Unverified Commit 52a34d74 authored by Ying Sheng's avatar Ying Sheng Committed by GitHub
Browse files

Add greedy verification kernel (#4383)

parent 06d12b39
...@@ -17,6 +17,8 @@ ...@@ -17,6 +17,8 @@
#include <ATen/ATen.h> #include <ATen/ATen.h>
#include <ATen/cuda/CUDAContext.h> #include <ATen/cuda/CUDAContext.h>
#include "pytorch_extension_utils.h"
// parent_list [bs, topk * (depth - 1) + 1)] // parent_list [bs, topk * (depth - 1) + 1)]
// selected_index [bs, draft_token_num - 1] // selected_index [bs, draft_token_num - 1]
// verified_seq_len [bs] // verified_seq_len [bs]
...@@ -72,8 +74,8 @@ __global__ void build_tree_efficient( ...@@ -72,8 +74,8 @@ __global__ void build_tree_efficient(
} }
if (parent_position == draft_token_num) { if (parent_position == draft_token_num) {
printf( printf(
"ERROR: invalid eagle tree!!! Detected a token with no parent token selected. Check the logprob. The token " "WARNING: invalid eagle tree!!! Detected a token with no parent token selected. "
"will be dropped."); "Please check if the logprob has nan. The token will be ignored to keep proceeding.\n");
continue; continue;
} }
...@@ -140,112 +142,141 @@ void build_tree_kernel_efficient( ...@@ -140,112 +142,141 @@ void build_tree_kernel_efficient(
int32_t(draft_token_num)); int32_t(draft_token_num));
} }
// parent_list [bs, topk * (depth - 1) + 1)] template <typename IdType>
// selected_index [bs, draft_token_num - 1] __global__ void VerifyTreeGreedy(
// verified_seq_len [bs] IdType* predicts,
// tree_mask [draft_token*(seq_len[0]+draft_token) | draft_token*(seq_len[1]+draft_token) | ..] = IdType* accept_index,
// [sum(verified_seq_len)*draft_token+bs*draft_token*draft_token] positions [bs * draft_token] retrive_index [b, IdType* accept_token_num, // mutable
// draft_token, depth + 2] IdType* candidates,
__global__ void build_tree( IdType* retrive_index,
int64_t* parent_list, IdType* retrive_next_token,
int64_t* selected_index, IdType* retrive_next_sibling,
int32_t* verified_seq_len, IdType* target_predict,
bool* tree_mask, uint32_t batch_size,
int64_t* positions, uint32_t num_speculative_tokens,
int64_t* retrive_index, uint32_t num_draft_tokens) {
int topk, uint32_t bx = blockIdx.x;
int depth,
int draft_token_num) {
int bid = blockIdx.x;
int tid = threadIdx.x;
if (tid >= draft_token_num) {
return;
}
int seq_tree_idx = draft_token_num * draft_token_num * bid;
for (int i = 0; i < bid; i++) {
seq_tree_idx += verified_seq_len[i] * draft_token_num;
}
int seq_len = verified_seq_len[bid];
int token_tree_idx = seq_tree_idx + (seq_len + draft_token_num) * tid + seq_len + 1;
for (int i = 0; i < draft_token_num - 1; i++) {
tree_mask[token_tree_idx + i] = false;
}
int position = 0; IdType last_accepted_retrive_idx = retrive_index[bx * num_draft_tokens];
if (tid == 0) { accept_index[bx * num_speculative_tokens] = last_accepted_retrive_idx;
positions[bid * draft_token_num] = seq_len; uint32_t num_accepted_tokens = 0;
retrive_index[bid * draft_token_num * (depth + 2)] = bid * draft_token_num; IdType cur_index = 0;
return;
}
int depends_order[10]; for (uint32_t j = 1; j < num_speculative_tokens; ++j) {
cur_index = retrive_next_token[bx * num_draft_tokens + cur_index];
while (cur_index != -1) {
IdType draft_index = retrive_index[bx * num_draft_tokens + cur_index];
IdType draft_token_id = candidates[bx * num_draft_tokens + cur_index];
IdType target_token_id = target_predict[last_accepted_retrive_idx];
int cur_position = tid - 1; if (draft_token_id == target_token_id) {
while (true) { // accept token
depends_order[position] = cur_position + 1; predicts[last_accepted_retrive_idx] = target_token_id;
position += 1; ++num_accepted_tokens;
tree_mask[token_tree_idx + cur_position] = true; accept_index[bx * num_speculative_tokens + num_accepted_tokens] = draft_index;
int parent_tb_idx = selected_index[bid * (draft_token_num - 1) + cur_position] / topk; last_accepted_retrive_idx = draft_index;
if (parent_tb_idx == 0) {
break; break;
} else {
cur_index = retrive_next_sibling[bx * num_draft_tokens + cur_index];
}
} }
if (cur_index == -1) break;
}
accept_token_num[bx] = num_accepted_tokens;
predicts[last_accepted_retrive_idx] = target_predict[last_accepted_retrive_idx];
}
int token_idx = parent_list[bid * (topk * (depth - 1) + 1) + parent_tb_idx]; // predicts: [tot_num_draft_tokens]
for (cur_position = 0; cur_position < draft_token_num; cur_position++) { // accept_index: [bs, num_spec_step]
if (selected_index[bid * (draft_token_num - 1) + cur_position] == token_idx) { // accept_token_num: [bs]
break; // candidates: [bs, num_draft_tokens]
// retrive_index: [bs, num_draft_tokens]
// retrive_next_token: [bs, num_draft_tokens]
// retrive_next_sibling: [bs, num_draft_tokens]
// target_predict: [bs, num_draft_tokens]
void verify_tree_greedy(
at::Tensor predicts,
at::Tensor accept_index,
at::Tensor accept_token_num, // mutable
at::Tensor candidates,
at::Tensor retrive_index,
at::Tensor retrive_next_token,
at::Tensor retrive_next_sibling,
at::Tensor target_predict,
int64_t cuda_stream = 0) {
CHECK_INPUT(candidates);
CHECK_INPUT(retrive_index);
CHECK_INPUT(retrive_next_token);
CHECK_INPUT(retrive_next_sibling);
CHECK_INPUT(target_predict);
auto device = target_predict.device();
CHECK_EQ(candidates.device(), device);
CHECK_EQ(retrive_index.device(), device);
CHECK_EQ(retrive_next_token.device(), device);
CHECK_EQ(retrive_next_sibling.device(), device);
CHECK_EQ(target_predict.device(), device);
CHECK_DIM(1, predicts);
CHECK_DIM(2, accept_index);
CHECK_DIM(1, accept_token_num);
CHECK_DIM(2, candidates);
CHECK_DIM(2, retrive_index);
CHECK_DIM(2, retrive_next_token);
CHECK_DIM(2, retrive_next_sibling);
CHECK_DIM(2, target_predict);
unsigned int batch_size = candidates.size(0);
unsigned int num_spec_step = accept_index.size(1);
unsigned int num_draft_tokens = candidates.size(1);
CHECK_EQ(batch_size, accept_index.size(0));
CHECK_EQ(batch_size, accept_token_num.size(0));
CHECK_EQ(batch_size, retrive_index.size(0));
CHECK_EQ(batch_size, retrive_next_token.size(0));
CHECK_EQ(batch_size, retrive_next_sibling.size(0));
CHECK_EQ(batch_size, target_predict.size(0));
CHECK_EQ(num_draft_tokens, retrive_index.size(1));
CHECK_EQ(num_draft_tokens, retrive_next_token.size(1));
CHECK_EQ(num_draft_tokens, retrive_next_sibling.size(1));
CHECK_EQ(num_draft_tokens, target_predict.size(1));
CHECK_EQ(batch_size, accept_index.size(0));
CHECK_EQ(batch_size, accept_token_num.size(0));
if (predicts.scalar_type() != at::kInt) {
throw std::runtime_error("Expected 'predicts' to be of type int (torch.int32).");
} }
if (accept_index.scalar_type() != at::kInt) {
throw std::runtime_error("Expected 'accept_index' to be of type int (torch.int32).");
} }
if (cur_position == draft_token_num) { if (accept_token_num.scalar_type() != at::kInt) {
printf( throw std::runtime_error("Expected 'accept_token_num' to be of type int (torch.int32).");
"ERROR: invalid eagle tree!!! Detected a token with no parent token selected. Check the logprob. The token "
"will be dropped.");
break;
} }
if (candidates.scalar_type() != at::kInt) {
throw std::runtime_error("Expected 'candidates' to be of type int (torch.int32).");
} }
positions[bid * draft_token_num + tid] = position + seq_len; if (retrive_index.scalar_type() != at::kInt) {
throw std::runtime_error("Expected 'retrive_index' to be of type int (torch.int32).");
int is_leaf = 0;
for (int i = 1; i < draft_token_num; i++) {
if (tree_mask[seq_tree_idx + i * (draft_token_num + seq_len) + seq_len + tid]) {
is_leaf++;
} }
if (retrive_next_token.scalar_type() != at::kInt) {
throw std::runtime_error("Expected 'retrive_next_token' to be of type int (torch.int32).");
} }
if (is_leaf == 1) { if (retrive_next_sibling.scalar_type() != at::kInt) {
for (int i = 0; i < position; i++) { throw std::runtime_error("Expected 'retrive_next_sibling' to be of type int (torch.int32).");
retrive_index[(bid * (draft_token_num) + tid) * (depth + 2) + position - i] =
depends_order[i] + bid * draft_token_num;
} }
retrive_index[(bid * (draft_token_num) + tid) * (depth + 2)] = bid * draft_token_num; if (target_predict.scalar_type() != at::kInt) {
throw std::runtime_error("Expected 'target_predict' to be of type int (torch.int32).");
} }
}
void build_tree_kernel( cudaStream_t stream = reinterpret_cast<cudaStream_t>(cuda_stream);
at::Tensor parent_list, dim3 grid(batch_size);
at::Tensor selected_index, dim3 block(1);
at::Tensor verified_seq_len,
at::Tensor tree_mask,
at::Tensor positions,
at::Tensor retrive_index,
int64_t topk,
int64_t depth,
int64_t draft_token_num) {
// TODO (ying) check shape
// TODO (ying) check type
int bs = parent_list.size(0);
dim3 grid(bs);
dim3 block(draft_token_num);
const cudaStream_t stream = at::cuda::getCurrentCUDAStream();
build_tree<<<grid, block, 0, stream>>>( VerifyTreeGreedy<int><<<grid, block, 0, stream>>>(
static_cast<int64_t*>(parent_list.data_ptr()), static_cast<int*>(predicts.data_ptr()),
static_cast<int64_t*>(selected_index.data_ptr()), static_cast<int*>(accept_index.data_ptr()),
static_cast<int32_t*>(verified_seq_len.data_ptr()), static_cast<int*>(accept_token_num.data_ptr()),
static_cast<bool*>(tree_mask.data_ptr()), static_cast<int*>(candidates.data_ptr()),
static_cast<int64_t*>(positions.data_ptr()), static_cast<int*>(retrive_index.data_ptr()),
static_cast<int64_t*>(retrive_index.data_ptr()), static_cast<int*>(retrive_next_token.data_ptr()),
int32_t(topk), static_cast<int*>(retrive_next_sibling.data_ptr()),
int32_t(depth), static_cast<int*>(target_predict.data_ptr()),
int32_t(draft_token_num)); batch_size,
num_spec_step,
num_draft_tokens);
} }
// This is only a pluggin used for flashinfer 0.1.6. The new version does not need it.
/*
* Copyright (c) 2025 by SGLang team.
* Copyright (c) 2025 by FlashInfer team.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#include <flashinfer/quantization.cuh>
#include "pytorch_extension_utils.h"
using namespace flashinfer;
// bitorder = "little"
void segment_packbits(
at::Tensor x, at::Tensor input_indptr, at::Tensor output_indptr, at::Tensor y, int64_t cuda_stream) {
CHECK_INPUT(x);
CHECK_INPUT(input_indptr);
CHECK_INPUT(output_indptr);
auto device = x.device();
CHECK_EQ(input_indptr.device(), device);
CHECK_EQ(output_indptr.device(), device);
CHECK_EQ(y.device(), device);
unsigned int batch_size = input_indptr.size(0) - 1;
CHECK_EQ(output_indptr.size(0), batch_size + 1);
cudaStream_t stream = reinterpret_cast<cudaStream_t>(cuda_stream);
cudaError_t status = quantization::SegmentPackBits(
static_cast<bool*>(x.data_ptr()),
static_cast<uint8_t*>(y.data_ptr()),
static_cast<int32_t*>(input_indptr.data_ptr()),
static_cast<int32_t*>(output_indptr.data_ptr()),
batch_size,
quantization::BitOrder::kLittle,
stream);
}
...@@ -14,7 +14,6 @@ ...@@ -14,7 +14,6 @@
* See the License for the specific language governing permissions and * See the License for the specific language governing permissions and
* limitations under the License. * limitations under the License.
*/ */
#include "pytorch_extension_utils.h" #include "pytorch_extension_utils.h"
#include "speculative_sampling.cuh" #include "speculative_sampling.cuh"
...@@ -40,7 +39,9 @@ void tree_speculative_sampling_target_only( ...@@ -40,7 +39,9 @@ void tree_speculative_sampling_target_only(
at::Tensor uniform_samples, at::Tensor uniform_samples,
at::Tensor target_probs, at::Tensor target_probs,
at::Tensor draft_probs, at::Tensor draft_probs,
bool deterministic, double threshold_single,
double threshold_acc,
bool deterministic = true,
int64_t cuda_stream = 0) { int64_t cuda_stream = 0) {
CHECK_INPUT(candidates); CHECK_INPUT(candidates);
CHECK_INPUT(retrive_index); CHECK_INPUT(retrive_index);
...@@ -112,6 +113,10 @@ void tree_speculative_sampling_target_only( ...@@ -112,6 +113,10 @@ void tree_speculative_sampling_target_only(
if (draft_probs.scalar_type() != at::kFloat) { if (draft_probs.scalar_type() != at::kFloat) {
throw std::runtime_error("Expected 'target_probs' to be of type float (torch.float32)."); throw std::runtime_error("Expected 'target_probs' to be of type float (torch.float32).");
} }
CHECK_GE(threshold_single, 0);
CHECK_GE(1, threshold_single);
CHECK_GE(threshold_acc, 0);
CHECK_GE(1, threshold_acc);
cudaStream_t stream = reinterpret_cast<cudaStream_t>(cuda_stream); cudaStream_t stream = reinterpret_cast<cudaStream_t>(cuda_stream);
cudaError_t status = sampling::TreeSpeculativeSamplingTargetOnly<float, int>( cudaError_t status = sampling::TreeSpeculativeSamplingTargetOnly<float, int>(
...@@ -129,6 +134,8 @@ void tree_speculative_sampling_target_only( ...@@ -129,6 +134,8 @@ void tree_speculative_sampling_target_only(
num_spec_step, num_spec_step,
num_draft_tokens, num_draft_tokens,
vocab_size, vocab_size,
static_cast<float>(threshold_single),
static_cast<float>(threshold_acc),
deterministic, deterministic,
stream); stream);
......
...@@ -49,7 +49,9 @@ __global__ void TreeSpeculativeSamplingTargetOnly( ...@@ -49,7 +49,9 @@ __global__ void TreeSpeculativeSamplingTargetOnly(
uint32_t batch_size, uint32_t batch_size,
uint32_t num_speculative_tokens, uint32_t num_speculative_tokens,
uint32_t num_draft_tokens, uint32_t num_draft_tokens,
uint32_t d) { uint32_t d,
DType threshold_single,
DType threshold_acc) {
const uint32_t bx = blockIdx.x, tx = threadIdx.x; const uint32_t bx = blockIdx.x, tx = threadIdx.x;
extern __shared__ __align__(alignof(SamplingTempStorage<DType, BLOCK_THREADS, SCAN_ALGORITHM, REDUCE_ALGORITHM>)) extern __shared__ __align__(alignof(SamplingTempStorage<DType, BLOCK_THREADS, SCAN_ALGORITHM, REDUCE_ALGORITHM>))
...@@ -70,9 +72,10 @@ __global__ void TreeSpeculativeSamplingTargetOnly( ...@@ -70,9 +72,10 @@ __global__ void TreeSpeculativeSamplingTargetOnly(
while (cur_index != -1) { while (cur_index != -1) {
IdType draft_index = retrive_index[bx * num_draft_tokens + cur_index]; IdType draft_index = retrive_index[bx * num_draft_tokens + cur_index];
IdType draft_token_id = candidates[bx * num_draft_tokens + cur_index]; IdType draft_token_id = candidates[bx * num_draft_tokens + cur_index];
prob_acc += target_probs[cur_prob_offset + draft_token_id]; DType target_prob_single = target_probs[cur_prob_offset + draft_token_id];
prob_acc += target_prob_single;
if (coin < prob_acc) { if (coin <= prob_acc / threshold_acc || target_prob_single >= threshold_single) {
// accept token // accept token
prob_acc = 0.; prob_acc = 0.;
cur_prob_offset = (bx * num_draft_tokens + cur_index) * d; cur_prob_offset = (bx * num_draft_tokens + cur_index) * d;
...@@ -169,7 +172,9 @@ cudaError_t TreeSpeculativeSamplingTargetOnly( ...@@ -169,7 +172,9 @@ cudaError_t TreeSpeculativeSamplingTargetOnly(
uint32_t num_speculative_tokens, uint32_t num_speculative_tokens,
uint32_t num_draft_tokens, uint32_t num_draft_tokens,
uint32_t d, uint32_t d,
bool deterministic, DType threshold_single = 1,
DType threshold_acc = 1,
bool deterministic = true,
cudaStream_t stream = 0) { cudaStream_t stream = 0) {
constexpr uint32_t BLOCK_THREADS = 1024; constexpr uint32_t BLOCK_THREADS = 1024;
const uint32_t vec_size = std::gcd(16 / sizeof(DType), d); const uint32_t vec_size = std::gcd(16 / sizeof(DType), d);
...@@ -177,6 +182,7 @@ cudaError_t TreeSpeculativeSamplingTargetOnly( ...@@ -177,6 +182,7 @@ cudaError_t TreeSpeculativeSamplingTargetOnly(
const uint32_t smem_size = sizeof(SamplingTempStorage<DType, BLOCK_THREADS, SCAN_ALGO, REDUCE_ALGO>); const uint32_t smem_size = sizeof(SamplingTempStorage<DType, BLOCK_THREADS, SCAN_ALGO, REDUCE_ALGO>);
dim3 nblks(batch_size); dim3 nblks(batch_size);
dim3 nthrs(BLOCK_THREADS); dim3 nthrs(BLOCK_THREADS);
float capped_threshold_acc = fmaxf(threshold_acc, 1e-9f);
void* args[] = { void* args[] = {
&predicts, &predicts,
&output_token_ids, &output_token_ids,
...@@ -191,7 +197,9 @@ cudaError_t TreeSpeculativeSamplingTargetOnly( ...@@ -191,7 +197,9 @@ cudaError_t TreeSpeculativeSamplingTargetOnly(
&batch_size, &batch_size,
&num_speculative_tokens, &num_speculative_tokens,
&num_draft_tokens, &num_draft_tokens,
&d}; &d,
&threshold_single,
&capped_threshold_acc};
DISPATCH_ALIGNED_VEC_SIZE( DISPATCH_ALIGNED_VEC_SIZE(
vec_size, VEC_SIZE, {DISPATCH_DETERMINISTIC(deterministic, DETERMINISTIC, { vec_size, VEC_SIZE, {DISPATCH_DETERMINISTIC(deterministic, DETERMINISTIC, {
auto kernel = TreeSpeculativeSamplingTargetOnly< auto kernel = TreeSpeculativeSamplingTargetOnly<
......
...@@ -129,21 +129,24 @@ TORCH_LIBRARY_EXPAND(sgl_kernel, m) { ...@@ -129,21 +129,24 @@ TORCH_LIBRARY_EXPAND(sgl_kernel, m) {
"tree_speculative_sampling_target_only(Tensor! predicts, Tensor! accept_index, Tensor! accept_token_num, " "tree_speculative_sampling_target_only(Tensor! predicts, Tensor! accept_index, Tensor! accept_token_num, "
"Tensor candidates, Tensor retrive_index, Tensor retrive_next_token, Tensor retrive_next_sibling, " "Tensor candidates, Tensor retrive_index, Tensor retrive_next_token, Tensor retrive_next_sibling, "
"Tensor uniform_samples, Tensor target_probs, Tensor draft_probs, " "Tensor uniform_samples, Tensor target_probs, Tensor draft_probs, "
"float threshold_single, float threshold_acc, "
"bool deterministic, int cuda_stream) -> ()"); "bool deterministic, int cuda_stream) -> ()");
m.impl("tree_speculative_sampling_target_only", torch::kCUDA, &tree_speculative_sampling_target_only); m.impl("tree_speculative_sampling_target_only", torch::kCUDA, &tree_speculative_sampling_target_only);
m.def(
"verify_tree_greedy(Tensor! predicts, Tensor! accept_index, Tensor! accept_token_num, "
"Tensor candidates, Tensor retrive_index, Tensor retrive_next_token, Tensor retrive_next_sibling, "
"Tensor target_predict, int cuda_stream) -> ()");
m.impl("verify_tree_greedy", torch::kCUDA, &verify_tree_greedy);
m.def( m.def(
"build_tree_kernel_efficient(Tensor parent_list, Tensor selected_index, Tensor verified_seq_len, " "build_tree_kernel_efficient(Tensor parent_list, Tensor selected_index, Tensor verified_seq_len, "
"Tensor! tree_mask, Tensor! positions, Tensor! retrive_index, Tensor! retrive_next_token, Tensor! " "Tensor! tree_mask, Tensor! positions, Tensor! retrive_index, Tensor! retrive_next_token, "
"retrive_next_sibling, " "Tensor! retrive_next_sibling, int topk, int depth, int draft_token_num) -> ()");
"int topk, int depth, int draft_token_num) -> ()");
m.impl("build_tree_kernel_efficient", torch::kCUDA, &build_tree_kernel_efficient); m.impl("build_tree_kernel_efficient", torch::kCUDA, &build_tree_kernel_efficient);
m.def( m.def("segment_packbits(Tensor x, Tensor input_indptr, Tensor output_indptr, Tensor! y, int cuda_stream) -> ()");
"build_tree_kernel(Tensor parent_list, Tensor selected_index, Tensor verified_seq_len, " m.impl("segment_packbits", torch::kCUDA, &segment_packbits);
"Tensor! tree_mask, Tensor! positions, Tensor! retrive_index, "
"int topk, int depth, int draft_token_num) -> ()");
m.impl("build_tree_kernel", torch::kCUDA, &build_tree_kernel);
/* /*
* From FlashInfer * From FlashInfer
......
...@@ -183,8 +183,8 @@ void topk_softmax( ...@@ -183,8 +183,8 @@ void topk_softmax(
* From csrc/speculative * From csrc/speculative
*/ */
void tree_speculative_sampling_target_only( void tree_speculative_sampling_target_only(
at::Tensor predicts, at::Tensor predicts, // mutable
at::Tensor accept_index, at::Tensor accept_index, // mutable
at::Tensor accept_token_num, // mutable at::Tensor accept_token_num, // mutable
at::Tensor candidates, at::Tensor candidates,
at::Tensor retrive_index, at::Tensor retrive_index,
...@@ -193,33 +193,38 @@ void tree_speculative_sampling_target_only( ...@@ -193,33 +193,38 @@ void tree_speculative_sampling_target_only(
at::Tensor uniform_samples, at::Tensor uniform_samples,
at::Tensor target_probs, at::Tensor target_probs,
at::Tensor draft_probs, at::Tensor draft_probs,
double threshold_single = 1,
double threshold_acc = 1,
bool deterministic = true, bool deterministic = true,
int64_t cuda_stream = 0); int64_t cuda_stream = 0);
void build_tree_kernel_efficient( void verify_tree_greedy(
at::Tensor parent_list, at::Tensor predicts, // mutable
at::Tensor selected_index, at::Tensor accept_index, // mutable
at::Tensor verified_seq_len, at::Tensor accept_token_num, // mutable
at::Tensor tree_mask, at::Tensor candidates,
at::Tensor positions,
at::Tensor retrive_index, at::Tensor retrive_index,
at::Tensor retrive_next_token, at::Tensor retrive_next_token,
at::Tensor retrive_next_sibling, at::Tensor retrive_next_sibling,
int64_t topk, at::Tensor target_predict,
int64_t depth, int64_t cuda_stream = 0);
int64_t draft_token_num);
void build_tree_kernel( void build_tree_kernel_efficient(
at::Tensor parent_list, at::Tensor parent_list,
at::Tensor selected_index, at::Tensor selected_index,
at::Tensor verified_seq_len, at::Tensor verified_seq_len,
at::Tensor tree_mask, at::Tensor tree_mask,
at::Tensor positions, at::Tensor positions,
at::Tensor retrive_index, at::Tensor retrive_index,
at::Tensor retrive_next_token,
at::Tensor retrive_next_sibling,
int64_t topk, int64_t topk,
int64_t depth, int64_t depth,
int64_t draft_token_num); int64_t draft_token_num);
void segment_packbits(
at::Tensor x, at::Tensor input_indptr, at::Tensor output_indptr, at::Tensor y, int64_t cuda_stream);
/* /*
* From FlashInfer * From FlashInfer
*/ */
......
...@@ -42,8 +42,13 @@ from sgl_kernel.sampling import ( ...@@ -42,8 +42,13 @@ from sgl_kernel.sampling import (
top_p_sampling_from_probs, top_p_sampling_from_probs,
) )
from sgl_kernel.speculative import ( from sgl_kernel.speculative import (
build_tree_kernel,
build_tree_kernel_efficient, build_tree_kernel_efficient,
segment_packbits,
tree_speculative_sampling_target_only, tree_speculative_sampling_target_only,
verify_tree_greedy,
) )
from sgl_kernel.version import __version__ from sgl_kernel.version import __version__
build_tree_kernel = (
None # TODO(ying): remove this after updating the sglang python code.
)
...@@ -13,6 +13,8 @@ def tree_speculative_sampling_target_only( ...@@ -13,6 +13,8 @@ def tree_speculative_sampling_target_only(
uniform_samples: torch.Tensor, uniform_samples: torch.Tensor,
target_probs: torch.Tensor, target_probs: torch.Tensor,
draft_probs: torch.Tensor, draft_probs: torch.Tensor,
threshold_single: float = 1.0,
threshold_acc: float = 1.0,
deterministic: bool = True, deterministic: bool = True,
) -> None: ) -> None:
torch.ops.sgl_kernel.tree_speculative_sampling_target_only( torch.ops.sgl_kernel.tree_speculative_sampling_target_only(
...@@ -26,58 +28,74 @@ def tree_speculative_sampling_target_only( ...@@ -26,58 +28,74 @@ def tree_speculative_sampling_target_only(
uniform_samples, uniform_samples,
target_probs, target_probs,
draft_probs, draft_probs,
threshold_single,
threshold_acc,
deterministic, deterministic,
get_cuda_stream(), get_cuda_stream(),
) )
def build_tree_kernel_efficient( def verify_tree_greedy(
parent_list: torch.Tensor, predicts: torch.Tensor, # mutable
selected_index: torch.Tensor, accept_index: torch.Tensor, # mutable
verified_seq_len: torch.Tensor, accept_token_num: torch.Tensor, # mutable
tree_mask: torch.Tensor, candidates: torch.Tensor,
positions: torch.Tensor,
retrive_index: torch.Tensor, retrive_index: torch.Tensor,
retrive_next_token: torch.Tensor, retrive_next_token: torch.Tensor,
retrive_next_sibling: torch.Tensor, retrive_next_sibling: torch.Tensor,
topk: int, target_predict: torch.Tensor,
depth: int,
draft_token_num: int,
) -> None: ) -> None:
torch.ops.sgl_kernel.build_tree_kernel_efficient( torch.ops.sgl_kernel.verify_tree_greedy(
parent_list, predicts,
selected_index, accept_index,
verified_seq_len, accept_token_num,
tree_mask, candidates,
positions,
retrive_index, retrive_index,
retrive_next_token, retrive_next_token,
retrive_next_sibling, retrive_next_sibling,
topk, target_predict,
depth, get_cuda_stream(),
draft_token_num,
) )
def build_tree_kernel( def build_tree_kernel_efficient(
parent_list: torch.Tensor, parent_list: torch.Tensor,
selected_index: torch.Tensor, selected_index: torch.Tensor,
verified_seq_len: torch.Tensor, verified_seq_len: torch.Tensor,
tree_mask: torch.Tensor, tree_mask: torch.Tensor,
positions: torch.Tensor, positions: torch.Tensor,
retrive_index: torch.Tensor, retrive_index: torch.Tensor,
retrive_next_token: torch.Tensor,
retrive_next_sibling: torch.Tensor,
topk: int, topk: int,
depth: int, depth: int,
draft_token_num: int, draft_token_num: int,
) -> None: ) -> None:
torch.ops.sgl_kernel.build_tree_kernel( torch.ops.sgl_kernel.build_tree_kernel_efficient(
parent_list, parent_list,
selected_index, selected_index,
verified_seq_len, verified_seq_len,
tree_mask, tree_mask,
positions, positions,
retrive_index, retrive_index,
retrive_next_token,
retrive_next_sibling,
topk, topk,
depth, depth,
draft_token_num, draft_token_num,
) )
def segment_packbits(
x: torch.Tensor,
input_indptr: torch.Tensor,
output_indptr: torch.Tensor,
y: torch.Tensor,
) -> None:
torch.ops.sgl_kernel.segment_packbits(
x,
input_indptr,
output_indptr,
y,
torch.cuda.current_stream().cuda_stream,
)
...@@ -209,6 +209,7 @@ sources = [ ...@@ -209,6 +209,7 @@ sources = [
"csrc/moe/moe_topk_softmax_kernels.cu", "csrc/moe/moe_topk_softmax_kernels.cu",
"csrc/speculative/eagle_utils.cu", "csrc/speculative/eagle_utils.cu",
"csrc/speculative/speculative_sampling.cu", "csrc/speculative/speculative_sampling.cu",
"csrc/speculative/packbit.cu",
"csrc/torch_extension.cc", "csrc/torch_extension.cc",
"3rdparty/flashinfer/csrc/norm.cu", "3rdparty/flashinfer/csrc/norm.cu",
"3rdparty/flashinfer/csrc/renorm.cu", "3rdparty/flashinfer/csrc/renorm.cu",
......
import torch
import torch.nn.functional as F
from sgl_kernel import verify_tree_greedy
def test_verify_tree_greedy():
candidates = torch.tensor(
[
[0, 1, 2, 3, 4, 5],
[7, 8, 9, 10, 11, 12],
],
dtype=torch.int32,
device="cuda",
)
retrive_index = torch.tensor(
[
[0, 1, 2, 3, 4, 5],
[6, 7, 8, 9, 10, 11],
],
dtype=torch.int32,
device="cuda",
)
retrive_next_token = torch.tensor(
[
[1, 2, -1, 4, 5, -1],
[4, 2, 3, -1, 5, -1],
],
dtype=torch.int32,
device="cuda",
)
retrive_next_sibling = torch.tensor(
[
[-1, 3, -1, -1, -1, -1],
[-1, -1, -1, -1, 1, -1],
],
dtype=torch.int32,
device="cuda",
)
target_logits = torch.full((2, 6, 20), 1, dtype=torch.float32, device="cuda")
target_logits[0, 0, 3] = 10
target_logits[0, 3, 4] = 10
target_logits[0, 4, 5] = 10
target_logits[1, 0, 11] = 10
target_logits[1, 4, 12] = 10
for i in range(target_logits.shape[0]):
for j in range(target_logits.shape[1]):
if torch.max(target_logits[i][j]) < 10:
target_logits[i][j][18] = 10
print(f"{target_logits=}")
target_predict = torch.argmax(target_logits, dim=-1).to(torch.int32)
predict_shape = (12,)
bs = candidates.shape[0]
num_spec_step = 4
num_draft_tokens = candidates.shape[1]
predicts = torch.full(
predict_shape, -1, dtype=torch.int32, device="cuda"
) # mutable
accept_index = torch.full(
(bs, num_spec_step), -1, dtype=torch.int32, device="cuda"
) # mutable
accept_token_num = torch.full((bs,), 0, dtype=torch.int32, device="cuda") # mutable
print(f"{candidates=}")
print(f"{retrive_index=}")
print(f"{retrive_next_token=}")
print(f"{retrive_next_sibling=}")
print(f"{target_predict=}")
verify_tree_greedy(
predicts=predicts,
accept_index=accept_index,
accept_token_num=accept_token_num,
candidates=candidates,
retrive_index=retrive_index,
retrive_next_token=retrive_next_token,
retrive_next_sibling=retrive_next_sibling,
target_predict=target_predict,
)
print(f"{predicts=}")
print(f"{accept_index=}")
print(f"{accept_token_num=}")
return predicts, accept_index, accept_token_num
if __name__ == "__main__":
predicts, accept_index, accept_token_num = test_verify_tree_greedy()
assert predicts.tolist() == [3, -1, -1, 4, 5, 18, 11, -1, -1, -1, 12, 18]
assert accept_index.tolist() == [
[0, 3, 4, 5],
[6, 10, 11, -1],
]
assert accept_token_num.tolist() == [3, 2]
...@@ -3,7 +3,10 @@ import torch.nn.functional as F ...@@ -3,7 +3,10 @@ import torch.nn.functional as F
from sgl_kernel import tree_speculative_sampling_target_only from sgl_kernel import tree_speculative_sampling_target_only
def test_tree_speculative_sampling_target_only(): def test_tree_speculative_sampling_target_only(threshold_single=1, threshold_acc=1):
print(
f"\n============= run test: {threshold_single=} {threshold_acc=} ==============\n"
)
candidates = torch.tensor( candidates = torch.tensor(
[ [
[0, 1, 2, 3, 4, 5], [0, 1, 2, 3, 4, 5],
...@@ -37,7 +40,7 @@ def test_tree_speculative_sampling_target_only(): ...@@ -37,7 +40,7 @@ def test_tree_speculative_sampling_target_only():
device="cuda", device="cuda",
) )
target_logits = torch.zeros((2, 6, 20), dtype=torch.float32, device="cuda") target_logits = torch.full((2, 6, 20), 1, dtype=torch.float32, device="cuda")
target_logits[0, 0, 3] = 10 target_logits[0, 0, 3] = 10
target_logits[0, 3, 4] = 10 target_logits[0, 3, 4] = 10
target_logits[0, 4, 5] = 10 target_logits[0, 4, 5] = 10
...@@ -85,6 +88,8 @@ def test_tree_speculative_sampling_target_only(): ...@@ -85,6 +88,8 @@ def test_tree_speculative_sampling_target_only():
uniform_samples=coins, uniform_samples=coins,
target_probs=target_probs, target_probs=target_probs,
draft_probs=draft_probs, draft_probs=draft_probs,
threshold_single=threshold_single,
threshold_acc=threshold_acc,
deterministic=True, deterministic=True,
) )
...@@ -92,6 +97,13 @@ def test_tree_speculative_sampling_target_only(): ...@@ -92,6 +97,13 @@ def test_tree_speculative_sampling_target_only():
print(f"{accept_index=}") print(f"{accept_index=}")
print(f"{accept_token_num=}") print(f"{accept_token_num=}")
return predicts, accept_index, accept_token_num
if __name__ == "__main__":
predicts, accept_index, accept_token_num = (
test_tree_speculative_sampling_target_only(threshold_single=1, threshold_acc=1)
)
assert predicts.tolist() == [3, -1, -1, 4, 5, 18, 11, -1, -1, -1, 12, 18] assert predicts.tolist() == [3, -1, -1, 4, 5, 18, 11, -1, -1, -1, 12, 18]
assert accept_index.tolist() == [ assert accept_index.tolist() == [
[0, 3, 4, 5], [0, 3, 4, 5],
...@@ -99,6 +111,12 @@ def test_tree_speculative_sampling_target_only(): ...@@ -99,6 +111,12 @@ def test_tree_speculative_sampling_target_only():
] ]
assert accept_token_num.tolist() == [3, 2] assert accept_token_num.tolist() == [3, 2]
predicts, accept_index, accept_token_num = (
if __name__ == "__main__": test_tree_speculative_sampling_target_only(threshold_single=0, threshold_acc=0)
test_tree_speculative_sampling_target_only() )
assert predicts.tolist() == [1, 2, 18, -1, -1, -1, 11, -1, -1, -1, 12, 18]
assert accept_index.tolist() == [
[0, 1, 2, -1],
[6, 10, 11, -1],
]
assert accept_token_num.tolist() == [2, 2]
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