Hugging Face ML Toolkit

Runs reproducible model evaluations against Hugging Face Hub models on local hardware using inspect-ai or lighteval, with guidance on choosing vLLM, Transformers, or accelerate backends.

Explores and extracts Hugging Face datasets through the read-only Dataset Viewer API - splits, paginated rows, search, filter, parquet links, and stats - plus zero/low-dependency upload flows.

Builds interactive Python web UIs and ML demos with Gradio - Interface, Blocks, ChatInterface, event listeners, layouts, and streaming - with accurate component signatures.

Fine-tunes language models (SFT, DPO, GRPO, reward modeling) on managed Hugging Face Jobs cloud GPUs using TRL or Unsloth, with dataset prep, cost estimation, Trackio monitoring, and GGUF export.

Finds llama.cpp-compatible GGUF models on the Hugging Face Hub, picks the right quantization, and runs them locally with llama-cli or an OpenAI-compatible llama-server on CPU, Metal, CUDA, or ROCm.

Publishes and manages research papers on the Hugging Face Hub - indexing from arXiv, claiming authorship, linking papers to models/datasets, and generating polished markdown research articles.

Looks up and reads Hugging Face / arXiv paper pages as markdown and queries the papers API for structured metadata (authors, linked models/datasets/spaces, GitHub repo) to summarize AI research.

Creates, deploys, debugs, and maintains apps on Hugging Face Spaces across Gradio, Docker, and Static SDKs - choosing hardware tiers (cpu-basic, ZeroGPU, dedicated GPU), model sourcing, and grants.

Builds reusable command-line scripts and utilities around the Hugging Face API and `hf` CLI - composable, pipeable tools that fetch, enrich, and process Hub data with proper auth.

Adds experiment tracking to ML training with Trackio - log metrics via the Python API, fire structured alerts on diagnostic conditions, and query runs/metrics/alerts via CLI with JSON output for agents.

Fine-tunes vision models - object detection (D-FINE, RT-DETR, DETR, YOLOS), image classification (timm/Transformers), and SAM/SAM2 segmentation - on Hugging Face Jobs GPUs with dataset validation and mAP/accuracy metrics.

Writes and debugs Gradio Spaces code for Hugging Face ZeroGPU - @spaces.GPU decorators, duration/quota tuning, process-isolation pitfalls, the CUDA availability model, and CUDA-wheel build constraints.

Trains sentence-transformers models - bi-encoders, CrossEncoder rerankers, and SPLADE sparse encoders - covering loss selection, hard-negative mining, evaluators, LoRA, Matryoshka, and Hub publishing.

Post-trains transformer language models with the TRL CLI - SFT, DPO, GRPO, KTO, RLOO, and reward modeling - covering dataset preparation, configuration, and the right command per method.