查找技能

Use SAM to identify object regions in images from points, boxes, or existing masks. The skill provides implementation examples for transparent cutouts, automatic mask generation, annotation tools, and model selection based on available resources.

中文简介正在准备中。当前显示英语简介。

• Extracting objects with transparent backgrounds
• Building annotation tools with point or box selection
• Generating object masks automatically

Investigate why a PyTorch model behaves a certain way by changing its internal values with pyvene. The skill guides experiments that swap or remove activations, train interventions, compare outputs, and save or share intervention configurations.

中文简介正在准备中。当前显示英语简介。

• Locating model components that affect an output
• Swapping activations to test behavioral hypotheses
• Training interventions to investigate causal structure

Reduce GPU memory use by loading language models in 8-bit or 4-bit formats with bitsandbytes and HuggingFace Transformers. Includes configuration examples, QLoRA fine-tuning of compressed models, and optimizers that reduce memory needed during training.

中文简介正在准备中。当前显示英语简介。

• Load larger models within limited GPU memory
• Compare 8-bit and 4-bit model settings
• Fine-tune a model with QLoRA

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.

中文简介正在准备中。当前显示英语简介。

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

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.

中文简介正在准备中。当前显示英语简介。

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

Configure large language model pretraining with torchtitan across multiple GPUs. Covers ways to split training work, launch jobs on computing clusters, enable Float8 numerical processing, and save or resume training through distributed checkpoints.

中文简介正在准备中。当前显示英语简介。

• Pretraining Llama 3.1 8B on eight GPUs
• Configuring multi-node training with SLURM
• Enabling Float8 training on H100 GPUs

Combine capabilities from multiple fine-tuned LLMs without retraining using mergekit. Choose a merging method, configure how models are blended, and evaluate the resulting model on tasks such as math, code generation, and conversation.

中文简介正在准备中。当前显示英语简介。

• Combining math, coding, and conversation capabilities
• Blending two models at different ratios
• Comparing variants made with different merging methods

Adapt PyTorch training scripts for multiple GPUs or machines with Accelerate. This skill guides code changes, configuration, and launch commands, including mixed precision, which combines numerical formats, and DeepSpeed or FSDP integration for distributed training.

中文简介正在准备中。当前显示英语简介。

• Moving single-GPU training to multiple GPUs
• Configuring mixed precision training
• Setting up DeepSpeed or FSDP

Build LlamaIndex applications that retrieve relevant material before generating answers, a method called RAG. Connect documents, websites, and databases to support document questions, conversational search, structured summaries, and checks of answer relevance and source consistency.

中文简介正在准备中。当前显示英语简介。

• Building question answering over internal documents
• Creating a chatbot that remembers the conversation
• Searching information across files, websites, and databases

Extend the amount of text a language model can process at once. Compares RoPE, YaRN, ALiBi, and position interpolation, with implementation examples, additional training guidance, and memory optimization approaches for long document processing.

中文简介正在准备中。当前显示英语简介。

• Extend an existing model to handle longer inputs
• Compare methods for representing token positions
• Plan additional training with long documents

Adapt machine learning training to run across multiple GPUs or machines with Ray Train. Covers changes to existing code, comparing training settings with Ray Tune, and saving checkpoints so training can resume after failures.

中文简介正在准备中。当前显示英语简介。

• Adapting existing PyTorch code for multiple GPUs
• Training HuggingFace models across multiple machines
• Comparing learning rates and batch sizes with Ray Tune

Evaluate code generation models with HumanEval, MBPP, and other benchmarks. Test generated programs, compare models across programming languages, and record pass@k scores, which estimate the chance that at least one of k solutions passes the tests.

中文简介正在准备中。当前显示英语简介。

• Comparing code generation models on shared benchmarks
• Measuring correctness across programming languages
• Evaluating models trained to follow instructions

Build and train Mixture of Experts models, which activate selected specialist networks for each input. Use DeepSpeed or HuggingFace workflows to configure routing, balance workloads, distribute experts across GPUs, and optimize inference.

中文简介正在准备中。当前显示英语简介。

• Designing MoE models for a limited compute budget
• Implementing Mixtral-style expert routing
• Adjusting expert workloads and training stability

Transcribe audio files through 9Router using models from providers such as OpenAI, Groq, and Deepgram. Receive plain text, structured results with timestamps, or SRT and VTT subtitles, depending on the selected model’s support.

中文简介正在准备中。当前显示英语简介。

• Transcribing recorded meetings or interviews
• Creating SRT or VTT subtitles from audio
• Getting transcriptions with timestamps

Audit whether an AI agent follows its intended purpose, business rules, and permission boundaries. The skill guides setup discovery, test configuration, execution against an agent endpoint or substitute model, and interpretation of scores from selected judging models.

中文简介正在准备中。当前显示英语简介。

• Auditing an agent's compliance with business rules
• Testing permission boundaries and dangerous tools
• Evaluating an agent through its HTTP endpoint

Explore code structure and history using an index of code relationships and meaning-based search. Trace calls, assess change impact, find likely bug locations, generate architecture reports, and review pull requests for risk and conflicts.

中文简介正在准备中。当前显示英语简介。

• Tracing function calls and change impact
• Investigating code locations related to a GitHub issue
• Generating architecture reports and class diagrams

Check that changes to existing PEFT methods still work with saved training checkpoints. The skill guides compatibility tests, configuration checks, benchmark comparisons, and reporting, including comparisons of model outputs before and after a change.

中文简介正在准备中。当前显示英语简介。

• Checking compatibility after a forward method rewrite
• Changing state_dict structure while supporting old checkpoints
• Reviewing defaults for new configuration fields

Guides migration of existing AI inference workloads to self-hosted GKE, Google’s managed Kubernetes service. Covers requirements, GPU sizing, deployment configuration, model storage, validation, and traffic cutover using gcloud and kubectl.

中文简介正在准备中。当前显示英语简介。

• Migrating an existing Cloud Run AI service to GKE
• Planning a self-hosted replacement for Gemini API inference
• Sizing GPUs and preparing deployment configuration

Prepare to fine-tune open models or Gemini models on Agent Platform. Check the environment, convert and validate training data, choose supported models and settings, and estimate tuning costs for open models.

中文简介正在准备中。当前显示英语简介。

• Checking whether an Agent Platform tuning environment is ready
• Converting and validating training datasets
• Choosing models and training settings for a task

Deploy Model Garden models to Agent Platform endpoints, the addresses used for model requests. Check supported hardware and estimated hourly costs, inspect deployment progress, test predictions, and remove deployed resources after use.

中文简介正在准备中。当前显示英语简介。

• Choose a deployable model and supported hardware
• Estimate hourly deployment costs before approval
• Check deployment progress and test predictions

Add or review research paper entries in TRL’s paper index. Organize each paper’s contribution, its connection to the implementation, and a Python configuration example using training settings supported by the paper.

中文简介正在准备中。当前显示英语简介。

• Documenting a newly implemented research method
• Checking paper index updates during PR review
• Recording training settings from a paper

Write and debug code for further training large language models with TRL. Match training methods to dataset formats, configure supervised or preference-based training, and set up reward functions through Python or the command line.

中文简介正在准备中。当前显示英语简介。

• Training a language model on example responses
• Training with preferred and rejected response pairs
• Debugging GRPO reward functions and batch settings

Train sentence-transformers models for search, similarity, and reranking, which reorders search results by relevance. Choose a model type, prepare data, configure training, compare results against a baseline, and publish the trained model to Hugging Face Hub.

中文简介正在准备中。当前显示英语简介。

• Fine-tuning an embedding model for domain-specific search
• Training a model to rerank search results
• Training SPLADE or ColBERT retrieval models

Answer questions about videos from URLs or local files using timestamped visual and spoken evidence. Choose extracted frames and transcripts, or let Gemini view the full video with a configured API key.

中文简介正在准备中。当前显示英语简介。

• Summarizing a video's visuals and speech
• Finding timestamps relevant to a question
• Checking a specific section of a long video