Price-effective AI picture era with PixArt-Σ inference on AWS Trainium and AWS Inferentia

PixArt-Sigma is a diffusion transformer mannequin that’s able to picture era at 4k decision. This mannequin reveals vital enhancements over earlier era PixArt fashions like Pixart-Alpha and different diffusion fashions via dataset and architectural enhancements. AWS Trainium and AWS Inferentia are purpose-built AI chips to speed up machine studying (ML) workloads, making them excellent for cost-effective deployment of huge generative fashions. Through the use of these AI chips, you possibly can obtain optimum efficiency and effectivity when working inference with diffusion transformer fashions like PixArt-Sigma.

This submit is the primary in a sequence the place we’ll run a number of diffusion transformers on Trainium and Inferentia-powered cases. On this submit, we present how one can deploy PixArt-Sigma to Trainium and Inferentia-powered cases.

Resolution overview

The steps outlined under might be used to deploy the PixArt-Sigma mannequin on AWS Trainium and run inference on it to generate high-quality pictures.

• Step 1 – Pre-requisites and setup
• Step 2 – Obtain and compile the PixArt-Sigma mannequin for AWS Trainium
• Step 3 – Deploy the mannequin on AWS Trainium to generate pictures

Step 1 – Stipulations and setup

To get began, you’ll need to arrange a improvement surroundings on a trn1, trn2, or inf2 host. Full the next steps:

1. Launch a trn1.32xlarge or trn2.48xlarge occasion with a Neuron DLAMI. For directions on tips on how to get began, seek advice from Get Began with Neuron on Ubuntu 22 with Neuron Multi-Framework DLAMI.
2. Launch a Jupyter Pocket book sever. For directions to arrange a Jupyter server, seek advice from the next consumer information.
3. Clone the aws-neuron-samples GitHub repository:

   git clone https://github.com/aws-neuron/aws-neuron-samples.git

4. Navigate to the hf_pretrained_pixart_sigma_1k_latency_optimized.ipynb pocket book:

   cd aws-neuron-samples/torch-neuronx/inference

The supplied instance script is designed to run on a Trn2 occasion, however you possibly can adapt it for Trn1 or Inf2 cases with minimal modifications. Particularly, inside the pocket book and in every of the part information below the neuron_pixart_sigma listing, you will see that commented-out adjustments to accommodate Trn1 or Inf2 configurations.

Step 2 – Obtain and compile the PixArt-Sigma mannequin for AWS Trainium

This part offers a step-by-step information to compiling PixArt-Sigma for AWS Trainium.

Obtain the mannequin

You will see that a helper operate in cache-hf-model.py in above talked about GitHub repository that reveals tips on how to obtain the PixArt-Sigma mannequin from Hugging Face. If you’re utilizing PixArt-Sigma in your personal workload, and decide to not use the script included on this submit, you should use the huggingface-cli to obtain the mannequin as a substitute.

The Neuron PixArt-Sigma implementation incorporates a number of scripts and courses. The assorted information and scrips are damaged down as follows:

├── compile_latency_optimized.sh # Full Mannequin Compilation script for Latency Optimized
├── compile_throughput_optimized.sh # Full Mannequin Compilation script for Throughput Optimized
├── hf_pretrained_pixart_sigma_1k_latency_optimized.ipynb # Pocket book to run Latency Optimized Pixart-Sigma
├── hf_pretrained_pixart_sigma_1k_throughput_optimized.ipynb # Pocket book to run Throughput Optimized Pixart-Sigma
├── neuron_pixart_sigma
│ ├── cache_hf_model.py # Mannequin downloading Script
│ ├── compile_decoder.py # Textual content Encoder Compilation Script and Wrapper Class
│ ├── compile_text_encoder.py # Textual content Encoder Compilation Script and Wrapper Class
│ ├── compile_transformer_latency_optimized.py # Latency Optimized Transformer Compilation Script and Wrapper Class
│ ├── compile_transformer_throughput_optimized.py # Throughput Optimized Transformer Compilation Script and Wrapper Class
│ ├── neuron_commons.py # Base Courses and Consideration Implementation
│ └── neuron_parallel_utils.py # Sharded Consideration Implementation
└── necessities.txt

This pocket book will assist you to to obtain the mannequin, compile the person part fashions, and invoke the era pipeline to generate a picture. Though the notebooks may be run as a standalone pattern, the subsequent few sections of this submit will stroll via the important thing implementation particulars inside the part information and scripts to assist working PixArt-Sigma on Neuron.

For every part of PixArt (T5, Transformer, and VAE), the instance makes use of Neuron particular wrapper courses. These wrapper courses serve two functions. The primary goal is it permits us to hint the fashions for compilation:

class InferenceTextEncoderWrapper(nn.Module):
    def __init__(self, dtype, t: T5EncoderModel, seqlen: int):
        tremendous().__init__()
        self.dtype = dtype
        self.gadget = t.gadget
        self.t = t
    def ahead(self, text_input_ids, attention_mask=None):
        return [self.t(text_input_ids, attention_mask)['last_hidden_state'].to(self.dtype)]

Please seek advice from the neuron_commons.py file for all wrapper modules and courses.

The second purpose for utilizing wrapper courses is to switch the eye implementation to run on Neuron. As a result of diffusion fashions like PixArt are usually compute-bound, you possibly can enhance efficiency by sharding the eye layer throughout a number of units. To do that, you substitute the linear layers with NeuronX Distributed’s RowParallelLinear and ColumnParallelLinear layers:

def shard_t5_self_attention(tp_degree: int, selfAttention: T5Attention):
    orig_inner_dim = selfAttention.q.out_features
    dim_head = orig_inner_dim // selfAttention.n_heads
    original_nheads = selfAttention.n_heads
    selfAttention.n_heads = selfAttention.n_heads // tp_degree
    selfAttention.inner_dim = dim_head * selfAttention.n_heads
    orig_q = selfAttention.q
    selfAttention.q = ColumnParallelLinear(
        selfAttention.q.in_features,
        selfAttention.q.out_features,
        bias=False, 
        gather_output=False)
    selfAttention.q.weight.knowledge = get_sharded_data(orig_q.weight.knowledge, 0)
    del(orig_q)
    orig_k = selfAttention.ok
    selfAttention.ok = ColumnParallelLinear(
        selfAttention.ok.in_features, 
        selfAttention.ok.out_features, 
        bias=(selfAttention.ok.bias shouldn't be None),
        gather_output=False)
    selfAttention.ok.weight.knowledge = get_sharded_data(orig_k.weight.knowledge, 0)
    del(orig_k)
    orig_v = selfAttention.v
    selfAttention.v = ColumnParallelLinear(
        selfAttention.v.in_features, 
        selfAttention.v.out_features, 
        bias=(selfAttention.v.bias shouldn't be None),
        gather_output=False)
    selfAttention.v.weight.knowledge = get_sharded_data(orig_v.weight.knowledge, 0)
    del(orig_v)
    orig_out = selfAttention.o
    selfAttention.o = RowParallelLinear(
        selfAttention.o.in_features,
        selfAttention.o.out_features,
        bias=(selfAttention.o.bias shouldn't be None),
        input_is_parallel=True)
    selfAttention.o.weight.knowledge = get_sharded_data(orig_out.weight.knowledge, 1)
    del(orig_out)
    return selfAttention

Please seek advice from the neuron_parallel_utils.py file for extra particulars on parallel consideration.

Compile particular person sub-models

The PixArt-Sigma mannequin consists of three parts. Every part is compiled so your complete era pipeline can run on Neuron:

• Textual content encoder – A 4-billion-parameter encoder, which interprets a human-readable immediate into an embedding. Within the textual content encoder, the eye layers are sharded, together with the feed-forward layers, with tensor parallelism.
• Denoising transformer mannequin – A 700-million-parameter transformer, which iteratively denoises a latent (a numerical illustration of a compressed picture). Within the transformer, the eye layers are sharded, together with the feed-forward layers, with tensor parallelism.
• Decoder – A VAE decoder that converts our denoiser-generated latent to an output picture. For the decoder, the mannequin is deployed with knowledge parallelism.

Now that the mannequin definition is prepared, you might want to hint a mannequin to run it on Trainium or Inferentia. You’ll be able to see tips on how to use the hint() operate to compile the decoder part mannequin for PixArt within the following code block:

compiled_decoder = torch_neuronx.hint(
    decoder,
    sample_inputs,
    compiler_workdir=f"{compiler_workdir}/decoder",
    compiler_args=compiler_flags,
    inline_weights_to_neff=False
)

Please seek advice from the compile_decoder.py file for extra on tips on how to instantiate and compile the decoder.

To run fashions with tensor parallelism, a way used to separate a tensor into chunks throughout a number of NeuronCores, you might want to hint with a pre-specified tp_degree. This tp_degree specifies the variety of NeuronCores to shard the mannequin throughout. It then makes use of the parallel_model_trace API to compile the encoder and transformer part fashions for PixArt:

compiled_text_encoder = neuronx_distributed.hint.parallel_model_trace(
    get_text_encoder_f,
    sample_inputs,
    compiler_workdir=f"{compiler_workdir}/text_encoder",
    compiler_args=compiler_flags,
    tp_degree=tp_degree,
)

Please seek advice from the compile_text_encoder.py file for extra particulars on tracing the encoder with tensor parallelism.

Lastly, you hint the transformer mannequin with tensor parallelism:

compiled_transformer = neuronx_distributed.hint.parallel_model_trace(
    get_transformer_model_f,
    sample_inputs,
    compiler_workdir=f"{compiler_workdir}/transformer",
    compiler_args=compiler_flags,
    tp_degree=tp_degree,
    inline_weights_to_neff=False,
)

Please seek advice from the compile_transformer_latency_optimized.py file for extra particulars on tracing the transformer with tensor parallelism.

You’ll use the compile_latency_optimized.sh script to compile all three fashions as described on this submit, so these features might be run robotically if you run via the pocket book.

Step 3 – Deploy the mannequin on AWS Trainium to generate pictures

This part will stroll us via the steps to run inference on PixArt-Sigma on AWS Trainium.

Create a diffusers pipeline object

The Hugging Face diffusers library is a library for pre-trained diffusion fashions, and consists of model-specific pipelines that bundle the parts (independently-trained fashions, schedulers, and processors) wanted to run a diffusion mannequin. The PixArtSigmaPipeline is particular to the PixArtSigma mannequin, and is instantiated as follows:

pipe: PixArtSigmaPipeline = PixArtSigmaPipeline.from_pretrained(
    "PixArt-alpha/PixArt-Sigma-XL-2-1024-MS",
    torch_dtype=torch.bfloat16,
    local_files_only=True,
    cache_dir="pixart_sigma_hf_cache_dir_1024")

Please seek advice from the hf_pretrained_pixart_sigma_1k_latency_optimized.ipynb pocket book for particulars on pipeline execution.

Load compiled part fashions into the era pipeline

After every part mannequin has been compiled, load them into the general era pipeline for picture era. The VAE mannequin is loaded with knowledge parallelism, which permits us to parallelize picture era for batch measurement or a number of pictures per immediate. For extra particulars, seek advice from the hf_pretrained_pixart_sigma_1k_latency_optimized.ipynb pocket book.

vae_decoder_wrapper.mannequin = torch_neuronx.DataParallel( 
    torch.jit.load(decoder_model_path), [0, 1, 2, 3], False
)

text_encoder_wrapper.t = neuronx_distributed.hint.parallel_model_load(
    text_encoder_model_path
)

Lastly, the loaded fashions are added to the era pipeline:

pipe.text_encoder = text_encoder_wrapper
pipe.transformer = transformer_wrapper
pipe.vae.decoder = vae_decoder_wrapper
pipe.vae.post_quant_conv = vae_post_quant_conv_wrapper

Compose a immediate

Now that the mannequin is prepared, you possibly can write a immediate to convey what sort of picture you need generated. When making a immediate, you need to at all times be as particular as attainable. You should utilize a optimistic immediate to convey what is needed in your new picture, together with a topic, motion, fashion, and site, and may use a adverse immediate to point options that needs to be eliminated.

For instance, you should use the next optimistic and adverse prompts to generate a photograph of an astronaut driving a horse on mars with out mountains:

# Topic: astronaut
# Motion: driving a horse
# Location: Mars
# Fashion: picture
immediate = "a photograph of an astronaut driving a horse on mars"
negative_prompt = "mountains"

Be happy to edit the immediate in your pocket book utilizing immediate engineering to generate a picture of your selecting.

Generate a picture

To generate a picture, you move the immediate to the PixArt mannequin pipeline, after which save the generated picture for later reference:

# pipe: variable holding the Pixart era pipeline with every of 
# the compiled part fashions
pictures = pipe(
        immediate=immediate,
        negative_prompt=negative_prompt,
        num_images_per_prompt=1,
        peak=1024, # variety of pixels
        width=1024, # variety of pixels
        num_inference_steps=25 # Variety of passes via the denoising mannequin
    ).pictures
    
    for idx, img in enumerate(pictures): 
        img.save(f"image_{idx}.png")

Cleanup

To keep away from incurring extra prices, cease your EC2 occasion utilizing both the AWS Administration Console or AWS Command Line Interface (AWS CLI).

Conclusion

On this submit, we walked via tips on how to deploy PixArt-Sigma, a state-of-the-art diffusion transformer, on Trainium cases. This submit is the primary in a sequence centered on working diffusion transformers for various era duties on Neuron. To be taught extra about working diffusion transformers fashions with Neuron, seek advice from Diffusion Transformers.

In regards to the Authors

Achintya Pinninti is a Options Architect at Amazon Internet Providers. He helps public sector clients, enabling them to attain their goals utilizing the cloud. He makes a speciality of constructing knowledge and machine studying options to resolve advanced issues.

Miriam Lebowitz is a Options Architect centered on empowering early-stage startups at AWS. She leverages her expertise with AI/ML to information firms to pick and implement the appropriate applied sciences for his or her enterprise goals, setting them up for scalable development and innovation within the aggressive startup world.

Sadaf Rasool is a Options Architect in Annapurna Labs at AWS. Sadaf collaborates with clients to design machine studying options that handle their crucial enterprise challenges. He helps clients prepare and deploy machine studying fashions leveraging AWS Trainium or AWS Inferentia chips to speed up their innovation journey.

John Grey is a Options Architect in Annapurna Labs, AWS, primarily based out of Seattle. On this position, John works with clients on their AI and machine studying use circumstances, architects options to cost-effectively clear up their enterprise issues, and helps them construct a scalable prototype utilizing AWS AI chips.