 welberr
|
e2b1000198
|
P1–P2 complete: routing strategies, streaming, discovery, observed metrics + role catalogs
Control plane:
- fallback_roles chain in resolve_route() with cycle protection
- round_robin and least_loaded routing strategies; default_strategy dispatches all three
- Streaming chat completions: async generator, eager route resolution, SSE reasoning-strip
- POST /v1/audio/transcriptions proxy (multipart, dedicated httpx path)
- ServiceProber background task: probes /health, falls back to /v1/models for vLLM
- ServiceObserved gains loaded_model_count and vram_used_bytes
- _runtime_signals exposes loaded_model_count to route scoring
Node agent:
- discover_protocol: "ollama"|"openai"|null per-service config field
- discovery.py: discover_ollama_assets (loaded: False), _get_ollama_ps_models helper,
query_ollama_ps, discover_openai_models, enrich_service_assets (two-phase Ollama,
corrects stale loaded state, populates observed metrics from /api/ps)
- Heartbeat zips service dicts with config to pass protocol; allocates discovery client
only when needed
Tests: 47 passing (up from 19)
Role catalogs (example configs):
- roles.surgical-team.example.yaml — Brooks/Mills surgical team (surg_ prefix, 9 roles)
- roles.belbin.example.yaml — Belbin team roles (belbin_ prefix, 9 roles)
- roles.sixhats.example.yaml — De Bono Six Thinking Hats (sixhats_ prefix, 6 roles)
- roles.disney.example.yaml — Disney creative strategy (disney_ prefix, 3 roles)
- roles.xp.example.yaml — XP team roles (xp_ prefix, 5 roles)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
|
2026-04-27 14:12:54 -04:00 |
|
welberr
|
a76c7e81f4
|
Revise architecture/roadmap docs and add LLM evaluation guide
- architecture.md: rewrite to describe the actual running system; remove
design-phase repo-naming discussion and initial-implementation-sequence
list; add data-flow diagram, scoring weights table, API status table
- roadmap.md: replace aspirational list with concrete completed/gap/next
structure; document four confirmed implementation gaps (transcription
stub, strategy field ignored, fallback_roles unimplemented, benchmark
quality score additive overflow); prioritise fixes as P0/P1/P2/P3
- docs/local_llm_evaluation.md: new document; role taxonomy (tier 1–3),
hardware inventory template, candidate model suggestions, three-phase
evaluation protocol, GenieHive integration steps, results template,
notes on Qwen3/Mistral/DeepSeek/Ollama embedding path quirks
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
|
2026-04-27 09:25:51 -04:00 |