Medhelp
Community@MedTiLab
I am a clinical doctor
Agent Skills by Medhelp
Showing 161 vetted skills indexed across 1 GitHub repositories.
MedTiLabsentence-transformers
Generate sentence, text, and image embeddings for semantic search and RAG.
MedTiLabpeft-fine-tuning
Fine-tune large language models with LoRA, QLoRA, and adapter methods using HuggingFace PEFT.
MedTiLabaxolotl
Configure and fine-tune LLMs with Axolotl YAML configs, LoRA, DPO, and multimodal training.
MedTiLabunsloth
Guides LLM fine-tuning with Unsloth using LoRA, QLoRA, and GRPO reinforcement learning.
MedTiLabnnsight-remote-interpretability
Interpret and manipulate PyTorch model internals locally or remotely via NDIF.
MedTiLabtransformer-lens-interpretability
Inspect and manipulate transformer internals using HookPoints and activation caching.
MedTiLabpyvene-interventions
Perform causal interventions on PyTorch models using pyvene's declarative intervention framework.
MedTiLabsparse-autoencoder-training
Train and analyze Sparse Autoencoders with SAELens to decompose neural network activations into interpretable features.
MedTiLabaudiocraft-audio-generation
Generate music and sound effects from text prompts using AudioCraft models.
MedTiLabclip
Performs zero-shot image classification and image-text retrieval using CLIP embeddings.
MedTiLabstable-diffusion-image-generation
Generate images from text prompts using Stable Diffusion models via HuggingFace Diffusers.
MedTiLabwhisper
Transcribe and translate multilingual audio into text using OpenAI Whisper models.
MedTiLabopenrlhf-training
Train large language models with distributed PPO, GRPO, RLOO, and DPO using Ray and vLLM.
MedTiLabsimpo-training
Trains LLMs with reference-free SimPO preference optimization using preference pair datasets.
MedTiLabslime-rl-training
Guides LLM post-training with reinforcement learning using Megatron-LM and SGLang.
MedTiLabfine-tuning-with-trl
Fine-tune LLMs with TRL using SFT, DPO, PPO, GRPO, and reward model training.
MedTiLabmiles-rl-training
Configures RL post-training for large MoE models with FP8, INT4, and speculative decoding.
MedTiLabtorchforge-rl-training
Implements PyTorch-native agentic RL training with GRPO, Monarch, and TorchTitan.
MedTiLabverl-rl-training
Implements reinforcement learning training for LLMs using verl with PPO, GRPO, and distributed backends.
MedTiLabevaluating-llms-harness
Evaluates LLMs across 60+ academic benchmarks using standardized prompts and metrics.
MedTiLabevaluating-code-models
Benchmarks code generation models on HumanEval, MBPP, and MultiPL-E with pass@k metrics.
MedTiLabimplementing-llms-litgpt
Implements and fine-tunes LLMs using LitGPT with LoRA, QLoRA, and FSDP training workflows.
MedTiLabmamba-architecture
Implements and benchmarks Mamba selective state-space models for linear-complexity sequence modeling.
MedTiLabnanogpt
Train and fine-tune minimalist GPT models in PyTorch from scratch.
Frequently Asked Questions About Medhelp
FAQPage SchemaWhat tasks can I accomplish with Medhelp's skills?ā¼
You can execute full medical systematic review and meta-analysis pipelines: literature search across PubMed/PMC/bioRxiv/OpenAlex, PDF parsing with MinerU, staged screening, effect-size extraction (HR/OR/RR/SMD/sensitivity/specificity), R-based diagnostic accuracy statistics, PRISMA manuscript writing, plus grant proposals, peer review, and Nature-style figure and citation preparation.
Who are Medhelp's skills designed for?ā¼
They target clinical researchers, epidemiologists, and academic authors. The manifest is built by a clinical doctor around MedHelp Meta projects, with skills for tumor biomarker meta-analysis, baseline Table 1 generation, clinical pre-analysis, NSF/NIH/NSFC grant writing, and journal submission workflows for medical and translational research teams.
How do the meta-analysis skills run in practice?ā¼
Start with meta-pipeline-planner or meta-analysis-workflow to create the research brief and task queue, then use pubmed-search-strategy, staged screening, legal PDF acquisition, MinerU parsing, meta-extraction, and meta-statistics-r. Skills use Read/Write/Bash/Glob/Grep permissions and enforce a standard Meta project directory contract.
Are Medhelp's skills open source and what do they cost?ā¼
Most skills carry permissive licenses (MIT, Apache-2.0, BSD-3-Clause); some use GPL-3.0 (scikit-survival, denario), and the docx/xlsx/pdf/pptx document skills are proprietary. The skills themselves are free; costs only arise from external services you configure, such as Gemini or OpenRouter API keys.
What prerequisites and dependencies do the skills require?ā¼
Deep research skills need API keys such as GEMINI_API_KEY or OpenRouter access. Statistical skills require R or Python libraries (scikit-survival, PyMC, statsmodels, polars). Local database skills expect datasets under $HOME/database (NHANES, HRS, MIMIC-IV), and Zotero integration uses the local Zotero API for full-text handoff.