and not use_cpu:
+        gpu_ids = _ask_field(
+            "What GPU(s) (by id) should be used for training on this machine as a comma-seperated list? [all]:",
+            default="all",
+        )
+
     if distributed_type != DistributedType.TPU:
         if distributed_type == DistributedType.DEEPSPEED and use_deepspeed_config:
             mixed_precision = "no"

src/accelerate/utils/modeling.py

@@ -258,7 +258,7 @@ def get_max_layer_size(
     modules_to_treat = modules.copy()
     while len(modules_to_treat) > 0:
         module_name, module = modules_to_treat.pop(0)
-        modules_children = list(module.named_children())
+        modules_children = list(module.named_children()) if isinstance(module, torch.nn.Module) else []
         if len(modules_children) == 0 or module.__class__.__name__ in no_split_module_classes:
             # No splitting this one so we compare to the max_size
             size = module_sizes[module_name]