fix: defer auto-round imports and fix tokenizer variable name

- Change default iters from 200 to 100 in usage comment
- Remove unused imports (load_dataset, get_rank_partition)
- Defer AutoRoundModifier import until after model loading
- Use get_dataset from auto_round.calib_dataset directly
- Fix model_name -> model_id for tokenizer loading

Signed-off-by: yiliu30 <yi4.liu@intel.com>

Author: yiliu30

experimental/ddp/ddp_qwen3_example.py

@@ -7,7 +7,7 @@
 # torchrun --nproc_per_node=2 ddp_qwen3_example.py \
 #     --model Qwen/Qwen3-8B \
 #     --nsamples 128 \
-#     --iters 200 \
+#     --iters 100 \
 #     --disable_torch_compile \
 #     --deterministic
 #############################################################################
@@ -19,13 +19,10 @@
 import torch
 import torch.distributed as dist
 from compressed_tensors.offload import dispatch_model, init_dist, load_offloaded_model
-from datasets import load_dataset
 from loguru import logger
 from transformers import AutoModelForCausalLM, AutoTokenizer
 import torch.distributed as dist
 from llmcompressor import oneshot
-from llmcompressor.datasets.utils import get_rank_partition
-from llmcompressor.modifiers.autoround import AutoRoundModifier
 
 
 def fix_everything(seed=42):
@@ -87,14 +84,19 @@ def config_deterministic():
     )
 ##################################
 
-tokenizer = AutoTokenizer.from_pretrained(model_name)
+tokenizer = AutoTokenizer.from_pretrained(model_id)
 
 # Select calibration dataset.
 NUM_CALIBRATION_SAMPLES = args.nsamples
 MAX_SEQUENCE_LENGTH = 2048
 ITERS = args.iters
-# Get aligned calibration dataset.
 
+# Make sure model are loaded before importing auto-round related code.
+# This requirement will be lifted once https://github.com/intel/auto-round/pull/1460 is merged.
+from llmcompressor.modifiers.autoround import AutoRoundModifier
+
+# Get aligned calibration dataset.
+from auto_round.calib_dataset import get_dataset
 ds = get_dataset(
     tokenizer=tokenizer,
     seqlen=MAX_SEQUENCE_LENGTH,