Skip to content

GitLab

Unverified Commit 928de468 authored by Woosuk Kwon's avatar Woosuk Kwon Committed by GitHub
Browse files

Implement PagedAttention V2 (#1348)

parent 29678cd2
Hide whitespace changes
Inline Side-by-side
Showing with 764 additions and 139 deletions
benchmarks/kernels/benchmark_paged_attention.py 0 → 100644
View file @ 928de468
import argparse
import random
import time
import torch
from vllm import attention_ops
NUM_BLOCKS = 1024
PARTITION_SIZE = 512
@torch.inference_mode()
def main(
version: str,
num_seqs: int,
context_len: int,
num_query_heads: int,
num_kv_heads: int,
head_size: int,
use_alibi: bool,
block_size: int,
dtype: torch.dtype,
seed: int,
do_profile: bool,
) -> None:
random.seed(seed)
torch.random.manual_seed(seed)
torch.cuda.manual_seed(seed)
scale = float(1.0 / (head_size**0.5))
query = torch.empty(num_seqs,
num_query_heads,
head_size,
dtype=dtype,
device="cuda")
query.uniform_(-scale, scale)
assert num_query_heads % num_kv_heads == 0
num_queries_per_kv = num_query_heads // num_kv_heads
head_mapping = torch.repeat_interleave(
torch.arange(num_kv_heads, dtype=torch.int32, device="cuda"),
num_queries_per_kv)
alibi_slopes = None
if use_alibi:
alibi_slopes = torch.randn(num_query_heads,
dtype=torch.float,
device="cuda")
context_lens = [context_len for _ in range(num_seqs)]
max_context_len = max(context_lens)
context_lens = torch.tensor(context_lens, dtype=torch.int, device="cuda")
# Create the block tables.
max_num_blocks_per_seq = (max_context_len + block_size - 1) // block_size
block_tables = []
for _ in range(num_seqs):
block_table = [
random.randint(0, NUM_BLOCKS - 1)
for _ in range(max_num_blocks_per_seq)
]
block_tables.append(block_table)
block_tables = torch.tensor(block_tables, dtype=torch.int, device="cuda")
# Create the KV cache.
x = 16 // torch.tensor([], dtype=dtype).element_size()
key_cache_shape = (NUM_BLOCKS, num_kv_heads, head_size // x, block_size, x)
key_cache = torch.empty(size=key_cache_shape, dtype=dtype, device="cuda")
key_cache.uniform_(-scale, scale)
value_cache_shape = (NUM_BLOCKS, num_kv_heads, head_size, block_size)
value_cache = torch.empty(size=value_cache_shape,
dtype=dtype,
device="cuda")
value_cache.uniform_(-scale, scale)
# Prepare for the paged attention kernel.
output = torch.empty_like(query)
if version == "v2":
num_partitions = ((max_context_len + PARTITION_SIZE - 1) //
PARTITION_SIZE)
tmp_output = torch.empty(
size=(num_seqs, num_query_heads, num_partitions, head_size),
dtype=output.dtype,
device=output.device,
)
exp_sums = torch.empty(
size=(num_seqs, num_query_heads, num_partitions),
dtype=torch.float32,
device=output.device,
)
max_logits = torch.empty_like(exp_sums)
def run_benchmark(num_iters: int, profile: bool = False) -> float:
torch.cuda.synchronize()
if profile:
torch.cuda.cudart().cudaProfilerStart()
start_time = time.perf_counter()
for _ in range(num_iters):
if version == "v1":
attention_ops.paged_attention_v1(
output,
query,
key_cache,
value_cache,
head_mapping,
scale,
block_tables,
context_lens,
block_size,
max_context_len,
alibi_slopes,
)
elif version == "v2":
attention_ops.paged_attention_v2(
output,
exp_sums,
max_logits,
tmp_output,
query,
key_cache,
value_cache,
head_mapping,
scale,
block_tables,
context_lens,
block_size,
max_context_len,
alibi_slopes,
)
else:
raise ValueError(f"Invalid version: {version}")
torch.cuda.synchronize()
end_time = time.perf_counter()
if profile:
torch.cuda.cudart().cudaProfilerStart()
return (end_time - start_time) / num_iters
# Warmup.
print("Warming up...")
run_benchmark(num_iters=3, profile=False)
# Benchmark.
if do_profile:
latency = run_benchmark(num_iters=1, profile=True)
else:
latency = run_benchmark(num_iters=100, profile=False)
print(f"Kernel running time: {latency * 1000000:.3f} us")
if __name__ == '__main__':
parser = argparse.ArgumentParser(
description="Benchmark the paged attention kernel.")
parser.add_argument("--version",
type=str,
choices=["v1", "v2"],
default="v2")
parser.add_argument("--batch-size", type=int, default=8)
parser.add_argument("--context-len", type=int, default=4096)
parser.add_argument("--num-query-heads", type=int, default=64)
parser.add_argument("--num-kv-heads", type=int, default=8)
parser.add_argument("--head-size",
type=int,
choices=[64, 80, 96, 112, 128, 256],
default=128)
parser.add_argument("--block-size", type=int, choices=[16, 32], default=16)
parser.add_argument("--use-alibi", action="store_true")
parser.add_argument("--dtype",
type=str,
choices=["half", "bfloat16", "float"],
default="half")
parser.add_argument("--seed", type=int, default=0)
parser.add_argument("--profile", action="store_true")
args = parser.parse_args()
print(args)
if args.num_query_heads % args.num_kv_heads != 0:
raise ValueError("num_query_heads must be divisible by num_kv_heads")
dtype_to_torch_dtype = {
"half": torch.half,
"bfloat16": torch.bfloat16,
"float": torch.float,
}
main(
version=args.version,
num_seqs=args.batch_size,
context_len=args.context_len,
num_query_heads=args.num_query_heads,
num_kv_heads=args.num_kv_heads,
head_size=args.head_size,
block_size=args.block_size,
use_alibi=args.use_alibi,
dtype=dtype_to_torch_dtype[args.dtype],
seed=args.seed,
do_profile=args.profile,
)
csrc/attention.cpp
View file @ 928de468
#include <torch/extension.h> #include <torch/extension.h>
#include <c10/util/Optional.h> #include <c10/util/Optional.h>
void single_query_cached_kv_attention( void paged_attention_v1(
torch::Tensor& out, torch::Tensor& out,
torch::Tensor& query, torch::Tensor& query,
torch::Tensor& key_cache, torch::Tensor& key_cache,
...@@ -14,9 +14,29 @@ void single_query_cached_kv_attention( ...@@ -14,9 +14,29 @@ void single_query_cached_kv_attention(
int max_context_len, int max_context_len,
const c10::optional<torch::Tensor>& alibi_slopes); const c10::optional<torch::Tensor>& alibi_slopes);
void paged_attention_v2(
torch::Tensor& out,
torch::Tensor& exp_sums,
torch::Tensor& max_logits,
torch::Tensor& tmp_out,
torch::Tensor& query,
torch::Tensor& key_cache,
torch::Tensor& value_cache,
torch::Tensor& head_mapping,
float scale,
torch::Tensor& block_tables,
torch::Tensor& context_lens,
int block_size,
int max_context_len,
const c10::optional<torch::Tensor>& alibi_slopes);
PYBIND11_MODULE(TORCH_EXTENSION_NAME, m) { PYBIND11_MODULE(TORCH_EXTENSION_NAME, m) {
m.def( m.def(
"single_query_cached_kv_attention", "paged_attention_v1",
&single_query_cached_kv_attention, &paged_attention_v1,
"Compute the attention between an input query and the cached key/value tensors"); "Compute the attention between an input query and the cached keys/values using PagedAttention.");
m.def(
"paged_attention_v2",
&paged_attention_v2,
"PagedAttention V2.");
} }
csrc/attention/attention_kernels.cu
View file @ 928de468
...@@ -26,6 +26,7 @@ ...@@ -26,6 +26,7 @@
#define WARP_SIZE 32 #define WARP_SIZE 32
#define MAX(a, b) ((a) > (b) ? (a) : (b)) #define MAX(a, b) ((a) > (b) ? (a) : (b))
#define MIN(a, b) ((a) < (b) ? (a) : (b)) #define MIN(a, b) ((a) < (b) ? (a) : (b))
#define DIVIDE_ROUND_UP(a, b) (((a) + (b) - 1) / (b))
namespace vllm { namespace vllm {
...@@ -65,14 +66,18 @@ inline __device__ float block_sum(float* red_smem, float sum) { ...@@ -65,14 +66,18 @@ inline __device__ float block_sum(float* red_smem, float sum) {
return __shfl_sync(uint32_t(-1), sum, 0); return __shfl_sync(uint32_t(-1), sum, 0);
} }
// Grid: (num_heads, num_seqs). // TODO(woosuk): Merge the last two dimensions of the grid.
// Grid: (num_heads, num_seqs, max_num_partitions).
template< template<
typename scalar_t, typename scalar_t,
int HEAD_SIZE, int HEAD_SIZE,
int BLOCK_SIZE, int BLOCK_SIZE,
int NUM_THREADS> int NUM_THREADS,
__global__ void single_query_cached_kv_attention_kernel( int PARTITION_SIZE = 0> // Zero means no partitioning.
scalar_t* __restrict__ out, // [num_seqs, num_heads, head_size] __device__ void paged_attention_kernel(
float* __restrict__ exp_sums, // [num_seqs, num_heads, max_num_partitions]
float* __restrict__ max_logits, // [num_seqs, num_heads, max_num_partitions]
scalar_t* __restrict__ out, // [num_seqs, num_heads, max_num_partitions, head_size]
const scalar_t* __restrict__ q, // [num_seqs, num_heads, head_size] const scalar_t* __restrict__ q, // [num_seqs, num_heads, head_size]
const scalar_t* __restrict__ k_cache, // [num_blocks, num_kv_heads, head_size/x, block_size, x] const scalar_t* __restrict__ k_cache, // [num_blocks, num_kv_heads, head_size/x, block_size, x]
const scalar_t* __restrict__ v_cache, // [num_blocks, num_kv_heads, head_size, block_size] const scalar_t* __restrict__ v_cache, // [num_blocks, num_kv_heads, head_size, block_size]
...@@ -85,10 +90,33 @@ __global__ void single_query_cached_kv_attention_kernel( ...@@ -85,10 +90,33 @@ __global__ void single_query_cached_kv_attention_kernel(
const int q_stride, const int q_stride,
const int kv_block_stride, const int kv_block_stride,
const int kv_head_stride) { const int kv_head_stride) {
const int seq_idx = blockIdx.y;
const int partition_idx = blockIdx.z;
const int max_num_partitions = gridDim.z;
constexpr bool USE_PARTITIONING = PARTITION_SIZE > 0;
const int context_len = context_lens[seq_idx];
if (USE_PARTITIONING && partition_idx * PARTITION_SIZE >= context_len) {
// No work to do. Terminate the thread block.
return;
}
const int num_context_blocks = DIVIDE_ROUND_UP(context_len, BLOCK_SIZE);
const int num_blocks_per_partition = USE_PARTITIONING ? PARTITION_SIZE / BLOCK_SIZE : num_context_blocks;
// [start_block_idx, end_block_idx) is the range of blocks to process.
const int start_block_idx = USE_PARTITIONING ? partition_idx * num_blocks_per_partition : 0;
const int end_block_idx = MIN(start_block_idx + num_blocks_per_partition, num_context_blocks);
const int num_blocks = end_block_idx - start_block_idx;
// [start_token_idx, end_token_idx) is the range of tokens to process.
const int start_token_idx = start_block_idx * BLOCK_SIZE;
const int end_token_idx = MIN(start_token_idx + num_blocks * BLOCK_SIZE, context_len);
const int num_tokens = end_token_idx - start_token_idx;
constexpr int THREAD_GROUP_SIZE = MAX(WARP_SIZE / BLOCK_SIZE, 1); constexpr int THREAD_GROUP_SIZE = MAX(WARP_SIZE / BLOCK_SIZE, 1);
constexpr int NUM_THREAD_GROUPS = NUM_THREADS / THREAD_GROUP_SIZE; // Note: This assumes THREAD_GROUP_SIZE divides NUM_THREADS constexpr int NUM_THREAD_GROUPS = NUM_THREADS / THREAD_GROUP_SIZE; // Note: This assumes THREAD_GROUP_SIZE divides NUM_THREADS
assert(NUM_THREADS % THREAD_GROUP_SIZE == 0); assert(NUM_THREADS % THREAD_GROUP_SIZE == 0);
constexpr int NUM_TOKENS_PER_THREAD_GROUP = (BLOCK_SIZE + WARP_SIZE - 1) / WARP_SIZE; constexpr int NUM_TOKENS_PER_THREAD_GROUP = DIVIDE_ROUND_UP(BLOCK_SIZE, WARP_SIZE);
constexpr int NUM_WARPS = NUM_THREADS / WARP_SIZE; constexpr int NUM_WARPS = NUM_THREADS / WARP_SIZE;
const int thread_idx = threadIdx.x; const int thread_idx = threadIdx.x;
const int warp_idx = thread_idx / WARP_SIZE; const int warp_idx = thread_idx / WARP_SIZE;
...@@ -97,7 +125,6 @@ __global__ void single_query_cached_kv_attention_kernel( ...@@ -97,7 +125,6 @@ __global__ void single_query_cached_kv_attention_kernel(
const int head_idx = blockIdx.x; const int head_idx = blockIdx.x;
const int num_heads = gridDim.x; const int num_heads = gridDim.x;
const int kv_head_idx = head_mapping[head_idx]; const int kv_head_idx = head_mapping[head_idx];
const int seq_idx = blockIdx.y;
const float alibi_slope = alibi_slopes == nullptr ? 0.f : alibi_slopes[head_idx]; const float alibi_slope = alibi_slopes == nullptr ? 0.f : alibi_slopes[head_idx];
// A vector type to store a part of a key or a query. // A vector type to store a part of a key or a query.
...@@ -142,15 +169,12 @@ __global__ void single_query_cached_kv_attention_kernel( ...@@ -142,15 +169,12 @@ __global__ void single_query_cached_kv_attention_kernel(
constexpr int x = 16 / sizeof(scalar_t); constexpr int x = 16 / sizeof(scalar_t);
float qk_max = -FLT_MAX; float qk_max = -FLT_MAX;
const int* block_table = block_tables + seq_idx * max_num_blocks_per_seq;
const int context_len = context_lens[seq_idx];
const int num_blocks = (context_len + BLOCK_SIZE - 1) / BLOCK_SIZE;
// Iterate over the key blocks. // Iterate over the key blocks.
// Each warp fetches a block of keys for each iteration. // Each warp fetches a block of keys for each iteration.
// Each thread group in a warp fetches a key from the block, and computes // Each thread group in a warp fetches a key from the block, and computes
// dot product with the query. // dot product with the query.
for (int block_idx = warp_idx; block_idx < num_blocks; block_idx += NUM_WARPS) { const int* block_table = block_tables + seq_idx * max_num_blocks_per_seq;
for (int block_idx = start_block_idx + warp_idx; block_idx < end_block_idx; block_idx += NUM_WARPS) {
const int physical_block_number = block_table[block_idx]; const int physical_block_number = block_table[block_idx];
// Load a key to registers. // Load a key to registers.
...@@ -184,7 +208,7 @@ __global__ void single_query_cached_kv_attention_kernel( ...@@ -184,7 +208,7 @@ __global__ void single_query_cached_kv_attention_kernel(
// Store the partial reductions to shared memory. // Store the partial reductions to shared memory.
// NOTE(woosuk): It is required to zero out the masked logits. // NOTE(woosuk): It is required to zero out the masked logits.
const bool mask = token_idx >= context_len; const bool mask = token_idx >= context_len;
logits[token_idx] = mask ? 0.f : qk; logits[token_idx - start_token_idx] = mask ? 0.f : qk;
// Update the max value. // Update the max value.
qk_max = mask ? qk_max : fmaxf(qk_max, qk); qk_max = mask ? qk_max : fmaxf(qk_max, qk);
} }
...@@ -215,7 +239,7 @@ __global__ void single_query_cached_kv_attention_kernel( ...@@ -215,7 +239,7 @@ __global__ void single_query_cached_kv_attention_kernel(
// Get the sum of the exp values. // Get the sum of the exp values.
float exp_sum = 0.f; float exp_sum = 0.f;
for (int i = thread_idx; i < context_len; i += NUM_THREADS) { for (int i = thread_idx; i < num_tokens; i += NUM_THREADS) {
float val = __expf(logits[i] - qk_max); float val = __expf(logits[i] - qk_max);
logits[i] = val; logits[i] = val;
exp_sum += val; exp_sum += val;
...@@ -224,11 +248,23 @@ __global__ void single_query_cached_kv_attention_kernel( ...@@ -224,11 +248,23 @@ __global__ void single_query_cached_kv_attention_kernel(
// Compute softmax. // Compute softmax.
const float inv_sum = __fdividef(1.f, exp_sum + 1e-6f); const float inv_sum = __fdividef(1.f, exp_sum + 1e-6f);
for (int i = thread_idx; i < context_len; i += NUM_THREADS) { for (int i = thread_idx; i < num_tokens; i += NUM_THREADS) {
logits[i] *= inv_sum; logits[i] *= inv_sum;
} }
__syncthreads(); __syncthreads();
// If partitioning is enabled, store the max logit and exp_sum.
if (USE_PARTITIONING && thread_idx == 0) {
float* max_logits_ptr = max_logits + seq_idx * num_heads * max_num_partitions
+ head_idx * max_num_partitions
+ partition_idx;
*max_logits_ptr = qk_max;
float* exp_sums_ptr = exp_sums + seq_idx * num_heads * max_num_partitions
+ head_idx * max_num_partitions
+ partition_idx;
*exp_sums_ptr = exp_sum;
}
// Each thread will fetch 16 bytes from the value cache at a time. // Each thread will fetch 16 bytes from the value cache at a time.
constexpr int V_VEC_SIZE = MIN(16 / sizeof(scalar_t), BLOCK_SIZE); constexpr int V_VEC_SIZE = MIN(16 / sizeof(scalar_t), BLOCK_SIZE);
using V_vec = typename Vec<scalar_t, V_VEC_SIZE>::Type; using V_vec = typename Vec<scalar_t, V_VEC_SIZE>::Type;
...@@ -237,7 +273,7 @@ __global__ void single_query_cached_kv_attention_kernel( ...@@ -237,7 +273,7 @@ __global__ void single_query_cached_kv_attention_kernel(
constexpr int NUM_V_VECS_PER_ROW = BLOCK_SIZE / V_VEC_SIZE; constexpr int NUM_V_VECS_PER_ROW = BLOCK_SIZE / V_VEC_SIZE;
constexpr int NUM_ROWS_PER_ITER = WARP_SIZE / NUM_V_VECS_PER_ROW; constexpr int NUM_ROWS_PER_ITER = WARP_SIZE / NUM_V_VECS_PER_ROW;
constexpr int NUM_ROWS_PER_THREAD = (HEAD_SIZE + NUM_ROWS_PER_ITER - 1) / NUM_ROWS_PER_ITER; constexpr int NUM_ROWS_PER_THREAD = DIVIDE_ROUND_UP(HEAD_SIZE, NUM_ROWS_PER_ITER);
// NOTE(woosuk): We use FP32 for the accumulator for better accuracy. // NOTE(woosuk): We use FP32 for the accumulator for better accuracy.
float accs[NUM_ROWS_PER_THREAD]; float accs[NUM_ROWS_PER_THREAD];
...@@ -248,12 +284,12 @@ __global__ void single_query_cached_kv_attention_kernel( ...@@ -248,12 +284,12 @@ __global__ void single_query_cached_kv_attention_kernel(
scalar_t zero_value; scalar_t zero_value;
zero(zero_value); zero(zero_value);
for (int block_idx = warp_idx; block_idx < num_blocks; block_idx += NUM_WARPS) { for (int block_idx = start_block_idx + warp_idx; block_idx < end_block_idx; block_idx += NUM_WARPS) {
const int physical_block_number = block_table[block_idx]; const int physical_block_number = block_table[block_idx];
const int physical_block_offset = (lane % NUM_V_VECS_PER_ROW) * V_VEC_SIZE; const int physical_block_offset = (lane % NUM_V_VECS_PER_ROW) * V_VEC_SIZE;
const int token_idx = block_idx * BLOCK_SIZE + physical_block_offset; const int token_idx = block_idx * BLOCK_SIZE + physical_block_offset;
L_vec logits_vec; L_vec logits_vec;
from_float(logits_vec, *reinterpret_cast<Float_L_vec*>(logits + token_idx)); from_float(logits_vec, *reinterpret_cast<Float_L_vec*>(logits + token_idx - start_token_idx));
const scalar_t* v_ptr = v_cache + physical_block_number * kv_block_stride const scalar_t* v_ptr = v_cache + physical_block_number * kv_block_stride
+ kv_head_idx * kv_head_stride; + kv_head_idx * kv_head_stride;
...@@ -263,7 +299,7 @@ __global__ void single_query_cached_kv_attention_kernel( ...@@ -263,7 +299,7 @@ __global__ void single_query_cached_kv_attention_kernel(
if (row_idx < HEAD_SIZE) { if (row_idx < HEAD_SIZE) {
const int offset = row_idx * BLOCK_SIZE + physical_block_offset; const int offset = row_idx * BLOCK_SIZE + physical_block_offset;
V_vec v_vec = *reinterpret_cast<const V_vec*>(v_ptr + offset); V_vec v_vec = *reinterpret_cast<const V_vec*>(v_ptr + offset);
if (block_idx == num_blocks - 1) { if (block_idx == num_context_blocks - 1) {
// NOTE(woosuk): When v_vec contains the tokens that are out of the context, // NOTE(woosuk): When v_vec contains the tokens that are out of the context,
// we should explicitly zero out the values since they may contain NaNs. // we should explicitly zero out the values since they may contain NaNs.
// See https://github.com/vllm-project/vllm/issues/641#issuecomment-1682544472 // See https://github.com/vllm-project/vllm/issues/641#issuecomment-1682544472
...@@ -327,7 +363,9 @@ __global__ void single_query_cached_kv_attention_kernel( ...@@ -327,7 +363,9 @@ __global__ void single_query_cached_kv_attention_kernel(
// Write the final output. // Write the final output.
if (warp_idx == 0) { if (warp_idx == 0) {
scalar_t* out_ptr = out + seq_idx * num_heads * HEAD_SIZE + head_idx * HEAD_SIZE; scalar_t* out_ptr = out + seq_idx * num_heads * max_num_partitions * HEAD_SIZE
+ head_idx * max_num_partitions * HEAD_SIZE
+ partition_idx * HEAD_SIZE;
#pragma unroll #pragma unroll
for (int i = 0; i < NUM_ROWS_PER_THREAD; i++) { for (int i = 0; i < NUM_ROWS_PER_THREAD; i++) {
const int row_idx = lane / NUM_V_VECS_PER_ROW + i * NUM_ROWS_PER_ITER; const int row_idx = lane / NUM_V_VECS_PER_ROW + i * NUM_ROWS_PER_ITER;
...@@ -338,13 +376,167 @@ __global__ void single_query_cached_kv_attention_kernel( ...@@ -338,13 +376,167 @@ __global__ void single_query_cached_kv_attention_kernel(
} }
} }
// Grid: (num_heads, num_seqs, 1).
template<
typename scalar_t,
int HEAD_SIZE,
int BLOCK_SIZE,
int NUM_THREADS>
__global__ void paged_attention_v1_kernel(
scalar_t* __restrict__ out, // [num_seqs, num_heads, head_size]
const scalar_t* __restrict__ q, // [num_seqs, num_heads, head_size]
const scalar_t* __restrict__ k_cache, // [num_blocks, num_kv_heads, head_size/x, block_size, x]
const scalar_t* __restrict__ v_cache, // [num_blocks, num_kv_heads, head_size, block_size]
const int* __restrict__ head_mapping, // [num_heads]
const float scale,
const int* __restrict__ block_tables, // [num_seqs, max_num_blocks_per_seq]
const int* __restrict__ context_lens, // [num_seqs]
const int max_num_blocks_per_seq,
const float* __restrict__ alibi_slopes, // [num_heads]
const int q_stride,
const int kv_block_stride,
const int kv_head_stride) {
paged_attention_kernel<scalar_t, HEAD_SIZE, BLOCK_SIZE, NUM_THREADS>(
/* exp_sums */ nullptr, /* max_logits */ nullptr,
out, q, k_cache, v_cache, head_mapping, scale, block_tables, context_lens,
max_num_blocks_per_seq, alibi_slopes, q_stride, kv_block_stride, kv_head_stride);
}
// Grid: (num_heads, num_seqs, max_num_partitions).
template<
typename scalar_t,
int HEAD_SIZE,
int BLOCK_SIZE,
int NUM_THREADS,
int PARTITION_SIZE>
__global__ void paged_attention_v2_kernel(
float* __restrict__ exp_sums, // [num_seqs, num_heads, max_num_partitions]
float* __restrict__ max_logits, // [num_seqs, num_heads, max_num_partitions]
scalar_t* __restrict__ tmp_out, // [num_seqs, num_heads, max_num_partitions, head_size]
const scalar_t* __restrict__ q, // [num_seqs, num_heads, head_size]
const scalar_t* __restrict__ k_cache, // [num_blocks, num_kv_heads, head_size/x, block_size, x]
const scalar_t* __restrict__ v_cache, // [num_blocks, num_kv_heads, head_size, block_size]
const int* __restrict__ head_mapping, // [num_heads]
const float scale,
const int* __restrict__ block_tables, // [num_seqs, max_num_blocks_per_seq]
const int* __restrict__ context_lens, // [num_seqs]
const int max_num_blocks_per_seq,
const float* __restrict__ alibi_slopes, // [num_heads]
const int q_stride,
const int kv_block_stride,
const int kv_head_stride) {
paged_attention_kernel<scalar_t, HEAD_SIZE, BLOCK_SIZE, NUM_THREADS, PARTITION_SIZE>(
exp_sums, max_logits, tmp_out, q, k_cache, v_cache, head_mapping, scale,
block_tables, context_lens, max_num_blocks_per_seq, alibi_slopes,
q_stride, kv_block_stride, kv_head_stride);
}
// Grid: (num_heads, num_seqs).
template<
typename scalar_t,
int HEAD_SIZE,
int NUM_THREADS,
int PARTITION_SIZE>
__global__ void paged_attention_v2_reduce_kernel(
scalar_t* __restrict__ out, // [num_seqs, num_heads, head_size]
const float* __restrict__ exp_sums, // [num_seqs, num_heads, max_num_partitions]
const float* __restrict__ max_logits, // [num_seqs, num_heads, max_num_partitions]
const scalar_t* __restrict__ tmp_out, // [num_seqs, num_heads, max_num_partitions, head_size]
const int* __restrict__ context_lens, // [num_seqs]
const int max_num_partitions) {
const int num_heads = gridDim.x;
const int head_idx = blockIdx.x;
const int seq_idx = blockIdx.y;
const int context_len = context_lens[seq_idx];
const int num_partitions = DIVIDE_ROUND_UP(context_len, PARTITION_SIZE);
if (num_partitions == 1) {
// No need to reduce. Only copy tmp_out to out.
scalar_t* out_ptr = out + seq_idx * num_heads * HEAD_SIZE + head_idx * HEAD_SIZE;
const scalar_t* tmp_out_ptr = tmp_out + seq_idx * num_heads * max_num_partitions * HEAD_SIZE
+ head_idx * max_num_partitions * HEAD_SIZE;
for (int i = threadIdx.x; i < HEAD_SIZE; i += blockDim.x) {
out_ptr[i] = tmp_out_ptr[i];
}
// Terminate the thread block.
return;
}
constexpr int NUM_WARPS = NUM_THREADS / WARP_SIZE;
const int warp_idx = threadIdx.x / WARP_SIZE;
const int lane = threadIdx.x % WARP_SIZE;
// Size: 2 * num_partitions.
extern __shared__ char shared_mem[];
// Workspace for reduction.
__shared__ float red_smem[2 * NUM_WARPS];
// Load max logits to shared memory.
float* shared_max_logits = reinterpret_cast<float*>(shared_mem);
const float* max_logits_ptr = max_logits + seq_idx * num_heads * max_num_partitions
+ head_idx * max_num_partitions;
float max_logit = -FLT_MAX;
for (int i = threadIdx.x; i < num_partitions; i += blockDim.x) {
const float l = max_logits_ptr[i];
shared_max_logits[i] = l;
max_logit = fmaxf(max_logit, l);
}
__syncthreads();
// Get the global max logit.
// Reduce within the warp.
#pragma unroll
for (int mask = WARP_SIZE / 2; mask >= 1; mask /= 2) {
max_logit = fmaxf(max_logit, __shfl_xor_sync(uint32_t(-1), max_logit, mask));
}
if (lane == 0) {
red_smem[warp_idx] = max_logit;
}
__syncthreads();
// Reduce across warps.
max_logit = lane < NUM_WARPS ? red_smem[lane] : -FLT_MAX;
#pragma unroll
for (int mask = NUM_WARPS / 2; mask >= 1; mask /= 2) {
max_logit = fmaxf(max_logit, __shfl_xor_sync(uint32_t(-1), max_logit, mask));
}
// Broadcast the max value to all threads.
max_logit = __shfl_sync(uint32_t(-1), max_logit, 0);
// Load rescaled exp sums to shared memory.
float* shared_exp_sums = reinterpret_cast<float*>(shared_mem + sizeof(float) * num_partitions);
const float* exp_sums_ptr = exp_sums + seq_idx * num_heads * max_num_partitions
+ head_idx * max_num_partitions;
float global_exp_sum = 0.0f;
for (int i = threadIdx.x; i < num_partitions; i += blockDim.x) {
float l = shared_max_logits[i];
float rescaled_exp_sum = exp_sums_ptr[i] * expf(l - max_logit);
global_exp_sum += rescaled_exp_sum;
shared_exp_sums[i] = rescaled_exp_sum;
}
__syncthreads();
global_exp_sum = block_sum<NUM_WARPS>(&red_smem[NUM_WARPS], global_exp_sum);
const float inv_global_exp_sum = __fdividef(1.0f, global_exp_sum + 1e-6f);
// Aggregate tmp_out to out.
const scalar_t* tmp_out_ptr = tmp_out + seq_idx * num_heads * max_num_partitions * HEAD_SIZE
+ head_idx * max_num_partitions * HEAD_SIZE;
scalar_t* out_ptr = out + seq_idx * num_heads * HEAD_SIZE + head_idx * HEAD_SIZE;
#pragma unroll
for (int i = threadIdx.x; i < HEAD_SIZE; i += NUM_THREADS) {
float acc = 0.0f;
for (int j = 0; j < num_partitions; ++j) {
acc += to_float(tmp_out_ptr[j * HEAD_SIZE + i]) * shared_exp_sums[j] * inv_global_exp_sum;
}
from_float(out_ptr[i], acc);
}
}
} // namespace vllm } // namespace vllm
#define LAUNCH_ATTENTION_KERNEL(T, HEAD_SIZE, BLOCK_SIZE, NUM_THREADS) \ #define LAUNCH_PAGED_ATTENTION_V1(HEAD_SIZE) \
cudaFuncSetAttribute( \ cudaFuncSetAttribute( \
vllm::single_query_cached_kv_attention_kernel<T, HEAD_SIZE, BLOCK_SIZE, NUM_THREADS>, \ vllm::paged_attention_v1_kernel<T, HEAD_SIZE, BLOCK_SIZE, NUM_THREADS>, \
cudaFuncAttributeMaxDynamicSharedMemorySize, shared_mem_size); \ cudaFuncAttributeMaxDynamicSharedMemorySize, shared_mem_size); \
vllm::single_query_cached_kv_attention_kernel<T, HEAD_SIZE, BLOCK_SIZE, NUM_THREADS> \ vllm::paged_attention_v1_kernel<T, HEAD_SIZE, BLOCK_SIZE, NUM_THREADS> \
<<<grid, block, shared_mem_size, stream>>>( \ <<<grid, block, shared_mem_size, stream>>>( \
out_ptr, \ out_ptr, \
query_ptr, \ query_ptr, \
...@@ -365,7 +557,7 @@ template< ...@@ -365,7 +557,7 @@ template<
typename T, typename T,
int BLOCK_SIZE, int BLOCK_SIZE,
int NUM_THREADS = 128> int NUM_THREADS = 128>
void single_query_cached_kv_attention_launcher( void paged_attention_v1_launcher(
torch::Tensor& out, torch::Tensor& out,
torch::Tensor& query, torch::Tensor& query,
torch::Tensor& key_cache, torch::Tensor& key_cache,
...@@ -401,45 +593,206 @@ void single_query_cached_kv_attention_launcher( ...@@ -401,45 +593,206 @@ void single_query_cached_kv_attention_launcher(
int* context_lens_ptr = context_lens.data_ptr<int>(); int* context_lens_ptr = context_lens.data_ptr<int>();
constexpr int NUM_WARPS = NUM_THREADS / WARP_SIZE; constexpr int NUM_WARPS = NUM_THREADS / WARP_SIZE;
int padded_max_context_len = ((max_context_len + BLOCK_SIZE - 1) / BLOCK_SIZE) * BLOCK_SIZE; int padded_max_context_len = DIVIDE_ROUND_UP(max_context_len, BLOCK_SIZE) * BLOCK_SIZE;
int logits_size = padded_max_context_len * sizeof(float); int logits_size = padded_max_context_len * sizeof(float);
int outputs_size = (NUM_WARPS / 2) * head_size * sizeof(float); int outputs_size = (NUM_WARPS / 2) * head_size * sizeof(float);
// Python-side check in vllm.worker.worker._check_if_can_support_max_seq_len // Python-side check in vllm.worker.worker._check_if_can_support_max_seq_len
// Keep that in sync with the logic here! // Keep that in sync with the logic here!
int shared_mem_size = std::max(logits_size, outputs_size); int shared_mem_size = std::max(logits_size, outputs_size);
dim3 grid(num_heads, num_seqs); dim3 grid(num_heads, num_seqs, 1);
dim3 block(NUM_THREADS);
const cudaStream_t stream = at::cuda::getCurrentCUDAStream();
switch (head_size) {
// NOTE(woosuk): To reduce the compilation time, we only compile for the
// head sizes that we use in the model. However, we can easily extend this
// to support any head size which is a multiple of 16.
case 64:
LAUNCH_PAGED_ATTENTION_V1(64);
break;
case 80:
LAUNCH_PAGED_ATTENTION_V1(80);
break;
case 96:
LAUNCH_PAGED_ATTENTION_V1(96);
break;
case 112:
LAUNCH_PAGED_ATTENTION_V1(112);
break;
case 128:
LAUNCH_PAGED_ATTENTION_V1(128);
break;
case 256:
LAUNCH_PAGED_ATTENTION_V1(256);
break;
default:
TORCH_CHECK(false, "Unsupported head size: ", head_size);
break;
}
}
#define CALL_V1_LAUNCHER(T, BLOCK_SIZE) \
paged_attention_v1_launcher<T, BLOCK_SIZE>( \
out, \
query, \
key_cache, \
value_cache, \
head_mapping, \
scale, \
block_tables, \
context_lens, \
max_context_len, \
alibi_slopes);
// NOTE(woosuk): To reduce the compilation time, we omitted block sizes
// 1, 2, 4, 64, 128, 256.
#define CALL_V1_LAUNCHER_BLOCK_SIZE(T) \
switch (block_size) { \
case 8: \
CALL_V1_LAUNCHER(T, 8); \
break; \
case 16: \
CALL_V1_LAUNCHER(T, 16); \
break; \
case 32: \
CALL_V1_LAUNCHER(T, 32); \
break; \
default: \
TORCH_CHECK(false, "Unsupported block size: ", block_size); \
break; \
}
void paged_attention_v1(
torch::Tensor& out, // [num_seqs, num_heads, head_size]
torch::Tensor& query, // [num_seqs, num_heads, head_size]
torch::Tensor& key_cache, // [num_blocks, num_heads, head_size/x, block_size, x]
torch::Tensor& value_cache, // [num_blocks, num_heads, head_size, block_size]
torch::Tensor& head_mapping, // [num_heads]
float scale,
torch::Tensor& block_tables, // [num_seqs, max_num_blocks_per_seq]
torch::Tensor& context_lens, // [num_seqs]
int block_size,
int max_context_len,
const c10::optional<torch::Tensor>& alibi_slopes) {
if (query.dtype() == at::ScalarType::Float) {
CALL_V1_LAUNCHER_BLOCK_SIZE(float);
} else if (query.dtype() == at::ScalarType::Half) {
CALL_V1_LAUNCHER_BLOCK_SIZE(uint16_t);
} else if (query.dtype() == at::ScalarType::BFloat16) {
CALL_V1_LAUNCHER_BLOCK_SIZE(__nv_bfloat16);
} else {
TORCH_CHECK(false, "Unsupported data type: ", query.dtype());
}
}
#define LAUNCH_PAGED_ATTENTION_V2(HEAD_SIZE) \
vllm::paged_attention_v2_kernel<T, HEAD_SIZE, BLOCK_SIZE, NUM_THREADS, PARTITION_SIZE> \
<<<grid, block, shared_mem_size, stream>>>( \
exp_sums_ptr, \
max_logits_ptr, \
tmp_out_ptr, \
query_ptr, \
key_cache_ptr, \
value_cache_ptr, \
head_mapping_ptr, \
scale, \
block_tables_ptr, \
context_lens_ptr, \
max_num_blocks_per_seq, \
alibi_slopes_ptr, \
q_stride, \
kv_block_stride, \
kv_head_stride); \
vllm::paged_attention_v2_reduce_kernel<T, HEAD_SIZE, NUM_THREADS, PARTITION_SIZE> \
<<<reduce_grid, block, reduce_shared_mem_size, stream>>>( \
out_ptr, \
exp_sums_ptr, \
max_logits_ptr, \
tmp_out_ptr, \
context_lens_ptr, \
max_num_partitions);
template<
typename T,
int BLOCK_SIZE,
int NUM_THREADS = 128,
int PARTITION_SIZE = 512>
void paged_attention_v2_launcher(
torch::Tensor& out,
torch::Tensor& exp_sums,
torch::Tensor& max_logits,
torch::Tensor& tmp_out,
torch::Tensor& query,
torch::Tensor& key_cache,
torch::Tensor& value_cache,
torch::Tensor& head_mapping,
float scale,
torch::Tensor& block_tables,
torch::Tensor& context_lens,
int max_context_len,
const c10::optional<torch::Tensor>& alibi_slopes) {
int num_seqs = query.size(0);
int num_heads = query.size(1);
int head_size = query.size(2);
int max_num_blocks_per_seq = block_tables.size(1);
int q_stride = query.stride(0);
int kv_block_stride = key_cache.stride(0);
int kv_head_stride = key_cache.stride(1);
int thread_group_size = MAX(WARP_SIZE / BLOCK_SIZE, 1);
assert(head_size % thread_group_size == 0);
// NOTE: alibi_slopes is optional.
const float* alibi_slopes_ptr = alibi_slopes ?
reinterpret_cast<const float*>(alibi_slopes.value().data_ptr())
: nullptr;
T* out_ptr = reinterpret_cast<T*>(out.data_ptr());
float* exp_sums_ptr = reinterpret_cast<float*>(exp_sums.data_ptr());
float* max_logits_ptr = reinterpret_cast<float*>(max_logits.data_ptr());
T* tmp_out_ptr = reinterpret_cast<T*>(tmp_out.data_ptr());
T* query_ptr = reinterpret_cast<T*>(query.data_ptr());
T* key_cache_ptr = reinterpret_cast<T*>(key_cache.data_ptr());
T* value_cache_ptr = reinterpret_cast<T*>(value_cache.data_ptr());
int* head_mapping_ptr = reinterpret_cast<int*>(head_mapping.data_ptr());
int* block_tables_ptr = block_tables.data_ptr<int>();
int* context_lens_ptr = context_lens.data_ptr<int>();
constexpr int NUM_WARPS = NUM_THREADS / WARP_SIZE;
int max_num_partitions = DIVIDE_ROUND_UP(max_context_len, PARTITION_SIZE);
int logits_size = PARTITION_SIZE * sizeof(float);
int outputs_size = (NUM_WARPS / 2) * head_size * sizeof(float);
// For paged attention v2 kernel.
dim3 grid(num_heads, num_seqs, max_num_partitions);
int shared_mem_size = std::max(logits_size, outputs_size);
// For paged attention v2 reduce kernel.
dim3 reduce_grid(num_heads, num_seqs);
int reduce_shared_mem_size = 2 * max_num_partitions * sizeof(float);
dim3 block(NUM_THREADS); dim3 block(NUM_THREADS);
const cudaStream_t stream = at::cuda::getCurrentCUDAStream(); const cudaStream_t stream = at::cuda::getCurrentCUDAStream();
switch (head_size) { switch (head_size) {
// NOTE(woosuk): To reduce the compilation time, we omitted head sizes // NOTE(woosuk): To reduce the compilation time, we only compile for the
// 32, 160, 192. // head sizes that we use in the model. However, we can easily extend this
// case 32: // to support any head size which is a multiple of 16.
// LAUNCH_ATTENTION_KERNEL(T, 32, BLOCK_SIZE, NUM_THREADS);
// break;
case 64: case 64:
LAUNCH_ATTENTION_KERNEL(T, 64, BLOCK_SIZE, NUM_THREADS); LAUNCH_PAGED_ATTENTION_V2(64);
break; break;
case 80: case 80:
LAUNCH_ATTENTION_KERNEL(T, 80, BLOCK_SIZE, NUM_THREADS); LAUNCH_PAGED_ATTENTION_V2(80);
break; break;
case 96: case 96:
LAUNCH_ATTENTION_KERNEL(T, 96, BLOCK_SIZE, NUM_THREADS); LAUNCH_PAGED_ATTENTION_V2(96);
break; break;
case 112: case 112:
LAUNCH_ATTENTION_KERNEL(T, 112, BLOCK_SIZE, NUM_THREADS); LAUNCH_PAGED_ATTENTION_V2(112);
break; break;
case 128: case 128:
LAUNCH_ATTENTION_KERNEL(T, 128, BLOCK_SIZE, NUM_THREADS); LAUNCH_PAGED_ATTENTION_V2(128);
break; break;
// case 160:
// LAUNCH_ATTENTION_KERNEL(T, 160, BLOCK_SIZE, NUM_THREADS);
// break;
// case 192:
// LAUNCH_ATTENTION_KERNEL(T, 192, BLOCK_SIZE, NUM_THREADS);
// break;
case 256: case 256:
LAUNCH_ATTENTION_KERNEL(T, 256, BLOCK_SIZE, NUM_THREADS); LAUNCH_PAGED_ATTENTION_V2(256);
break; break;
default: default:
TORCH_CHECK(false, "Unsupported head size: ", head_size); TORCH_CHECK(false, "Unsupported head size: ", head_size);
...@@ -447,9 +800,12 @@ void single_query_cached_kv_attention_launcher( ...@@ -447,9 +800,12 @@ void single_query_cached_kv_attention_launcher(
} }
} }
#define CALL_KERNEL_LAUNCHER(T, BLOCK_SIZE) \ #define CALL_V2_LAUNCHER(T, BLOCK_SIZE) \
single_query_cached_kv_attention_launcher<T, BLOCK_SIZE>( \ paged_attention_v2_launcher<T, BLOCK_SIZE>( \
out, \ out, \
exp_sums, \
max_logits, \
tmp_out, \
query, \ query, \
key_cache, \ key_cache, \
value_cache, \ value_cache, \
...@@ -462,42 +818,27 @@ void single_query_cached_kv_attention_launcher( ...@@ -462,42 +818,27 @@ void single_query_cached_kv_attention_launcher(
// NOTE(woosuk): To reduce the compilation time, we omitted block sizes // NOTE(woosuk): To reduce the compilation time, we omitted block sizes
// 1, 2, 4, 64, 128, 256. // 1, 2, 4, 64, 128, 256.
#define CALL_KERNEL_LAUNCHER_BLOCK_SIZE(T) \ #define CALL_V2_LAUNCHER_BLOCK_SIZE(T) \
switch (block_size) { \ switch (block_size) { \
/* case 1: */ \
/* CALL_KERNEL_LAUNCHER(T, 1); */ \
/* break; */ \
/* case 2: */ \
/* CALL_KERNEL_LAUNCHER(T, 2); */ \
/* break; */ \
/* case 4: */ \
/* CALL_KERNEL_LAUNCHER(T, 4); */ \
/* break; */ \
case 8: \ case 8: \
CALL_KERNEL_LAUNCHER(T, 8); \ CALL_V2_LAUNCHER(T, 8); \
break; \ break; \
case 16: \ case 16: \
CALL_KERNEL_LAUNCHER(T, 16); \ CALL_V2_LAUNCHER(T, 16); \
break; \ break; \
case 32: \ case 32: \
CALL_KERNEL_LAUNCHER(T, 32); \ CALL_V2_LAUNCHER(T, 32); \
break; \ break; \
/* case 64: */ \
/* CALL_KERNEL_LAUNCHER(T, 64); */ \
/* break; */ \
/* case 128: */ \
/* CALL_KERNEL_LAUNCHER(T, 128); */ \
/* break; */ \
/* case 256: */ \
/* CALL_KERNEL_LAUNCHER(T, 256); */ \
/* break; */ \
default: \ default: \
TORCH_CHECK(false, "Unsupported block size: ", block_size); \ TORCH_CHECK(false, "Unsupported block size: ", block_size); \
break; \ break; \
} }
void single_query_cached_kv_attention( void paged_attention_v2(
torch::Tensor& out, // [num_seqs, num_heads, head_size] torch::Tensor& out, // [num_seqs, num_heads, head_size]
torch::Tensor& exp_sums, // [num_seqs, num_heads, max_num_partitions]
torch::Tensor& max_logits, // [num_seqs, num_heads, max_num_partitions]
torch::Tensor& tmp_out, // [num_seqs, num_heads, max_num_partitions, head_size]
torch::Tensor& query, // [num_seqs, num_heads, head_size] torch::Tensor& query, // [num_seqs, num_heads, head_size]
torch::Tensor& key_cache, // [num_blocks, num_heads, head_size/x, block_size, x] torch::Tensor& key_cache, // [num_blocks, num_heads, head_size/x, block_size, x]
torch::Tensor& value_cache, // [num_blocks, num_heads, head_size, block_size] torch::Tensor& value_cache, // [num_blocks, num_heads, head_size, block_size]
...@@ -509,11 +850,11 @@ void single_query_cached_kv_attention( ...@@ -509,11 +850,11 @@ void single_query_cached_kv_attention(
int max_context_len, int max_context_len,
const c10::optional<torch::Tensor>& alibi_slopes) { const c10::optional<torch::Tensor>& alibi_slopes) {
if (query.dtype() == at::ScalarType::Float) { if (query.dtype() == at::ScalarType::Float) {
CALL_KERNEL_LAUNCHER_BLOCK_SIZE(float); CALL_V2_LAUNCHER_BLOCK_SIZE(float);
} else if (query.dtype() == at::ScalarType::Half) { } else if (query.dtype() == at::ScalarType::Half) {
CALL_KERNEL_LAUNCHER_BLOCK_SIZE(uint16_t); CALL_V2_LAUNCHER_BLOCK_SIZE(uint16_t);
} else if (query.dtype() == at::ScalarType::BFloat16) { } else if (query.dtype() == at::ScalarType::BFloat16) {
CALL_KERNEL_LAUNCHER_BLOCK_SIZE(__nv_bfloat16); CALL_V2_LAUNCHER_BLOCK_SIZE(__nv_bfloat16);
} else { } else {
TORCH_CHECK(false, "Unsupported data type: ", query.dtype()); TORCH_CHECK(false, "Unsupported data type: ", query.dtype());
} }
...@@ -522,3 +863,4 @@ void single_query_cached_kv_attention( ...@@ -522,3 +863,4 @@ void single_query_cached_kv_attention(
#undef WARP_SIZE #undef WARP_SIZE
#undef MAX #undef MAX
#undef MIN #undef MIN
#undef DIVIDE_ROUND_UP
csrc/attention/dtype_bfloat16.cuh
View file @ 928de468
...@@ -420,6 +420,11 @@ inline __device__ void from_float(bf16_8_t& dst, Float8_ src) { ...@@ -420,6 +420,11 @@ inline __device__ void from_float(bf16_8_t& dst, Float8_ src) {
#endif #endif
} }
// From bfloat16 to float32.
inline __device__ float to_float(__nv_bfloat16 u) {
return __bfloat162float(u);
}
// Zero-out a variable. // Zero-out a variable.
inline __device__ void zero(__nv_bfloat16& dst) { inline __device__ void zero(__nv_bfloat16& dst) {
#if defined(__CUDA_ARCH__) && __CUDA_ARCH__ < 800 #if defined(__CUDA_ARCH__) && __CUDA_ARCH__ < 800
......
tests/kernels/test_attention.py
View file @ 928de468
...@@ -14,13 +14,14 @@ FLOAT32_BYTES = torch.finfo(torch.float).bits // 8 ...@@ -14,13 +14,14 @@ FLOAT32_BYTES = torch.finfo(torch.float).bits // 8
# - 512 as a buffer # - 512 as a buffer
MAX_SEQ_LEN = get_max_shared_memory_bytes() // FLOAT32_BYTES - 512 MAX_SEQ_LEN = get_max_shared_memory_bytes() // FLOAT32_BYTES - 512
NUM_BLOCKS = 128 # Arbitrary values for testing NUM_BLOCKS = 128 # Arbitrary values for testing
PARTITION_SIZE = 512
DTYPES = [torch.half, torch.bfloat16, torch.float] DTYPES = [torch.half, torch.bfloat16, torch.float]
NUM_GEN_SEQS = [7] # Arbitrary values for testing NUM_GEN_SEQS = [7] # Arbitrary values for testing
NUM_PREFILL_SEQS = [1, 3, 7] # Arbitrary values for testing NUM_PREFILL_SEQS = [3] # Arbitrary values for testing
NUM_HEADS = [(40, 40), (64, 8)] # Arbitrary values for testing NUM_HEADS = [(40, 40), (64, 8)] # Arbitrary values for testing
HEAD_SIZES = [64, 80, 96, 112, 128, 256] HEAD_SIZES = [64, 80, 96, 112, 128, 256]
BLOCK_SIZES = [8, 16, 32] BLOCK_SIZES = [16, 32]
USE_ALIBI = [False, True] USE_ALIBI = [False, True]
SEEDS = [0] SEEDS = [0]
...@@ -96,6 +97,7 @@ def ref_single_query_cached_kv_attention( ...@@ -96,6 +97,7 @@ def ref_single_query_cached_kv_attention(
output[i].copy_(out, non_blocking=True) output[i].copy_(out, non_blocking=True)
@pytest.mark.parametrize("version", ["v1", "v2"])
@pytest.mark.parametrize("num_seqs", NUM_GEN_SEQS) @pytest.mark.parametrize("num_seqs", NUM_GEN_SEQS)
@pytest.mark.parametrize("num_heads", NUM_HEADS) @pytest.mark.parametrize("num_heads", NUM_HEADS)
@pytest.mark.parametrize("head_size", HEAD_SIZES) @pytest.mark.parametrize("head_size", HEAD_SIZES)
...@@ -103,9 +105,9 @@ def ref_single_query_cached_kv_attention( ...@@ -103,9 +105,9 @@ def ref_single_query_cached_kv_attention(
@pytest.mark.parametrize("block_size", BLOCK_SIZES) @pytest.mark.parametrize("block_size", BLOCK_SIZES)
@pytest.mark.parametrize("dtype", DTYPES) @pytest.mark.parametrize("dtype", DTYPES)
@pytest.mark.parametrize("seed", SEEDS) @pytest.mark.parametrize("seed", SEEDS)
@torch.inference_mode() def test_paged_attention(
def test_single_query_cached_kv_attention(
kv_cache_factory, kv_cache_factory,
version: str,
num_seqs: int, num_seqs: int,
num_heads: Tuple[int, int], num_heads: Tuple[int, int],
head_size: int, head_size: int,
...@@ -162,19 +164,54 @@ def test_single_query_cached_kv_attention( ...@@ -162,19 +164,54 @@ def test_single_query_cached_kv_attention(
# Call the paged attention kernel. # Call the paged attention kernel.
output = torch.empty_like(query) output = torch.empty_like(query)
attention_ops.single_query_cached_kv_attention( if version == "v1":
output, attention_ops.paged_attention_v1(
query, output,
key_cache, query,
value_cache, key_cache,
head_mapping, value_cache,
scale, head_mapping,
block_tables, scale,
context_lens, block_tables,
block_size, context_lens,
max_context_len, block_size,
alibi_slopes, max_context_len,
) alibi_slopes,
)
elif version == "v2":
num_partitions = ((max_context_len + PARTITION_SIZE - 1) //
PARTITION_SIZE)
assert PARTITION_SIZE % block_size == 0
num_seqs, num_heads, head_size = output.shape
tmp_output = torch.empty(
size=(num_seqs, num_heads, num_partitions, head_size),
dtype=output.dtype,
device=output.device,
)
exp_sums = torch.empty(
size=(num_seqs, num_heads, num_partitions),
dtype=torch.float32,
device=output.device,
)
max_logits = torch.empty_like(exp_sums)
attention_ops.paged_attention_v2(
output,
exp_sums,
max_logits,
tmp_output,
query,
key_cache,
value_cache,
head_mapping,
scale,
block_tables,
context_lens,
block_size,
max_context_len,
alibi_slopes,
)
else:
assert False, f"Unknown version: {version}"
# Run the reference implementation. # Run the reference implementation.
ref_output = torch.empty_like(query) ref_output = torch.empty_like(query)
......
vllm/model_executor/layers/attention.py
View file @ 928de468
...@@ -15,6 +15,8 @@ from vllm.model_executor.layers.rotary_embedding import ( ...@@ -15,6 +15,8 @@ from vllm.model_executor.layers.rotary_embedding import (
RotaryEmbedding) RotaryEmbedding)
_SUPPORTED_HEAD_SIZES = [64, 80, 96, 112, 128, 256] _SUPPORTED_HEAD_SIZES = [64, 80, 96, 112, 128, 256]
# Should be the same as PARTITION_SIZE in `paged_attention_v2_launcher`.
_PARTITION_SIZE = 512
class PagedAttention(nn.Module): class PagedAttention(nn.Module):
...@@ -130,6 +132,14 @@ class PagedAttention(nn.Module): ...@@ -130,6 +132,14 @@ class PagedAttention(nn.Module):
output.copy_(out.squeeze(0)) output.copy_(out.squeeze(0))
return output return output
def get_alibi_slopes(self) -> Optional[torch.Tensor]:
"""Returns the slopes for the alibi attention bias.
Returns:
slopes: shape = [num_heads]
"""
return None
def single_query_cached_kv_attention( def single_query_cached_kv_attention(
self, self,
output: torch.Tensor, output: torch.Tensor,
...@@ -137,6 +147,7 @@ class PagedAttention(nn.Module): ...@@ -137,6 +147,7 @@ class PagedAttention(nn.Module):
key_cache: torch.Tensor, key_cache: torch.Tensor,
value_cache: torch.Tensor, value_cache: torch.Tensor,
input_metadata: InputMetadata, input_metadata: InputMetadata,
alibi_slopes: Optional[torch.Tensor],
) -> None: ) -> None:
"""PagedAttention for the generation tokens. """PagedAttention for the generation tokens.
...@@ -148,21 +159,65 @@ class PagedAttention(nn.Module): ...@@ -148,21 +159,65 @@ class PagedAttention(nn.Module):
value_cache: shape = [num_blocks, num_kv_heads, head_size, value_cache: shape = [num_blocks, num_kv_heads, head_size,
block_size] block_size]
input_metadata: metadata for paged attention. input_metadata: metadata for paged attention.
alibi_slopes: shape = [num_heads]
""" """
block_size = value_cache.shape[3] block_size = value_cache.shape[3]
attention_ops.single_query_cached_kv_attention( num_seqs, num_heads, head_size = query.shape
output, max_num_partitions = (
query, (input_metadata.max_context_len + _PARTITION_SIZE - 1) //
key_cache, _PARTITION_SIZE)
value_cache, # NOTE(woosuk): We use a simple heuristic to decide whether to use
self.head_mapping, # PagedAttention V1 or V2. If the number of partitions is 1, we use
self.scale, # V1 to avoid the overhead of reduction. Also, if the number of
input_metadata.block_tables, # sequences or heads is large, we use V1 since there is enough work
input_metadata.context_lens, # to parallelize.
block_size, # TODO(woosuk): Tune this heuristic.
input_metadata.max_context_len, use_v1 = max_num_partitions == 1 or num_seqs * num_heads > 512
None, # alibi_slopes if use_v1:
) # Run PagedAttention V1.
attention_ops.paged_attention_v1(
output,
query,
key_cache,
value_cache,
self.head_mapping,
self.scale,
input_metadata.block_tables,
input_metadata.context_lens,
block_size,
input_metadata.max_context_len,
alibi_slopes,
)
else:
# Run PagedAttention V2.
assert _PARTITION_SIZE % block_size == 0
tmp_output = torch.empty(
size=(num_seqs, num_heads, max_num_partitions, head_size),
dtype=output.dtype,
device=output.device,
)
exp_sums = torch.empty(
size=(num_seqs, num_heads, max_num_partitions),
dtype=torch.float32,
device=output.device,
)
max_logits = torch.empty_like(exp_sums)
attention_ops.paged_attention_v2(
output,
exp_sums,
max_logits,
tmp_output,
query,
key_cache,
value_cache,
self.head_mapping,
self.scale,
input_metadata.block_tables,
input_metadata.context_lens,
block_size,
input_metadata.max_context_len,
alibi_slopes,
)
def forward( def forward(
self, self,
...@@ -253,7 +308,7 @@ class PagedAttention(nn.Module): ...@@ -253,7 +308,7 @@ class PagedAttention(nn.Module):
self.single_query_cached_kv_attention( self.single_query_cached_kv_attention(
output[num_prompt_tokens:num_valid_tokens], output[num_prompt_tokens:num_valid_tokens],
query[num_prompt_tokens:num_valid_tokens], key_cache, query[num_prompt_tokens:num_valid_tokens], key_cache,
value_cache, input_metadata) value_cache, input_metadata, self.get_alibi_slopes())
# Reshape the output tensor. # Reshape the output tensor.
# NOTE(woosuk): The output tensor may include paddings. # NOTE(woosuk): The output tensor may include paddings.
...@@ -431,36 +486,5 @@ class PagedAttentionWithALiBi(PagedAttention): ...@@ -431,36 +486,5 @@ class PagedAttentionWithALiBi(PagedAttention):
start += prompt_len start += prompt_len
return output return output
def single_query_cached_kv_attention( def get_alibi_slopes(self) -> Optional[torch.Tensor]:
self, return self.alibi_slopes
output: torch.Tensor,
query: torch.Tensor,
key_cache: torch.Tensor,
value_cache: torch.Tensor,
input_metadata: InputMetadata,
) -> None:
"""PagedAttention with ALiBi bias for the generation tokens.
Args:
output: shape = [num_generation_tokens, num_heads, head_size]
query: shape = [num_generation_tokens, num_heads, head_size]
key_cache: shape = [num_blocks, num_kv_heads, head_size/x,
block_size, x]
value_cache: shape = [num_blocks, num_kv_heads, head_size,
block_size]
input_metadata: metadata for paged attention.
"""
block_size = value_cache.shape[3]
attention_ops.single_query_cached_kv_attention(
output,
query,
key_cache,
value_cache,
self.head_mapping,
self.scale,
input_metadata.block_tables,
input_metadata.context_lens,
block_size,
input_metadata.max_context_len,
self.alibi_slopes,
)
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment