Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Scan for outdated or missing drivers - takes under a minute3Clear out junk files and repair common Windows errorsShopify’s reported loop takes weak conversations from its merchant-facing GraphQL agent, diagnoses and repairs them, then uses successful repairs to update the model’s weights. In the system Shopify describes, PyTorch distributes the training work and vLLM serves the resulting model; they are separate parts of the pipeline, not competing tools.
What Shopify’s continual-learning loop changes
Prompts, retrieval examples, routing rules and orchestration code can improve an agent while its model weights remain unchanged. Shopify’s approach aims to go further: turn experience from production use into training data, then use that data to change the weights of the deployed model.
As an Amazon Associate I earn from qualifying purchases.
The example is a merchant-facing GraphQL agent. It generates and runs queries against Shopify’s Admin GraphQL API, then explains the results in natural language. Shopify describes the loop as a recurring process rather than a one-time fine-tune. The company’s account of Sidekick’s continual-learning loop calls it a way to compress production experience into the model’s weights.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
How does Shopify’s continual-learning loop work?
- Define the behavior to reward. Shopify starts with a rubric for completeness, execution, response quality and safety. The rubric needs concrete score anchors so annotators and an automated judge can distinguish, for example, a complete answer from one that omits a material result.
- Check whether people apply the rubric consistently. The account recommends having two experienced annotators blindly score 25 randomly sampled production conversations, then calculating inter-annotator agreement with Cohen’s kappa. Agreement around 0.2 is presented as a warning that the rubric may be ambiguous. These are Shopify’s methodological recommendations, not a reported measurement of its own annotators.
- Calibrate an LLM judge. Shopify backtests judge scores against outcomes from earlier A/B tests and runs targeted degradation tests: deliberately make a behavior worse and verify that the relevant rubric criterion responds. It recommends focused judges for specific behaviors instead of one judge tasked with evaluating everything. The judge is still an offline proxy, so its scores need to be checked against online outcomes.
- Find difficult production cases. After improvements to prompts, tool definitions and the agent harness plateau, Shopify mines anonymized production traffic for hard negatives—conversations where the agent performed poorly.
- Diagnose and attempt a repair. A panel of frontier reasoning models critiques each failure. An arbiter combines their critiques into a repair instruction, called a hint. Shopify replays the conversation with the hint and scores the result.
- Turn successful repairs into training trajectories. If the replay succeeds, Shopify treats it as a reinforcement-learning trajectory. Cases that still fail go to human experts for annotation; Shopify names Toloka as the source of those annotators.
- Train and repeat. Shopify describes a daily pipeline with supervised fine-tuning followed by GRPO. Full-parameter fine-tuning uses accumulated earlier trajectories as well as new ones, which the company says helps limit drift and catastrophic forgetting.
Why the quality judge matters
The judge defines what counts as a successful repair and, later, supplies the reward signal for GRPO. That makes evaluation a control point: if the rubric is vague or the judge rewards the wrong behavior, training can reinforce the wrong behavior at scale. Human scoring, consistency checks, historical A/B outcomes and targeted degradation tests address different parts of that risk; none alone proves that an offline score predicts the experience merchants will have online.
#1 Best Overall
- High-Performance AI Processor：The MS-02 Ultra features an Intel Core Ultra 9 285HX (24C/24T, up to 5.5 GHz, 13 TOPS NPU), delivering fast and efficient performance for AI inference, algorithm development, and media workloads. A PCIe x16 expansion slot supports desktop-class GPU upgrades for advanced model training and accelerated computing tasks. It's ideal for creators, engineers, and teams handling intensive parallel workloads.
- 4 × M.2 PCIe 4.0 + 4 × DDR5 SODIMM slots:Four DDR5 SODIMM slots support up to 256 GB of memory, while ECC helps maintain data integrity in mission-critical environments. Four PCIe 4.0 M.2 slots support up to 24 TB of storage, supporting RAID 0/1/5/10, combining high-speed performance with data protection. It allows for the creation of independent scratch disks, media libraries, and project drives, providing high-throughput for production workflows.
- PCIe & USB 4.0 v2: Up to three PCIe slots can be equipped, including a dual-slot x16 GPU. The main slot supports PCIe 5.0, meeting the needs of high-bandwidth creative and computing workloads. USB 4.0 v2 (80Gbps) supports high-bandwidth external storage and displays.
- Ultra-fast Networking: Wi-Fi 7 further enhances wireless performance with next-generation speeds and low-latency stability. Intelligent bandwidth switching optimizes throughput in different network environments, ensuring optimal performance for enterprise or local networks. Dual 25GbE ports (providing up to approximately 3.125 GB/s bandwidth, about 25 times faster than traditional 1GbE), enabling seamless large-scale file transfers and parallel computing. 10GbE and 2.5GbE ports, with support for Intel vPro technology, ensure enterprise-grade remote management and deployment flexibility.
- Server-grade thermal architecture: Utilizing a dedicated CPU/GPU airflow design, equipped with a 6-pipe dual-fan cooler, it maintains stable performance even under sustained loads, delivering up to 140W Turbo power while maintaining a 100W TDP, and operating with noise levels as low as 36 dB. An integrated 350W power supply ensures stable and reliable output for demanding computing tasks and fully loaded extended configurations.
Shopify says the judge scores repaired conversations, but the available account does not provide a detailed model configuration, audited training log or enough comparative-evaluation detail to independently verify its quality claims. The company characterizes the loop’s results as higher quality than frontier models; that should be read as Shopify’s characterization, not as an independently established general ranking.
What SFT and GRPO do in the pipeline
Supervised fine-tuning distills repaired examples
In the first stage, supervised fine-tuning teaches a smaller model from successful, “healed” trajectories. Shopify says the trajectories include their reasoning, so the model learns from the sequence that produced a good outcome, not just a final answer.
GRPO uses judge scores as rewards
In the second stage, Group Relative Policy Optimization (GRPO) samples groups of responses and uses the calibrated judge’s scores as the reward signal. In this setup, the judge turns the rubric into feedback the training procedure can optimize. The account does not publish detailed settings or enough information to reproduce the run, so it does not support conclusions about how GRPO would perform with a different judge, task or training configuration.
How PyTorch and vLLM fit together
PyTorch is the training foundation in Shopify’s description. The company says it distributes training across GPUs using tensor, context and data parallelism to make full-parameter fine-tuning practical. In Shopify’s words: “We use PyTorch to distribute training across GPUs with tensor, context and data parallelism, making full parameter fine-tuning practical at scale.”
Rank #3
- Professional AI & Creator Workstation: AMD Radeon AI PRO R9700 GPU with 32GB GDDR6 is engineered for AI development, professional content creation, and compute-intensive workloads.
- Massive 32GB Memory Capacity: 32GB of GDDR6 memory on a 256-bit bus provides ample bandwidth for large AI models, 8K video editing, and complex 3D rendering.
- Advanced RDNA 4 with AI Accelerators: 64 Compute Units with 3rd Gen Ray Tracing and dedicated 2nd Gen AI Accelerators for groundbreaking AI performance and visual computing.
- Professional Blower Cooling: Efficient single blower design exhausts heat directly out of the chassis, ideal for multi-GPU workstation and server configurations.
- Enterprise-Grade Thermal Solution: Vapor chamber heatsink with industrial Honeywell PTM7950 thermal interface material ensures reliable cooling under sustained professional loads.
vLLM handles serving, not weight training. Shopify’s account describes its continuous batching as a fit for the agent’s tool-call-heavy workload: it helps serve model requests while the agent works through calls to its GraphQL tool. The PyTorch case study, adapted from Shopify Engineering and published September 22, 2026, describes the training and serving roles. It credits Cody Mazza-Anthony and Andrew McNamara.
What results did Shopify report?
The figures below are Shopify’s 2026 claims, not independently audited results. The source provides some test conditions, but not enough detail to reproduce the full comparison.
Rank #4
- FAST RUNS IN THE FAMILY — The 16-inch MacBook Pro with the M5 Pro or M5 Max chip brings next-generation speed and powerful on-device AI to personal, professional, and creative tasks. With all-day battery life, double the starting storage,* and a breathtaking Liquid Retina XDR display, it’s pro in every way.*
- BUCKLE UP — Along with a next-generation CPU, faster unified memory, and up to 2x faster SSD storage,* M5 Pro and M5 Max feature a more powerful GPU with a Neural Accelerator built into each core, delivering faster AI performance and on-device training capabilities. So you can blaze through demanding workloads at mind-bending speeds.
- BUILT FOR AI — Apple silicon, and every major component that powers it, is designed to run demanding on-device AI workloads like LLM inference and training. And Apple Intelligence helps you write, express yourself, and get things done effortlessly with groundbreaking privacy protections at every step.*
- ALL-DAY BATTERY LIFE — MacBook Pro delivers the same exceptional performance whether it’s running on battery or plugged in.*
- MACOS RUNS APPS FAST — All your go-to apps run lightning fast in macOS, including built-in apps like FaceTime and Messages. Plus, built-in virus protection and free software updates help keep your Mac running smoothly and securely.
| Measure | Shopify’s reported result | Qualification |
|---|---|---|
| Production capacity | Up to 2,000 requests per minute | Reported capacity for the GraphQL agent; Shopify does not provide further load-test conditions with this figure. |
| Latency | About 19% lower time to first token and about 38% lower end-to-end latency | Shopify reports these as load-test results at 350 requests per minute. |
| Throughput with gist compression | About 16% more requests per second and about 12% more output tokens per second on identical GPUs | Shopify associates these results with reducing the prompt from about 6,000 tokens to about 1,500 learned gist tokens. |
| GPUs for the same traffic | Approximately 14% fewer GPUs | Shopify reports this alongside gist compression; the account does not provide a hardware configuration here. |
| Annual serving cost scenario | About $27 million for a frontier-model serving scenario versus closer to $1 million for the fine-tuned model; Shopify states a 96% reduction | The $27 million figure is an estimate based on average token costs, not audited actual spend. It is not a universal savings estimate. |
| Task quality versus frontier models | Shopify says the loop delivers higher quality | The account does not provide sufficient independent comparative-evaluation detail to establish this as a general result. |
What the account does not establish
Shopify says it uses anonymized production conversations, but its account does not describe the complete privacy, retention, access-control or consent design. Anonymization alone is not a substitute for those details, so the description does not establish what safeguards apply across the full data lifecycle.
Free tools Windows power users keep installed
One-click scans. No signup required.
Likewise, the reported latency, throughput, capacity and cost figures should be treated as company-reported outcomes under the stated or available conditions. They do not establish that another agent, workload or deployment would see the same results. Shopify’s cost comparison depends on its serving scenario and token-cost assumptions, while the quality claim lacks enough published evaluation detail for independent reproduction.
Quick Recap
Best Value
- 【High-Performance APU】The MS-S1 MAX features an AMD Ryzen AI Max+ 395 APU, integrating a Zen 5 architecture CPU (up to 5.1GHz, 16C/32T, 64M L3 Cache), an RDNA 3.5 GPU, and an NPU (50 TOPS). The total system output is 126 TOPS. It provides powerful parallel computing capabilities for demanding AI workflows. It is ideal for running local LLMs, multimodal models, and computationally intensive tasks
- 【128GB UMA Memory】Equipped with up to 128GB of LPDDR5x-8000MT/s unified memory, it enables the CPU and GPU to access a shared, high-bandwidth memory pool with extremely low latency. Ideal for large-scale AI inference, 3D workloads, and complex timelines in video editing. It eliminates traditional VRAM bottlenecks, ensuring smoother data transfer during high-intensity computations. The UMA design maximizes performance stability under high loads
- 【Flexible Expansion】The MS-S1 MAX features USB4 V2 (up to 80Gbps), dual 10GbE LAN, HDMI 2.1 (up to 8K60), a full-length PCIe x16 expansion slot, and dual M.2 slots supporting up to 16TB RAID 0/1. Wi-Fi 7 provides stronger signal coverage and a more stable wireless experience. The slide-out design facilitates upgrades and maintenance. It easily adapts to personal, studio, or rack-mount enterprise environments
- 【High-Efficiency Cooling System】Utilizing an aerospace-grade aluminum alloy chassis, copper base plate, six heat pipes, dual turbine fans, and advanced PCM thermal conductive material, it maintains stable cooling performance even under continuous load. This system supports 130W continuous power and 160W peak power operation, with a built-in 320W power supply. It boasts multiple global certifications including CCC, FCC, UL, CE, and UKCA, ensuring stable and reliable operation in various environments
- 【Cluster Design】Two MS-S1 MAX units can be configured as a dual-unit cluster to run a large 235B Q4 model locally, achieving an output speed of 10.87 tok/s. Supporting 2U rack deployment, multiple MS-S1 MAX units can be cascaded into a distributed cluster to create a high-efficiency AI computing center. A cluster of four MS-S1 MAX units successfully ran a DeepSeek-R1 671B Q4 large model. A reserved cluster power-on interface allows for unified start-up and shutdown
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
