Generation parameters configuration (top-level defaults). More...

#include <entropic/types/config.h>

Public Attributes
int	max_tokens = 4096
	Default max tokens (64–32768)

float	default_temperature = 0.7f
	Default temperature (0.0–2.0)

float	default_top_p = 0.9f
	Default top_p (0.0–1.0)

std::string	budget_mode = "off"
	gh#80 (v2.5.0) thinking-budget mode: "off" (default), "tokens", or "wall_clock".

int	budget_limit = 0
	gh#80 (v2.5.0) budget ceiling: generated tokens (budget_mode "tokens") or wall-clock seconds (budget_mode "wall_clock") of tool-call-free generation before the engine nudges-then-hard-cuts.

bool	stream_output = true
	gh#110 (v2.9.6) agent-loop token delivery mode: true streams tokens via the per-token callback path, false batches the full response.

Detailed Description

Generation parameters configuration (top-level defaults).

Version: 1.8.1

Definition at line 757 of file config.h.

Member Data Documentation

◆ budget_limit

int entropic::GenerationConfig::budget_limit = 0

gh#80 (v2.5.0) budget ceiling: generated tokens (budget_mode "tokens") or wall-clock seconds (budget_mode "wall_clock") of tool-call-free generation before the engine nudges-then-hard-cuts.

Must be > 0 to engage.

Version: 2.5.0

Definition at line 775 of file config.h.

◆ budget_mode

std::string entropic::GenerationConfig::budget_mode = "off"

gh#80 (v2.5.0) thinking-budget mode: "off" (default), "tokens", or "wall_clock".

Stored as a string here to keep types/config.h free of the core engine_types BudgetMode enum; the facade's build_loop_config maps it. Unknown values resolve to "off" with a warning.

Version: 2.5.0

Definition at line 768 of file config.h.

◆ default_temperature

float entropic::GenerationConfig::default_temperature = 0.7f

Default temperature (0.0–2.0)

Definition at line 759 of file config.h.

◆ default_top_p

float entropic::GenerationConfig::default_top_p = 0.9f

Default top_p (0.0–1.0)

Definition at line 760 of file config.h.

◆ max_tokens

int entropic::GenerationConfig::max_tokens = 4096

Default max tokens (64–32768)

Definition at line 758 of file config.h.

◆ stream_output

bool entropic::GenerationConfig::stream_output = true

gh#110 (v2.9.6) agent-loop token delivery mode: true streams tokens via the per-token callback path, false batches the full response.

Default true preserves existing behavior. MTP/speculative decoding rejects streaming calls (the thinking-channel strip is a post-buffer operation) — set this false to make the agent loop reachable by inference.speculative.mtp.

Version: 2.9.6

Definition at line 785 of file config.h.

The documentation for this struct was generated from the following file:

entropic/types/config.h

Public Attributes