Typically quantization algorithms will have different schemes for how the activation and weights are quantized so A16W8 for instance means the activations are quantized to 16 bits wheras the weights are quantized to 8 bits. Trying out different quantization schemes in torchao is generally a 1 line change. Note: exact APIs are not stable, we may change them in the future.
Benchmarks are run on a machine with a single A100 GPU in torchtune.
| Model | Technique | wikitext-perplexity | Tokens/Second | Memory Bandwidth (GB/s) |
|---|---|---|---|---|
| Llama-2-7B | Base (bfloat16) | 8.789390849382297 | 20.16 | 316.26 |
| 8-bit | 8.788388896424118 | 27.81 | 251.81 | |
| 4-bit (G=256) | 9.618806853408442 | 64.23 | 375.77 | |
| 4-bit GPTQ (G=256) | 9.1791455391884 | 64.81 | 379.19 |
The autoquant api can be used to quickly and accurately quantize your model. When used as in the example below, the api first identifies the shapes
of the activations that the different linear layers see, it then benchmarks these shapes across different types of quantized and non-quantized layers in order to pick the fastest one, attempting to take into account fusions where possible. Finally once the best class is found for each layer, it swaps the linear. Currently this api chooses between no quantization, int8 dynamic quantization and int8 weight only quantization for each layer.
import torch
import torchao
# inductor settings which improve torch.compile performance for quantized modules
torch._inductor.config.force_fuse_int_mm_with_mul = True
torch._inductor.config.use_mixed_mm = True
# Plug in your model and example input
model = torch.nn.Sequential(torch.nn.Linear(32, 64)).cuda().to(torch.bfloat16)
input = torch.randn(32,32, dtype=torch.bfloat16, device='cuda')
# perform autoquantization
torchao.autoquant(model, (input))
# compile the model to improve performance
model = torch.compile(model, mode='max-autotune')
model(input)# Fuse the int8*int8 -> int32 matmul and subsequent mul op avoiding materialization of the int32 intermediary tensor
torch._inductor.config.force_fuse_int_mm_with_mul = True
from torchao.quantization import quant_api
# convert linear modules to quantized tensor subclasses
quant_api.change_linear_weights_to_int8_dqtensors(model)from torchao.quantization import quant_api
quant_api.change_linear_weights_to_int8_woqtensors(model)This technique works best when the torch._inductor.config.use_mixed_mm option is enabled. This avoids dequantizing the weight tensor before the matmul, instead fusing the dequantization into the matmul, thereby avoiding materialization of a large floating point weight tensor.
from torchao.quantization import quant_api
quant_api.change_linear_weights_to_int4_woqtensors(model)Note: The quantization error incurred by applying int4 quantization to your model can be fairly significant, so using external techniques like GPTQ may be necessary to obtain a usable model.
from torchao.quantization.GPTQ import Int4WeightOnlyGPTQQuantizer, InputRecorder, TransformerEvalWrapper
precision = torch.bfloat16
device = "cuda"
checkpoint_file_name = "../gpt-fast/checkpoints/meta-llama/Llama-2-7b-chat-hf/model.pth"
checkpoint_path = Path(checkpoint_file_name)
model = Transformer.from_name(checkpoint_path.parent.name)
checkpoint = torch.load(str(checkpoint_path), mmap=True, weights_only=True)
model.load_state_dict(checkpoint, assign=True)
model = model.to(dtype=precision, device="cpu")
model.eval()
tokenizer_path = checkpoint_path.parent / "tokenizer.model"
assert tokenizer_path.is_file(), tokenizer_path
tokenizer = SentencePieceProcessor( # pyre-ignore[28]
model_file=str(tokenizer_path)
)
blocksize = 128
percdamp = 0.01
groupsize = 128
calibration_tasks = ["wikitext"]
calibration_limit = 1
calibration_seq_length = 100
input_prep_func = prepare_inputs_for_model
pad_calibration_inputs = False
inputs = InputRecorder(
tokenizer,
calibration_seq_length,
input_prep_func,
pad_calibration_inputs,
model.config.vocab_size,
device="cpu",
).record_inputs(
calibration_tasks,
calibration_limit,
).get_inputs()
quantizer = Int4WeightOnlyGPTQQuantizer(
blocksize,
percdamp,
groupsize,
)
model.setup_caches(max_batch_size=1, max_seq_length=calibration_seq_length)
model = quantizer.quantize(model, inputs).cuda()from torchao.quantization.quant_api import Int8DynActInt4WeightQuantizer
quantizer = Int8DynActInt4WeightQuantizer(groupsize=32)
model = quantizer.quantize(model)This is used in ExecuTorch to quantize llama model right now.
We've also implemented a version of smoothquant with the same GEMM format as above. Due to requiring calibration, the API is more complicated.
Example
import torch
from torchao.quantization.smoothquant import swap_linear_with_smooth_fq_linear, smooth_fq_linear_to_inference
# Fuse the int8*int8 -> int32 matmul and subsequent mul op avoiding materialization of the int32 intermediary tensor
torch._inductor.config.force_fuse_int_mm_with_mul = True
# plug in your model
model = get_model()
# convert linear modules to smoothquant
# linear module in calibration mode
swap_linear_with_smooth_fq_linear(model)
# Create a data loader for calibration
calibration_data = get_calibration_data()
calibration_dataset = MyDataset(calibration_data)
calibration_loader = DataLoader(calibration_dataset, batch_size=32, shuffle=True)
# Calibrate the model
model.train()
for batch in calibration_loader:
inputs = batch
model(inputs)
# set it to inference mode
smooth_fq_linear_to_inference(model)
# compile the model to improve performance
model = torch.compile(model, mode='max-autotune')
model(input)- APIs have been hardware tested on A100 and T4(colab)
- While these techniques are designed to improve model performance, in some cases the opposite can occur. This is because quantization adds additional overhead to the model that is hopefully made up for by faster matmuls (dynamic quantization) or loading weights faster (weight-only quantization). If your matmuls are small enough or your non-quantized perf isn't bottlenecked by weight load time, these techniques may reduce performance.
- Use the PyTorch nightlies so you can leverage tensor subclasses which is preferred over older module swap based methods because it doesn't modify the graph and is generally more composable and flexible.