How do you test for catastrophic forgetting after fine-tuning?
A higher overall score can hide the loss of familiar capabilities. Use a paired retention test to decide whether a fine-tuned model is ready.
Nodus blog
Notes on running AI compute
Guides, prices and benchmarks for training, inference and coding agents, from the people building Nodus.
fine-tuning6 min read
How do you test for catastrophic forgetting after fine-tuning?
A higher overall score can hide the loss of familiar capabilities. Use a paired retention test to decide whether a fine-tuned model is ready.
Read the postcapacity6 min read
Nodus vs AWS SageMaker: which is better for training models?
Compare Nodus and AWS SageMaker for training: checkpoint ownership, total run cost, migration work and when staying in AWS makes more sense.
Read the postfine-tuning5 min read
How should you split fine-tuning data when prompts repeat?
A random row split can put the same question in training and validation. Choose the holdout unit that matches the claim you want to test.
Read the postagents6 min read
Can an agent's pass rate rise while it completes fewer tasks?
A higher agent pass rate can hide fewer completed tasks. Compare scored pass rate, scoring coverage and confirmed successes before ranking two evaluations.
Read the postagents7 min read
How can Claude Code train a model on a cloud GPU from one prompt?
Name the model, data, stop rule, dollar cap and outputs in one prompt, and Claude Code can fine-tune on a cloud GPU. The prompt, the manifest and what to check.
Read the postsandboxes8 min read
Nodus vs E2B: which sandbox should AI agents use?
Nodus vs E2B for AI agent sandboxes: isolation, what the defaults do when an agent forgets, network access, GPUs and what 2,000 sandbox-hours cost.
Read the postgpu9 min read
How much VRAM do you need to run or fine-tune a 70B model?
A 70B model needs 40 to 141 GB for weights plus a KV cache to serve, and 50 GB to 1.1 TB to fine-tune. The math, GPU setups and a script to check yours.
Read the posttraining9 min read
Nodus vs Modal: which is better for GPU training?
Nodus vs Modal for GPU training: listed rates, preemption, the 24-hour limit, multi-node clusters, agent support and what the same jobs cost on each.
Read the postagents7 min read
How do you stop an AI agent from overspending on GPUs?
A prompt is not a limit. Cap every run, put a blocking budget around agent work, scope its credentials and approve writes. Setup for Claude Code and Codex.
Read the postbenchmarks8 min read
Continued pretraining on Nodus: 42.5% lower charges
Both platforms reduced perplexity by 21.8% and produced identical weights. Our pilot compares what that improvement cost and how long delivery took.
Read the postbenchmarks8 min read
Fine-tuning on Nodus: 67% lower charges
Nodus and Modal reached 91.74% accuracy with identical trained weights. We compare successful-run charges, retry costs and delivery time.
Read the postcapacity6 min read
Spot vs on-demand GPUs: what breaks, and how do you make spot safe?
Spot GPUs cost up to 90% less but can vanish with 30 seconds of notice. What breaks, each cloud's notice window, the cost math and a safe setup.
Read the posttraining8 min read
What are the best Modal alternatives for GPU training in 2026?
Modal alternatives for GPU training compared on H100 price, preemption, run limits and multi-node scale, with the cost of one 80 GPU-hour run worked out.
Read the postpricing3 min read
Cloud GPU pricing, October 2026: what an H100, H200, B200 or B300 costs per hour
Listed per GPU-hour prices for B300, B200, H200, H100, A100, L40S, L4 and RTX cards as of October 8, 2026, with a training-run cost example.
Read the postmcp3 min read
How to give Claude Code, Codex or Cursor access to GPUs (MCP setup)
Connect one MCP server and your coding agent can estimate, launch and watch GPU jobs on your account, with dry runs and spending limits.
Read the postcheckpointing3 min read
How to make GPU training survive spot preemption (without babysitting it)
Save model, optimizer and step to one directory, write atomically, resume on start, and let the platform carry that directory across a reclaim.
Read the postNodus vs AWS SageMaker: which is better for training models?
Compare Nodus and AWS SageMaker for training: checkpoint ownership, total run cost, migration work and when staying in AWS makes more sense.
How should you split fine-tuning data when prompts repeat?
A random row split can put the same question in training and validation. Choose the holdout unit that matches the claim you want to test.
Can an agent's pass rate rise while it completes fewer tasks?
A higher agent pass rate can hide fewer completed tasks. Compare scored pass rate, scoring coverage and confirmed successes before ranking two evaluations.
How can Claude Code train a model on a cloud GPU from one prompt?
Name the model, data, stop rule, dollar cap and outputs in one prompt, and Claude Code can fine-tune on a cloud GPU. The prompt, the manifest and what to check.
Nodus vs E2B: which sandbox should AI agents use?
Nodus vs E2B for AI agent sandboxes: isolation, what the defaults do when an agent forgets, network access, GPUs and what 2,000 sandbox-hours cost.
How much VRAM do you need to run or fine-tune a 70B model?
A 70B model needs 40 to 141 GB for weights plus a KV cache to serve, and 50 GB to 1.1 TB to fine-tune. The math, GPU setups and a script to check yours.
Nodus vs Modal: which is better for GPU training?
Nodus vs Modal for GPU training: listed rates, preemption, the 24-hour limit, multi-node clusters, agent support and what the same jobs cost on each.
Continued pretraining on Nodus: 42.5% lower charges
Both platforms reduced perplexity by 21.8% and produced identical weights. Our pilot compares what that improvement cost and how long delivery took.
Fine-tuning on Nodus: 67% lower charges
Nodus and Modal reached 91.74% accuracy with identical trained weights. We compare successful-run charges, retry costs and delivery time.
How do you stop an AI agent from overspending on GPUs?
A prompt is not a limit. Cap every run, put a blocking budget around agent work, scope its credentials and approve writes. Setup for Claude Code and Codex.
Spot vs on-demand GPUs: what breaks, and how do you make spot safe?
Spot GPUs cost up to 90% less but can vanish with 30 seconds of notice. What breaks, each cloud's notice window, the cost math and a safe setup.
What are the best Modal alternatives for GPU training in 2026?
Modal alternatives for GPU training compared on H100 price, preemption, run limits and multi-node scale, with the cost of one 80 GPU-hour run worked out.
Cloud GPU pricing, October 2026: what an H100, H200, B200 or B300 costs per hour
Listed per GPU-hour prices for B300, B200, H200, H100, A100, L40S, L4 and RTX cards as of October 8, 2026, with a training-run cost example.
How to give Claude Code, Codex or Cursor access to GPUs (MCP setup)
Connect one MCP server and your coding agent can estimate, launch and watch GPU jobs on your account, with dry runs and spending limits.
How to make GPU training survive spot preemption (without babysitting it)
Save model, optimizer and step to one directory, write atomically, resume on start, and let the platform carry that directory across a reclaim.