Download

Skill UI

Browse and discover 5096+ curated skills

All Development Artificial Intelligence Design & Creative Product & Business Data Science Marketing Soft Skills Productivity Engineering Languages

Search vLLM , found 10 results

Default Newest Most Downloaded

AWQ Weight Quantization

awq-quantization

Orchestra-Research/AI-Research-SKILLs

AWQ provides activation-aware 4-bit quantization for large language models, delivering ~3x inference speedup and sub-5% accuracy loss to deploy instruction-tuned or multimodal models on memory-constrained GPUs with vLLM integration and Marlin kernels.

LLM Evaluation Harness

evaluating-llms-harness

Orchestra-Research/AI-Research-SKILLs

Runs lm-evaluation-harness across 60+ academic benchmarks such as MMLU, HumanEval, GSM8K, TruthfulQA, and HellaSwag to benchmark HuggingFace/vLLM/API models, compare variants, track training checkpoints, and save standardized reports for research labs.

Half Quadratic Quantization

hqq-quantization

Orchestra-Research/AI-Research-SKILLs

Calibrate-free half-quadratic quantization for LLMs that delivers 1-8 bit model compression with configurable backends (PyTorch, TorchAO, BitBlas, Marlin). Ideal for fast quantization workflows, zero-data deployment, and PEFT/LoRA fine-tuning with HuggingFace or vLLM stacks.

LlamaGuard Content Moderation

Orchestra-Research/AI-Research-SKILLs

LlamaGuard is Meta’s 7–8B safety-specialized LLM that filters both prompts and responses by classifying six threat categories, enabling fast inference via vLLM/SageMaker and integration into NeMo Guardrails for end-to-end moderation.

OpenRLHF High Performance Training

openrlhf-training

Orchestra-Research/AI-Research-SKILLs

OpenRLHF is a Ray-based RLHF framework that accelerates large-model PPO, GRPO, RLOO, and DPO training with vLLM inference and ZeRO-3 resource sharing, targeting distributed GPU clusters handling 7B-70B+ parameter models.

Structured Text Generation

Orchestra-Research/AI-Research-SKILLs

Outlines guarantees valid JSON/XML/code generation via CFG-driven FSM filtering, Pydantic schemas, and fast local or API models (Transformers, vLLM, llama.cpp), making structured inference safe and high-performance.

HighThroughput LLMS

serving-llms-vllm

Orchestra-Research/AI-Research-SKILLs

Deploy LLMs with vLLM to maximize throughput and minimize latency through PagedAttention, continuous batching, quantization, and tensor parallelism for production APIs or batch inference on constrained GPUs.

Torchforge RL Training

torchforge-rl-training

Orchestra-Research/AI-Research-SKILLs

Torchforge RL Training documents Meta's PyTorch-native torchforge library, which separates algorithms from infrastructure so researchers can explore GRPO/SFT losses, manage Monarch actors, tap TorchTitan and vLLM for scalable distributed runs, and monitor KL/loss curves.

Verl RL Training

verl-rl-training

Orchestra-Research/AI-Research-SKILLs

Guides RLHF-style LLM fine-tuning with verl's HybridFlow stack, scaling PPO/GRPO/DAPO across FSDP/Megatron/vLLM backends for math reasoning, agentic multiturn, and vision-language workflows.

Hugging Face Eval Manager

hugging-face-evaluation

sickn33/antigravity-awesome-skills

Manages evaluation results in Hugging Face model cards by scraping tables, importing Artificial Analysis benchmarks, and running custom vLLM/lighteval evaluations, with metadata formatted for model-index and Jobs integration.

1

Language