Fix for deepspeed training

finetune/accelerate_config_machine_single.yaml

@@ -1,21 +1,17 @@
 compute_environment: LOCAL_MACHINE
+gpu_ids: "0"
 debug: false
 deepspeed_config:
-  gradient_accumulation_steps: 1
+  deepspeed_config_file: ds_config.json
-  gradient_clipping: 1.0
-  offload_optimizer_device: none
-  offload_param_device: none
   zero3_init_flag: false
-  zero_stage: 2
 distributed_type: DEEPSPEED
 downcast_bf16: 'no'
 enable_cpu_affinity: false
 machine_rank: 0
 main_training_function: main
 dynamo_backend: 'no'
-mixed_precision: 'no'
 num_machines: 1
-num_processes: 8
+num_processes: 1
 rdzv_backend: static
 same_network: true
 tpu_env: []

finetune/ds_config.json

@@ -0,0 +1,20 @@
+{
+  "scheduler": {
+    "type": "WarmupDecayLR",
+    "params": {
+        "warmup_min_lr": "auto",
+        "warmup_max_lr": "auto",
+        "warmup_num_steps": "auto",
+        "total_num_steps": "auto"
+    }
+  },
+  "zero_optimization": {
+      "stage": 2,
+      "allgather_partitions": true,
+      "allgather_bucket_size": 2e8,
+      "overlap_comm": true,
+      "reduce_scatter": true,
+      "reduce_bucket_size": 1e8,
+      "contiguous_gradients": true
+  }
+}

finetune/train_cogvideox_lora.py

@@ -948,6 +948,10 @@ def main(args):
 
     logging_dir = Path(args.output_dir, args.logging_dir)
 
+    expected_midxed_precision = "bf16" if "5b" in args.pretrained_model_name_or_path.lower() else "fp16"
+    if args.mixed_precision != expected_midxed_precision:
+        raise ValueError(f"Mixed precision {args.mixed_precision} does not match the model precision, should be {expected_midxed_precision}")
+
     accelerator_project_config = ProjectConfiguration(project_dir=args.output_dir, logging_dir=logging_dir)
     kwargs = DistributedDataParallelKwargs(find_unused_parameters=True)
     accelerator = Accelerator(
@@ -958,6 +962,28 @@ def main(args):
         kwargs_handlers=[kwargs],
     )
 
+    if accelerator.state.deepspeed_plugin:
+        # Set deepspeed config according to args
+        config = {
+            'optimizer': {
+                'type': args.optimizer,
+                'params': {
+                    'lr': args.learning_rate,
+                    'betas': [args.adam_beta1, args.adam_beta2]
+                },
+                'torch_adam': True
+            },
+            'bf16': {
+                'enabled': True if args.mixed_precision == "bf16" else False
+            },
+            'fp16': {
+                'enabled': True if args.mixed_precision == "fp16" else False
+            },
+            'gradient_accumulation_steps': args.gradient_accumulation_steps,
+            'train_batch_size': args.train_batch_size
+        }
+        accelerator.state.deepspeed_plugin.deepspeed_config.update(config)
+
     # Disable AMP for MPS.
     if torch.backends.mps.is_available():
         accelerator.native_amp = False
@@ -1045,7 +1071,7 @@ def main(args):
             "bf16" in accelerator.state.deepspeed_plugin.deepspeed_config
             and accelerator.state.deepspeed_plugin.deepspeed_config["bf16"]["enabled"]
         ):
-            weight_dtype = torch.float16
+            weight_dtype = torch.bfloat16
     else:
         if accelerator.mixed_precision == "fp16":
             weight_dtype = torch.float16
@@ -1158,11 +1184,11 @@ def main(args):
 
     use_deepspeed_optimizer = (
         accelerator.state.deepspeed_plugin is not None
-        and accelerator.state.deepspeed_plugin.deepspeed_config.get("optimizer", "none").lower() == "none"
+        and "optimizer" in accelerator.state.deepspeed_plugin.deepspeed_config
     )
     use_deepspeed_scheduler = (
         accelerator.state.deepspeed_plugin is not None
-        and accelerator.state.deepspeed_plugin.deepspeed_config.get("scheduler", "none").lower() == "none"
+        and "scheduler" in accelerator.state.deepspeed_plugin.deepspeed_config
     )
 
     optimizer = get_optimizer(args, params_to_optimize, use_deepspeed=use_deepspeed_optimizer)