You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Challenge: Deep Research settings that actually finish on small, medium and large local systems
#7289
🔬 Challenge: Deep Research settings that actually finish on small, medium and large local systems
ROADMAP: "Deep Research presets: explore suitable model and parameter configurations for small, medium, and large local systems." Related reports: #4620 (times out frequently), #7012 (reasoning models fail before any search; the engine's max time ignores research_run_timeout_seconds; helper budgets hard-coded), #331 (configurable timeouts), with fixes in flight in #7013 and #6987.
Before anyone proposes presets, it would help to know which combinations finish on real hardware. So: a scoreboard.
Scoreboard
Completed reports out of 5 pinned questions, plus median wall time. Search results vary, so completion first, speed second. Report quality is not scored.
Pinned questions (boring on purpose, stable public sources):
What is the difference between GQA and MQA attention?
Summarise the licensing differences between SQLite and PostgreSQL.
What changed between HTTP/2 and HTTP/3?
How does the Raft consensus algorithm handle leader election?
What were the main causes of the 1970s oil crises?
| tier (8 / 12-16 / 24+ GB / CPU) | backend + build | model/quant | search provider | settings changed from default | completed /5 (A, B) | median wall min | step that stalled, if any | Odysseus commit |
Baseline: the defaults on your machine, run twice.
If a tier converges on settings that clearly work, that becomes a Feature Request for named presets with this data attached. Presets change defaults, so they'd go through the normal product process.
Drafted with the help of an AI agent; settings checked against dev@0096533c.
reacted with thumbs up emoji reacted with thumbs down emoji reacted with laugh emoji reacted with hooray emoji reacted with confused emoji reacted with heart emoji reacted with rocket emoji reacted with eyes emoji
Uh oh!
There was an error while loading. Please reload this page.
🔬 Challenge: Deep Research settings that actually finish on small, medium and large local systems
ROADMAP: "Deep Research presets: explore suitable model and parameter configurations for small, medium, and large local systems." Related reports: #4620 (times out frequently), #7012 (reasoning models fail before any search; the engine's max time ignores
research_run_timeout_seconds; helper budgets hard-coded), #331 (configurable timeouts), with fixes in flight in #7013 and #6987.The knobs already exist in
src/settings.py:research_max_tokensresearch_planning_timeout_seconds/research_query_timeout_seconds/research_extraction_timeout_secondsresearch_extraction_concurrencyresearch_run_timeout_secondsresearch_model,research_search_providerBefore anyone proposes presets, it would help to know which combinations finish on real hardware. So: a scoreboard.
Scoreboard
Completed reports out of 5 pinned questions, plus median wall time. Search results vary, so completion first, speed second. Report quality is not scored.
Pinned questions (boring on purpose, stable public sources):
If a tier converges on settings that clearly work, that becomes a Feature Request for named presets with this data attached. Presets change defaults, so they'd go through the normal product process.
Drafted with the help of an AI agent; settings checked against
dev@0096533c.All reactions