Automate ML post-training research, code generation, and experiment orchestration.
npx skills add https://github.com/AlexiosBluffMara/mercury --skill ml-intern
Please help me install this Agent Skill. Skill: ml-intern Source: https://github.com/AlexiosBluffMara/mercury/tree/main/skills/self-evolution/ml-intern Command: npx skills add https://github.com/AlexiosBluffMara/mercury --skill ml-intern
ml-intern is an open-source AI agent that autonomously researches, writes, and ships ML related code using the HuggingFace ecosystem to conduct post-training experiments, literature review, dataset discovery, training scripts, and iterative evaluation. It enables automated Cortex Gemma-4-e4b fine-tuning, ablation studies, and hyperparameter sweeps without writing a bespoke harness.
Run ml-intern to start an autonomous post-training experiment workflow.
High-intent search queries and answers about installing and using this skill.
Autonomous ML post-training uses a single-agent loop to orchestrate HuggingFace fine-tuning and ablation studies automatically. It researches datasets, generates training scripts, and executes experiments on hosted compute without requiring a bespoke harness.
Running autonomous ML post-training experiments requires HF_TOKEN, ANTHROPIC_API_KEY, GITHUB_TOKEN, and OPENAI_API_KEY. These environment variables authenticate provider-agnostic backends via litellm, HuggingFace Hub access, and GitHub repository search.
Context management during long hyperparameter sweeps uses auto-compaction at 170k tokens to preserve relevant context. A doom-loop detector also monitors repeated patterns and injects corrective prompts to prevent the agent from getting stuck.
The maximum duration for an autonomous ML post-training session is capped at 300 iterations within a single-agent loop. This limit ensures controlled experiment orchestration while allowing sufficient time for literature review, training, and evaluation.
Autonomous ML post-training supports automated Cortex Gemma-4-e4b fine-tuning directly within the HuggingFace ecosystem. It handles dataset discovery, training script generation, and iterative evaluation specifically for these model architectures.
Completing an autonomous ablation study uploads a final session to the HuggingFace Hub containing the full execution trace, trained model artifacts, and generated model cards. This ensures all experimental results are reproducible and shareable.
kevinnftConfigure Hermes Agent for autonomous task management with persistent memory and multi-agent orchestration.
ScientiaCapitalExport fine-tuned models to GGUF and deploy with vLLM or Ollama.
zengbaochengConnect to MCP servers and register their tools in Hermes Agent.
Clay-HHKTrack and visualize ML experiments and artifacts with wandb.
