-
Notifications
You must be signed in to change notification settings - Fork 119
Expand file tree
/
Copy pathsimple.yaml
More file actions
55 lines (53 loc) · 2.09 KB
/
Copy pathsimple.yaml
File metadata and controls
55 lines (53 loc) · 2.09 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
# Deterministic agent-team baseline: keep five recent tool results and never
# call an LLM to summarize history.
llm:
provider: ${OPENAI_PROVIDER:-openai}
model: ${OPENAI_MODEL}
base_url: ${OPENAI_BASE_URL}
api_key: ${OPENAI_API_KEY}
# Not 0.0: at temperature 0 sglang takes the argmax path, which turns a
# reasoning runaway into a deterministic deadlock the loop cannot leave (a
# runaway context resampled 4x at temp=0 returned byte-identical output).
# See workflows/agent_team/profiles/benchmark.yaml for the full measurement.
# Values are the model's generation_config.json defaults. Moving temperature
# is the load-bearing part — top_p/top_k are inert while it is 0.
temperature: 1.0
top_p: 0.95
extra_body:
top_k: 20
max_tokens: ${OPENAI_MAX_TOKENS:-32768}
report_llm:
provider: ${OPENAI_PROVIDER:-openai}
model: ${OPENAI_MODEL}
base_url: ${OPENAI_BASE_URL}
api_key: ${OPENAI_API_KEY}
# Deliberately 0.0: one tool-free synthesis call, no loop to deadlock in.
temperature: 0.0
max_tokens: ${OPENAI_MAX_TOKENS:-32768}
llm_timeout_s: 600
agent:
reporter_backend: fast
main_max_turns: 200
sub_max_turns: 100
tool_timeout_s: 360
llm_timeout_s: 600
reasoning_only_timeout_s: 120
reasoning_only_max_tokens: 16384
logical_call_timeout_s: 900
thinking_in_history: true
# `off` only blanks older tool-result bodies, keeping the latest five. It
# never summarizes user/assistant history, so very long runs can still fill
# the context window.
context_compaction: "off"
main_keep_tool_result: 5
sub_keep_tool_result: 5
compaction_spill: false
max_input_tokens: ${OPENAI_MAX_INPUT_TOKENS:-229376}
team_effort: max
planning_mode: false
planning_max_turns: 40
fresh_execution_context: false
web_search_impl: original
web_fetch_impl: aligned
main_agent_tools: [create_subagent, assign_task, collect_reports, stop_subagent, add_task, update_task, finish_planning, grep_search, glob_search, web_search]
sub_agent_tools: [web_search, web_fetch, submit_report, bash, download_file, grep_search, glob_search, read_file, create_file, recover_result]