Orchestra Research

Build text tokenizers, which split text into units for AI models, directly from raw multilingual text. Learn custom vocabularies with BPE or Unigram, convert text to pieces or numeric IDs, and decode the results.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Preparing text for multilingual AI models
• Training a vocabulary from Japanese text
• Reusing a tokenizer for reproducible text splitting

Configure large language model training across multiple NVIDIA GPUs with Megatron-Core. The skill helps choose how to split models and workloads, set training parameters, monitor performance, and address memory shortages or low GPU utilization.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Planning distributed training for a LLaMA-style model
• Distributing MoE expert networks across GPUs
• Tuning GPU utilization and training throughput

Evaluate code generation models with HumanEval, MBPP, and multilingual benchmarks. The skill guides test execution and model comparisons using pass@k, an estimate of whether at least one of k generated solutions passes the tests.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Compare code generation models on shared benchmarks
• Measure code generation performance across languages
• Evaluate instruction-tuned coding models

Compress large language model weights to four bits with AWQ, a method that uses model activation patterns to limit quality loss. Follow workflows for quantization, loading compressed models, and serving them with vLLM using less GPU memory.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Running large language models with limited GPU memory
• Quantizing a model with domain-specific text
• Serving an AWQ model with vLLM

Investigate how transformer language models produce predictions with TransformerLens. This skill guides inspection of intermediate values and attention patterns, plus experiments that replace internal values to test their effects and identify components involved in model behavior.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Inspecting attention patterns and intermediate values
• Testing output changes by replacing internal activations
• Analyzing attention head contributions to predictions

Convert language models to GGUF and reduce their size through quantization, which stores model weights at lower precision. Guides cover compression choices, local inference on CPUs and Apple Silicon, and integration with tools such as Ollama.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Running a language model on a CPU-only computer
• Choosing model compression for available memory
• Configuring Metal acceleration on Apple Silicon

Adapt existing PyTorch training scripts to run across multiple GPUs or machines with Accelerate. This skill covers device setup, distributed training configuration, and mixed precision, which uses lower-precision numbers to reduce memory use.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Adapt a single-GPU training script for multiple GPUs
• Configure mixed precision to reduce training memory use
• Set up training with DeepSpeed or FSDP

Guide language model training with SimPO using pairs of preferred and rejected answers. Covers configuration, learning rate tuning, and memory troubleshooting for preference alignment, which adjusts model responses toward human preferences without a separate reference model.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Training on preferred and rejected answer pairs
• Exploring an alternative to DPO without a reference model
• Tuning preference training for math and code tasks

Guides experiments that change internal values in PyTorch models to test what causes their behavior. Uses pyvene to swap or remove activations, locate influential components, train interventions, and save experiments for reuse.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Locating model components that affect an answer
• Comparing outputs after swapping internal activations
• Training interventions to investigate causal structure

Helps adapt machine learning training to multiple GPUs and machines with Ray Train. Covers resource allocation, recovery from saved training states, and Ray Tune searches that compare settings such as learning rate and batch size.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Adapt existing PyTorch training to multiple GPUs
• Run HuggingFace training across multiple machines
• Compare learning rates and batch sizes with Ray Tune

Set up reinforcement learning, training guided by rewards, for language models with slime. The skill covers Megatron-LM training, SGLang response generation, data preparation, asynchronous execution, and custom workflows for agents that use tools across multiple turns.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Prepare GRPO training for GLM or Qwen3 models
• Overlap response generation and model training
• Train agents with multi-turn tool use

Assess language models with over 60 academic benchmark tests using consistent prompts and scoring. Compare models, track performance during training, and organize research results with support for HuggingFace models, vLLM, and APIs.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Compare models on standard knowledge and reasoning tests
• Track model performance across training checkpoints
• Organize evaluation results for research reports

Use SAELens to train and analyze sparse autoencoders, which break model activations into a small set of interpretable features. Explore pretrained features, evaluate reconstruction quality, troubleshoot training, and test how individual features influence generated text.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Exploring features in a pretrained SAE
• Training an SAE for a selected model layer
• Investigating inactive features and reconstruction errors

Add safety checks to LLM applications with NeMo Guardrails. This skill guides input and output validation, detection of attempts to bypass safety rules, personal information masking, and checks for unsupported or inaccurate answers.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Detect attempts to bypass an AI app's safety rules
• Mask personal information before processing
• Add fact-checking to generated answers

Guides large language model reinforcement learning with miles, including MoE models that route work to specialized components. Covers low-precision training, matching training and inference behavior, and accelerating training response generation with draft models.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Setting up reinforcement learning for large MoE models
• Configuring FP8 or INT4 training
• Checking consistency between training and inference

Learn how GPT language models work by preparing text, training a small model, and generating samples. Covers CPU experiments, custom datasets, further training of pretrained GPT-2 models, and reproducing GPT-2 with multiple GPUs.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Train a small model on Shakespeare using a CPU
• Prepare custom text for model training
• Further train a pretrained GPT-2 model

Optimize transformer attention, the calculation that relates tokens to one another, using Flash Attention. The skill guides PyTorch integration, GPU memory reduction, performance measurement, and output comparisons for training or inference with long sequences.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Reducing GPU memory use with long sequences
• Integrating Flash Attention into PyTorch models
• Comparing runtime and outputs before and after changes

Prepare large text collections for AI models with fast tokenizers, which split text into model inputs. Train custom vocabularies, trace tokens to original text positions, adjust input lengths, and integrate the results with transformers.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Tokenizing large text collections
• Training a custom vocabulary
• Mapping extracted names or answers back to original text

Prepare training datasets for large language models with GPU-assisted cleaning. Covers duplicate removal, quality filtering, and personal information redaction, with workflows for text, images, video, and audio using NeMo Curator.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Clean web data for model training
• Remove exact and near duplicates from large text collections
• Redact personal information from training text

Add PyTorch FSDP2 to training scripts so model state can be split across GPUs. Covers distributed initialization, mixed precision, CPU offloading, and checkpoint saving and loading for models that exceed a single GPU's memory.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Training a model that exceeds one GPU's memory
• Adding FSDP2 to an existing training script
• Configuring mixed precision and CPU offloading

Guides LLM fine-tuning with Axolotl, from YAML configuration and dataset formats to distributed training across GPUs. Covers LoRA and QLoRA, methods that train on preferences, multimodal support, and troubleshooting using documentation references and code examples.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Prepare YAML settings for LLM fine-tuning
• Explore LoRA, QLoRA, and preference training options
• Check dataset formats and input handling

Use LitGPT to explore language model implementations and train models on custom data. Covers pretrained model loading, LoRA fine-tuning that updates a small set of parameters, pretraining, and preparing trained models for deployment.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Studying language model architecture through readable code
• Fine-tuning a model on custom instruction data
• Configuring LoRA training for a single GPU

Coordinate AI research from literature review and hypothesis testing to findings and paper drafts. The skill alternates focused experiments with broader reflection, maintains project records, and directs specialized skills to carry out research tasks.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Turning a research question into testable hypotheses
• Managing repeated experiments across multiple hypotheses
• Resuming research from recorded findings and project state

Helps reduce GPU memory use by loading large language models in 4-bit or 8-bit formats, a technique called quantization. Covers bitsandbytes configuration, memory checks, QLoRA fine-tuning, and training with memory-saving optimizers.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Load a larger model within available GPU memory
• Compare 4-bit and 8-bit quantization settings
• Fine-tune a quantized model with QLoRA