sentencepiece
orchestra-research
Build text tokenizers that work across languages.
Natural language processing
Machine learning & data science skills for AI agents
Data comes first here: analysis and visualisation Skills explore datasets, test hypotheses and build dashboards. Machine learning Skills fine-tune and evaluate models, train them across many GPUs and serve them, with smaller groups for computer vision, language and vector search.
Sort by
Most installed
532 Skills
Showing Skills in Machine learning & data science.- 895
- 894
moe-training
orchestra-research
Train larger models with fewer active experts.
Machine learning
- 894
training-llms-megatron
orchestra-research
Train large language models across NVIDIA GPUs.
Language model training
- 893
llama-factory
orchestra-research
Fine-tune models with LLaMA-Factory guidance.
Model fine-tuning
- 893
pyvene-interventions
orchestra-research
Find what drives your neural network's behavior.
Model interpretability
- 893
transformer-lens-interpretability
orchestra-research
Find what drives a transformer model's answers.
Mechanistic interpretability
+1 - 892
evaluating-code-models
orchestra-research
Compare coding models with tests, not guesses.
Model evaluation
+1 - 892
model-merging
orchestra-research
Blend language models without training again.
Machine learning
- 892
simpo-training
orchestra-research
Align language models without a reference model.
Language model training
- 892
sparse-autoencoder-training
orchestra-research
Find interpretable features in language models.
Mechanistic interpretability
+1 - 890
mamba-architecture
orchestra-research
Build Mamba models for long-sequence workloads.
Machine learning
- 890
nemo-curator
orchestra-research
Clean training data with GPU-powered filters.
Training data curation
- 890
torchforge-rl-training
orchestra-research
Build RL training code apart from infrastructure.
Reinforcement learning
- 888
miles-rl-training
orchestra-research
Train large AI models with less memory in miles.
Reinforcement learning
- 888
model-pruning
orchestra-research
Compress language models without retraining.
Model compression
LL 
887maplibre-tile-sources
maplibre
Choose the right data source for your MapLibre map.
Web mapping
- 887
nnsight-remote-interpretability
orchestra-research
Probe large neural networks without local GPUs.
Mechanistic interpretability
- 884
blip-2-vision-language
orchestra-research
Build apps that describe images with BLIP-2.
Computer vision
884eval-model
modular
Benchmark your MAX model against standard tests.
Model evaluation
- 883
nemo-evaluator-sdk
orchestra-research
Compare language models on repeatable benchmarks.
Model evaluation
+2 
882d3-viz
chrisvoncsefalvay
Custom D3.js charts with precise visual control.
Data visualisation
874he4rt-twitter-engagement-tracker
reason-machines
Capture Twitter engagement as you browse.
Community analytics
866amazon-location-service
cline
Build location-aware apps with Amazon Location Service.
Geospatial development
+1
866pandas-best-practices
mindrally
Write efficient, reproducible Pandas analysis.
Data analysis
Yes, on hardware you provide or a cloud you connect. Training Skills set up the run, watch it, and save the evidence so results can be compared.
No, but the Skills show their working: which test they ran, on what data and with what assumptions, so you can judge the result.
Under Research & science, with the other domain-science Skills. This category keeps general data analysis and machine learning.