harn-vm 0.10.94

Async bytecode virtual machine for the Harn programming language
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

//! LLM usage/tracing snapshots and delta accounting.

use crate::orchestration::LlmUsageRecord;

#[derive(Clone, Debug, Default, serde::Serialize, serde::Deserialize)]
pub(super) struct UsageSnapshot {
    pub(super) input_tokens: i64,
    pub(super) output_tokens: i64,
    pub(super) total_duration_ms: i64,
    pub(super) call_count: i64,
    pub(super) total_cost: f64,
    pub(super) unpriced_calls: i64,
    pub(super) usage_unknown_calls: i64,
    pub(super) trace_len: usize,
}

pub(super) fn llm_usage_snapshot() -> UsageSnapshot {
    let (input_tokens, output_tokens, total_duration_ms, call_count) =
        crate::llm::peek_trace_summary();
    let trace = crate::llm::peek_trace();
    let certainty =
        crate::llm::usage::summarize_usage_cost_certainty(trace.iter().map(|entry| &entry.usage));
    let trace_len = trace.len();
    UsageSnapshot {
        input_tokens,
        output_tokens,
        total_duration_ms,
        call_count,
        total_cost: certainty.known_cost_usd,
        unpriced_calls: certainty.unpriced_calls,
        usage_unknown_calls: certainty.usage_unknown_calls,
        trace_len,
    }
}

pub(super) fn llm_usage_delta(before: &UsageSnapshot, after: &UsageSnapshot) -> LlmUsageRecord {
    let trace = crate::llm::peek_trace();
    let start = before.trace_len.min(trace.len());
    let models = trace[start..]
        .iter()
        .map(|entry| entry.model.clone())
        .filter(|model| !model.is_empty())
        .fold(Vec::<String>::new(), |mut acc, model| {
            if !acc.iter().any(|existing| existing == &model) {
                acc.push(model);
            }
            acc
        });

    let known_cost_usd = (after.total_cost - before.total_cost).max(0.0);
    let unpriced_calls = after.unpriced_calls.saturating_sub(before.unpriced_calls);
    LlmUsageRecord {
        input_tokens: after.input_tokens.saturating_sub(before.input_tokens),
        output_tokens: after.output_tokens.saturating_sub(before.output_tokens),
        total_duration_ms: after
            .total_duration_ms
            .saturating_sub(before.total_duration_ms),
        call_count: after.call_count.saturating_sub(before.call_count),
        unpriced_calls,
        usage_unknown_calls: after
            .usage_unknown_calls
            .saturating_sub(before.usage_unknown_calls),
        cost_usd: (unpriced_calls == 0).then_some(known_cost_usd),
        known_cost_usd,
        total_cost: known_cost_usd,
        models,
    }
}