Ready to supercharge your AI agents?

Install any skill in seconds. Free to start, no credit card required.

Learning Rate Scheduler · dicklesworthstone | DecimalAI

dicklesworthstone/learning-rate-scheduler

SkillScore100Maintained

Manage learning rate scheduler operations. Auto-activating skill for ML Training. Triggers on: learning rate scheduler, learning rate scheduler Part of the ML Training skill category. Use when working with learning rate scheduler functionality. Trigger with phrases like "learning rate scheduler", "learning scheduler", "learning".

by dicklesworthstone

Links this skill to your agents and keeps it current with the author’s updates.

Automatic

AI Engineer pack

1 · One-time — install the CLI (docs ↗)

pip install decimalai

2 · Get the skill — writes .claude/skills/dicklesworthstone-learning-rate-scheduler/SKILL.md

decimalai skills pull dicklesworthstone-learning-rate-scheduler --out .claude/skills/

No CLI? Get the same file — curl · raw file

curl -fsSL --create-dirs -o .claude/skills/dicklesworthstone-learning-rate-scheduler/SKILL.md https://app.decimal.ai/s/dicklesworthstone-learning-rate-scheduler/SKILL.md

Raw file: /s/dicklesworthstone-learning-rate-scheduler/SKILL.md ↗ · machine-readable summary at /s/dicklesworthstone-learning-rate-scheduler.json ↗

In productionGlobal

how this skill behaves on real traffic · last 30 days

No production data for this skill yet — usage numbers appear here once agents start running it.

Anonymized aggregate across every org using this skill — proof from real production use, not a demo.

Does this skill help? · Benchmark: with vs. withoutsign in to run this on your agents

22 test cases from the publisher’s eval.yaml · each run twice on gemini-3.6-flash — with the skill injected, then without · last run 22d ago

Yes — +5% lift+24% tokens0% turns100% of whole cases pass with skill (22/22)

Test case	Without → With	Effect	Δ tokens	Δ turns
case-10	✗→✓	▲ Improved	-4%	0%
case-01	✓→✓	＝ Same ✓	17%	0%
case-02	✓→✓	＝ Same ✓	68%	0%
case-03	✓→✓	＝ Same ✓	15%	0%
case-04	✓→✓	＝ Same ✓	45%	0%

Showing 5 of 22 cases

SkillSafety

✓ 3 of 3 passed

Code scan15 of 15 checks clean

0 critical · 0 warning · 0 info — scanner v4 — 8/10/2026

live_secretremote_code_executionreverse_shelldata_exfiltrationssrf_metadataagent_snoopinganti_refusaldestructive_commandsobfuscated_unicodeunicode_homoglyphopaque_blobsuspicious_urltool_overreachtrigger_abuseprompt_injection_phrasing

IntentSafe

LLM judge · does the skill do what it claims

ContentClean

LLM judge · policy & harmful-content review

Source

Imported

dicklesworthstone / pi_agent_rust

tests/ext_conformance/artifacts/plugins-community/skills/07-ml-training/learning-rate-scheduler/SKILL.md

View on GitHub ↗

License MITImported 8/10/2026Version v1

Skill contents

SKILL.md

Learning Rate Scheduler

Overview

This skill provides automated assistance for learning rate scheduler tasks within the ML Training domain.

When to Use

This skill activates automatically when you:

Mention "learning rate scheduler" in your request
Ask about learning rate scheduler patterns or best practices
Need help with machine learning training skills covering data preparation, model training, hyperparameter tuning, and experiment tracking.

Instructions

Provides step-by-step guidance for learning rate scheduler
Follows industry best practices and patterns
Generates production-ready code and configurations
Validates outputs against common standards

Examples

Example: Basic Usage Request: "Help me with learning rate scheduler" Result: Provides step-by-step guidance and generates appropriate configurations

Prerequisites

Relevant development environment configured
Access to necessary tools and services
Basic understanding of ml training concepts

Output

Generated configurations and code
Best practice recommendations
Validation results

Error Handling

| Error | Cause | Solution | |-------|-------|----------| | Configuration invalid | Missing required fields | Check documentation for required parameters | | Tool not found | Dependency not installed | Install required tools per prerequisites | | Permission denied | Insufficient access | Verify credentials and permissions |

Resources

Official documentation for related tools
Best practices guides
Community examples and tutorials

Related Skills

Part of the ML Training skill category. Tags: ml, training, pytorch, tensorflow, sklearn

Illustrative Want to see what dicklesworthstone/learning-rate-scheduler would put in front of your model? Preview the rendered body and check its trigger phrases — no LLM call, no signup.

Latest benchmark · publisher runVerified

8/30/2026, 06:05 UTC · 22 cases

Lift

+5%

100% vs 95%

Agent turns

1.0 vs 1.0

Tokens

+24%

2,366.73 vs 1,914.18

fail → pass · 1pass → pass · 21

Per-case results (22)

Case	Status	Duration (ms)			Turns			Tokens			Tool calls
Case	Status	Without	With	Δ	Without	With	Δ	Without	With	Δ	Without	With	Δ
case-01 I am training a ResNet model in PyTorch for 100 epochs with cosine annealing decay. I usually call the scheduler step function once per epoch, but my colleague suggested calling it after every mini-batch iteration to get smoother decay curves. How should I update the scheduler and call scheduler.step() in my PyTorch training loop?	pass→pass	15,395	15,209	-1%	1	1	0%	2,117	2,479	+17%	0	0	—
case-02 I am configuring PyTorch's ReduceLROnPlateau to lower the learning rate when training slows down. Should I call scheduler.step() without arguments at the end of every epoch like standard schedulers, or pass a metric to it?	pass→pass	11,805	14,729	+25%	1	1	0%	1,384	2,325	+68%	0	0	—
case-03 I am training a Transformer model using AdamW from scratch. Should I start immediately with a peak learning rate of 1e-3, or should I start small and scale up? How should the schedule be structured?	pass→pass	20,027	20,469	+2%	1	1	0%	2,779	3,197	+15%	0	0	—
case-04 I am using PyTorch's OneCycleLR for fast convergence on an image classifier. I set my base learning rate to 1e-4 in the optimizer, but OneCycleLR throws an error or behaves unexpectedly regarding peak learning rate. What parameter must be explicitly provided to OneCycleLR to define the peak rate?	pass→pass	12,021	13,634	+13%	1	1	0%	1,441	2,095	+45%	0	0	—
case-05 I am running mixed-precision training in PyTorch using torch.cuda.amp.GradScaler. I noticed my learning rate scheduler step throws warnings or skips steps. Should I call scheduler.step() before scaler.step(optimizer) or after scaler.update()?	pass→pass	15,920	17,477	+10%	1	1	0%	2,037	2,667	+31%	0	0	—
case-06 I want to train a model with CyclicLR in PyTorch and want the peak learning rate amplitude to decrease over time rather than remaining constant throughout training. Which mode parameter setting should I use?	pass→pass	13,702	11,541	-16%	1	1	0%	1,641	1,582	-4%	0	0	—
case-07 I am building a semantic segmentation model in PyTorch and want a learning rate decay schedule where the rate drops quickly early on and flattens out near the end, following a power curve. What scheduler function or power parameter configuration should I use?	pass→pass	19,341	22,913	+18%	1	1	0%	2,870	3,382	+18%	0	0	—
case-08 I am training a classification network for 90 epochs and want to drop the learning rate by 10x at epoch 30 and epoch 60. How should I configure MultiStepLR in PyTorch?	pass→pass	16,002	13,139	-18%	1	1	0%	2,156	2,082	-3%	0	0	—
case-09 I am using ExponentialLR in PyTorch to decay the learning rate by 5% every epoch. What parameter value should I pass for gamma?	pass→pass	9,391	8,880	-5%	1	1	0%	845	1,107	+31%	0	0	—
case-10 I want to create a custom decay rule in PyTorch where the learning rate scales linearly from 1.0 down to 0.1 over 100 epochs based on the current epoch index. How should I use LambdaLR for this task?	fail→pass	16,123	14,590	-10%	1	1	0%	2,315	2,223	-4%	0	0	—
case-11 I am configuring PyTorch CosineAnnealingLR for a 200-epoch training run where scheduler.step() is called once per epoch. What value should T_max be set to, and what does eta_min represent?	pass→pass	13,878	17,347	+25%	1	1	0%	1,766	2,390	+35%	0	0	—
case-12 I want to use CosineAnnealingWarmRestarts in PyTorch so that my learning rate resets every 10 epochs initially, and the interval doubles after each restart. What parameters control this behavior?	pass→pass	11,698	13,537	+16%	1	1	0%	1,510	2,127	+41%	0	0	—
case-13 I am trying to find the optimal initial learning rate for a new architecture before full training. Should I test discrete learning rates in separate short runs, or use an automated learning rate range test?	pass→pass	14,119	22,032	+56%	1	1	0%	2,340	3,643	+56%	0	0	—
case-14 I am using gradient accumulation across 4 mini-batches before calling optimizer.step(). Should I call scheduler.step() after every mini-batch or only when optimizer.step() is executed?	pass→pass	8,819	14,272	+62%	1	1	0%	1,615	2,092	+30%	0	0	—
case-15 In PyTorch 1.10+, I want to combine a LinearLR warmup for 5 epochs with a CosineAnnealingLR for the remaining 45 epochs. What built-in scheduler utility should I use to chain them sequentially?	pass→pass	12,857	7,675	-40%	1	1	0%	1,625	1,789	+10%	0	0	—
case-16 When saving a training checkpoint to resume later in PyTorch, I saved model.state_dict() and optimizer.state_dict(), but when resuming, my learning rate jumped back to the initial value. What step did I miss?	pass→pass	8,306	9,054	+9%	1	1	0%	1,668	2,165	+30%	0	0	—
case-17 I am fine-tuning a BERT model using Hugging Face Transformers and PyTorch. Instead of manual PyTorch schedulers, what Hugging Face helper function builds a standard linear warmup with cosine decay schedule?	pass→pass	9,274	12,057	+30%	1	1	0%	1,414	1,760	+24%	0	0	—
case-18 I am writing a custom PyTorch training script. I instantiated CosineAnnealingLR and noticed epoch 0 learning rate seems wrong or missed. Should I call scheduler.step() before the first optimizer.step() or after?	pass→pass	14,521	12,049	-17%	1	1	0%	1,984	2,738	+38%	0	0	—
case-19 During LLM pretraining with AdamW, should the learning rate decay all the way to absolute zero at the end of training, or should it decay to a non-zero floor value like 10% of peak LR?	pass→pass	15,330	18,762	+22%	1	1	0%	2,618	2,874	+10%	0	0	—
case-20 I am training a PyTorch image segmentation network with severe class imbalance where background pixels outnumber target pixels 100:1. What loss function adjustment should I make in PyTorch CrossEntropyLoss?	pass→pass	14,027	11,030	-21%	1	1	0%	2,118	2,509	+18%	0	0	—
case-21 What is the key mathematical difference between how standard Adam with L2 regularization handles weight decay versus AdamW in PyTorch?	pass→pass	17,309	18,587	+7%	1	1	0%	2,427	2,993	+23%	0	0	—
case-22 I am building a vision dataset pipeline using Albumentations in Python. How do I construct a Compose pipeline that randomly flips images horizontally and applies random brightness contrast?	pass→pass	12,305	12,813	+4%	1	1	0%	1,442	1,849	+28%	0	0	—

These cases were also screened by the skill’s intent / safety pass.See the Trust review ↗Reproduce this benchmark in Colab ↗

How this skill was measured

DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +5 percentage points is the difference between those two pass rates over the 22 comparable cases.

Model measured: gemini-3.6-flash
Graded by: gemini-3.6-flash
Grading method: judged
Verification: verified
Pass rate without the skill: 95.5%
Pass rate with the skill: 100%
Cases attempted: 22
Cases counted toward lift: 22
Cases that regressed: 0
Skill version measured: v1
Run completed: 8/30/2026, 06:05 UTC

Measured example — same prompt, with vs. without the skill · from the benchmark run · case case-10

Prompt

I want to create a custom decay rule in PyTorch where the learning rate scales linearly from 1.0 down to 0.1 over 100 epochs based on the current epoch index. How should I use LambdaLR for this task?

Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.

Show both answers

SkillScore blends 4 signals1 of 4 complete

BenchmarkPass rate WITH skill on a verified run100

Live evalNeeds ≥10 evaluated traces in 30 days

AI ratingNeeds ≥10 rated traces

AdoptionNeeds ≥10 serves across ≥2 orgs in 30 days

Efficiency · separate axis (at equal correctness)+0% turns+25% tokens

Shown for transparency; not credited unless a verified run shows no correctness regression.

Backed by 1 of 4 signals — each added signal makes the score more robust.

Verified on gemini-3.6-flash · re-tested 22 days ago

Other measured skills in the registry, with their headline benchmark lift.