Skip to content

fix: Use eval_duration for output TPS calculations in Ollama LLM provider #4568

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 6 commits into
base: master
Choose a base branch
from
Open

fix: Use eval_duration for output TPS calculations in Ollama LLM provider #4568

Show file tree
Hide file tree
Changes from 4 commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
ca905f9
fix: Use eval_duration for output TPS calculations and add as a metri…
jonathanortega2023 Oct 21, 2025
5a62844
refactor usage of eval_duration from ollama metrics
shatfield4 Oct 22, 2025
a233373
Merge branch 'master' into pr-4568-branch
shatfield4 Oct 22, 2025
809e5ac
move eval_duration to usage
shatfield4 Oct 22, 2025
7831b16
overwrite duration in ollama provider wip measureAsyncFunction option…
shatfield4 Oct 25, 2025
b33c84d
allow for overloaded duration in measureAsyncFunction
shatfield4 Oct 27, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
7 changes: 6 additions & 1 deletion server/utils/AiProviders/ollama/index.js
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -263,6 +263,7 @@ class OllamaAILLM {
prompt_tokens: res.prompt_eval_count,
completion_tokens: res.eval_count,
total_tokens: res.prompt_eval_count + res.eval_count,
eval_duration: res.eval_duration / 1e9,
},
};
})
Expand All @@ -282,7 +283,10 @@ class OllamaAILLM {
prompt_tokens: result.output.usage.prompt_tokens,
completion_tokens: result.output.usage.completion_tokens,
total_tokens: result.output.usage.total_tokens,
outputTps: result.output.usage.completion_tokens / result.duration,
outputTps:
result.output.usage.completion_tokens /
result.output.usage.eval_duration,
eval_duration: result.output.usage.eval_duration,
duration: result.duration,
},
};
Expand Down Expand Up @@ -349,6 +353,7 @@ class OllamaAILLM {
if (chunk.done) {
usage.prompt_tokens = chunk.prompt_eval_count;
usage.completion_tokens = chunk.eval_count;
usage.eval_duration = chunk.eval_duration / 1e9;
writeResponseChunk(response, {
uuid,
sources,
Expand Down
8 changes: 7 additions & 1 deletion server/utils/helpers/chat/LLMPerformanceMonitor.js
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -13,6 +13,7 @@ const { TokenManager } = require("../tiktoken");
* @property {number} total_tokens - the total number of tokens
* @property {number} outputTps - the tokens per second of the output
* @property {number} duration - the duration of the stream
* @property {number} [eval_duration] - optional eval duration from providers (e.g., Ollama) used for more accurate outputTps calculation
*/

/**
Expand Down Expand Up @@ -88,7 +89,12 @@ class LLMPerformanceMonitor {

stream.metrics.total_tokens =
stream.metrics.prompt_tokens + (stream.metrics.completion_tokens || 0);
stream.metrics.outputTps = stream.metrics.completion_tokens / duration;

// Use eval_duration if provided (for providers like Ollama that report it)
// otherwise fall back to total request duration
stream.metrics.outputTps =
stream.metrics.completion_tokens /
(stream.metrics.eval_duration || duration);
stream.metrics.duration = duration;
return stream.metrics;
};
Expand Down