atman-runtime 1.5.0

atman flow execution runtime: evaluator, tool dispatch, provider dispatch, executor, memory stores
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
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185

use std::collections::HashMap;

use crate::event::{Event, LlmCallStatus};
use crate::provider::TokenUsage;

#[derive(Debug, Default, Clone)]
pub struct CostSummary {
    pub calls: u64,
    pub failures: u64,
    pub usage: TokenUsage,
    pub wallclock_ms: u64,
}

impl CostSummary {
    fn accumulate(&mut self, usage: &TokenUsage, wallclock_ms: u64, failed: bool) {
        self.calls += 1;
        if failed {
            self.failures += 1;
        }
        self.usage.input = self.usage.input.saturating_add(usage.input);
        self.usage.cached_input = self.usage.cached_input.saturating_add(usage.cached_input);
        self.usage.output = self.usage.output.saturating_add(usage.output);
        self.usage.cache_write = self.usage.cache_write.saturating_add(usage.cache_write);
        self.wallclock_ms = self.wallclock_ms.saturating_add(wallclock_ms);
    }
}

pub fn summarize_by_model(events: &[Event]) -> HashMap<String, CostSummary> {
    let mut out: HashMap<String, CostSummary> = HashMap::new();
    for e in events {
        if let Event::LlmCall {
            model,
            usage,
            wallclock_ms,
            status,
            ..
        } = e
        {
            let entry = out.entry(model.clone()).or_default();
            entry.accumulate(
                usage,
                *wallclock_ms,
                matches!(status, LlmCallStatus::Errored { .. }),
            );
        }
    }
    out
}

pub fn summarize_by_provider(events: &[Event]) -> HashMap<String, CostSummary> {
    let mut out: HashMap<String, CostSummary> = HashMap::new();
    for e in events {
        if let Event::LlmCall {
            provider,
            usage,
            wallclock_ms,
            status,
            ..
        } = e
        {
            let entry = out.entry(provider.clone()).or_default();
            entry.accumulate(
                usage,
                *wallclock_ms,
                matches!(status, LlmCallStatus::Errored { .. }),
            );
        }
    }
    out
}

pub fn total(events: &[Event]) -> CostSummary {
    let mut acc = CostSummary::default();
    for e in events {
        if let Event::LlmCall {
            usage,
            wallclock_ms,
            status,
            ..
        } = e
        {
            acc.accumulate(
                usage,
                *wallclock_ms,
                matches!(status, LlmCallStatus::Errored { .. }),
            );
        }
    }
    acc
}

#[cfg(test)]
mod tests {
    use super::*;

    fn ok_call(model: &str, provider: &str, in_tok: u64, out_tok: u64) -> Event {
        Event::LlmCall {
            model: model.into(),
            provider: provider.into(),
            usage: TokenUsage {
                input: in_tok,
                output: out_tok,
                ..Default::default()
            },
            wallclock_ms: 100,
            status: LlmCallStatus::Ok,
            ttft_ms: None,
            tokens_per_second: None,
            run_id: None,
            node_id: None,
        }
    }

    fn err_call(model: &str) -> Event {
        Event::LlmCall {
            model: model.into(),
            provider: "p".into(),
            usage: TokenUsage {
                input: 5,
                ..Default::default()
            },
            wallclock_ms: 50,
            status: LlmCallStatus::Errored {
                message: "boom".into(),
            },
            ttft_ms: None,
            tokens_per_second: None,
            run_id: None,
            node_id: None,
        }
    }

    #[test]
    fn summarize_by_model_groups_multiple_calls() {
        let events = vec![
            ok_call("opus", "anthropic", 10, 20),
            ok_call("opus", "anthropic", 5, 8),
            ok_call("mini", "openai", 3, 4),
        ];
        let s = summarize_by_model(&events);
        assert_eq!(s.get("opus").unwrap().calls, 2);
        assert_eq!(s.get("opus").unwrap().usage.input, 15);
        assert_eq!(s.get("opus").unwrap().usage.output, 28);
        assert_eq!(s.get("mini").unwrap().calls, 1);
    }

    #[test]
    fn failures_counted_separately() {
        let events = vec![ok_call("m", "p", 1, 2), err_call("m"), err_call("m")];
        let s = summarize_by_model(&events);
        assert_eq!(s.get("m").unwrap().calls, 3);
        assert_eq!(s.get("m").unwrap().failures, 2);
    }

    #[test]
    fn total_sums_wallclock() {
        let events = vec![ok_call("a", "p", 1, 1), ok_call("b", "p", 2, 2)];
        let t = total(&events);
        assert_eq!(t.calls, 2);
        assert_eq!(t.wallclock_ms, 200);
    }

    #[test]
    fn non_llm_events_ignored() {
        use crate::event::{FlowRunId, FlowStatus};
        let run_id = FlowRunId::now();
        let events = vec![
            Event::FlowStart {
                run_id: run_id.clone(),
                flow_name: "t".into(),
                parent_run_id: None,
                parent_node_id: None,
            },
            ok_call("m", "p", 5, 5),
            Event::FlowEnd {
                run_id,
                flow_name: "t".into(),
                status: FlowStatus::Ok,
            },
        ];
        let t = total(&events);
        assert_eq!(t.calls, 1);
        assert_eq!(t.usage.input, 5);
    }
}