Results for “models”

Detect poisoned training data and backdoored models across the ML pipeline using statistical analysis, activation clustering, and spectral signatures.

Fine-tune large language models by training less than 1% of parameters using LoRA, QLoRA, and 25+ adapter methods, enabling efficient adaptation on consumer GPUs.

Convert and quantize models to GGUF format for efficient CPU/GPU inference with llama.cpp, supporting 2-8 bit quantization and Apple Silicon acceleration.

Train reinforcement learning models using torchforge, Meta's PyTorch-native RL library for scalable, algorithm-focused experimentation with GRPO, DAPO, and custom loss functions.

Evaluates text-to-image models on spatial relationship accuracy using the VISOR metric, separating object detection from spatial correctness to reveal biases like object priority and merging.

Evaluates social bias in question-answering models using the BBQ benchmark, measuring accuracy and a bias score across ambiguous and disambiguated contexts to reveal reliance on stereotypes.

Fine-tune, evaluate, run inference, and export Cosmos-Embed1 video-text embedding models for tasks like text-to-video retrieval and semantic deduplication.

Merge multiple fine-tuned models using mergekit to combine capabilities without retraining, covering SLERP, TIES-Merging, DARE, Task Arithmetic, linear merging, and production deployment strategies.

Train AI models to be harmless through self-critique and AI feedback using a set of constitutional principles, without requiring human labels for harmful outputs.

Fine-tune Gemma models locally using QLoRA, Unsloth, or TRL for SFT, DPO, and reward modeling, with dataset preparation and conversion to GGUF or LiteRT-LM.

Run ML workloads on Modal's serverless GPU cloud: deploy models as auto-scaling APIs, run batch jobs, and schedule tasks with pay-per-second GPU pricing.

Benchmarks automatic post-editing (APE) models on WMT'18 SMT, SubEdits, and MLQE-PE datasets, reporting BLEU, ChrF, and TER scores computed with SacreBLEU and TERCOM.

Manage Hugging Face Hub resources: download/upload models, datasets, spaces; manage repos, buckets, collections, discussions, and cache; run SQL queries on datasets; authenticate and manage tokens.

Trains, evaluates, exports, and runs inference for NVPanoptix3D models that perform panoptic 3D scene reconstruction from posed RGB images, producing 3D panoptic segmentation with occupancy completion.

Provides an adaptive methodology for adding machine learning models to existing codebases, covering problem framing, data readiness, architectural decoupling, and baseline model integration.

Quantize large language models to 8/4/3/2/1-bit precision without calibration data, using multiple optimized backends for deployment with vLLM or HuggingFace Transformers.

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

Loads and processes whole-slide pathology images, builds spatial graphs, trains deep learning models, and analyzes multiplexed immunofluorescence data across 160+ slide formats.

Evaluates speaker verification models by computing true positive rate at fixed false positive rate thresholds, probing embedding space separation of same-speaker versus different-speaker pairs.

Benchmarks large language models on symbolic music understanding and instruction following using text-based ABC notation, covering syntax parsing, error detection, segment-level reasoning, and sequence-level musical analysis.

Trains, evaluates, exports, prunes, quantizes, retrains, and runs inference for TAO OCRNet models for scene text recognition from cropped text-region images, supporting CTC and attention-based decoders.

Fine-tune and deploy CLIP vision-language models for zero-shot classification, image-text retrieval, and embedding extraction with ONNX and TensorRT support.

Train, evaluate, export, and run inference for monocular depth estimation models using Metric Depth Anything v2 or Relative Depth Anything architectures via the TAO toolkit.

Train, evaluate, export, and run inference on TAO action-recognition models for classifying temporal actions in video clips using RGB, optical flow, or joint input.

Provides practical guidance for training Mixture-of-Experts Vision-Language Models in Megatron Bridge, comparing FSDP and 3D-parallel approaches with lessons from recent multimodal experiments.

Build clinical deep-learning pipelines with PyHealth: load EHR, signal, and imaging datasets, define prediction tasks, instantiate models, train with the PyHealth Trainer, and compute clinical metrics.

Detects AI-generated deepfake audio used in voice phishing (vishing) attacks by extracting spectral features and classifying samples with machine learning models.

Quantifies the minimum accuracy gap needed between two models for a sampled micro-benchmark to reliably preserve their ranking, using the MDAD metric from Yauney et al. (2025).

Benchmarks large audio-language models against adversarial audio attacks using the CAA dataset, computing WER, ROUGE-L, cosine similarity, and coherence scores to assess robustness in conversational settings.

Trains and fine-tunes vision models for object detection, image classification, and segmentation using Hugging Face Transformers on cloud GPUs, with automatic dataset validation and Hub persistence.

Trains, evaluates, exports, prunes, quantizes, retrains, and runs inference for OCDNet scene text detection models using TAO, detecting arbitrary-oriented text regions in natural images.

Trains, evaluates, and runs inference for Mask Auto-Label (MAL) weakly-supervised segmentation models using ViT-MAE backbones with minimal point or box annotations.

Generate high-quality sentence and text embeddings for semantic similarity, clustering, and retrieval using 5000+ pre-trained models. Supports multilingual and domain-specific embeddings for RAG and semantic search.

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.

Manage Hugging Face Hub resources via the `hf` CLI: download and upload models, datasets, and spaces; manage buckets, cache, collections, discussions, and inference endpoints; run SQL queries on datasets.

Benchmarks long-context language models across 20 sub-tasks spanning 3k–200k tokens, covering retrieval, reasoning, summarization, and instruction understanding, with exact-match accuracy as the primary metric.