Results for “llm-pretraining”

Pretrains large language models at scale using PyTorch-native torchtitan with 4D parallelism, Float8, and distributed checkpointing.

Provides battle-tested PyTorch training recipes for LLMs, vision, diffusion, and biomedical domains, covering training loops, optimizer selection, LR scheduling, mixed precision, and debugging.

Large Language Model development, training, fine-tuning, and deployment best practices.

Provides guidance for LLM post-training with RL using slime, a Megatron+SGLang framework. Use when training GLM models, implementing custom data generation workflows, or needing tight Megatron-LM integration for RL scaling.

Post-train LLMs with reinforcement learning using the slime framework, which integrates Megatron-LM for training and SGLang for rollout generation.

Implements and trains LLMs using Lightning AI's LitGPT with 20+ pretrained architectures (Llama, Gemma, Phi, Qwen, Mistral). Use when need clean model implementations, educational understanding of architectures, or production fine-tuning with LoRA/QLoRA. Single-file implementations, no abstraction layers.

Provides guidance for LLM post-training with RL using slime, a Megatron+SGLang framework. Use when training GLM models, implementing custom data generation workflows, or needing tight Megatron-LM integration for RL scaling.

Pretrains large language models from scratch using PyTorch-native distributed training with 4D parallelism (FSDP2, TP, PP, CP) and Float8 support on H100 GPUs.

Provides guidance for LLM post-training with RL using slime, a Megatron+SGLang framework. Use when training GLM models, implementing custom data generation workflows, or needing tight Megatron-LM integration for RL scaling.

Provides guidance for LLM post-training with RL using slime, a Megatron+SGLang framework. Use when training GLM models, implementing custom data generation workflows, or needing tight Megatron-LM integration for RL scaling.

Implements and trains LLMs using Lightning AI's LitGPT with 20+ pretrained architectures (Llama, Gemma, Phi, Qwen, Mistral). Use when need clean model implementations, educational understanding of architectures, or production fine-tuning with LoRA/QLoRA. Single-file implementations, no abstraction layers.

Guides LLM post-training with RL using slime, a Megatron+SGLang framework for training GLM, Qwen, DeepSeek, and Llama models with GRPO, async, and multi-turn workflows.

Run Megatron-LM (MLM) and Megatron Bridge training with mock or real data, covering correlation testing, available recipes, and multi-GPU examples.

Validate and configure packed sequences and long-context training in Megatron-Bridge, distinguishing offline packed SFT for LLMs from in-batch packing for VLMs with correct context parallelism constraints.

Provides guidance for enterprise-grade RL training using miles, a production-ready fork of slime. Use when training large MoE models with FP8/INT4, needing train-inference alignment, or requiring speculative RL for maximum throughput.

Entrena modelos de machine learning con Scikit-learn, LightGBM y XGBoost, desde un baseline hasta un modelo productivo con validación robusta y explicabilidad.

Train, fine-tune, and deploy LLMs using LitGPT's clean implementations of 20+ architectures like Llama, Gemma, and Phi.

Guides machine learning development with experiment tracking, hyperparameter optimization, model registry, and MLOps pipeline integration.

Push the LLM to reconsider, refine, and improve its recent output. Use when user asks for deeper critique or mentions a known deeper critique method, e.g. socratic, first principles, pre-mortem, red team.

Push the LLM to reconsider, refine, and improve its recent output. Use when user asks for deeper critique or mentions a known deeper critique method, e.g. socratic, first principles, pre-mortem, red team.

Push the LLM to reconsider, refine, and improve its recent output. Use when user asks for deeper critique or mentions a known deeper critique method, e.g. socratic, first principles, pre-mortem, red team.

Jailbreak API-served LLMs using G0DM0D3 techniques — Parseltongue input obfuscation (33 techniques), GODMODE CLASSIC system prompt templates, ULTRAPLINIAN multi-model racing, encoding escalation, and Hermes-native prefill/system prompt integration. Use when a user wants to bypass safety filters, jailbreak a model, or red-team model robustness.

Deploy Llama Guard, NeMo Guardrails, and LLM Guard as runtime input/output scanners to block jailbreaks, prompt injection, and toxic content in production LLM applications.

Trains LLMs with SimPO, a reference-free preference optimization method that outperforms DPO, using configurable hyperparameters and workflows for various models and tasks.

Use this skill whenever a developer wants to turn an idea into a complete, production-ready, end-to-end system plan BEFORE writing any code. Trigger on 'plan this system', 'design the architecture for', 'help me blueprint', 'deep plan for X', 'break this idea into components', 'expand into action points', 'full implementation plan', or when the user pastes a project idea wanting architecture, components, pipelines, and file-level execution mapped out. Casual phrasing also triggers: 'help me think this through end-to-end', 'plan before coding'. Also covers living-workspace patterns: self-improving knowledge bases, reflection loops with auditor agents, four-agent teams, schema-as-code, wiki health scoring. **Resume triggers**: 'where did we leave off', 'continue this plan', 'resume my blueprint' - rehydrates state from disk via SESSION.md/NEXT.md/decisions/. Web research is mandatory every invocation.

Train large-scale MoE models with FP8/INT4 low-precision RL, speculative decoding, and train-inference alignment using the miles framework.

Visual Prompt Tuning

Organizes PyTorch code with a Trainer class, automatic distributed training (DDP/FSDP/DeepSpeed), callbacks, and minimal boilerplate. Scales from laptop to supercomputer with the same code.

Serves LLMs with high throughput using vLLM's PagedAttention and continuous batching. Use when deploying production LLM APIs, optimizing inference latency/throughput, or serving models with limited GPU memory. Supports OpenAI-compatible endpoints, quantization (GPTQ/AWQ/FP8), and tensor parallelism.

Simple Preference Optimization for LLM alignment. Reference-free alternative to DPO with better performance (+6.4 points on AlpacaEval 2.0). No reference model needed, more efficient than DPO. Use for preference alignment when want simpler, faster training than DPO/PPO.

Fine-tune LLMs using reinforcement learning with TRL - SFT for instruction tuning, DPO for preference alignment, PPO/GRPO for reward optimization, and reward model training. Use when need RLHF, align model with preferences, or train from human feedback. Works with HuggingFace Transformers.

Train machine learning models end-to-end, covering data loading, preprocessing, architecture selection, training loops, validation, and checkpointing. Use when the user requests model training or provides relevant inputs for this workflow.

Compress large language models using knowledge distillation from teacher to student models, covering temperature scaling, soft targets, reverse KLD, logit distillation, and MiniLLM training strategies.

Simple Preference Optimization for LLM alignment. Reference-free alternative to DPO with better performance (+6.4 points on AlpacaEval 2.0). No reference model needed, more efficient than DPO. Use for preference alignment when want simpler, faster training than DPO/PPO.