140 skills · sorted by relevance
Fine-tune large language models with LoRA and QLoRA, methods that train small additions to a base model while limiting GPU memory use. Configure training, save task-specific adapters, switch between them, and merge them for deployment.
Prepare text for language models by splitting it into tokens, small units represented by numbers. The skill covers batch processing, training custom vocabularies, tracking positions in the original text, and adjusting input lengths.
Review code that manages the conversation history and extra information sent to Codex models. Apply rules for incremental history, size limits, cache stability, manual review of large additions, and required implementation structures.
Organize single-cell measurements with cell and gene metadata, read and write h5ad or Zarr files, and prepare subsets or combined datasets. Includes guidance on handling large annotated matrices without loading all measurement values into memory.
Analyze bid–ask spreads, order books, and buying and selling imbalances to assess liquidity, or how easily an asset trades. Estimate large-order price impact and transaction costs, with additional guidance on Chinese A-share auction and block trades.
Prepare text for language models using HuggingFace Tokenizers, which splits text into model-readable units. Train custom vocabularies, process batches, track original text positions, and configure input lengths for use with transformers.
Build image captioning, visual question answering, and text-based image search with BLIP-2. The skill explains model selection, batch processing, and memory-saving settings, with Python examples for connecting image understanding to language models.
Get guidance on adapting large language models with Axolotl. Covers YAML configuration files, LoRA and QLoRA methods for training fewer parameters, distributed training settings, dataset formats, and debugging training code.
Build image-based conversations with LLaVA, an AI model that combines image understanding with language. Covers setup, model selection, image descriptions, follow-up questions, and memory-saving options for running models on a GPU.
Guides experiments that inspect and change internal model values using nnsight. Covers local PyTorch models and remote experiments through NDIF, including replacing intermediate values to test how particular layers affect a model’s predictions.
Adapt large language models to specific tasks using LoRA and QLoRA, methods that train a small set of added parameters. Get guidance on preparing examples, choosing training settings, evaluating outputs, and saving or merging trained adapters.
Deploy language model APIs and run batch generation with vLLM. Configure GPU memory use, quantization—lower precision model storage—and distribution across GPUs, then test response times and monitor how many requests the server handles.
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.
Implement and troubleshoot large-model training with DeepSpeed across multiple GPUs. The skill covers memory optimization, parallel training, and numerical precision choices, alongside examples for storage transfers and models that divide work among specialized components.
Evaluate language models on knowledge, math, coding, safety, and image understanding benchmarks. The skill guides reproducible runs in containers using local Docker, Slurm clusters, or cloud execution, with shared configurations for model comparison and result export.
Predict B-cell and T-cell epitopes for vaccine antigen design and epitope mapping with BepiPred-3.0, DiscoTope-3.0, the IEDB tools, and EL-mode MHC presentation. Encodes the load-bearing asymmetry that T-cell epitope prediction is mature (it reduces to MHC presentation, AUC>0.9) while B-cell prediction is unreliable (linear predictors ~AUC 0.6 because ~90% of real epitopes are conformational) — so structure-based DiscoTope-3.0 on AlphaFold models is the only defensible B-cell path, propensity scales are obsolete, and NetChop is largely redundant on EL-trained models. Use when mapping epitopes or selecting vaccine antigens. MHC binding lives in mhc-binding-prediction.
The English summary is being prepared. Showing the original description.
Fit linear models to overdispersed count data. The package can estimate the overdispersion and fit repeated models for matrix input. It is designed to handle large input datasets as they typically occur in single cell RNA-seq experiments.
The English summary is being prepared. Showing the original description.
Delegates tasks to Qwen CLI via delegation-core for Alibaba's models. Use when delegation-core selects Qwen or large-context batch processing is needed.
The English summary is being prepared. Showing the original description.
Use when training models across multiple GPUs or nodes, handling large models that don't fit in memory, or optimizing training throughput - covers DDP, FSDP, DeepSpeed ZeRO, model/data parallelism, and gradient checkpointingUse when ", " mentioned.
The English summary is being prepared. Showing the original description.
Expert guide for integrating Large Language Models (LLMs), Model Context Protocol (MCP v1.x), hybrid reasoning models, RAG architecture, vector databases, and AI agents / Panduan ahli untuk integrasi LLM, Model Context Protocol (MCP), model hybrid reasoning, arsitektur RAG, vector database, dan agen AI.
The English summary is being prepared. Showing the original description.
Decompose dense codebase-wide, multi-document, PDF, and aggregation work even when the input fits the context window, following Recursive Language Models (Zhang, Kraska, Khattab, 2025). Use when the user asks to analyse all files, a whole repo, all docs, large PDFs, or to aggregate or multi-hop across scattered sources. Skip one file, one function, a single needle, or a one-page PDF conversion. Triggers: long context, context rot, large codebase, many files, all files, big document, multi-document, PDF, aggregate, summarize everything, codebase-wide, multi-hop, recursive, sub-agents, map-reduce.
The English summary is being prepared. Showing the original description.
Our large-scale experiment with 19 LLMs reveals that current models degrade documents during delegation: even frontier models (Gemini 3.
The English summary is being prepared. Showing the original description.
I-A LLMs Large Language Models (LLMs) are extensive, general-purpose models that can be pre-trained and adapted for specific tasks.
The English summary is being prepared. Showing the original description.
Activation-aware weight quantization for 4-bit LLM compression with 3x speedup and minimal accuracy loss. Use when deploying large models (7B-70B) on limited GPU memory, when you need faster inference than GPTQ with better accuracy preservation, or for instruction-tuned and multimodal models. MLSys 2024 Best Paper Award winner.
The English summary is being prepared. Showing the original description.