Simple Fairscale Model Parallelization works locally, but using Sagemaker SMP gives me errors

I’ve posted the stacktrace from sagemaker, but essentially i get ValueError: not enough values to unpack (expected 2, got 1). I can train fine locally on multiple gpus using fairscale simple.

2022-03-02 15:14:03 Starting - Starting the training job...
2022-03-02 15:14:27 Starting - Launching requested ML instancesProfilerReport-1646234043: InProgress
.........
smdistributed/modelparallel/backend/split.py:166] Non-splittable object of type <class 'NoneType'> passed to smp.step. If this object contains tensors that need to be split across microbatches, implement a 'smp_slice' method for this class. See SMP documentation for further information.
[1,mpirank:2,algo-1]<stdout>:[2022-03-02 15:22:36.545 algo-1:54 INFO utils.py:27] RULE_JOB_STOP_SIGNAL_FILENAME: None
[1,mpirank:0,algo-1]<stderr>:Using apex fp16 backend
[1,mpirank:0,algo-1]<stdout>:Using apex fp16 backend
[1,mpirank:6,algo-1]<stdout>:[2022-03-02 15:22:36.560 algo-1:58 INFO profiler_config_parser.py:102] User has disabled profiler.
[1,mpirank:6,algo-1]<stdout>:[2022-03-02 15:22:36.560 algo-1:58 INFO json_config.py:91] Creating hook from json_config at /opt/ml/input/config/debughookconfig.json.
[1,mpirank:6,algo-1]<stdout>:[2022-03-02 15:22:36.561 algo-1:58 INFO hook.py:200] tensorboard_dir has not been set for the hook. SMDebug will not be exporting tensorboard summaries.
[1,mpirank:6,algo-1]<stdout>:[2022-03-02 15:22:36.561 algo-1:58 INFO hook.py:255] Saving to /opt/ml/output/tensors
[1,mpirank:6,algo-1]<stdout>:[2022-03-02 15:22:36.561 algo-1:58 INFO state_store.py:77] The checkpoint config file /opt/ml/input/config/checkpointconfig.json does not exist.
[1,mpirank:0,algo-1]<stderr>:***** Running training *****
[1,mpirank:0,algo-1]<stdout>:***** Running training *****
[1,mpirank:0,algo-1]<stderr>:  Num examples = 32
[1,mpirank:0,algo-1]<stderr>:  Num Epochs = 1
[1,mpirank:0,algo-1]<stderr>:  Instantaneous batch size per device = 1
[1,mpirank:0,algo-1]<stdout>:  Num examples = 32
[1,mpirank:0,algo-1]<stdout>:  Num Epochs = 1
[1,mpirank:0,algo-1]<stdout>:  Instantaneous batch size per device = 1
[1,mpirank:0,algo-1]<stdout>:  Total train batch size (w. parallel, distributed & accumulation) = 8
[1,mpirank:0,algo-1]<stderr>:  Total train batch size (w. parallel, distributed & accumulation) = 8
[1,mpirank:0,algo-1]<stderr>:  Gradient Accumulation steps = 4
[1,mpirank:0,algo-1]<stderr>:  Total optimization steps = 4
[1,mpirank:0,algo-1]<stdout>:  Gradient Accumulation steps = 4
[1,mpirank:0,algo-1]<stdout>:  Total optimization steps = 4
[1,mpirank:0,algo-1]<stderr>:#015  0%|          | 0/4 [00:00<?, ?it/s]
[
[1,mpirank:0,algo-1]<stdout>:[2022-03-02 15:22:36.670 algo-1:52 INFO hook.py:200] tensorboard_dir has not been set for the hook. SMDebug will not be exporting tensorboard summaries.
[1,mpirank:0,algo-1]<stdout>:[2022-03-02 15:22:36.671 algo-1:52 INFO hook.py:255] Saving to /opt/ml/output/tensors
[1,mpirank:0,algo-1]<stdout>:[2022-03-02 15:22:36.671 algo-1:52 INFO state_store.py:77] The checkpoint config file /opt/ml/input/config/checkpointconfig.json does not exist.
[1,mpirank:0,algo-1]<stdout>:[2022-03-02 15:22:36.678: W smdistributed/modelparallel/backend/split.py:166] Non-splittable object of type <class 'NoneType'> passed to smp.step. If this object contains tensors that need to be split across microbatches, implement a 'smp_slice' method for this class. See SMP documentation for further information.
[1,mpirank:0,algo-1]<stdout>:[2022-03-02 15:22:36.679: I smdistributed/modelparallel/torch/worker.py:280] Tracing on GPU. If the model parameters do not fit in a single GPU, you can set trace_device to `cpu`.
[1,mpirank:0,algo-1]<stdout>:[2022-03-02 15:22:36.995 algo-1:52 INFO hook.py:591] name:led.shared.weight count_params:51470336
[1,mpirank:0,algo-1]<stdout>:[2022-03-02 15:22:36.996 algo-1:52 INFO hook.py:591] name:led.encoder.embed_positions.weight count_params:16777216
[1,mpirank:0,algo-1]<stdout>:[2022-03-02 15:22:37.014 algo-1:52 INFO hook.py:593] Total Trainable Params: 359020544
[1,mpirank:0,algo-1]<stdout>:[2022-03-02 15:22:37.014 algo-1:52 INFO hook.py:424] Monitoring the collections: losses
[1,mpirank:0,algo-1]<stdout>:[2022-03-02 15:22:37.064: C smdistributed/modelparallel/torch/worker.py:105] [0] Hit an exception for 0/0 on thread 0: not enough values to unpack (expected 2, got 1)
[1,mpirank:0,algo-1]<stdout>:[2022-03-02 15:22:37.074: C smdistributed/modelparallel/torch/worker.py:110] [0]   File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/worker.py", line 469, in _thread_compute
[1,mpirank:0,algo-1]<stdout>:    self.thread_execute_tracing(req)
[1,mpirank:0,algo-1]<stdout>:  File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/worker.py", line 286, in thread_execute_tracing
[1,mpirank:0,algo-1]<stdout>:    self._exec_trace_on_device(req, device)
[1,mpirank:0,algo-1]<stdout>:  File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/worker.py", line 250, in _exec_trace_on_device
[1,mpirank:0,algo-1]<stdout>:    outputs = step_fn(*args, **kwargs)
[1,mpirank:0,algo-1]<stdout>:  File "/opt/conda/lib/python3.8/site-packages/transformers/trainer_pt_utils.py", line 1014, in smp_forward_backward
[1,mpirank:0,algo-1]<stdout>:    outputs = model(**inputs)
[1,mpirank:0,algo-1]<stdout>:  File "/opt/conda/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1100, in _call_impl
[1,mpirank:0,algo-1]<stdout>:    result = forward_call(*input, **kwargs)
[1,mpirank:0,algo-1]<stdout>:  File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/patches/tracing.py", line 68, in trace_forward
[1,mpirank:0,algo-1]<stdout>:    raise e
[1,mpirank:0,algo-1]<stdout>:  File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/patches/tracing.py", line 51, in trace_forward
[1,mpirank:0,algo-1]<stdout>:    output = original_forward(self, *args, **kwargs)
[1,mpirank:0,algo-1]<stdout>:  File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/model.py", line 436, in forward
[1,mpirank:0,algo-1]<stdout>:    return self.module(*args, **kwargs)
[1,mpirank:0,algo-1]<stdout>:  File "/opt/conda/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1100, in _call_impl
[1,mpirank:0,algo-1]<stdout>:    result = forward_call(*input, **kwargs)
[1,mpirank:0,algo-1]<stdout>:  File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/patches/tracing.py", line 68, in trace_forward
[1,mpirank:0,algo-1]<stdout>:    raise e
[1,mpirank:0,algo-1]<stdout>:  File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/patches/tracing.py", line 51, in trace_forward
[1,mpirank:0,algo-1]<stdout>:    output = original_forward(self, *args, **kwargs)
[1,mpirank:0,algo-1]<stdout>:  File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/ddp_model.py", line 222, in forward
[1,mpirank:0,algo-1]<stdout>:    return self.module(*args, **kwargs)
[1,mpirank:0,algo-1]<stdout>:  File "/opt/conda/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1100, in _call_impl
[1,mpirank:0,algo-1]<stdout>:    result = forward_call(*input, **kwargs)
[1,mpirank:0,algo-1]<stdout>:  File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/patches/tracing.py", line 68, in trace_forward
[1,mpirank:0,algo-1]<stdout>:    raise e
[1,mpirank:0,algo-1]<stdout>:  File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/patches/tracing.py", line 51, in trace_forward
[1,mpirank:0,algo-1]<stdout>:    output = original_forward(self, *args, **kwargs)
[1,mpirank:0,algo-1]<stdout>:  File "/opt/conda/lib/python3.8/site-packages/transformers/models/led/modeling_led.py", line 2365, in forward
[1,mpirank:0,algo-1]<stdout>:    outputs = self.led(
[1,mpirank:0,algo-1]<stdout>:  File "/opt/conda/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1100, in _call_impl
[1,mpirank:0,algo-1]<stdout>:    result = forward_call(*input, **kwargs)
[1,mpirank:0,algo-1]<stdout>:  File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/patches/tracing.py", line 68, in trace_forward
[1,mpirank:0,algo-1]<stdout>:    raise e
[1,mpirank:0,algo-1]<stdout>:  File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/patches/tracing.py", line 51, in trace_forward
[1,mpirank:0,algo-1]<stdout>:    output = original_forward(self, *args, **kwargs)
[1,mpirank:0,algo-1]<stdout>:  File "/opt/conda/lib/python3.8/site-packages/transformers/models/led/modeling_led.py", line 2217, in forward
[1,mpirank:0,algo-1]<stdout>:    encoder_outputs = self.encoder(
[1,mpirank:0,algo-1]<stdout>:  File "/opt/conda/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1100, in _call_impl
[1,mpirank:0,algo-1]<stdout>:    result = forward_call(*input, **kwargs)
[1,mpirank:0,algo-1]<stdout>:  File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/patches/tracing.py", line 68, in trace_forward
[1,mpirank:0,algo-1]<stdout>:    raise e
[1,mpirank:0,algo-1]<stdout>:  File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/patches/tracing.py", line 51, in trace_forward
[1,mpirank:0,algo-1]<stdout>:    output = original_forward(self, *args, **kwargs)
[1,mpirank:0,algo-1]<stdout>:  File "/opt/conda/lib/python3.8/site-packages/transformers/models/led/modeling_led.py", line 1794, in forward
[1,mpirank:0,algo-1]<stdout>:    embed_pos = self.embed_positions(input_shape)
[1,mpirank:0,algo-1]<stdout>:  File "/opt/conda/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1100, in _call_impl
[1,mpirank:0,algo-1]<stdout>:    result = forward_call(*input, **kwargs)
[1,mpirank:0,algo-1]<stdout>:  File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/patches/tracing.py", line 68, in trace_forward
[1,mpirank:0,algo-1]<stdout>:    raise e
[1,mpirank:0,algo-1]<stdout>:  File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/patches/tracing.py", line 51, in trace_forward
[1,mpirank:0,algo-1]<stdout>:    output = original_forward(self, *args, **kwargs)
[1,mpirank:0,algo-1]<stdout>:  File "/opt/conda/lib/python3.8/site-packages/transformers/models/led/modeling_led.py", line 124, in forward
[1,mpirank:0,algo-1]<stdout>:    return super().forward(positions)
[1,mpirank:0,algo-1]<stdout>:  File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/patches/tracing.py", line 68, in trace_forward
[1,mpirank:0,algo-1]<stdout>:    raise e
[1,mpirank:0,algo-1]<stdout>:  File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/patches/tracing.py", line 51, in trace_forward
[1,mpirank:0,algo-1]<stdout>:    output = original_forward(self, *args, **kwargs)
[1,mpirank:0,algo-1]<stdout>:  File "/opt/conda/lib/python3.8/site-packages/transformers/models/led/modeling_led.py", line 124, in forward
[1,mpirank:0,algo-1]<stdout>:    return super().forward(positions)
[1,mpirank:0,algo-1]<stdout>:  File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/patches/tracing.py", line 68, in trace_forward
[1,mpirank:0,algo-1]<stdout>:    raise e
[1,mpirank:0,algo-1]<stdout>:  File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/patches/tracing.py", line 51, in trace_forward
[1,mpirank:0,algo-1]<stdout>:    output = original_forward(self, *args, **kwargs)
[1,mpirank:0,algo-1]<stdout>:  File "/opt/conda/lib/python3.8/site-packages/transformers/models/led/modeling_led.py", line 120, in forward
[1,mpirank:0,algo-1]<stdout>:    bsz, seq_len = input_ids_shape[:2]
[1,mpirank:0,algo-1]<stdout>:
[1,mpirank:0,algo-1]<stdout>:[2022-03-02 15:22:37.075: C smdistributed/modelparallel/torch/worker.py:111] [0] Parent exec stack []
[1,mpirank:0,algo-1]<stdout>:[2022-03-02 15:22:37.075: C smdistributed/modelparallel/torch/worker.py:112] [0] Req <TraceReq::mb:0, requester:0>
[1,mpirank:0,algo-1]<stderr>:Traceback (most recent call last):
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/runpy.py", line 194, in _run_module_as_main
[1,mpirank:0,algo-1]<stderr>:    return _run_code(code, main_globals, None,
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/runpy.py", line 87, in _run_code
[1,mpirank:0,algo-1]<stderr>:    exec(code, run_globals)
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/site-packages/mpi4py/__main__.py", line 7, in <module>
[1,mpirank:0,algo-1]<stderr>:    main()
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/site-packages/mpi4py/run.py", line 196, in main
[1,mpirank:0,algo-1]<stderr>:    run_command_line(args)
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/site-packages/mpi4py/run.py", line 47, in run_command_line
[1,mpirank:0,algo-1]<stderr>:    run_path(sys.argv[0], run_name='__main__')
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/runpy.py", line 265, in run_path
[1,mpirank:0,algo-1]<stderr>:    return _run_module_code(code, init_globals, run_name,
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/runpy.py", line 97, in _run_module_code
[1,mpirank:0,algo-1]<stderr>:    _run_code(code, mod_globals, init_globals,
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/runpy.py", line 87, in _run_code
[1,mpirank:0,algo-1]<stderr>:    exec(code, run_globals)
[1,mpirank:0,algo-1]<stderr>:  File "ledFinalTrainer.py", line 253, in <module>
[1,mpirank:0,algo-1]<stderr>:    trainer.train()
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/site-packages/transformers/trainer.py", line 1316, in train
[1,mpirank:0,algo-1]<stderr>:    tr_loss_step = self.training_step(model, inputs)
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/site-packages/transformers/trainer.py", line 1842, in training_step
[1,mpirank:0,algo-1]<stderr>:    loss_mb = smp_forward_backward(model, inputs, self.args.gradient_accumulation_steps, scaler=scaler)
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/step.py", line 193, in __call__
[1,mpirank:0,algo-1]<stderr>:    state.exec_server.run_step_leader(mb_args, mb_kwargs, self.id)
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/server.py", line 332, in run_step_leader
[1,mpirank:0,algo-1]<stderr>:    self.execute_request(
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/server.py", line 108, in execute_request
[1,mpirank:0,algo-1]<stderr>:    chosen_worker.execute(req)
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/worker.py", line 150, in execute
[1,mpirank:0,algo-1]<stderr>:    self._resume_thread_common()
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/worker.py", line 181, in _resume_thread_common
[1,mpirank:0,algo-1]<stderr>:    self._check_queue_after_thread_return()
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/worker.py", line 116, in _check_queue_after_thread_return
[1,mpirank:0,algo-1]<stderr>:    self._check_exception()
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/worker.py", line 113, in _check_exception
[1,mpirank:0,algo-1]<stderr>:    raise self.exception
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/worker.py", line 469, in _thread_compute
[1,mpirank:0,algo-1]<stderr>:    self.thread_execute_tracing(req)
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/worker.py", line 286, in thread_execute_tracing
[1,mpirank:0,algo-1]<stderr>:    self._exec_trace_on_device(req, device)
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/worker.py", line 250, in _exec_trace_on_device
[1,mpirank:0,algo-1]<stderr>:    outputs = step_fn(*args, **kwargs)
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/site-packages/transformers/trainer_pt_utils.py", line 1014, in smp_forward_backward
[1,mpirank:0,algo-1]<stderr>:    outputs = model(**inputs)
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1100, in _call_impl
[1,mpirank:0,algo-1]<stderr>:    result = forward_call(*input, **kwargs)
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/patches/tracing.py", line 68, in trace_forward
[1,mpirank:0,algo-1]<stderr>:    raise e
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/patches/tracing.py", line 51, in trace_forward
[1,mpirank:0,algo-1]<stderr>:    output = original_forward(self, *args, **kwargs)
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/model.py", line 436, in forward
[1,mpirank:0,algo-1]<stderr>:    return self.module(*args, **kwargs)
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1100, in _call_impl
[1,mpirank:0,algo-1]<stderr>:    result = forward_call(*input, **kwargs)
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/patches/tracing.py", line 68, in trace_forward
[1,mpirank:0,algo-1]<stderr>:    raise e
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/patches/tracing.py", line 51, in trace_forward
[1,mpirank:0,algo-1]<stderr>:    output = original_forward(self, *args, **kwargs)
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/ddp_model.py", line 222, in forward
[1,mpirank:0,algo-1]<stderr>:    return self.module(*args, **kwargs)
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1100, in _call_impl
[1,mpirank:0,algo-1]<stderr>:    result = forward_call(*input, **kwargs)
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/patches/tracing.py", line 68, in trace_forward
[1,mpirank:0,algo-1]<stderr>:    raise e
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/patches/tracing.py", line 51, in trace_forward
[1,mpirank:0,algo-1]<stderr>:    output = original_forward(self, *args, **kwargs)
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/site-packages/transformers/models/led/modeling_led.py", line 2365, in forward
[1,mpirank:0,algo-1]<stderr>:    outputs = self.led(
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1100, in _call_impl
[1,mpirank:0,algo-1]<stderr>:    result = forward_call(*input, **kwargs)
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/patches/tracing.py", line 68, in trace_forward
[1,mpirank:0,algo-1]<stderr>:    raise e
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/patches/tracing.py", line 51, in trace_forward
[1,mpirank:0,algo-1]<stderr>:    output = original_forward(self, *args, **kwargs)
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/site-packages/transformers/models/led/modeling_led.py", line 2217, in forward
[1,mpirank:0,algo-1]<stderr>:    encoder_outputs = self.encoder(
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1100, in _call_impl
[1,mpirank:0,algo-1]<stderr>:    result = forward_call(*input, **kwargs)
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/patches/tracing.py", line 68, in trace_forward
[1,mpirank:0,algo-1]<stderr>:    raise e
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/patches/tracing.py", line 51, in trace_forward
[1,mpirank:0,algo-1]<stderr>:    output = original_forward(self, *args, **kwargs)
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/site-packages/transformers/models/led/modeling_led.py", line 1794, in forward
[1,mpirank:0,algo-1]<stderr>:    embed_pos = self.embed_positions(input_shape)
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1100, in _call_impl
[1,mpirank:0,algo-1]<stderr>:    result = forward_call(*input, **kwargs)
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/patches/tracing.py", line 68, in trace_forward
[1,mpirank:0,algo-1]<stderr>:    raise e
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/patches/tracing.py", line 51, in trace_forward
[1,mpirank:0,algo-1]<stderr>:    output = original_forward(self, *args, **kwargs)
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/site-packages/transformers/models/led/modeling_led.py", line 124, in forward
[1,mpirank:0,algo-1]<stderr>:    return super().forward(positions)
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/patches/tracing.py", line 68, in trace_forward
[1,mpirank:0,algo-1]<stderr>:    raise e
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/patches/tracing.py", line 51, in trace_forward
[1,mpirank:0,algo-1]<stderr>:    output = original_forward(self, *args, **kwargs)
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/site-packages/transformers/models/led/modeling_led.py", line 124, in forward
[1,mpirank:0,algo-1]<stderr>:    return super().forward(positions)
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/patches/tracing.py", line 68, in trace_forward
[1,mpirank:0,algo-1]<stderr>:    raise e
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/site-packages/smdistributed/modelparallel/torch/patches/tracing.py", line 51, in trace_forward
[1,mpirank:0,algo-1]<stderr>:    output = original_forward(self, *args, **kwargs)
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/site-packages/transformers/models/led/modeling_led.py", line 120, in forward
[1,mpirank:0,algo-1]<stderr>:    bsz, seq_len = input_ids_shape[:2]
[1,mpirank:0,algo-1]<stderr>:ValueError: not enough values to unpack (expected 2, got 1)
--------------------------------------------------------------------------
Primary job  terminated normally, but 1 process returned
a non-zero exit code. Per user-direction, the job has been aborted.
--------------------------------------------------------------------------
--------------------------------------------------------------------------
mpirun.real detected that one or more processes exited with non-zero status, thus causing
the job to be terminated. The first process to do so was:
  Process name: [[41174,1],0]
  Exit code:    1
--------------------------------------------------------------------------
2022-03-02 15:22:39,556 sagemaker-training-toolkit ERROR    Reporting training FAILURE
2022-03-02 15:22:39,557 sagemaker-training-toolkit ERROR    ExecuteUserScriptError:
ExitCode 1
ErrorMessage ":ValueError: not enough values to unpack (expected 2, got 1)
 -------------------------------------------------------------------------- Primary job  terminated normally, but 1 process returned a non-zero exit code. Per user-direction, the job has been aborted. mpirun.real detected that one or more processes exited with non-zero status, thus causing the job to be terminated. The first process to do so was:    Process name: [[41174,1],0]   Exit code:    1"
Command "mpirun --host algo-1:8 -np 8 --allow-run-as-root --display-map --tag-output -mca btl_tcp_if_include eth0 -mca oob_tcp_if_include eth0 -mca plm_rsh_no_tree_spawn 1 -bind-to none -map-by slot -mca pml ob1 -mca btl ^openib -mca orte_abort_on_non_zero_status 1 -mca btl_vader_single_copy_mechanism none -x NCCL_MIN_NRINGS=4 -x NCCL_SOCKET_IFNAME=eth0 -x NCCL_DEBUG=INFO -x LD_LIBRARY_PATH -x PATH -x LD_PRELOAD=/opt/conda/lib/python3.8/site-packages/gethostname.cpython-38-x86_64-linux-gnu.so -x SM_HOSTS -x SM_NETWORK_INTERFACE_NAME -x SM_HPS -x SM_USER_ENTRY_POINT -x SM_FRAMEWORK_PARAMS -x SM_RESOURCE_CONFIG -x SM_INPUT_DATA_CONFIG -x SM_OUTPUT_DATA_DIR -x SM_CHANNELS -x SM_CURRENT_HOST -x SM_MODULE_NAME -x SM_LOG_LEVEL -x SM_FRAMEWORK_MODULE -x SM_INPUT_DIR -x SM_INPUT_CONFIG_DIR -x SM_OUTPUT_DIR -x SM_NUM_CPUS -x SM_NUM_GPUS -x SM_MODEL_DIR -x SM_MODULE_DIR -x SM_TRAINING_ENV -x SM_USER_ARGS -x SM_OUTPUT_INTERMEDIATE_DIR -x SM_CHANNEL_TEST -x SM_CHANNEL_TRAIN -x SM_HP_EVALUATION_STRATEGY -x SM_HP_EVAL_BATCH_SIZE -x SM_HP_GRADIENT_ACCUMULATION_STEPS -x SM_HP_TRAIN_BATCH_SIZE -x SM_HP_MODEL_NAME -x SM_HP_WARMUP_STEPS -x SM_HP_OUTPUT_DIR -x SM_HP_EPOCHS -x SM_HP_LOGGING_STEPS -x SM_HP_MP_PARAMETERS -x PYTHONPATH /opt/conda/bin/python3.8 -m mpi4py ledFinalTrainer.py --epochs 1 --eval_batch_size 1 --evaluation_strategy steps --gradient_accumulation_steps 4 --logging_steps 100 --model_name HHousen/distil-led-large-cnn-16384 --mp_parameters ddp=True,microbatches=1,optimize=speed,partitions=4,pipeline=interleaved,placement_strategy=spread --output_dir /opt/ml/model --train_batch_size 1 --warmup_steps 25"
2022-03-02 15:22:39,557 sagemaker-training-toolkit ERROR    Encountered exit_code 1

2022-03-02 15:22:53 Uploading - Uploading generated training model
2022-03-02 15:22:53 Failed - Training job failed
---------------------------------------------------------------------------
UnexpectedStatusException                 Traceback (most recent call last)
<ipython-input-47-aba514c9e2f8> in <module>
----> 1 huggingface_estimator.fit({"train":"s3://decisions-data/train","test":"s3://decisions-data/test"})

~/anaconda3/envs/pytorch_p36/lib/python3.6/site-packages/sagemaker/estimator.py in fit(self, inputs, wait, logs, job_name, experiment_config)
    690         self.jobs.append(self.latest_training_job)
    691         if wait:
--> 692             self.latest_training_job.wait(logs=logs)
    693 
    694     def _compilation_job_name(self):

~/anaconda3/envs/pytorch_p36/lib/python3.6/site-packages/sagemaker/estimator.py in wait(self, logs)
   1653         # If logs are requested, call logs_for_jobs.
   1654         if logs != "None":
-> 1655             self.sagemaker_session.logs_for_job(self.job_name, wait=True, log_type=logs)
   1656         else:
   1657             self.sagemaker_session.wait_for_job(self.job_name)

~/anaconda3/envs/pytorch_p36/lib/python3.6/site-packages/sagemaker/session.py in logs_for_job(self, job_name, wait, poll, log_type)
   3777 
   3778         if wait:
-> 3779             self._check_job_status(job_name, description, "TrainingJobStatus")
   3780             if dot:
   3781                 print()

~/anaconda3/envs/pytorch_p36/lib/python3.6/site-packages/sagemaker/session.py in _check_job_status(self, job, desc, status_key_name)
   3336                 ),
   3337                 allowed_statuses=["Completed", "Stopped"],
-> 3338                 actual_status=status,
   3339             )
   3340 

UnexpectedStatusException: Error for Training job huggingface-pytorch-training-2022-03-02-15-14-03-282: Failed. Reason: AlgorithmError: ExecuteUserScriptError:
ExitCode 1
ErrorMessage ":ValueError: not enough values to unpack (expected 2, got 1)
 -------------------------------------------------------------------------- Primary job  terminated normally, but 1 process returned a non-zero exit code. Per user-direction, the job has been aborted. mpirun.real detected that one or more processes exited with non-zero status, thus causing the job to be terminated. The first process to do so was:    Process name: [[41174,1],0]   Exit code:    1"
Command "mpirun --host algo-1:8 -np 8 --allow-run-as-root --display-map --tag-output -mca btl_tcp_if_include eth0 -mca oob_tcp_if_include eth0 -mca plm_rsh_no_tree_spawn 1 -bind-to none -map-by slot -mca pml ob1 -mca btl ^openib -mca orte_abort_on_non_zero_status 1 -mca btl_vader_single_copy_mechanism none -x NCCL_MIN_NRINGS=4 -x NCCL_SOCKET_IFNAME=eth0 -x NCCL_DEBUG=INFO -x LD_LIBRARY_PATH -x PATH -x LD_PRELOAD=/opt/conda/lib/python3.8/site-packages/gethostname.cpython-38-x86_64-linux-gnu.so -```

Hello @anwarika,

could you please provide some more information on what you want to do? how did you start the training job? which model, dataset, transformers version, pytorch version etc…?

Thanks @philschmid for the reply. I am trying to FineTune HHousen/distil-led-large-cnn-16384 against a custom data set. I was able to finetune it on a p3dn.24xlarge EC2 instance with a huggingface training image and fairscale simple (essentially same params). Using sagemaker here are my params:
mpi_options = { "enabled" : True, "processes_per_host" : 8 }

smp_options = { "enabled":True, "parameters": { "microbatches": 1, "placement_strategy": "spread", "pipeline": "interleaved", "optimize": "speed", "partitions": 4, "ddp": True, } }

distribution={ "smdistributed": {"modelparallel": smp_options}, "mpi": mpi_options }

hyperparameters={'epochs': 1, 'train_batch_size': 1, 'eval_batch_size': 1, 'model_name':MODEL, 'output_dir': '/opt/ml/model', 'warmup_steps': 25, 'logging_steps':100, 'evaluation_strategy':"steps", 'gradient_accumulation_steps':4 }
huggingface_estimator = HuggingFace(entry_point='train.py', source_dir='./scripts', instance_type='ml.p3dn.24xlarge', instance_count=1, role=role, volume=300, py_version='py38', transformers_version='4.12.3', pytorch_version='1.9.1', hyperparameters=hyperparameters, distribution=distribution)
huggingface_estimator.fit({"train":f"s3://{s3_bucket}/train","test":f"s3://{s3_bucket}/test"})

Sorry in advanced for the poor formatting

Did you use the same transformers version and pytorch version on ec2?

transformers-4.16.2
torch-1.10.2
fairscale-0.4.5
py37

So sagemaker doesn’t support these versions, should I just use a custom image_uri or does huggingface have one that you think could work?

And how does your training script look like? Are you using SageMaker Parallelism or are you trying to use fairscale?

Locally or on single EC2 instance i’m using fairscale, but trying to use sagemaker estimator to manage the training so using Sagemaker Parallelism

training_args = TrainingArguments(
    evaluation_strategy="steps",
    per_device_train_batch_size=batch_size,
    per_device_eval_batch_size=batch_size,
    fp16=True,
    output_dir="./",
    logging_steps=5,
    eval_steps=10,
    save_steps=10,
    save_total_limit=2,
    gradient_accumulation_steps=4,
    num_train_epochs=1
)
trainer = Trainer(
    model=MODEL,
    tokenizer=tokenizer,
    args=training_args,
    compute_metrics=compute_metrics,
    train_dataset=train_dataset,
    eval_dataset=eval_dataset,
)```

Got it thanks!
We have created 2 blog posts showing how to do distributed training:

As well as 2 gettings started notebooks for that

If this doesn’t help you. Feel free to open an issue in the GitHub transformers repository.

Hi,

This is a follow up on this post with the same title. We are trying to fix the issue and are still getting the same error after trying out several things including matching the python, transformers, and pytorch versions according to the recommendations (3.8, 4.16.2, and 1.10.2, respectively):
-ValueError: not enough values to unpack (expected 2, got 1)

The error is in the “modeling_led” within the transformers module expecting a different input_ids shape. We tried unsqueezing the input_ids and attention_masks but it didn’t fix the error.

I’d greatly appreciate any feedback.

...
[1,mpirank:0,algo-1]<stderr>:  File "/opt/conda/lib/python3.8/site-packages/transformers/models/led/modeling_led.py", line 120, in forward
[1,mpirank:0,algo-1]<stderr>:    bsz, seq_len = input_ids_shape[:2]
[1,mpirank:0,algo-1]<stderr>:ValueError: not enough values to unpack (expected 2, got 1)
--------------------------------------------------------------------------
Primary job  terminated normally, but 1 process returned
a non-zero exit code. Per user-direction, the job has been aborted.
--------------------------------------------------------------------------
--------------------------------------------------------------------------
mpirun.real detected that one or more processes exited with non-zero status, thus causing
the job to be terminated. The first process to do so was:
  Process name: [[41174,1],0]
  Exit code:    1
--------------------------------------------------------------------------
2022-03-02 15:22:39,556 sagemaker-training-toolkit ERROR    Reporting training FAILURE
2022-03-02 15:22:39,557 sagemaker-training-toolkit ERROR    ExecuteUserScriptError:
ExitCode 1
ErrorMessage ":ValueError: not enough values to unpack (expected 2, got 1)
 -------------------------------------------------------------------------- Primary job  terminated normally, but 1 process returned a non-zero exit code. Per user-direction, the job has been aborted. mpirun.real detected that one or more processes exited with non-zero status, thus causing the job to be terminated. The first process to do so was:    Process name: [[41174,1],0]   Exit code:    1"
Command "mpirun --host algo-1:8 -np 8 --allow-run-as-root --display-map --tag-output -mca btl_tcp_if_include eth0 -mca oob_tcp_if_include eth0 -mca plm_rsh_no_tree_spawn 1 -bind-to none -map-by slot -mca pml ob1 -mca btl ^openib -mca orte_abort_on_non_zero_status 1 -mca btl_vader_single_copy_mechanism none -x NCCL_MIN_NRINGS=4 -x NCCL_SOCKET_IFNAME=eth0 -x NCCL_DEBUG=INFO -x LD_LIBRARY_PATH -x PATH -x LD_PRELOAD=/opt/conda/lib/python3.8/site-packages/gethostname.cpython-38-x86_64-linux-gnu.so -x SM_HOSTS -x SM_NETWORK_INTERFACE_NAME -x SM_HPS -x SM_USER_ENTRY_POINT -x SM_FRAMEWORK_PARAMS -x SM_RESOURCE_CONFIG -x SM_INPUT_DATA_CONFIG -x SM_OUTPUT_DATA_DIR -x SM_CHANNELS -x SM_CURRENT_HOST -x SM_MODULE_NAME -x SM_LOG_LEVEL -x SM_FRAMEWORK_MODULE -x SM_INPUT_DIR -x SM_INPUT_CONFIG_DIR -x SM_OUTPUT_DIR -x SM_NUM_CPUS -x SM_NUM_GPUS -x SM_MODEL_DIR -x SM_MODULE_DIR -x SM_TRAINING_ENV -x SM_USER_ARGS -x SM_OUTPUT_INTERMEDIATE_DIR -x SM_CHANNEL_TEST -x SM_CHANNEL_TRAIN -x SM_HP_EVALUATION_STRATEGY -x SM_HP_EVAL_BATCH_SIZE -x SM_HP_GRADIENT_ACCUMULATION_STEPS -x SM_HP_TRAIN_BATCH_SIZE -x SM_HP_MODEL_NAME -x SM_HP_WARMUP_STEPS -x SM_HP_OUTPUT_DIR -x SM_HP_EPOCHS -x SM_HP_LOGGING_STEPS -x SM_HP_MP_PARAMETERS -x PYTHONPATH /opt/conda/bin/python3.8 -m mpi4py ledFinalTrainer.py --epochs 1 --eval_batch_size 1 --evaluation_strategy steps --gradient_accumulation_steps 4 --logging_steps 100 --model_name HHousen/distil-led-large-cnn-16384 --mp_parameters ddp=True,microbatches=1,optimize=speed,partitions=4,pipeline=interleaved,placement_strategy=spread --output_dir /opt/ml/model --train_batch_size 1 --warmup_steps 25"
2022-03-02 15:22:39,557 sagemaker-training-toolkit ERROR    Encountered exit_code 1

2022-03-02 15:22:53 Uploading - Uploading generated training model
2022-03-02 15:22:53 Failed - Training job failed
---------------------------------------------------------------------------
UnexpectedStatusException                 Traceback (most recent call last)
<ipython-input-47-aba514c9e2f8> in <module>
----> 1 huggingface_estimator.fit({"train":"s3://decisions-data/train","test":"s3://decisions-data/test"})

~/anaconda3/envs/pytorch_p36/lib/python3.6/site-packages/sagemaker/estimator.py in fit(self, inputs, wait, logs, job_name, experiment_config)
    690         self.jobs.append(self.latest_training_job)
    691         if wait:
--> 692             self.latest_training_job.wait(logs=logs)
    693 
    694     def _compilation_job_name(self):

~/anaconda3/envs/pytorch_p36/lib/python3.6/site-packages/sagemaker/estimator.py in wait(self, logs)
   1653         # If logs are requested, call logs_for_jobs.
   1654         if logs != "None":
-> 1655             self.sagemaker_session.logs_for_job(self.job_name, wait=True, log_type=logs)
   1656         else:
   1657             self.sagemaker_session.wait_for_job(self.job_name)

~/anaconda3/envs/pytorch_p36/lib/python3.6/site-packages/sagemaker/session.py in logs_for_job(self, job_name, wait, poll, log_type)
   3777 
   3778         if wait:
-> 3779             self._check_job_status(job_name, description, "TrainingJobStatus")
   3780             if dot:
   3781                 print()

~/anaconda3/envs/pytorch_p36/lib/python3.6/site-packages/sagemaker/session.py in _check_job_status(self, job, desc, status_key_name)
   3336                 ),
   3337                 allowed_statuses=["Completed", "Stopped"],
-> 3338                 actual_status=status,
   3339             )
   3340 

UnexpectedStatusException: Error for Training job huggingface-pytorch-training-2022-03-02-15-14-03-282: Failed. Reason: AlgorithmError: ExecuteUserScriptError:
ExitCode 1
ErrorMessage ":ValueError: not enough values to unpack (expected 2, got 1)
 -------------------------------------------------------------------------- Primary job  terminated normally, but 1 process returned a non-zero exit code. Per user-direction, the job has been aborted. mpirun.real detected that one or more processes exited with non-zero status, thus causing the job to be terminated. The first process to do so was:    Process name: [[41174,1],0]   Exit code:    1"
Command "mpirun --host algo-1:8 -np 8 --allow-run-as-root --display-map --tag-output -mca btl_tcp_if_include eth0 -mca oob_tcp_if_include eth0 -mca plm_rsh_no_tree_spawn 1 -bind-to none -map-by slot -mca pml ob1 -mca btl ^openib -mca orte_abort_on_non_zero_status 1 -mca btl_vader_single_copy_mechanism none -x NCCL_MIN_NRINGS=4 -x NCCL_SOCKET_IFNAME=eth0 -x NCCL_DEBUG=INFO -x LD_LIBRARY_PATH -x PATH -x LD_PRELOAD=/opt/conda/lib/python3.8/site-packages/gethostname.cpython-38-x86_64-linux-gnu.so

Could you please open an issue in transformers for that? If you already have done could you share it?

Dear Philipp,

Thanks for your response. Yes, I opened an issue and CCed you (I opened an existing issue that my colleague opened in April). Please find the link below:

Best regards,
Omid

1 Like