Mirrors
/
hf_text-generation-inference
mirror of https://github.com/huggingface/text-generation-inference.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

OVerride the env in server tests.

This commit is contained in:
Nicolas Patry 2024-08-26 15:25:03 +02:00
parent 682db34b6a
commit 1568e82548
No known key found for this signature in database
GPG Key ID: 64AF4752B2967863
2 changed files with 10 additions and 35 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

40
backends/v3/src/radix.rs
View File

@ -577,14 +577,14 @@ where
mod tests {
use std::sync::Arc;
use super::RadixAllocator;
use super::*;
#[test]
fn allocator_reuses_prefixes() {
let mut cache = RadixAllocator::new(1, 12, None, true);
let mut cache = RadixAllocator::new(1, 12, None);
let allocation = cache.allocate(8, Some(Arc::new(vec![0, 1, 2, 3]))).unwrap();
assert_eq!(allocation.blocks, vec![4, 5, 6, 7, 8, 9, 10, 11]);
assert_eq!(allocation.slots, allocation.slots);
assert_eq!(allocation.blocks, allocation.slots);
assert_eq!(allocation.prefix_len, 0);
cache.free(allocation.blocks.clone(), allocation.allocation_id);
@ -593,37 +593,9 @@ mod tests {
assert_eq!(allocation.prefix_len, 4);
}
#[test]
fn allocator_doesnt_reuses_prefixes() {
let mut cache = RadixAllocator::new(1, 12, None, false);
let allocation = cache.allocate(8, Some(Arc::new(vec![0, 1, 2, 3]))).unwrap();
assert_eq!(allocation.blocks, vec![4, 5, 6, 7, 8, 9, 10, 11]);
assert_eq!(allocation.slots, allocation.slots);
assert_eq!(allocation.prefix_len, 0);
cache.free(allocation.blocks.clone(), allocation.allocation_id);
let allocation = cache.allocate(8, Some(Arc::new(vec![0, 1, 2, 3]))).unwrap();
assert_eq!(allocation.blocks, vec![4, 5, 6, 7, 8, 9, 10, 11]);
assert_eq!(allocation.prefix_len, 0);
}
#[test]
fn allocator_block_size() {
let mut cache = RadixAllocator::new(256, 12, None, false);
let allocation = cache.allocate(8, Some(Arc::new(vec![0, 1, 2, 3]))).unwrap();
assert_eq!(allocation.blocks, vec![11]);
assert_eq!(allocation.slots, allocation.slots);
assert_eq!(allocation.prefix_len, 0);
cache.free(allocation.blocks.clone(), allocation.allocation_id);
let allocation = cache.allocate(8, Some(Arc::new(vec![0, 1, 2, 3]))).unwrap();
assert_eq!(allocation.blocks, vec![11]);
assert_eq!(allocation.prefix_len, 0);
}
#[test]
fn allocator_collects_older_prefixes_first() {
let mut cache = RadixAllocator::new(1, 7, None, true);
let mut cache = RadixAllocator::new(1, 7, None);
let allocation1 = cache.allocate(4, Some(Arc::new(vec![0, 1, 2, 3]))).unwrap();
assert_eq!(allocation1.blocks, vec![3, 4, 5, 6]);
assert_eq!(allocation1.prefix_len, 0);
@ -643,7 +615,7 @@ mod tests {
#[test]
fn allocator_frees_fully_overlapping_prefills() {
let mut cache = RadixAllocator::new(1, 10, None, true);
let mut cache = RadixAllocator::new(1, 10, None);
let allocation1 = cache.allocate(4, Some(Arc::new(vec![0, 1, 2, 3]))).unwrap();
let allocation2 = cache.allocate(4, Some(Arc::new(vec![0, 1, 2, 3]))).unwrap();
@ -659,7 +631,7 @@ mod tests {
#[test]
fn allocator_frees_partially_overlapping_prefills() {
let mut cache = RadixAllocator::new(1, 20, None, true);
let mut cache = RadixAllocator::new(1, 20, None);
let allocation1 = cache.allocate(4, Some(Arc::new(vec![0, 1]))).unwrap();
assert_eq!(allocation1.blocks, vec![16, 17, 18, 19]);
assert_eq!(allocation1.prefix_len, 0);

5
server/tests/conftest.py
View File

@ -1,7 +1,10 @@
import pytest
import os
from text_generation_server.pb import generate_pb2
os.environ["USE_PREFIX_CACHING"] = "0"
os.environ["ATTENTION"] = "flashdecoding"
@pytest.fixture
def default_pb_parameters():