Skip to content

Qualcomm AI Engine Direct - Optimize QNN embedding op for llama #6725

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
wants to merge 2 commits into from
Closed

Qualcomm AI Engine Direct - Optimize QNN embedding op for llama #6725

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
73e2f7c
Qualcomm AI Engine Direct - Optimize QNN embedding op for llama
shewu-quic Nov 8, 2024
5aac8b6
set default value for use_int32_token
shewu-quic Nov 11, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
9 changes: 9 additions & 0 deletions examples/models/llama/export_llama_lib.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -519,6 +519,7 @@ def _prepare_for_llama_export(modelname: str, args) -> LLMEdgeManager:
checkpoint=checkpoint_path,
checkpoint_dir=checkpoint_dir,
params_path=params_path,
use_int32_token=True if args.qnn else False,
use_kv_cache=args.use_kv_cache,
use_sdpa_with_kv_cache=args.use_sdpa_with_kv_cache,
generate_full_logits=args.generate_full_logits,
Expand Down Expand Up @@ -746,6 +747,7 @@ def _export_llama(modelname, args) -> LLMEdgeManager: # noqa: C901

def _load_llama_model_metadata(
weight_type: WeightType,
use_int32_token: bool,
use_kv_cache: bool,
use_sdpa_with_kv_cache: bool,
enable_dynamic_shape: bool,
Expand All @@ -759,6 +761,7 @@ def _load_llama_model_metadata(
"get_max_seq_len": model_args.max_seq_len,
"get_n_layers": model_args.n_layers,
"get_vocab_size": model_args.vocab_size,
"use_int32_token": use_int32_token,
"use_kv_cache": use_kv_cache,
"use_sdpa_with_kv_cache": use_sdpa_with_kv_cache,
"enable_dynamic_shape": enable_dynamic_shape,
Expand All @@ -779,6 +782,7 @@ def _load_llama_model(
checkpoint: Optional[str] = None,
checkpoint_dir: Optional[str] = None,
params_path: str,
use_int32_token: bool = False,
use_kv_cache: bool = False,
use_sdpa_with_kv_cache: bool = False,
generate_full_logits: bool = False,
Expand Down Expand Up @@ -852,6 +856,10 @@ def _load_llama_model(
else:
raise ValueError(f"Unsupported dtype {dtype}")

if use_int32_token:
token = example_inputs[0].to(torch.int32)
example_inputs = (token,) + example_inputs[1:]

return LLMEdgeManager(
model=model,
modelname=modelname,
Expand All @@ -870,6 +878,7 @@ def _load_llama_model(
verbose=verbose,
metadata=_load_llama_model_metadata(
weight_type,
use_int32_token,
use_kv_cache,
use_sdpa_with_kv_cache,
enable_dynamic_shape,
Expand Down
4 changes: 4 additions & 0 deletions examples/models/llama/runner/runner.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -34,6 +34,7 @@ static constexpr auto kMaxSeqLen = "get_max_seq_len";
static constexpr auto kVocabSize = "get_vocab_size";
static constexpr auto kUseKVCache = "use_kv_cache";
static constexpr auto kUseSDPAWithKVCache = "use_sdpa_with_kv_cache";
static constexpr auto kUseInt32Token = "use_int32_token";
} // namespace

Runner::Runner(
Expand All @@ -51,6 +52,7 @@ Runner::Runner(
{kMaxSeqLen, 128},
{kUseKVCache, true},
{kUseSDPAWithKVCache, false},
{kUseInt32Token, true},
}) {
ET_LOG(
Info,
Expand Down Expand Up @@ -127,12 +129,14 @@ Error Runner::load() {
temperature_);
text_prefiller_ = std::make_unique<llm::TextPrefiller>(
text_decoder_runner_.get(),
metadata_.at(kUseInt32Token),
metadata_.at(kUseKVCache),
metadata_.at(kEnableDynamicShape));

text_token_generator_ = std::make_unique<llm::TextTokenGenerator>(
tokenizer_.get(),
text_decoder_runner_.get(),
metadata_.at(kUseInt32Token),
metadata_.at(kUseKVCache),
std::move(eos_ids),
&stats_);
Expand Down
13 changes: 8 additions & 5 deletions extension/llm/runner/text_prefiller.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,9 +17,11 @@ namespace llm {

TextPrefiller::TextPrefiller(
TextDecoderRunner* text_decoder_runner,
bool use_int32_token,
Copy link
Contributor

@cccclai cccclai Nov 11, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

oh I meant more like

TextPrefiller::TextPrefiller(
    TextDecoderRunner* text_decoder_runner,
    bool use_kv_cache,
    bool enable_parallel_prefill,
    bool use_int32_token=False)

so we don't need to update all callsite...

Copy link
Collaborator Author

@shewu-quic shewu-quic Nov 11, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Sorry about misunderstanding. I have updated. Thanks!

bool use_kv_cache,
bool enable_parallel_prefill)
: text_decoder_runner_(text_decoder_runner),
use_int32_token_(use_int32_token),
use_kv_cache_(use_kv_cache),
enable_parallel_prefill_(enable_parallel_prefill) {}

Expand All @@ -36,12 +38,13 @@ ::executorch::runtime::Result<uint64_t> TextPrefiller::prefill(

// store the token
uint64_t cur_token;
exec_aten::ScalarType token_type = use_int32_token_
? exec_aten::ScalarType::Int
: exec_aten::ScalarType::Long;
if (enable_parallel_prefill_ || !use_kv_cache_) {
// initialize tensor wrappers
auto tokens = from_blob(
prompt_tokens.data(),
{1, num_prompt_tokens},
exec_aten::ScalarType::Long);
auto tokens =
from_blob(prompt_tokens.data(), {1, num_prompt_tokens}, token_type);

auto start_pos_tensor =
from_blob(&start_pos, {1}, exec_aten::ScalarType::Long);
Expand All @@ -60,7 +63,7 @@ ::executorch::runtime::Result<uint64_t> TextPrefiller::prefill(
cur_token = prompt_tokens[0];

// initialize tensor wrappers
auto tokens = from_blob(&cur_token, {1, 1}, exec_aten::ScalarType::Long);
auto tokens = from_blob(&cur_token, {1, 1}, token_type);

auto start_pos_tensor =
from_blob(&start_pos, {1}, exec_aten::ScalarType::Long);
Expand Down
2 changes: 2 additions & 0 deletions extension/llm/runner/text_prefiller.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -24,6 +24,7 @@ class ET_EXPERIMENTAL TextPrefiller {
public:
TextPrefiller(
TextDecoderRunner* text_decoder_runner,
bool use_int32_token,
bool use_kv_cache_,
bool enable_parallel_prefill);
/**
Expand All @@ -40,6 +41,7 @@ class ET_EXPERIMENTAL TextPrefiller {

private:
TextDecoderRunner* text_decoder_runner_;
bool use_int32_token_;
bool use_kv_cache_;
bool enable_parallel_prefill_;
};
Expand Down
9 changes: 7 additions & 2 deletions extension/llm/runner/text_token_generator.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,12 +23,14 @@ class ET_EXPERIMENTAL TextTokenGenerator {
TextTokenGenerator(
Tokenizer* tokenizer,
TextDecoderRunner* text_decoder_runner,
bool use_int32_token,
Copy link
Contributor

@cccclai cccclai Nov 11, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This is causing BC breaking and causes CI failure. Can we set a default value so it's BC compatible? Also assert when it doesn't match metadata.

Copy link
Contributor

@cccclai cccclai Nov 11, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The CI breaking stack trace can be found here

bool use_kv_cache,
std::unique_ptr<std::unordered_set<uint64_t>>&& eos_ids,
Stats* stats)
: tokenizer_(tokenizer),
text_decoder_runner_(text_decoder_runner),
eos_ids_(std::move(eos_ids)),
use_int32_token_(use_int32_token),
use_kv_cache_(use_kv_cache),
stats_(stats) {}

Expand All @@ -54,6 +56,9 @@ class ET_EXPERIMENTAL TextTokenGenerator {

std::vector<uint64_t> token_data; // allocate space for the tokens
std::vector<executorch::aten::SizesType> token_shape;
exec_aten::ScalarType token_type = use_int32_token_
? exec_aten::ScalarType::Int
: exec_aten::ScalarType::Long;

// Token after prefill
uint64_t cur_token = tokens.back();
Expand All @@ -70,8 +75,7 @@ class ET_EXPERIMENTAL TextTokenGenerator {
}

// initialize tensor wrappers
auto tokens_managed = from_blob(
token_data.data(), token_shape, executorch::aten::ScalarType::Long);
auto tokens_managed = from_blob(token_data.data(), token_shape, token_type);
auto start_pos_managed =
from_blob(&pos, {1}, executorch::aten::ScalarType::Long);

Expand Down Expand Up @@ -133,6 +137,7 @@ class ET_EXPERIMENTAL TextTokenGenerator {
Tokenizer* tokenizer_;
TextDecoderRunner* text_decoder_runner_;
std::unique_ptr<std::unordered_set<uint64_t>> eos_ids_;
bool use_int32_token_;
bool use_kv_cache_;

// state machine
Expand Down
Loading