|
Entropic 2.14.3
Local-first agentic inference engine
|
One generation's metrics, without its text (gh#154). More...
#include <entropic/types/generation_result.h>

Public Attributes | |
| std::string | finish_reason |
| "stop", "length", "error" | |
| int | token_count = 0 |
| Tokens generated. | |
| int | prefill_tokens = 0 |
| Prompt tokens actually decoded (gh#144) | |
| double | throughput_tok_s = 0.0 |
| Measured decode throughput. | |
| int | n_drafted = 0 |
| Speculative/MTP tokens proposed. | |
| int | n_accepted = 0 |
| Of those, accepted by the target. | |
| GrammarProvenance | grammar |
| What constrained the decode. | |
One generation's metrics, without its text (gh#154).
GenerationResult is rich and per-call; this is what survives the call and reaches a consumer through entropic_metrics_json. Content and tool calls are deliberately excluded — the engine keeps a bounded ring of these, and keeping the text would make it a transcript.
Definition at line 156 of file generation_result.h.
| std::string entropic::GenerationRecord::finish_reason |
"stop", "length", "error"
Definition at line 157 of file generation_result.h.
| GrammarProvenance entropic::GenerationRecord::grammar |
What constrained the decode.
Definition at line 163 of file generation_result.h.
| int entropic::GenerationRecord::n_accepted = 0 |
Of those, accepted by the target.
Definition at line 162 of file generation_result.h.
| int entropic::GenerationRecord::n_drafted = 0 |
Speculative/MTP tokens proposed.
Definition at line 161 of file generation_result.h.
| int entropic::GenerationRecord::prefill_tokens = 0 |
Prompt tokens actually decoded (gh#144)
Definition at line 159 of file generation_result.h.
| double entropic::GenerationRecord::throughput_tok_s = 0.0 |
Measured decode throughput.
Definition at line 160 of file generation_result.h.
| int entropic::GenerationRecord::token_count = 0 |
Tokens generated.
Definition at line 158 of file generation_result.h.