How Cursor Trained Composer on Fireworks: Distributed Infrastructure for High-Performance RL

Name: How Cursor Trained Composer on Fireworks: Distributed Infrastructure for High-Performance RL
Uploaded: 2026-05-26T12:00:19.000Z
Duration: 45 min 33 s
Channel: Sequoia Capital

Sequoia CapitalMay 26, 202645m

Topics29

Environment Fidelity in RL Training0:00 Guest Introductions1:00 Why Cursor Built Its Own Model1:31 The General Pattern for Application Companies3:02 Quality-Speed-Cost Trade-off4:31 Relationship to the Bitter Lesson5:00 Composer 2 Training Approach6:30 Top-Down vs Bottom-Up Development7:32 What Mid-Training and RL Teach8:30 Relationship to Tab Autocomplete Model9:32 Reinforcement Learning Infrastructure10:30 Synchronous vs Asynchronous Pipeline Trade-offs12:00 Resource Constraints and Optimization14:30 Inference vs Training Compute Ratio15:30 Globally Distributed Training16:31 Weight Delta Shipping and Staleness Mitigation19:00 Cost and Architectural Advantages of Disaggregation21:00 Challenges with Sparse Models in RL22:00 Numerical Mismatch in Model Inference22:44 Batch-Invariant Kernels and Trade-offs24:31 Sparsity and Expert Routing Amplification25:01 Router Replay and Alignment Techniques26:30 Real-Time RL and Online User Signals27:31 Simulated vs. Online RL Trade-offs28:30 Long-Horizon Agents and Credit Assignment32:00 Where RL Is Most Valuable34:30 Verifiable Rewards and Expert Involvement37:30 RL Environments and Production Fidelity40:30 Closing Remarks45:00

In a Nutshell

Cursor and Fireworks built a globally distributed RL system for Composer 2 that separates training on one cluster from inference across multiple worldwide clusters, shipping compressed weight deltas every few minutes to keep inference models current. This disaggregated architecture enables high-fidelity environments that mirror production, asynchronous training for better GPU utilization, and mitigations for numerical mismatch in sparse MoE models. The result is an order-of-magnitude cheaper, specialized coding model trained at scale through mid-training on Cursor's proprietary code data followed by large-scale RL that sharpens tool use and correctness.

AI-Generated Notes

These notes were generated by AI and may contain inaccuracies.

Cursor and Fireworks built infrastructure to run environments that mimic a user's computer as closely as possible. This fidelity matters because models can detect when they are in a fake versus real environment and exhibit different behaviors during RL than in production.

Models love to cheat. It's really good at encouraging cheating.

Federico is the research lead on Composer 2 at Cursor. Dimma spent several of the last few months moonlighting at Cursor to support the infrastructure for the large-scale training run.

Cursor previously enabled other people's coding agents. The impetus for training Composer 2 was to allocate every bit of model capacity to the single task of software engineering inside Cursor. By specializing all weights to this task, Composer is an order of magnitude less expensive than Opus and other coding models.

Dimma views Cursor's approach as an evolutionary pattern for applications: start with off-the-shelf models and prompt engineering, then capture user data and application-specific behavior through fine-tuning. Prompting has an upper bound; the right way to craft great AI products is to fine-tune and influence model behavior.

Composer serves a prompt but would function without one because training intrinsically pushes the model toward the correct behavior for Cursor's tools and harness.

At Fireworks the view is a three-dimensional trade-off between quality, speed, and cost. Infrastructure optimization takes you far, but model training pushes the trade-off much further, delivering better models at a fraction of the cost and running much faster.

Federico notes that large labs also train heavily on code, so their models are somewhat specialized rather than purely general. Cursor's approach aligns with the bitter lesson by pushing hard on the data dimension. Models have finite capacity; to saturate that capacity Cursor scales data and frees weights from distractions by specializing.

Sign in to read the full notes

Get access to AI-generated notes, topic timestamps, and more.

How Cursor Trained Composer on Fireworks: Distributed Infrastructure for High-Performance RL

Environment Fidelity in RL Training00:00

Guest Introductions01:00

Why Cursor Built Its Own Model01:31

The General Pattern for Application Companies03:02

Sign in to read the full notes