Research: local Ollama deployment, harness selection, and model sizing #183

Merged
didericis merged 1 commits from research-local-ollama-harness into main 2026-06-03 21:37:56 -04:00
Collaborator

Summary

  • Covers native vs Docker vs LXC vs KVM/passthrough overhead for Ollama (TL;DR: LXC <1%, Docker ~1–2%, KVM passthrough 2–5% — none significant; CPU fallback without passthrough is the real cost)
  • Evaluates harnesses for agentic coding loops: OpenCode (best bot-bottle fit), Aider (git-native terminal), Continue.dev (IDE plugin + FIM autocomplete), Cline (IDE agent)
  • Model recommendations for RTX 3070 8 GB VRAM / 30 GB RAM: qwen2.5-coder:7b as primary (full VRAM, FIM, 35–55 tok/sec), qwen2.5-coder:14b as secondary via RAM offload, gemma4:e4b as general-purpose alternative
  • Confirms Gemma 4 exists (released 2026-04-02, Apache 2.0) but only the E4B fits in 8 GB

Doc: 8f05226a4a/docs/research/local-ollama-harness-and-model-selection.md

## Summary - Covers native vs Docker vs LXC vs KVM/passthrough overhead for Ollama (TL;DR: LXC <1%, Docker ~1–2%, KVM passthrough 2–5% — none significant; CPU fallback without passthrough is the real cost) - Evaluates harnesses for agentic coding loops: OpenCode (best bot-bottle fit), Aider (git-native terminal), Continue.dev (IDE plugin + FIM autocomplete), Cline (IDE agent) - Model recommendations for RTX 3070 8 GB VRAM / 30 GB RAM: `qwen2.5-coder:7b` as primary (full VRAM, FIM, 35–55 tok/sec), `qwen2.5-coder:14b` as secondary via RAM offload, `gemma4:e4b` as general-purpose alternative - Confirms Gemma 4 exists (released 2026-04-02, Apache 2.0) but only the E4B fits in 8 GB Doc: https://gitea.dideric.is/didericis/bot-bottle/src/commit/8f05226a4aafbd5b30a4df7e87e084d889a9d4e1/docs/research/local-ollama-harness-and-model-selection.md
didericis-claude added 1 commit 2026-06-03 21:26:37 -04:00
docs(research): local ollama deployment, harness selection, and model sizing
test / unit (pull_request) Successful in 38s
test / integration (pull_request) Successful in 51s
8f05226a4a
didericis merged commit d1556f4659 into main 2026-06-03 21:37:56 -04:00
didericis deleted branch research-local-ollama-harness 2026-06-03 21:37:56 -04:00
Sign in to join this conversation.