repo_name
stringlengths
1
62
dataset
stringclasses
1 value
lang
stringclasses
11 values
pr_id
int64
1
20.1k
owner
stringlengths
2
34
reviewer
stringlengths
2
39
diff_hunk
stringlengths
15
262k
code_review_comment
stringlengths
1
99.6k
openvino-ai-plugins-gimp
github_2023
python
91
intel
gblong1
@@ -113,89 +124,106 @@ def run(model_name,device_name): log.info('Initializing Inference Engine...') log.info('Model Path: %s',model_path ) - log.info('device_name: %s',device_name) + log.info('supported device_name: %s',device_name) + + device = ["CPU","GPU","GPU"] + device_int8 = ["CPU","GPU"...
remove un-needed commented lines.
openvino-ai-plugins-gimp
github_2023
python
91
intel
gblong1
@@ -232,78 +260,45 @@ def run(model_name,device_name): data_output = json.load(file) prompt = data_output["prompt"] - scheduler = data_output["scheduler"] + #scheduler = data_output["scheduler"] ...
Let's gate this on not being the LCM model, since the LCM model doesn't have a negative prompt: ``` if model_name != "SD_1.5_square_lcm": log.info('negative_prompt: %s',negative_prompt) ```
openvino-ai-plugins-gimp
github_2023
python
91
intel
gblong1
@@ -232,78 +260,45 @@ def run(model_name,device_name): data_output = json.load(file) prompt = data_output["prompt"] - scheduler = data_output["scheduler"] + #scheduler = data_output["scheduler"] ...
remove commented out line that is not needed.
openvino-ai-plugins-gimp
github_2023
python
91
intel
gblong1
@@ -414,19 +409,30 @@ def run(model_name,device_name): end_time = time.time() print("Image generated from Stable-Diffusion in ", end_time - start_time, " seconds.") + image = "sd_cache.png" + if model_name == "SD_1.5_squ...
commented line not needed
openvino-ai-plugins-gimp
github_2023
others
76
intel
RyanMetcalfeInt8
@@ -0,0 +1,258 @@ + + +# User guide for Windows Users + +## Install GIMP AI plugins with OpenVINO Backend on GIMP + +### Pre-requisite for execute NPU + +The following configurations are required: + +- Hardware: + - Intel Meteor Lake platform
Hmm, I wouldn't say that Meteor Lake is required. In general, these plugins will run on any configuration where OpenVINO is supported. So maybe give a link to supported OpenVINO configuration matrix here instead?
openvino-ai-plugins-gimp
github_2023
others
76
intel
RyanMetcalfeInt8
@@ -0,0 +1,258 @@ + + +# User guide for Windows Users + +## Install GIMP AI plugins with OpenVINO Backend on GIMP + +### Pre-requisite for execute NPU + +The following configurations are required: + +- Hardware: + - Intel Meteor Lake platform + - 16GB system memory as minimum requirement + - internet connection is ...
As per the comment above about Meteor Lake -- it should be clear that this requirement only applies to platforms that include NPU support (right now, only Intel Core Ultra)
openvino-ai-plugins-gimp
github_2023
others
76
intel
RyanMetcalfeInt8
@@ -0,0 +1,258 @@ + + +# User guide for Windows Users + +## Install GIMP AI plugins with OpenVINO Backend on GIMP + +### Pre-requisite for execute NPU + +The following configurations are required: + +- Hardware: + - Intel Meteor Lake platform + - 16GB system memory as minimum requirement + - internet connection is ...
1. Typo: 'download' 2. There's a mismatch between what is recommended just above this under 'Software and Packages', and here. Above gives a link to public distribution of 2023.1 OpenVINO, and here gives a link to Intel distribution. And this section is recommending 2023.2, while previous version recommended 2023.1.
openvino-ai-plugins-gimp
github_2023
python
4
intel
RyanMetcalfeInt8
@@ -49,6 +50,12 @@ def run(device,prompt,negative_prompt,num_infer_steps,guidance_scale,init_image, if seed is not None: np.random.seed(int(seed)) log.info('Seed: %s',seed) + else: + ran_seed = random.randrange(9999999999) #4294967294
😆 that's a big number
openvino-ai-plugins-gimp
github_2023
python
12
intel
arisha07
@@ -31,51 +30,20 @@ def setup_python_weights(install_location=None): step = 1 print("\n##########") - if os.name == "nt": # windows - print( - "{}>> Reference Models present in weights folder. You can also download them from open model zoo.".format( - step - )...
change "GIMP-OV" to "openvino-ai-plugins-gimp"
openvino-ai-plugins-gimp
github_2023
python
10
intel
RyanMetcalfeInt8
@@ -23,17 +23,19 @@ import os import random +from PIL import Image # scheduler -from diffusers import LMSDiscreteScheduler, PNDMScheduler +from diffusers.schedulers import LMSDiscreteScheduler, PNDMScheduler, EulerDiscreteScheduler # utils import cv2 import numpy as np +from gimpopenvino.tools.tools_utils imp...
So is the 'old' stable_diffusion_engine still used anywhere? If not, maybe it's better to replace stable_diffusion_engine.py with stable_diffusion_engine_NEW.py, so that we don't keep dead code around.
intel-lpmd
github_2023
others
29
intel
yselkowitz
@@ -55,24 +55,9 @@ GETTEXT_PACKAGE=intel_lpmd AC_SUBST([GETTEXT_PACKAGE]) AC_DEFINE_UNQUOTED([GETTEXT_PACKAGE], ["$GETTEXT_PACKAGE"], [Gettext package]) -dnl -dnl Checks for new dbus-glib property access function -dnl -PKG_CHECK_MODULES([DBUS_GLIB], [dbus-glib-1 >= 0.94], - [dbus_glib_global_set_disable_legacy_prop...
Why is glib 2.80 required?
intel-lpmd
github_2023
others
23
intel
eli-schwartz
@@ -1,4 +1,4 @@ -#!/bin/bash +#!/usr/bin/sh
> This script is pure POSIX sh so it do not need to be as bash script. This is the wrong solution to marking a script as POSIX sh. Please use /bin/sh exclusively for this, as that is the portable filename for the "sh" command.
intel-lpmd
github_2023
c
14
intel
zhang-rui
@@ -513,9 +526,11 @@ static int proc_message(message_capsul_t *msg) case LPM_FORCE_OFF: // Never enter LPM mode exit_lpm (USER_EXIT); + poll_for_wlt(0); break; case LPM_AUTO: // Enable oppotunistic LPM + poll_for_wlt(1);
Maybe we should hook poll_for_wlt() call into enter_lpm/exit_lpm to cover more cases. Say, we should disable wlt polling when switch from LPM_AUTO to LPM_FORCE_ON.
intel-lpmd
github_2023
c
14
intel
zhang-rui
@@ -188,6 +188,9 @@ static void update_one_cpu(struct perf_cap *perf_cap) if (perf_cap->cpu < 0) return; + if (!perf_cap->cpu) + reset_cpus (CPUMASK_HFI);
so here we assume 1, HFI table always start with CPU0 2. as a result, the netlink event always start with CPU0 right?
intel-lpmd
github_2023
c
14
intel
zhang-rui
@@ -190,7 +190,10 @@ static int lpm_can_process(enum lpm_command cmd) lpm_state &= ~LPM_USER_ON;
in changelog, s/exot/exit
intel-lpmd
github_2023
c
14
intel
zhang-rui
@@ -179,6 +189,7 @@ static int lpm_can_process(enum lpm_command cmd) case USER_AUTO: lpm_state &= ~LPM_USER_ON; lpm_state &= ~LPM_USER_OFF; + lpm_state |= LPM_UTIL_ON;
This means we always want utilization monitor in AUTO mode. Previously, if we use HFI hints only, then we never pull. So this is a functional change.
openvino-plugins-ai-audacity
github_2023
others
335
intel
gblong1
@@ -291,115 +291,26 @@ cd openvino-models wget https://storage.openvinotoolkit.org/repositories/open_model_zoo/2023.0/models_bin/1/noise-suppression-denseunet-ll-0001/FP16/noise-suppression-denseunet-ll-0001.xml wget https://storage.openvinotoolkit.org/repositories/open_model_zoo/2023.0/models_bin/1/noise-suppression...
Slashes are backwards for Linux in this section. Should be / vs current \ . Modifying this in the README will make cut-n-paste easier.
openvino-plugins-ai-audacity
github_2023
others
312
intel
RyanMetcalfeInt8
@@ -28,15 +31,26 @@ cd .. source setupvars.sh ``` * OpenVINO Tokenizers Extension - Download package from [here](https://storage.openvinotoolkit.org/repositories/openvino_tokenizers/packages/2024.3.0.0/). For these instructions, we will use ```openvino_tokenizers_ubuntu22_2024.3.0.0_x86_64.tar.gz```. +Again, if don'...
This cmd seems to assume that user has created ~/audacity-openvino and downloaded things there. Did you mean to do that?
openvino-plugins-ai-audacity
github_2023
others
312
intel
RyanMetcalfeInt8
@@ -131,8 +147,9 @@ git clone https://github.com/intel/openvino-plugins-ai-audacity.git ``` We need to copy the ```mod-openvino``` folder into the Audacity source tree. -i.e. Copy ```openvino-plugins-ai-audacity/mod-openvino``` folder to ```audacity/modules```. +i.e. Copy ```openvino-plugins-ai-audacity/mod-openvin...
Same as above comment about ~/audacity-openvino
openvino-plugins-ai-audacity
github_2023
cpp
243
intel
gblong1
@@ -291,6 +294,8 @@ bool EffectOVMusicGenerationLLM::Process(EffectInstance&, EffectSettings& settin // If they have chosen a stereo model, then most likely their intention / // expecation is to generate a stereo track. So, in that case we replace // the mono track that was added with a new stereo one. + ...
What is the FIXME comment referring to?
openvino-plugins-ai-audacity
github_2023
cpp
243
intel
gblong1
@@ -475,6 +475,8 @@ bool EffectOVNoiseSuppression::Process(EffectInstance&, EffectSettings&) return false; } + std::cout << "Going into WaveTrack Selection loop!" << std::endl;
Is this a debug statement that should be removed?
openvino-plugins-ai-audacity
github_2023
cpp
243
intel
gblong1
@@ -211,6 +212,7 @@ bool EffectOVMusicGenerationLLM::DoEffect( return EffectBase::DoEffect(settings, finder, projectRate, list, factory, selectedRegion, flags, pAccess); } +#if 0
Just noticed this one. #if 0 code should be removed, unless there is a strong reason to keep it
openvino-plugins-ai-audacity
github_2023
cpp
80
intel
gblong1
@@ -202,8 +337,10 @@ bool EffectOVMusicGenerationV2::Process(EffectInstance&, EffectSettings& setting continuation_context = ov_musicgen::MusicGenConfig::ContinuationContext::TEN_SECONDS; } + bool bFP16 = (model_selection_str.find("fp16") != std::string::npos);
Any risk of case sensitivity here? i think we do control the model name, but feels risky...
openvino-plugins-ai-audacity
github_2023
cpp
67
intel
gblong1
@@ -0,0 +1,390 @@ +// Copyright (C) 2024 Intel Corporation +// SPDX-License-Identifier: GPL-3.0-only + +#include "deepfilter.h" +#include "musicgen_utils.h" + + +namespace ov_deepfilternet +{ + DeepFilter::DeepFilter(std::string model_folder, std::string device, ModelSelection model_selection, std::optional<std::stri...
Any reason to keep ifdef 0 code?
openvino-plugins-ai-audacity
github_2023
cpp
67
intel
gblong1
@@ -0,0 +1,18 @@ +#pragma once + +#include <torch/torch.h> + +namespace ov_deepfilternet +{ + class DFNet + { + public: + + virtual torch::Tensor + forward(torch::Tensor spec, torch::Tensor feat_erb, torch::Tensor feat_spec) = 0; + + + virtual int64_t num_static_hops() = 0; + + }; +}
Missing newline at the end of this file?
ScalableVectorSearch
github_2023
python
96
intel
mihaic
@@ -281,12 +281,12 @@ def test_basic(self): self._test_basic(loader, matcher, first_iter = first_iter) first_iter = False - def test_deprecation(self): - with warnings.catch_warnings(record = True) as w: - p = svs.VamanaBuildParameters(num_threads = 1) - self....
If there is no particular reason to keep this commented code, let's remove it. ```suggestion ```
ScalableVectorSearch
github_2023
cpp
96
intel
ibhati
@@ -959,4 +962,74 @@ auto auto_assemble( index.apply(config); return index; } + +/// @brief Verify parameters and set defaults if needed +template <typename Dist> +void verify_or_set_default_index_parameters( + VamanaBuildParameters& parameters, Dist distance_function +) { + // Set default values + ...
```suggestion if (parameters.alpha <= 0.0f) { ```
ScalableVectorSearch
github_2023
cpp
96
intel
ibhati
@@ -16,10 +16,95 @@ // SVS #include "svs/orchestrators/vamana.h" +#include "svs/index/vamana/build_params.h" // Catch2 #include "catch2/catch_test_macros.hpp" +#include <catch2/catch_approx.hpp> + +// svsbenchmark +#include "svs-benchmark/benchmark.h" + +// tests +#include "tests/utils/test_dataset.h" +#include...
I think these tests should go to svs/index/vamana/index.cpp where the logic is implemented. Also, can we add similar tests for dynamic_index?
ScalableVectorSearch
github_2023
cpp
96
intel
mihaic
@@ -420,40 +421,21 @@ void wrap(py::module& m) { size_t window_size, size_t max_candidate_pool_size, size_t prune_to, - size_t num_threads) { - if (num_threads != std::numeric_limits<size_t>::max()) { - ...
It should be completely clear that we are using a placeholder value, so let's just use "placeholder" in the name. Please rename everywhere. ```suggestion py::arg("alpha") = svs::FLOAT_PLACEHOLDER, ```
ScalableVectorSearch
github_2023
cpp
96
intel
mihaic
@@ -420,40 +421,21 @@ void wrap(py::module& m) { size_t window_size, size_t max_candidate_pool_size, size_t prune_to, - size_t num_threads) { - if (num_threads != std::numeric_limits<size_t>::max()) { - ...
We are not setting `graph_max_degree` based on other parameters. The default is always 32. Let's leave it as 32 so users immediately see what the default is. ```suggestion py::arg("graph_max_degree") = 32, ```
ScalableVectorSearch
github_2023
cpp
96
intel
mihaic
@@ -420,40 +421,21 @@ void wrap(py::module& m) { size_t window_size, size_t max_candidate_pool_size, size_t prune_to, - size_t num_threads) { - if (num_threads != std::numeric_limits<size_t>::max()) { - ...
Same as `graph_max_degree`. ```suggestion py::arg("window_size") = 64, ```
ScalableVectorSearch
github_2023
cpp
96
intel
mihaic
@@ -420,40 +421,21 @@ void wrap(py::module& m) { size_t window_size, size_t max_candidate_pool_size, size_t prune_to, - size_t num_threads) { - if (num_threads != std::numeric_limits<size_t>::max()) { - ...
Same as alpha. ```suggestion py::arg("max_candidate_pool_size") = svs::UNSIGNED_INTEGER_PLACEHOLDER, ```
ScalableVectorSearch
github_2023
cpp
96
intel
mihaic
@@ -420,40 +421,21 @@ void wrap(py::module& m) { size_t window_size, size_t max_candidate_pool_size, size_t prune_to, - size_t num_threads) { - if (num_threads != std::numeric_limits<size_t>::max()) { - ...
```suggestion py::arg("prune_to") = svs::UNSIGNED_INTEGER_PLACEHOLDER, ```
ScalableVectorSearch
github_2023
cpp
96
intel
mihaic
@@ -159,3 +162,12 @@ inline constexpr bool have_avx512_avx2 = true; #endif } // namespace svs::arch + +namespace svs { +// Maximum values used as default initializers +inline constexpr size_t UNSIGNED_INTEGER_PLACEHOLDER = std::numeric_limits<size_t>::max(); +inline constexpr float FLOAT_PLACEHOLDER = std::numeric_...
Can you please look at the docs corresponding to the PR and confirm these `*_DEFAULT` values are visible? Users should be able to see the default value without reading the code. Ideally, there should be numbers visible in the `VamanaBuildParameters` doc [as currently](https://intel.github.io/ScalableVectorSearch/python...
ScalableVectorSearch
github_2023
cpp
96
intel
mihaic
@@ -420,40 +421,21 @@ void wrap(py::module& m) { size_t window_size, size_t max_candidate_pool_size, size_t prune_to, - size_t num_threads) { - if (num_threads != std::numeric_limits<size_t>::max()) { - ...
Could you please add a description in the docs (right below, PR line 442) for the `*_PLACEHOLDER` values? It could be something like this: > Default is `graph_max_degree * 2`.
ScalableVectorSearch
github_2023
cpp
96
intel
mihaic
@@ -959,4 +962,66 @@ auto auto_assemble( index.apply(config); return index; } + +/// @brief Verify parameters and set defaults if needed +template <typename Dist> +void verify_and_set_default_index_parameters( + VamanaBuildParameters& parameters, Dist distance_function +) { + // Set default values + ...
Good idea with the `*_DEFAULT` values! (I was only suggesting numerical values). What do you think about adding two more for the two alpha values for maintainability, e.g., `ALPHA_MAXIMIZE_DEFAULT` and `ALPHA_MINIMIZE_DEFAULT`? Feel free to say no.
ScalableVectorSearch
github_2023
cpp
96
intel
mihaic
@@ -420,40 +421,21 @@ void wrap(py::module& m) { size_t window_size, size_t max_candidate_pool_size, size_t prune_to, - size_t num_threads) { - if (num_threads != std::numeric_limits<size_t>::max()) {
This deprecation expiry should be noted in the release notes. I guess we don't have a formal way to record in-progress release notes (issue #98), so at lease it should be prominently noted in the PR description so it shows up in the PR squash commit message.
ScalableVectorSearch
github_2023
cpp
96
intel
ibhati
@@ -462,18 +448,26 @@ void wrap(py::module& m) { For distance types favoring minimization, set this to a number greater than 1.0 (typically, 1.2 is sufficient). For distance types preferring maximization, set to a value less than 1.0 (such as 0.95). + ...
```suggestion The default value is `graph_max_degree` * 2. ```
ScalableVectorSearch
github_2023
cpp
96
intel
ibhati
@@ -462,18 +448,26 @@ void wrap(py::module& m) { For distance types favoring minimization, set this to a number greater than 1.0 (typically, 1.2 is sufficient). For distance types preferring maximization, set to a value less than 1.0 (such as 0.95). + ...
```suggestion `graph_max_degree`. The default value is `graph_max_degree` - 4 if ```
ScalableVectorSearch
github_2023
cpp
96
intel
ibhati
@@ -462,18 +448,26 @@ void wrap(py::module& m) { For distance types favoring minimization, set this to a number greater than 1.0 (typically, 1.2 is sufficient). For distance types preferring maximization, set to a value less than 1.0 (such as 0.95). + ...
```suggestion `graph_max_degree` is at least 16, otherwise it equals `graph_max_degree`. ```
ScalableVectorSearch
github_2023
cpp
96
intel
dian-lun-lin
@@ -159,3 +162,14 @@ inline constexpr bool have_avx512_avx2 = true; #endif } // namespace svs::arch + +namespace svs {
Thank you for the good work. Since some of these parameters are vamana-specific, could we move these parameters to svs::index::vamana namespace? Or change the name to Vamana specific. For example, GRAPH_MAX_DEGREE_DEFAULT -> VAMANA_GRAPH_MAX_DEGREE_DEFAULT. ALPHA_*, USE_FULL_SEARCH_HISTORY_DEFAULT, WINDOW_SIZE...
ScalableVectorSearch
github_2023
cpp
96
intel
mihaic
@@ -462,18 +449,26 @@ void wrap(py::module& m) { For distance types favoring minimization, set this to a number greater than 1.0 (typically, 1.2 is sufficient). For distance types preferring maximization, set to a value less than 1.0 (such as 0.95). + ...
```suggestion The default value is 1.2 for L2 distance type and 0.95 for MIP/Cosine. ```
ScalableVectorSearch
github_2023
cpp
96
intel
mihaic
@@ -462,18 +449,26 @@ void wrap(py::module& m) { For distance types favoring minimization, set this to a number greater than 1.0 (typically, 1.2 is sufficient). For distance types preferring maximization, set to a value less than 1.0 (such as 0.95). + ...
```suggestion of a larger memory footprint. ```
ScalableVectorSearch
github_2023
cpp
96
intel
mihaic
@@ -462,18 +449,26 @@ void wrap(py::module& m) { For distance types favoring minimization, set this to a number greater than 1.0 (typically, 1.2 is sufficient). For distance types preferring maximization, set to a value less than 1.0 (such as 0.95). + ...
```suggestion ```
ScalableVectorSearch
github_2023
cpp
96
intel
mihaic
@@ -420,40 +421,26 @@ void wrap(py::module& m) { size_t window_size, size_t max_candidate_pool_size, size_t prune_to, - size_t num_threads) { - if (num_threads != std::numeric_limits<size_t>::max()) { - ...
```suggestion ```
ScalableVectorSearch
github_2023
cpp
96
intel
mihaic
@@ -420,40 +421,26 @@ void wrap(py::module& m) { size_t window_size, size_t max_candidate_pool_size, size_t prune_to, - size_t num_threads) { - if (num_threads != std::numeric_limits<size_t>::max()) { - ...
```suggestion ```
ScalableVectorSearch
github_2023
cpp
96
intel
mihaic
@@ -420,40 +421,26 @@ void wrap(py::module& m) { size_t window_size, size_t max_candidate_pool_size, size_t prune_to, - size_t num_threads) { - if (num_threads != std::numeric_limits<size_t>::max()) { - ...
```suggestion ```
ScalableVectorSearch
github_2023
cpp
96
intel
mihaic
@@ -462,18 +449,26 @@ void wrap(py::module& m) { For distance types favoring minimization, set this to a number greater than 1.0 (typically, 1.2 is sufficient). For distance types preferring maximization, set to a value less than 1.0 (such as 0.95). + ...
```suggestion The default value is ``graph_max_degree`` * 2. ```
ScalableVectorSearch
github_2023
cpp
96
intel
mihaic
@@ -462,18 +449,26 @@ void wrap(py::module& m) { For distance types favoring minimization, set this to a number greater than 1.0 (typically, 1.2 is sufficient). For distance types preferring maximization, set to a value less than 1.0 (such as 0.95). + ...
```suggestion ` `graph_max_degree`` - 4 if ``graph_max_degree`` is at least 16, otherwise ``graph_max_degree``. ```
ScalableVectorSearch
github_2023
cpp
96
intel
mihaic
@@ -462,18 +449,26 @@ void wrap(py::module& m) { For distance types favoring minimization, set this to a number greater than 1.0 (typically, 1.2 is sufficient). For distance types preferring maximization, set to a value less than 1.0 (such as 0.95). + ...
```suggestion ```
ScalableVectorSearch
github_2023
cpp
96
intel
mihaic
@@ -210,12 +214,20 @@ class MutableVamanaIndex { , distance_(std::move(distance_function)) , threadpool_(threads::as_threadpool(std::move(threadpool_proto))) , search_parameters_(vamana::construct_default_search_parameters(data_)) - , construction_window_size_(parameters.window_size) -...
Can you please rephrase this comment? I'm not sure what it means.
ScalableVectorSearch
github_2023
cpp
96
intel
mihaic
@@ -462,18 +449,26 @@ void wrap(py::module& m) { For distance types favoring minimization, set this to a number greater than 1.0 (typically, 1.2 is sufficient). For distance types preferring maximization, set to a value less than 1.0 (such as 0.95). + ...
```suggestion ``graph_max_degree`` will yield faster index building times. Default: ```
ScalableVectorSearch
github_2023
cpp
87
intel
dian-lun-lin
@@ -36,213 +36,270 @@ // tests #include "tests/utils/test_dataset.h" #include "tests/utils/utils.h" - -namespace { -template <typename T> auto copy_dataset(const T& data) { - auto copy = svs::data::SimplePolymorphicData<typename T::element_type, T::extent>{ - data.size(), data.dimensions()}; - for (size...
Why do we comment out this test?
ScalableVectorSearch
github_2023
cpp
87
intel
dian-lun-lin
@@ -104,7 +105,63 @@ CATCH_TEST_CASE("Vamana Index Parameters", "[index][vamana]") { CATCH_SECTION("Current version") { auto p = VamanaIndexParameters{ - 128, {12.4f, 478, 13, 4, 10, false}, {{10, 20}, true, 1, 1}}; + 128, {12.4f, 478, 13, 4, 10, false}, {{10, 20}, true, 1, 1} + ...
I'm not really sure the purpose of static in every `test_log_impl`. I think putting this function and the struct into [anonymous namespace](https://github.com/intel/ScalableVectorSearch/blob/a8221adaceebc493e23058694391cb14bc610b2b/tests/svs/index/vamana/index.cpp#L26) is enough?
ScalableVectorSearch
github_2023
cpp
87
intel
dian-lun-lin
@@ -36,213 +36,270 @@ // tests #include "tests/utils/test_dataset.h" #include "tests/utils/utils.h" - -namespace { -template <typename T> auto copy_dataset(const T& data) { - auto copy = svs::data::SimplePolymorphicData<typename T::element_type, T::extent>{ - data.size(), data.dimensions()}; - for (size...
Could you also add a test for not setting the callback (i.e., not calling `set_global_log_callback`)? Adding this test in dynamic index should be enough.
ScalableVectorSearch
github_2023
cpp
87
intel
dian-lun-lin
@@ -580,10 +593,10 @@ class MutableVamanaIndex { /// @brief Add the points with the given external IDs to the dataset. // /// When `delete_entries` is called, a soft deletion is performed, marking the entries - /// as `deleted`. When `consolidate` is called, the state of these deleted entries becomes ...
Just a side note that I notice clang-format keeps changing the format of this paragraph. We might want to consider disabling clang-format for this specific section.
ScalableVectorSearch
github_2023
cpp
87
intel
ibhati
@@ -60,14 +60,21 @@ inline constexpr std::array<Level, 7> all_levels = { Level::Warn, Level::Error, Level::Critical, - Level::Off}; + Level::Off +};
Can you please run these commands from the repository's root to ensure all the formatting is consistent? sudo apt install clang-format-15 bash tools/clang-format.sh clang-format-15
ScalableVectorSearch
github_2023
cpp
87
intel
ibhati
@@ -302,6 +309,13 @@ inline void set_level(Level level) { set_level(logger, level); } +/// @brief Function to set the global log callback +inline void set_global_log_callback(log_callback_function callback) { + global_log_callback = callback; + std::cout << "[DEBUG] set_global_log_callback(" << reinterpre...
I think we should remove the std::cout print statements.
ScalableVectorSearch
github_2023
cpp
87
intel
ibhati
@@ -328,6 +342,21 @@ void log(Level level, fmt::format_string<Args...> fmt, Args&&... args) { logging::log(global_logger, level, fmt, SVS_FWD(args)...); } +/// @brief Each index can pass its own ctx pointer, but +/// the function pointer is always global_log_callback. +inline void log(void* ctx, const ch...
Similarly, please remove the std::cout print statements from this section. Additionally, in the else part, I suggest calling the default log(...) function located above on line 340.
ScalableVectorSearch
github_2023
cpp
87
intel
ibhati
@@ -246,6 +243,8 @@ VamanaSearchParameters construct_default_search_parameters(const Data& data) { /// @tparam PostOp The cleanup operation that is performed after the initial graph /// search. This may be included to perform operations like reranking for quantized /// datasets. +/// @tparam log_callback_ctx...
We can remove the logger from here as it is not the template parameter
ScalableVectorSearch
github_2023
cpp
87
intel
rfsaliev
@@ -189,22 +192,36 @@ class FlatIndex { /// instance or an integer specifying the number of threads to use. In the latter /// case, a new default thread pool will be constructed using ``threadpool_proto`` /// as the number of threads to create. + /// @param log_callback_ctx A pointer to a ...
IMHO, `svs::logger_ptr` would be better.
ScalableVectorSearch
github_2023
cpp
87
intel
dian-lun-lin
@@ -414,10 +422,13 @@ class VamanaIndex { extensions::estimate_prefetch_parameters(data_) ); - builder.construct(1.0F, entry_point_[0]); - builder.construct(parameters.alpha, entry_point_[0]); + builder.construct(1.0F, entry_point_[0], logging::Level::Info, logger); + ...
Should we have this `get_logger` for each index?
ScalableVectorSearch
github_2023
cpp
87
intel
dian-lun-lin
@@ -199,3 +203,73 @@ CATCH_TEST_CASE( ++num_threads; } } + +// Helper function to create a logger with a callback +std::shared_ptr<spdlog::logger> create_test_logger(std::vector<std::string>& captured_logs +) { + auto callback_sink = std::make_shared<spdlog::sinks::callback_sink_mt>( + [&captur...
Please also test here to check if the global logger is not used.
ScalableVectorSearch
github_2023
cpp
87
intel
ibhati
@@ -940,7 +956,12 @@ auto auto_assemble( // Extract the index type of the provided graph. using I = typename decltype(graph)::index_type; auto index = VamanaIndex{ - std::move(graph), std::move(data), I{}, std::move(distance), std::move(threadpool)}; + std::move(graph), + std::move(d...
```suggestion std::move(logger)}; ``` should we move instead of copy?
ScalableVectorSearch
github_2023
others
90
intel
ibhati
@@ -1 +1,15 @@ # SVS 0.0.7 Release Notes + +## Additions and Changes + +* Implemented batch iterator support for hybrid search + +* Added support for custom threading and memory allocation + +* Introduced a timeout feature for search calls + +* Introduced `reuse_empty` flag in dynamic Vamana, enabling users to choose ...
maybe we can remove the last point as it's covered by custom threading (2nd point)
ScalableVectorSearch
github_2023
python
88
intel
mihaic
@@ -85,6 +85,7 @@ def resolve_compiler(name: str): aliases = { "GNU": "gcc", "Clang": "clang", + "IntelLLVM": "intel"
I think `intel` is the Intel® C++ Compiler Classic and that the name should be `oneapi` unless I misunderstood your setup. https://github.com/archspec/archspec-json/pull/47 ```suggestion "IntelLLVM": "oneapi" ```
ScalableVectorSearch
github_2023
cpp
78
intel
mihaic
@@ -45,4 +45,11 @@ template <typename T> inline T bound_with(T nearest, T epsilon, svs::DistanceIP) return nearest / (1 + epsilon); } +template <typename T> +inline T bound_with(T nearest, T epsilon, svs::DistanceCosineSimilarity) { + // TODO: This is just copy/paste from DistanceIP - is it correct?
@ibhati, do we need this TODO?
ScalableVectorSearch
github_2023
cpp
84
intel
ibhati
@@ -507,4 +507,121 @@ class MemoryMapper { return MMapPtr<void>(base, bytes.value()); } }; + +namespace detail { + +template <typename Alloc> +concept HasValueType = requires { typename Alloc::value_type; }; + +// clang-format off +template <typename Alloc> +concept Allocator = HasValueType<Alloc> && req...
Is there something unique about the AllocatorHandle when it comes to LVQ/Leanvec datasets? From what I understand, its primary use case is to enable these custom datasets to work with a shared library. However, I believe that the AllocatorHandle is more versatile and can be applied to a variety of other scenarios as we...
ScalableVectorSearch
github_2023
cpp
84
intel
ibhati
@@ -826,4 +826,44 @@ template <typename T, size_t Extent = Dynamic, typename Alloc = lib::Allocator<T using BlockedData = SimpleData<T, Extent, Blocked<Alloc>>; } // namespace data + +/// +/// @brief Creates a `Blocked` object with an `AllocatorHandle` from a given allocator. +/// +/// @tparam Alloc The type of the...
I think the order of the arguments in the doc string is reversed. First parameters and then alloc?
ScalableVectorSearch
github_2023
cpp
84
intel
ibhati
@@ -137,4 +137,64 @@ CATCH_TEST_CASE("Testing Allocator", "[allocators]") { } } } + + CATCH_SECTION("Testing `AllocatorHandle`") {
Should we add test for Blocked AllocatorHandle as well?
ScalableVectorSearch
github_2023
cpp
84
intel
rfsaliev
@@ -507,4 +507,121 @@ class MemoryMapper { return MMapPtr<void>(base, bytes.value()); } }; + +namespace detail { + +template <typename Alloc> +concept HasValueType = requires { typename Alloc::value_type; }; + +// clang-format off +template <typename Alloc> +concept Allocator = HasValueType<Alloc> && req...
IMHO, to avoid misusage, would be good to add requirements `same_as<T, Impl::value_type>`, elsewhere there is the risk of wrong amount of bytes allocation.
ScalableVectorSearch
github_2023
cpp
84
intel
rfsaliev
@@ -507,4 +507,121 @@ class MemoryMapper { return MMapPtr<void>(base, bytes.value()); } }; + +namespace detail { + +template <typename Alloc> +concept HasValueType = requires { typename Alloc::value_type; }; + +// clang-format off +template <typename Alloc> +concept Allocator = HasValueType<Alloc> && req...
As I understand, this means that `Impl` type should be 'copyable', i.e. `Impl::Impl(const Impl& other)` or `Impl::Impl(Impl other)` should be defined, right?
ScalableVectorSearch
github_2023
cpp
84
intel
rfsaliev
@@ -507,4 +507,121 @@ class MemoryMapper { return MMapPtr<void>(base, bytes.value()); } }; + +namespace detail { + +template <typename Alloc> +concept HasValueType = requires { typename Alloc::value_type; }; + +// clang-format off +template <typename Alloc> +concept Allocator = HasValueType<Alloc> && req...
nitpick: different allocation methods for `AllocatorImpl`: * In `AllocatorHandle` .ctor used `make_unique(...)` * When here is: `unique_ptr::unique_ptr(new ...)` I also do not understand, why C++ compiler allows this construction which looks like: ```cpp std::unique_ptr<AllocatorInterface>::operator=(AllocatorIn...
ScalableVectorSearch
github_2023
cpp
75
intel
mihaic
@@ -92,7 +92,8 @@ template <typename ElementType> void add_points( svs::DynamicVamana& index, const py_contiguous_array_t<ElementType>& py_data, - const py_contiguous_array_t<size_t>& ids + const py_contiguous_array_t<size_t>& ids, + bool reuse_empty = false
Users may be familiar with the similar argument in HNSWlib called `replace_deleted`. Should we use the same terminology?
ScalableVectorSearch
github_2023
cpp
75
intel
mihaic
@@ -117,6 +118,8 @@ each point. the index. ids: Vector of ids to assign to each row in ``points``. Must have the same number of elements as ``points`` has rows. + reuse_empty: A flag that determines whether to reuse empty slots that may exist after deletion and consolidation.
I could only find the term `slot` once in the SVS documentation, in [`delete_entries`](https://intel.github.io/ScalableVectorSearch/cpp/internal/index_impl/dynamic_vamana.html#_CPPv4I0EN3svs5index6vamana18MutableVamanaIndex14delete_entriesEvRK1T), without a clear definition. I think it would be better to describe the n...
ScalableVectorSearch
github_2023
cpp
75
intel
mihaic
@@ -117,10 +118,16 @@ each point. the index. ids: Vector of ids to assign to each row in ``points``. Must have the same number of elements as ``points`` has rows. + reuse_empty: A flag that determines whether to reuse empty entries that may exist after deletion and consolidation. When enabled,...
```suggestion When ``consolidate`` is called, the state of these deleted entries becomes ``empty``. ```
ScalableVectorSearch
github_2023
cpp
75
intel
mihaic
@@ -574,13 +578,24 @@ class MutableVamanaIndex { /// /// @brief Add the points with the given external IDs to the dataset. + // + /// When `delete_entries` is called, a soft deletion is performed, marking the entries + /// as `deleted`. The state of these deleted entries becomes `empty` as the user...
```suggestion /// as `deleted`. When `consolidate` is called, the state of these deleted entries becomes `empty`. /// When `add_points` is called with the `reuse_empty` flag enabled, the ```
ScalableVectorSearch
github_2023
cpp
75
intel
mihaic
@@ -219,6 +219,13 @@ class VamanaBuilder { size_t num_batches = std::max( size_t{40}, lib::div_round_up(num_nodes, lib::narrow_cast<size_t>(64 * 64)) ); + + // If num_batches is greater than num_nodes (num_batches will be 40), set
The reason is clear without the specific number. Plus, if we ever change `40` in the code, we will likely forget to change it in this comment. :) ```suggestion // If num_batches is greater than num_nodes, set ```
ScalableVectorSearch
github_2023
cpp
73
intel
rfsaliev
@@ -255,6 +255,31 @@ auto create_on_nodes(InterNUMAThreadPool& threadpool, F&& f) } #endif +///// +///// A thread pool that dynamically switches between single-threaded and multi-threaded execution. +///// - If `n == 1`, the task will be executed on the main thread without any locking mechanism. +///// - For `n > 1...
According to the [`NativeThreadPoolBase ::parallel_for()`](https://github.com/intel/ScalableVectorSearch/blob/main/include/svs/lib/threads/threadpool.h#L189) implementation, a task is executed in the current thread in case of `n == 1`. Why we cannot just fix existing code rather than making one more class?
ScalableVectorSearch
github_2023
cpp
69
intel
mihaic
@@ -486,7 +486,8 @@ void wrap(py::module& m) { .def_readwrite( "max_candidate_pool_size", &svs::index::vamana::VamanaBuildParameters::max_candidate_pool_size - ); + ) + .def_readwrite("prune_to", &svs::index::vamana::VamanaBuildParameters::prune_to);
Are all the other changes in the C++ API covered by the Python API?
ScalableVectorSearch
github_2023
others
62
intel
mihaic
@@ -0,0 +1,41 @@ +# Copyright 2023 Intel Corporation +# +# 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 applicab...
How about using the tag as a refernce? ```suggestion uses: actions/checkout@v4.2.2 ```
ScalableVectorSearch
github_2023
others
62
intel
mihaic
@@ -0,0 +1,41 @@ +# Copyright 2023 Intel Corporation +# +# 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 applicab...
```suggestion uses: apache/skywalking-eyes/header@v0.6.0 ```
ScalableVectorSearch
github_2023
others
62
intel
napetrov
@@ -0,0 +1,41 @@ +# Copyright 2023 Intel Corporation
Should it be 2024?
ScalableVectorSearch
github_2023
others
61
intel
napetrov
@@ -1,661 +1,202 @@ - GNU AFFERO GENERAL PUBLIC LICENSE - Version 3, 19 November 2007 - - Copyright (C) 2007 Free Software Foundation, Inc. <https://fsf.org/> - Everyone is permitted to copy and distribute verbatim copies - of this license document, but changing it is not allowe...
```suggestion limitations under the License. ============================================================================ Intel® oneAPI Math Kernel Library (Intel® oneMKL): Copyright 2022 Intel Corporation Intel® Scalable Vector Search: Copyright (C) Intel Corporation provides shared library in bin...
mac-registration-provider
github_2023
go
32
beeper
jetfir3
@@ -238,6 +238,8 @@ var offsets = map[[32]byte]imdOffsetTuple{ hexToByte32("fff8db27fef2a2b874f7bc6fb303a98e3e3b8aceb8dd4c5bfa2bad7b76ea438a"): offsets_13_6, // macOS 13.6.3 hexToByte32("2c674438d30bf489695f2d1b8520afc30cbfb183af82d2fc53d74ce39a25b24e"): offsets_13_6, + // macOS 13.6.4 + hexToByte32("8f22d...
This should be `offsets_13_6,` and not `offsets_13_6_4,` (due to the recent removal of superfluous offsets)
mac-registration-provider
github_2023
go
32
beeper
xChickens
@@ -238,6 +238,8 @@ var offsets = map[[32]byte]imdOffsetTuple{ hexToByte32("fff8db27fef2a2b874f7bc6fb303a98e3e3b8aceb8dd4c5bfa2bad7b76ea438a"): offsets_13_6, // macOS 13.6.3 hexToByte32("2c674438d30bf489695f2d1b8520afc30cbfb183af82d2fc53d74ce39a25b24e"): offsets_13_6, + // macOS 13.6.4 + hexToByte32("8f22d...
Spaced using spaces instead of tabs
mac-registration-provider
github_2023
go
12
beeper
0xdevalias
@@ -168,4 +239,4 @@ func hexToByte32(val string) [32]byte { panic(fmt.Errorf("expected 32 bytes, got %d", len(out))) } return *(*[32]byte)(out) -} +}
Seems the newline at the end of the file got removed in this change.
mac-registration-provider
github_2023
others
12
beeper
0xdevalias
@@ -6,7 +6,7 @@ registration code and use it in Beeper Mini. ## Supported MacOS versions The tool is currently quite hacky, so it only works on specific versions of macOS. -* Intel: 11.5 - 11.7, 13.3.1, 13.5 - 13.6, 14.0 - 14.3 +* Intel: 10.14.6, 10.15.1 - 10.15.7, 11.5 - 11.7, 13.3.1, 13.5 - 13.6, 14.0 - 14.3
Since the High Sierra offsets were added in [`e018389` (#12)](https://github.com/beeper/mac-registration-provider/pull/12/commits/e01838920a4c5adacd813c22dd53e113d269e95b), this should be updated: ```suggestion * Intel: 10.13.6, 10.14.6, 10.15.1 - 10.15.7, 11.5 - 11.7, 13.3.1, 13.5 - 13.6, 14.0 - 14.3 ```
phone-registration-provider
github_2023
others
15
beeper
itsjunetime
@@ -1,7 +1,13 @@ -TARGET := iphone:clang:latest:16.0 - -THEOS_PACKAGE_SCHEME = rootless -ARCHS = arm64 +ARCHS = arm64 arm64e + +ifeq ($(THEOS_PACKAGE_SCHEME), rootless) + TARGET := iphone:clang:latest:16.0
```suggestion TARGET := iphone:clang:latest:15.0 ``` Since you recommended using 15 instead of 16, I'll try that out in the CI of this branch to see if it seems to compile fine.
phone-registration-provider
github_2023
others
16
beeper
itsjunetime
@@ -11,6 +11,8 @@ // cheap globals 'cause IPC is stupid and we'll figure it out later static NSError *currentError; +static BPState *currentState;
Could we get a comment above this global describing what it's used for? If I understand correctly, it's used in both identityservicesd and Settings - in identityservicesd, it's set anytime the state is updated (the websocket disconnects, we get a code, etc) and then is read, serialized, and sent to settings when an upd...
phone-registration-provider
github_2023
others
16
beeper
itsjunetime
@@ -25,6 +26,27 @@ static NSString *stateFile = ROOT_PATH_NS(@"/var/mobile/.beepserv_state"); return [NSString stringWithFormat:@"<BPState code:%@ secret:%@ connected:%d error:%@", self.code, self.secret, self.connected, self.error]; } +- (void)broadcast { + NSMutableDictionary *state = @{
Since this struct is serialized to a dictionary multiple times, could we extract that into a method so we can reuse it in both the 'broadcast' and 'writeToDisk' methods?
phone-registration-provider
github_2023
others
16
beeper
itsjunetime
@@ -228,12 +230,16 @@ void log_impl(NSString *logStr) { LOG(@"Couldn't send identifiers: %@", sendErr); } -- (void)writeToStateWithCode:(NSString * __nullable)code +- (void)saveStateWithCode:(NSString * __nullable)code secret:(NSString * __nullable)secret connected:(BOOL)connected ...
It seems like this is an error left over from my changes, but I'm guessing `error` is not supposed to be always nil. Does that seem correct to you as well? ```suggestion currentState = [BPState.alloc initWithCode:code secret:secret connected:connected error:error]; ```
phone-registration-provider
github_2023
others
16
beeper
itsjunetime
@@ -415,8 +416,38 @@ NSDictionary *getIdentifiers() { // This %ctor will be called every time identityservicesd or the settings app is restarted. // So we only want it to try to reinitialize stuff if it's in identityservicesd - if (![bundleID isEqualToString:@"com.apple.identityservicesd"]) + if (![bundleID isEqu...
Could we have the notification observer names (`com.beeper.beepserv/updateState` and `com.beeper.beepserv/requestStateUpdate`) be placed into global statics? It would just make it easier to ensure we don't cause any issues with typos and such.
phone-registration-provider
github_2023
others
16
beeper
itsjunetime
@@ -415,8 +416,38 @@ NSDictionary *getIdentifiers() { // This %ctor will be called every time identityservicesd or the settings app is restarted. // So we only want it to try to reinitialize stuff if it's in identityservicesd - if (![bundleID isEqualToString:@"com.apple.identityservicesd"]) + if (![bundleID isEqu...
```suggestion currentState = [BPState.alloc initWithCode:state[kCode] secret:state[kSecret] connected:((NSNumber *)state[kConnected]).boolValue error:state[kError]]; ```
phone-registration-provider
github_2023
others
16
beeper
itsjunetime
@@ -415,8 +416,38 @@ NSDictionary *getIdentifiers() { // This %ctor will be called every time identityservicesd or the settings app is restarted. // So we only want it to try to reinitialize stuff if it's in identityservicesd - if (![bundleID isEqualToString:@"com.apple.identityservicesd"]) + if (![bundleID isEqu...
is `defaultCenter` guaranteed to never be de-allocated? And do you know if there's a possibility that these listener blocks might be deallocated, causing us to run into an issue where Settings and identityservicesd can't communicate with each other?
phone-registration-provider
github_2023
others
1
beeper
itsjunetime
@@ -15,7 +15,11 @@ jobs: uses: Randomblock1/theos-action@v1 - name: Build package - run: make package + run: | + export PACKAGE_BUILDNAME=$(git rev-parse --short HEAD)-rootless + make package THEOS_PACKAGE_SCHEME=rootless FINALPACKAGE=1
For some reason that we have yet to understand, this package doesn't seem to work on any of our test devices when we build with `FINALPACKAGE=1`. Have you tested that this works on one of your test devices with that option enabled?
phone-registration-provider
github_2023
others
1
beeper
itsjunetime
@@ -15,7 +15,11 @@ jobs: uses: Randomblock1/theos-action@v1 - name: Build package - run: make package + run: | + export PACKAGE_BUILDNAME=$(git rev-parse --short HEAD)-rootless + make clean package THEOS_PACKAGE_SCHEME=rootless FINALPACKAGE=1 + export PACKAGE_B...
```suggestion make clean package FINALPACKAGE=1 ```
phone-registration-provider
github_2023
others
14
beeper
Yumekui
@@ -418,7 +418,7 @@ NSDictionary *getIdentifiers() { if (![bundleID isEqualToString:@"com.apple.identityservicesd"]) return; - NSString *filePath = ROOT_PATH_NS(@"%@/.beepserv_wsurl"); + NSString *filePath = ROOT_PATH_NS(@"/.beepserv_wsurl");
```suggestion NSString *filePath = ROOT_PATH_NS(@"/var/mobile/.beepserv_wsurl"); ``` Everything else is being stored in /var/mobile, so this should probably be too.
bridge-manager
github_2023
others
31
beeper
holysoles
@@ -85,30 +85,24 @@ jobs: username: ${{ github.actor }} password: ${{ secrets.GITHUB_TOKEN }} + - name: Extract Docker metadata + id: metadata + uses: docker/metadata-action@v5 + with: + images: ghcr.io/${{ github.repository }} + tags: | + t...
Based on the example [here](https://github.com/docker/metadata-action?tab=readme-ov-file#latest-tag), I'd suggest we simplify this to match the intent, but it should be functionally the same ```suggestion type=raw,value=latest,enable={{is_default_branch}} ```
bridge-manager
github_2023
others
17
beeper
tulir
@@ -218,7 +218,7 @@ bridge: caption_in_message: true # Should the bridge explicitly set the avatar and room name for private chat portal rooms? # This is implicitly enabled in encrypted rooms. - private_chat_portal_meta: never + private_chat_portal_meta: auto
`auto` is not a valid value, the options are `always`, `default` and `never` should probably also have all the bridges use the same value 🤔