rho-coding-agent 1.40.0

A lightweight agent harness inspired by Pi
Documentation 
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
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127

use std::{collections::BTreeMap, time::Duration};

use rho_sdk::{ModelCallMetrics, ModelCallProfile};

const MIN_GENERATION_OUTPUT_TOKENS: u64 = 32;
const MIN_GENERATION_TIME: Duration = Duration::from_millis(500);

#[derive(Clone, Copy, Debug, PartialEq, Eq)]
pub(super) enum GenerationOutputTokens {
    AggregateFallback,
    Reported(u64),
    Unavailable,
}

impl GenerationOutputTokens {
    fn resolve(self, aggregate_output_tokens: Option<u64>) -> Option<u64> {
        match self {
            Self::AggregateFallback => aggregate_output_tokens,
            Self::Reported(tokens) => Some(tokens),
            Self::Unavailable => None,
        }
    }
}

#[derive(Clone, Copy, Debug, PartialEq)]
pub(super) struct ModelCallPerformance {
    pub(super) metrics: ModelCallMetrics,
    pub(super) generation_output_tokens: GenerationOutputTokens,
}

impl ModelCallPerformance {
    pub(super) fn throughput_output_tokens(self) -> Option<u64> {
        self.generation_output_tokens
            .resolve(self.metrics.output_tokens)
    }
}

#[derive(Clone, Copy, Debug, Default, PartialEq)]
pub(super) struct ModelPerformanceSummary {
    pub(super) latest_call: Option<ModelCallPerformance>,
    /// Token-weighted average of generation throughput (`tokens / generation_time`).
    pub(super) average_generation_tokens_per_second: Option<f64>,
    pub(super) eligible_calls: u64,
}

#[derive(Default)]
pub(super) struct ModelPerformanceTracker {
    profiles: BTreeMap<ModelCallProfile, ModelPerformanceAggregate>,
}

impl ModelPerformanceTracker {
    pub(super) fn record(
        &mut self,
        profile: ModelCallProfile,
        metrics: ModelCallMetrics,
        generation_output_tokens: GenerationOutputTokens,
    ) {
        self.profiles
            .entry(profile)
            .or_default()
            .record(metrics, generation_output_tokens);
    }

    pub(super) fn summary(&self, profile: &ModelCallProfile) -> ModelPerformanceSummary {
        self.profiles
            .get(profile)
            .map(ModelPerformanceAggregate::summary)
            .unwrap_or_default()
    }

    pub(super) fn clear(&mut self) {
        self.profiles.clear();
    }
}

#[derive(Default)]
struct ModelPerformanceAggregate {
    latest_call: Option<ModelCallPerformance>,
    generation_output_tokens: u64,
    generation_time: Duration,
    eligible_calls: u64,
}

impl ModelPerformanceAggregate {
    fn record(
        &mut self,
        metrics: ModelCallMetrics,
        generation_output_tokens: GenerationOutputTokens,
    ) {
        let latest_call = ModelCallPerformance {
            metrics,
            generation_output_tokens,
        };
        let Some(generation_output_tokens) = latest_call.throughput_output_tokens() else {
            self.latest_call = Some(latest_call);
            return;
        };
        self.latest_call = Some(latest_call);
        let Some(generation_time) = metrics.generation_time else {
            return;
        };
        if generation_output_tokens < MIN_GENERATION_OUTPUT_TOKENS
            || generation_time < MIN_GENERATION_TIME
        {
            return;
        }

        self.generation_output_tokens = self
            .generation_output_tokens
            .saturating_add(generation_output_tokens);
        self.generation_time = self.generation_time.saturating_add(generation_time);
        self.eligible_calls = self.eligible_calls.saturating_add(1);
    }

    fn summary(&self) -> ModelPerformanceSummary {
        ModelPerformanceSummary {
            latest_call: self.latest_call,
            average_generation_tokens_per_second: (self.eligible_calls > 0)
                .then(|| self.generation_output_tokens as f64 / self.generation_time.as_secs_f64()),
            eligible_calls: self.eligible_calls,
        }
    }
}

#[cfg(test)]
#[path = "model_performance_tests.rs"]
mod tests;