Guides

Task-oriented how-tos for using ColabHive: call models, import your own, train, and serve.

---

Use a model

• Import a Model from HuggingFace — search, check compatibility, and register any compatible HF model (in addition to the curated base catalog).
• Choosing a Model — curated catalog vs. import, by task and size.
• Tool Calling (Function Calling) — OpenAI-compatible tools / tool_calls.
• Generative Models — image, audio, and video generation.

Train and iterate

• Preparing Datasets — formats, examples, and data-volume guidance.
• LLM Fine-Tuning (QLoRA / LoRA) — transformers + PEFT + bitsandbytes.
• Fine-tune a CLM head — train a domain head for the CLM-8B scorer.
• Hyperparameter Tuning — per-model parameter reference.
• Merge and Retrain — combine and build on top of models (the flywheel).

Serve

• Register a Trained Model for Inference — turn a training run into an endpoint (account-private or public).
• Inference Lifecycle — warm/cold paths, model states, VRAM management.

Integrate

• Coding Agents — run OpenCode or another coding agent against ColabHive: colabhive agents init configures it in one command.
• Use ColabHive with SuperClaw / OpenCode — ColabHive as the cloud model tier.
• Plugin & Workflow Integration — the Actions API from n8n, agents, MCP.

More

• Model Selection — choose a compatible configuration from the live catalog.
• Troubleshooting
• Quick Start (Inference)