show peak memory usage (#14485)

This commit is contained in:
Patrick Devine
2026-02-26 18:38:27 -08:00
committed by GitHub
parent cc90a035a0
commit 79917cf80b
7 changed files with 26 additions and 0 deletions

View File

@@ -557,6 +557,7 @@ func (s *Server) GenerateHandler(c *gin.Context) {
PromptEvalDuration: cr.PromptEvalDuration,
EvalCount: cr.EvalCount,
EvalDuration: cr.EvalDuration,
PeakMemory: cr.PeakMemory,
},
Logprobs: toAPILogprobs(cr.Logprobs),
}
@@ -2309,6 +2310,7 @@ func (s *Server) ChatHandler(c *gin.Context) {
PromptEvalDuration: r.PromptEvalDuration,
EvalCount: r.EvalCount,
EvalDuration: r.EvalDuration,
PeakMemory: r.PeakMemory,
},
Logprobs: toAPILogprobs(r.Logprobs),
}