mirror of
https://github.com/BerriAI/litellm.git
synced 2026-11-18 09:45:23 +00:00
test(cost): pin allocation-free pricing lookups and overflow strictness
A counting-allocator test asserts ModelPricing lookups on the request path allocate nothing, keeping the Copy-able parsed-struct invariant as the Value-based modules migrate over. The anthropic iteration sums error on u64 overflow and compile rejects negative rates, both documented as the deliberate strictness divergence class where Python yields huge or negative numbers
This commit is contained in:
parent
9ca9795354
commit
0bdf8abf1e
3 changed files with 125 additions and 0 deletions
|
|
@ -224,6 +224,12 @@ fn compile_rejects_ambiguous_rates() {
|
|||
);
|
||||
let invalid = pricing(rates(Rate::Value(f64::NAN), Rate::Value(1.0)));
|
||||
assert_eq!(compile(&invalid).err(), Some(PricingError::InvalidRate));
|
||||
let negative = pricing(rates(Rate::Value(-1.0), Rate::Value(1.0)));
|
||||
assert_eq!(
|
||||
compile(&negative).err(),
|
||||
Some(PricingError::InvalidRate),
|
||||
"divergence: Python bills negative rates as negative costs; Rust refuses them"
|
||||
);
|
||||
}
|
||||
|
||||
#[test]
|
||||
|
|
|
|||
97
litellm-rust/crates/cost/tests/pricing_allocations.rs
Normal file
97
litellm-rust/crates/cost/tests/pricing_allocations.rs
Normal file
|
|
@ -0,0 +1,97 @@
|
|||
// mirrors: crates/cost/tests/generate_python_fixtures.py (catalog embed invariants)
|
||||
|
||||
#![allow(clippy::disallowed_types)]
|
||||
|
||||
use std::alloc::{GlobalAlloc, Layout, System};
|
||||
use std::sync::atomic::{AtomicUsize, Ordering};
|
||||
|
||||
use litellm_cost::pricing::{Metric, ModelPricing, ServiceTier};
|
||||
use litellm_cost::wire::RawCatalogEntry;
|
||||
|
||||
static ALLOCATIONS: AtomicUsize = AtomicUsize::new(0);
|
||||
|
||||
struct CountingAllocator;
|
||||
|
||||
unsafe impl GlobalAlloc for CountingAllocator {
|
||||
unsafe fn alloc(&self, layout: Layout) -> *mut u8 {
|
||||
ALLOCATIONS.fetch_add(1, Ordering::Relaxed);
|
||||
unsafe { System.alloc(layout) }
|
||||
}
|
||||
|
||||
unsafe fn dealloc(&self, ptr: *mut u8, layout: Layout) {
|
||||
unsafe { System.dealloc(ptr, layout) }
|
||||
}
|
||||
}
|
||||
|
||||
#[global_allocator]
|
||||
static ALLOCATOR: CountingAllocator = CountingAllocator;
|
||||
|
||||
fn parsed_pricing() -> ModelPricing {
|
||||
let entry: RawCatalogEntry = serde_json::from_value(serde_json::json!({
|
||||
"input_cost_per_token": 3e-6,
|
||||
"output_cost_per_token": 15e-6,
|
||||
"cache_read_input_token_cost": 3e-7,
|
||||
"cache_creation_input_token_cost_above_1hr": 4e-6,
|
||||
"input_cost_per_token_priority": 6e-6,
|
||||
"input_cost_per_token_batches": 1.5e-6,
|
||||
"input_cost_per_token_above_272k_tokens": 1.5e-6,
|
||||
"input_cost_per_token_above_272k_tokens_flex": 7.5e-7,
|
||||
"cache_read_input_token_cost_above_272k_tokens_batches": 7.5e-8,
|
||||
"provider_specific_entry": {"us": 1.1},
|
||||
"search_context_cost_per_query": {"search_context_size_medium": 3e-6},
|
||||
"litellm_provider": "synthetic",
|
||||
"max_tokens": 8192
|
||||
}))
|
||||
.expect("synthetic entry parses");
|
||||
entry.pricing()
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn model_pricing_lookups_allocate_nothing_on_the_request_path() {
|
||||
let pricing = parsed_pricing();
|
||||
let tiers = [
|
||||
ServiceTier::Priority,
|
||||
ServiceTier::Flex,
|
||||
ServiceTier::Auto,
|
||||
ServiceTier::Fast,
|
||||
ServiceTier::Ultrafast,
|
||||
];
|
||||
ALLOCATIONS.store(0, Ordering::Relaxed);
|
||||
let mut sink = 0.0_f64;
|
||||
for metric in [
|
||||
Metric::InputPerToken,
|
||||
Metric::OutputPerToken,
|
||||
Metric::CacheReadToken,
|
||||
Metric::CacheCreationToken1hr,
|
||||
Metric::OutputReasoningToken,
|
||||
] {
|
||||
sink += pricing.rate(metric).value().unwrap_or(0.0);
|
||||
for tier in tiers {
|
||||
sink += pricing.tier_rate(metric, tier).value().unwrap_or(0.0);
|
||||
}
|
||||
sink += pricing.batch_rate(metric).value().unwrap_or(0.0);
|
||||
}
|
||||
for threshold in &pricing.thresholds {
|
||||
sink += threshold.above_tokens as f64;
|
||||
for metric in [Metric::InputPerToken, Metric::CacheCreationToken1hr] {
|
||||
sink += threshold
|
||||
.standard
|
||||
.get(&metric)
|
||||
.and_then(|rate| rate.value())
|
||||
.unwrap_or(0.0);
|
||||
for tier in tiers {
|
||||
sink += threshold
|
||||
.tiers
|
||||
.get(&(metric, tier))
|
||||
.and_then(|rate| rate.value())
|
||||
.unwrap_or(0.0);
|
||||
}
|
||||
}
|
||||
}
|
||||
sink += pricing
|
||||
.provider_specific_entry
|
||||
.get("us")
|
||||
.and_then(|rate| rate.value())
|
||||
.unwrap_or(0.0);
|
||||
assert_eq!(ALLOCATIONS.load(Ordering::Relaxed), 0, "sink {sink}");
|
||||
}
|
||||
|
|
@ -5,6 +5,7 @@
|
|||
use litellm_cost::anthropic_usage::{
|
||||
is_anthropic_usage_object, transform_anthropic_usage_to_chat_usage,
|
||||
};
|
||||
use litellm_cost::responses_usage::UsageError;
|
||||
use litellm_cost::{Pricing, Rate, Rates, Request, ServiceTier, ThresholdPolicy, calculate};
|
||||
use rstest::rstest;
|
||||
use serde_json::{Value, json};
|
||||
|
|
@ -141,3 +142,24 @@ fn calculate_usage_sums_reported_reasoning_across_iterations() {
|
|||
assert_eq!(details.reasoning_tokens, Some(150));
|
||||
assert_eq!(details.text_tokens, Some(150));
|
||||
}
|
||||
|
||||
#[rstest]
|
||||
fn iteration_sums_overflow_errors_where_python_yields_a_big_int() {
|
||||
let usage = transform_anthropic_usage_to_chat_usage(
|
||||
&json!({
|
||||
"input_tokens": 0,
|
||||
"output_tokens": 0,
|
||||
"iterations": [
|
||||
{"input_tokens": u64::MAX, "output_tokens": 1},
|
||||
{"input_tokens": 1, "output_tokens": 1}
|
||||
]
|
||||
}),
|
||||
None,
|
||||
false,
|
||||
);
|
||||
assert_eq!(
|
||||
usage.unwrap_err(),
|
||||
UsageError::TokenCountOverflow,
|
||||
"divergence: Python sums arbitrary-precision ints and returns a huge cost"
|
||||
);
|
||||
}
|
||||
|
|
|
|||
Loading…
Add table
Reference in a new issue