SmoothQuant doesn't respect ignored modules for VLMs

[mgoin]

I am trying to apply SmoothQuant during W8A8 quantization of meta-llama/Llama-3.2-11B-Vision-Instruct where I ignore all of the modules except for language_model. However I find that it crashes when going through the vision model that I have chosen to ignore

ignore=["re:.*lm_head", "re:multi_modal_projector.*", "re:vision_model.*"]
recipe = [
    SmoothQuantModifier(smoothing_strength=0.8, ignore=ignore),
    GPTQModifier(targets="Linear", scheme="W8A8", ignore=ignore),
]

Error:

  File "/home/mgoin/code/llm-compressor/src/llmcompressor/modifiers/smoothquant/base.py", line 276, in _apply_smoothing
    self.scales_[mapping.smooth_name].max_channel_vals
KeyError: 'vision_model.transformer.layers.0.input_layernorm'

Code to trigger:

from datasets import load_dataset
from transformers import AutoTokenizer, MllamaForConditionalGeneration

from llmcompressor.modifiers.quantization import GPTQModifier
from llmcompressor.modifiers.smoothquant import SmoothQuantModifier
from llmcompressor.transformers import oneshot, wrap_hf_model_class

# Select model and load it.
MODEL_ID = "meta-llama/Llama-3.2-11B-Vision-Instruct"
model_class = wrap_hf_model_class(MllamaForConditionalGeneration)
model = model_class.from_pretrained(
    MODEL_ID,
    device_map="auto",
    torch_dtype="auto",
)
processor = AutoTokenizer.from_pretrained(MODEL_ID)

# Select calibration dataset.
DATASET_ID = "HuggingFaceH4/ultrachat_200k"
DATASET_SPLIT = "train_sft"

# Select number of samples. 512 samples is a good place to start.
# Increasing the number of samples can improve accuracy.
NUM_CALIBRATION_SAMPLES = 4
MAX_SEQUENCE_LENGTH = 2048

# Load dataset and preprocess.
ds = load_dataset(DATASET_ID, split=DATASET_SPLIT)
ds = ds.shuffle(seed=42).select(range(NUM_CALIBRATION_SAMPLES))


def preprocess(example):
    return {
        "text": processor.apply_chat_template(
            example["messages"],
            tokenize=False,
        )
    }


ds = ds.map(preprocess)


# Tokenize inputs.
def tokenize(sample):
    return processor(
        sample["text"],
        padding=False,
        max_length=MAX_SEQUENCE_LENGTH,
        truncation=True,
        add_special_tokens=False,
    )


ds = ds.map(tokenize, remove_columns=ds.column_names)
print(ds)

# Configure algorithms. In this case, we:
#   * apply SmoothQuant to make the activations easier to quantize
#   * quantize the weights to int8 with GPTQ (static per channel)
#   * quantize the activations to int8 (dynamic per token)
# Note: set sequential_update: true in the recipe to reduce memory
ignore=["re:.*lm_head", "re:multi_modal_projector.*", "re:vision_model.*"]
recipe = [
    SmoothQuantModifier(smoothing_strength=0.8, ignore=ignore),
    GPTQModifier(targets="Linear", scheme="W8A8", ignore=ignore),
]

# Apply algorithms.
oneshot(
    model=model,
    dataset=ds,
    recipe=recipe,
    max_seq_length=MAX_SEQUENCE_LENGTH,
    num_calibration_samples=NUM_CALIBRATION_SAMPLES,
)

# Confirm generations of the quantized model look sane.
print("\n\n")
print("========== SAMPLE GENERATION ==============")
input_ids = processor("Hello my name is", return_tensors="pt").input_ids.to("cuda")
output = model.generate(input_ids, max_new_tokens=100)
print(processor.decode(output[0]))
print("==========================================\n\n")

# Save to disk compressed.
SAVE_DIR = MODEL_ID.split("/")[1] + "-W8A8-Dynamic-Per-Token"
model.save_pretrained(SAVE_DIR, save_compressed=True)
processor.save_pretrained(SAVE_DIR)

Full log and error:

python llama3.2_vision_example.py
The model weights are not tied. Please use the `tie_weights` method before using the `infer_auto_device` function.
Loading checkpoint shards: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 5/5 [00:05<00:00,  1.19s/it]
Dataset({
    features: ['input_ids', 'attention_mask'],
    num_rows: 4
})
2024-09-26T21:08:35.974072+0000 | main | WARNING - Process rank: 0, device: cuda:0, n_gpu: 1, distributed training: True, 16-bits training: False
2024-09-26T21:08:35.975208+0000 | main | INFO - Training/evaluation parameters TrainingArguments(
_n_gpu=1,
accelerator_config={'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None, 'use_configured_state': False},
adafactor=False,
adam_beta1=0.9,
adam_beta2=0.999,
adam_epsilon=1e-08,
auto_find_batch_size=False,
batch_eval_metrics=False,
bf16=False,
bf16_full_eval=False,
clear_sparse_session=False,
data_seed=None,
dataloader_drop_last=False,
dataloader_num_workers=0,
dataloader_persistent_workers=False,
dataloader_pin_memory=True,
dataloader_prefetch_factor=None,
ddp_backend=None,
ddp_broadcast_buffers=None,
ddp_bucket_cap_mb=None,
ddp_find_unused_parameters=None,
ddp_timeout=1800,
debug=[],
deepspeed=None,
disable_tqdm=False,
dispatch_batches=None,
do_eval=False,
do_oneshot=True,
do_predict=False,
do_train=False,
eval_accumulation_steps=None,
eval_delay=0,
eval_do_concat_batches=True,
eval_on_start=False,
eval_steps=None,
eval_strategy=no,
eval_use_gather_object=False,
evaluation_strategy=None,
fp16=False,
fp16_backend=auto,
fp16_full_eval=False,
fp16_opt_level=O1,
fsdp=[],
fsdp_config={'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False},
fsdp_min_num_params=0,
fsdp_transformer_layer_cls_to_wrap=None,
full_determinism=False,
gradient_accumulation_steps=1,
gradient_checkpointing=False,
gradient_checkpointing_kwargs=None,
greater_is_better=None,
group_by_length=False,
half_precision_backend=auto,
hub_always_push=False,
hub_model_id=None,
hub_private_repo=False,
hub_strategy=every_save,
hub_token=<HUB_TOKEN>,
ignore_data_skip=False,
include_inputs_for_metrics=False,
include_num_input_tokens_seen=False,
include_tokens_per_second=False,
jit_mode_eval=False,
label_names=None,
label_smoothing_factor=0.0,
learning_rate=5e-05,
length_column_name=length,
load_best_model_at_end=False,
local_rank=0,
log_level=passive,
log_level_replica=warning,
log_on_each_node=True,
logging_dir=./output/runs/Sep26_21-08-35_beaker,
logging_first_step=False,
logging_nan_inf_filter=True,
logging_steps=500,
logging_strategy=steps,
lr_scheduler_kwargs={},
lr_scheduler_type=linear,
max_grad_norm=1.0,
max_steps=-1,
metric_for_best_model=None,
mp_parameters=,
neftune_noise_alpha=None,
no_cuda=False,
num_train_epochs=3.0,
oneshot_device=cuda:0,
optim=adamw_torch,
optim_args=None,
optim_target_modules=None,
output_dir=./output,
overwrite_output_dir=False,
past_index=-1,
per_device_eval_batch_size=8,
per_device_train_batch_size=8,
prediction_loss_only=False,
push_to_hub=False,
push_to_hub_model_id=None,
push_to_hub_organization=None,
push_to_hub_token=<PUSH_TO_HUB_TOKEN>,
ray_scope=last,
recipe=[SmoothQuantModifier(index=None, group=None, start=None, end=None, update=None, initialized_structure_=False, initialized_=False, finalized_=False, started_=False, ended_=False, smoothing_strength=0.8, mappings=[[['re:.*q_proj', 're:.*k_proj', 're:.*v_proj'], 're:.*input_layernorm'], [['re:.*gate_proj', 're:.*up_proj'], 're:.*post_attention_layernorm']], ignore=['re:.*lm_head', 're:multi_modal_projector.*', 're:vision_model.*'], num_calibration_steps=None, calibration_function=None, hooks_=None, resolved_mappings_=None, scales_=None), GPTQModifier(index=None, group=None, start=None, end=None, update=None, initialized_structure_=False, initialized_=False, finalized_=False, started_=False, ended_=False, sequential_update=True, targets='Linear', sequential_targets=None, block_size=128, quantize=True, dampening_frac=0.01, config_groups=None, ignore=['re:.*lm_head', 're:multi_modal_projector.*', 're:vision_model.*'], disable_quantization_observer_epoch=None, num_calibration_steps=None, scheme='W8A8', model=None, layer_compressors_=None, compressible_layers_=None, quantization_modifier_=None)],
recipe_args=None,
remove_unused_columns=True,
report_to=[],
restore_callback_states_from_checkpoint=False,
resume_from_checkpoint=None,
run_name=./output,
run_stages=False,
save_compressed=True,
save_on_each_node=False,
save_only_model=False,
save_safetensors=True,
save_steps=500,
save_strategy=steps,
save_total_limit=None,
seed=42,
skip_memory_metrics=True,
split_batches=None,
tf32=None,
torch_compile=False,
torch_compile_backend=None,
torch_compile_mode=None,
torch_empty_cache_steps=None,
torchdynamo=None,
tpu_metrics_debug=False,
tpu_num_cores=None,
use_cpu=False,
use_ipex=False,
use_legacy_prediction_loop=False,
use_liger_kernel=False,
use_mps_device=False,
warmup_ratio=0.0,
warmup_steps=0,
weight_decay=0.0,
)
2024-09-26T21:08:36.353807+0000 | _check_create_state | INFO - State created for compression lifecycle
2024-09-26T21:08:36.354551+0000 | pre_initialize_structure | INFO - Compression lifecycle structure pre-initialized for 0 modifiers
2024-09-26T21:08:36.354703+0000 | pre_initialize_structure | INFO - Compression lifecycle structure pre-initialized for 0 modifiers
2024-09-26T21:08:36.391736+0000 | one_shot | INFO - *** One Shot ***
2024-09-26T21:08:36.395763+0000 | from_modifiers | INFO - Creating recipe from modifiers
2024-09-26T21:08:36.427963+0000 | _check_compile_recipe | INFO - Recipe compiled and 1 modifiers created
2024-09-26T21:08:41.103626+0000 | _calibrate | INFO - Running SmoothQuantModifier calibration with 4 samples...
  0%|                                                                                                                                                    | 0/4 [00:00<?, ?it/s]torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
 25%|███████████████████████████████████                                                                                                         | 1/4 [00:00<00:00,  3.39it/s]torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 2048, 4096])
torch.Size([1, 977, 4096])
torch.Size([1, 977, 4096])
torch.Size([1, 977, 4096])
torch.Size([1, 977, 4096])
torch.Size([1, 977, 4096])
torch.Size([1, 977, 4096])
torch.Size([1, 977, 4096])
torch.Size([1, 977, 4096])
torch.Size([1, 977, 4096])
torch.Size([1, 977, 4096])
torch.Size([1, 977, 4096])
torch.Size([1, 977, 4096])
torch.Size([1, 977, 4096])
torch.Size([1, 977, 4096])
torch.Size([1, 977, 4096])
torch.Size([1, 977, 4096])
torch.Size([1, 977, 4096])
torch.Size([1, 977, 4096])
torch.Size([1, 977, 4096])
torch.Size([1, 977, 4096])
torch.Size([1, 977, 4096])
torch.Size([1, 977, 4096])
torch.Size([1, 977, 4096])
torch.Size([1, 977, 4096])
torch.Size([1, 977, 4096])
torch.Size([1, 977, 4096])
torch.Size([1, 977, 4096])
torch.Size([1, 977, 4096])
torch.Size([1, 977, 4096])
torch.Size([1, 977, 4096])
torch.Size([1, 977, 4096])
torch.Size([1, 977, 4096])
 75%|█████████████████████████████████████████████████████████████████████████████████████████████████████████                                   | 3/4 [00:00<00:00,  7.69it/s]torch.Size([1, 1119, 4096])
torch.Size([1, 1119, 4096])
torch.Size([1, 1119, 4096])
torch.Size([1, 1119, 4096])
torch.Size([1, 1119, 4096])
torch.Size([1, 1119, 4096])
torch.Size([1, 1119, 4096])
torch.Size([1, 1119, 4096])
torch.Size([1, 1119, 4096])
torch.Size([1, 1119, 4096])
torch.Size([1, 1119, 4096])
torch.Size([1, 1119, 4096])
torch.Size([1, 1119, 4096])
torch.Size([1, 1119, 4096])
torch.Size([1, 1119, 4096])
torch.Size([1, 1119, 4096])
torch.Size([1, 1119, 4096])
torch.Size([1, 1119, 4096])
torch.Size([1, 1119, 4096])
torch.Size([1, 1119, 4096])
torch.Size([1, 1119, 4096])
torch.Size([1, 1119, 4096])
torch.Size([1, 1119, 4096])
torch.Size([1, 1119, 4096])
torch.Size([1, 1119, 4096])
torch.Size([1, 1119, 4096])
torch.Size([1, 1119, 4096])
torch.Size([1, 1119, 4096])
torch.Size([1, 1119, 4096])
torch.Size([1, 1119, 4096])
torch.Size([1, 1119, 4096])
torch.Size([1, 1119, 4096])
100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 4/4 [00:00<00:00,  8.13it/s]
2024-09-26T21:08:41.598186+0000 | _apply_smoothing | INFO - Smoothing activation scales...
Traceback (most recent call last):
  File "/home/mgoin/code/llm-compressor/examples/quantization_w8a8_int8/llama3.2_vision_example.py", line 70, in <module>
    oneshot(
  File "/home/mgoin/code/llm-compressor/src/llmcompressor/transformers/finetune/text_generation.py", line 76, in oneshot
    main(model_args, data_args, training_args)
  File "/home/mgoin/code/llm-compressor/src/llmcompressor/transformers/finetune/text_generation.py", line 364, in main
    stage_runner.one_shot()
  File "/home/mgoin/code/llm-compressor/src/llmcompressor/transformers/finetune/runner.py", line 171, in one_shot
    self.trainer.one_shot(calibration_data=calib_data, stage=stage)
  File "/home/mgoin/code/llm-compressor/src/llmcompressor/transformers/finetune/session_mixin.py", line 401, in one_shot
    apply(
  File "/home/mgoin/code/llm-compressor/src/llmcompressor/core/session_functions.py", line 184, in apply
    return active_session().apply(
  File "/home/mgoin/code/llm-compressor/src/llmcompressor/core/session.py", line 210, in apply
    self.initialize(**kwargs)
  File "/home/mgoin/code/llm-compressor/src/llmcompressor/core/session.py", line 156, in initialize
    mod_data = self._lifecycle.initialize(
  File "/home/mgoin/code/llm-compressor/src/llmcompressor/core/lifecycle.py", line 126, in initialize
    data = mod.initialize(state=self.state, **extras)
  File "/home/mgoin/code/llm-compressor/src/llmcompressor/modifiers/stage.py", line 124, in initialize
    modifier.initialize(state, **kwargs)
  File "/home/mgoin/code/llm-compressor/src/llmcompressor/modifiers/modifier.py", line 118, in initialize
    initialized = self.on_initialize(state=state, **kwargs)
  File "/home/mgoin/code/llm-compressor/src/llmcompressor/modifiers/smoothquant/base.py", line 135, in on_initialize
    self._apply_smoothing(state.model)
  File "/home/mgoin/venvs/vllm/lib/python3.10/site-packages/torch/utils/_contextlib.py", line 116, in decorate_context
    return func(*args, **kwargs)
  File "/home/mgoin/code/llm-compressor/src/llmcompressor/modifiers/smoothquant/base.py", line 276, in _apply_smoothing
    self.scales_[mapping.smooth_name].max_channel_vals
KeyError: 'vision_model.transformer.layers.0.input_layernorm'

[markurtz]

@dsikka @kylesayrs is this one included in the bug fixes for the vision pipelines we're working on?

[mohammhn]

@markurtz @dsikka @kylesayrs Hey, just wanted to follow up on this. I'm facing the same issue on my end. Is this issue supposed to be resolved, or is it currently unsupported?

[robertgshaw2-redhat]

@kylesayrs - any updates on this?

[dsikka]

Closing, opening in asana: https://app.asana.com/0/1208216504433080/1209031311964416/f

[Reichenbachian]

Hey! I don't believe Asana is public access. Any word on progress for this?

[kylesayrs]

Hi @Reichenbachian! This was fixed by #1015