On-Point-RND · ZhMax · Apr 7, 2025 · Apr 15, 2025
diff --git a/Dockerfile b/Dockerfile
@@ -15,16 +15,9 @@ RUN pip install --no-cache-dir --upgrade pip wheel poetry==1.5.1 poetry-dynamic-
 COPY . RePlay-Accelerated/
 RUN cd RePlay-Accelerated && ./poetry_wrapper.sh install --all-extras
 
-RUN pip install --upgrade torch
+RUN pip install --upgrade torch==2.5.1
 RUN pip install rs_datasets
 RUN pip install Ninja==1.11.1.1
 RUN pip install -U tensorboard
 
-RUN pip3 install triton
-RUN pip3 install bitsandbytes
-RUN sed -i 's/tl\.libdevice\.llrint/tl\.extra\.cuda\.libdevice\.llrint/g' \
-    /opt/conda/lib/python3.11/site-packages/bitsandbytes/triton/quantize_global.py \
-    /opt/conda/lib/python3.11/site-packages/bitsandbytes/triton/quantize_rowwise.py \
-    /opt/conda/lib/python3.11/site-packages/bitsandbytes/triton/quantize_columnwise_and_transpose.py
-
 CMD ["bash"]
diff --git a/README.md b/README.md
@@ -1,214 +1,63 @@
 <img src="docs/images/replay_logo_color.svg" height="50"/>
 <br>
 
-[![GitHub License](https://img.shields.io/github/license/sb-ai-lab/RePlay)](https://github.com/sb-ai-lab/RePlay/blob/main/LICENSE)
-[![PyPI - Version](https://img.shields.io/pypi/v/replay-rec)](https://pypi.org/project/replay-rec)
-[![Documentation](https://img.shields.io/badge/docs-latest-brightgreen.svg)](https://sb-ai-lab.github.io/RePlay/)
-[![PyPI - Downloads](https://img.shields.io/pypi/dm/replay-rec)](https://pypistats.org/packages/replay-rec)
-<br>
-[![GitHub Workflow Status (with event)](https://img.shields.io/github/actions/workflow/status/sb-ai-lab/replay/main.yml)](https://github.com/sb-ai-lab/RePlay/actions/workflows/main.yml?query=branch%3Amain)
-[![Ruff](https://img.shields.io/endpoint?url=https://raw.githubusercontent.com/charliermarsh/ruff/main/assets/badge/v2.json)](https://github.com/astral-sh/ruff)
-[![Python Versions](https://img.shields.io/pypi/pyversions/replay-rec.svg?logo=python&logoColor=white)](https://pypi.org/project/replay-rec)
-[![Join the community on GitHub Discussions](https://badgen.net/badge/join%20the%20discussion/on%20github/black?icon=github)](https://github.com/sb-ai-lab/RePlay/discussions)
-
-
-RePlay is an advanced framework designed to facilitate the development and evaluation of recommendation systems. It provides a robust set of tools covering the entire lifecycle of a recommendation system pipeline:
-
-## 🚀 Features:
-* **Data Preprocessing and Splitting:** Streamlines the data preparation process for recommendation systems, ensuring optimal data structure and format for efficient processing.
-* **Wide Range of Recommendation Models:** Enables building of recommendation models from State-of-the-Art to commonly-used baselines and evaluate their performance and quality.
-* **Hyperparameter Optimization:** Offers tools for fine-tuning model parameters to achieve the best possible performance, reducing the complexity of the optimization process.
-* **Comprehensive Evaluation Metrics:** Incorporates a wide range of evaluation metrics to assess the accuracy and effectiveness of recommendation models.
-* **Model Ensemble and Hybridization:** Supports combining predictions from multiple models and creating two-level (ensemble) models to enhance the quality of recommendations.
-* **Seamless Mode Transition:** Facilitates easy transition from offline experimentation to online production environments, ensuring scalability and flexibility.
-
-## 💻 Hardware and Environment Compatibility:
-1. **Diverse Hardware Support:** Compatible with various hardware configurations including CPU, GPU, Multi-GPU.
-2. **Cluster Computing Integration:** Integrating with PySpark for distributed computing, enabling scalability for large-scale recommendation systems.
-
-<a name="toc"></a>
-# Table of Contents
-
-* [Quickstart](#quickstart)
-* [Installation](#installation)
-* [Resources](#examples)
-* [Contributing to RePlay](#contributing)
+This repository is a fork of the RePlay library, containing implementation Cut Cross Entropy (CCE) and Cut Cross Entropy with Negative Sampling (CCE-) for RecSys. Triton kernels are available in 
+`kernels/cut_cross_entropy`. Implementation of SASRec with CCE and CCE- can be found in `replay/models/nn/sequential/sasrec/lightning.py`. Experiment pipeline is located in `replay_benchmarks`.
 
+<a name="installation"></a>
+## Installation
 
-<a name="quickstart"></a>
-## 📈 Quickstart
+Installation via `poetry` package manager is recommended by default:
 
 ```bash
-pip install replay-rec[all]
-```
-
-Pyspark-based model and [fast](https://github.com/sb-ai-lab/RePlay/blob/main/examples/11_sasrec_dataframes_comparison.ipynb) polars-based data preprocessing:
-```python
-from polars import from_pandas
-from rs_datasets import MovieLens
-
-from replay.data import Dataset, FeatureHint, FeatureInfo, FeatureSchema, FeatureType
-from replay.data.dataset_utils import DatasetLabelEncoder
-from replay.metrics import HitRate, NDCG, Experiment
-from replay.models import ItemKNN
-from replay.utils.spark_utils import convert2spark
-from replay.utils.session_handler import State
-from replay.splitters import RatioSplitter
-
-spark = State().session
-
-ml_1m = MovieLens("1m")
-K = 10
-
-# convert data to polars
-interactions = from_pandas(ml_1m.ratings)
-
-# data splitting
-splitter = RatioSplitter(
-    test_size=0.3,
-    divide_column="user_id",
-    query_column="user_id",
-    item_column="item_id",
-    timestamp_column="timestamp",
-    drop_cold_items=True,
-    drop_cold_users=True,
-)
-train, test = splitter.split(interactions)
-
-# datasets creation
-feature_schema = FeatureSchema(
-    [
-        FeatureInfo(
-            column="user_id",
-            feature_type=FeatureType.CATEGORICAL,
-            feature_hint=FeatureHint.QUERY_ID,
-        ),
-        FeatureInfo(
-            column="item_id",
-            feature_type=FeatureType.CATEGORICAL,
-            feature_hint=FeatureHint.ITEM_ID,
-        ),
-        FeatureInfo(
-            column="rating",
-            feature_type=FeatureType.NUMERICAL,
-            feature_hint=FeatureHint.RATING,
-        ),
-        FeatureInfo(
-            column="timestamp",
-            feature_type=FeatureType.NUMERICAL,
-            feature_hint=FeatureHint.TIMESTAMP,
-        ),
-    ]
-)
-
-train_dataset = Dataset(feature_schema=feature_schema, interactions=train)
-test_dataset = Dataset(feature_schema=feature_schema, interactions=test)
-
-# data encoding
-encoder = DatasetLabelEncoder()
-train_dataset = encoder.fit_transform(train_dataset)
-test_dataset = encoder.transform(test_dataset)
-
-# convert datasets to spark
-train_dataset.to_spark()
-test_dataset.to_spark()
-
-# model training
-model = ItemKNN()
-model.fit(train_dataset)
+pip install --no-cache-dir --upgrade pip wheel poetry==1.5.1 poetry-dynamic-versioning \
+    && python -m poetry config virtualenvs.create false
+./poetry_wrapper.sh install --all-extras
 
-# model inference
-encoded_recs = model.predict(
-    dataset=train_dataset,
-    k=K,
-    queries=test_dataset.query_ids,
-    filter_seen_items=True,
-)
-
-recs = encoder.query_and_item_id_encoder.inverse_transform(encoded_recs)
-
-# model evaluation
-metrics = Experiment(
-    [NDCG(K), HitRate(K)],
-    test,
-    query_column="user_id",
-    item_column="item_id",
-    rating_column="rating",
-)
-metrics.add_result("ItemKNN", recs)
-print(metrics.results)
 ```
-
-<a name="installation"></a>
-## 🔧 Installation
-
-Installation via `pip` package manager is recommended by default:
-
+After installing replay, it is required to update torch and install additional packages:
 ```bash
-pip install replay-rec
+pip install --upgrade torch==2.5.1
+pip install --upgrade pytorch-lightning==2.5.1
+pip install numpy==1.24.4
+pip install rs_datasets
+pip install -U tensorboard
 ```
 
-In this case it will be installed the `core` package without `PySpark` and `PyTorch` dependencies.
-Also `experimental` submodule will not be installed.
-
-To install `experimental` submodule please specify the version with `rc0` suffix.
-For example:
+<a name="examples"></a>
+## Usage
 
+To run the experiments for training SASRec, use the following command from the RePlay-Accelerated directory:
 ```bash
-pip install replay-rec==XX.YY.ZZrc0
+python main.py
 ```
 
-### Extras
+Experiment parameters are defined in `.yaml` files located in the configs directory. 
+The dataset name is specified in the `config.yaml` file as follows:
 
-In addition to the core package, several extras are also provided, including:
-- `[spark]`: Install PySpark functionality
-- `[torch]`: Install PyTorch and Lightning functionality
-- `[all]`: `[spark]` `[torch]`
-
-Example:
-```bash
-# Install core package with PySpark dependency
-pip install replay-rec[spark]
-
-# Install package with experimental submodule and PySpark dependency
-pip install replay-rec[spark]==XX.YY.ZZrc0
+Parameters for the experiments are defined by `.yaml` files in `configs` directory.
+Name of the dataset in determined in the file `config.yaml`:
 ```
+defaults:
+  - dataset: <dataset_name>
+  - model: sasrec_<dataset_name>
+```
+The following datasets are available `movielens_20m`, `beauty`, `30music`, `zvuk`, `megamarket`. 
 
-To build RePlay from sources please use the [instruction](CONTRIBUTING.md#installing-from-the-source).
-
-
-<a name="examples"></a>
-## 📑  Resources
-
-### Usage examples
-1. [01_replay_basics.ipynb](https://github.com/sb-ai-lab/RePlay/blob/main/examples/01_replay_basics.ipynb) - get started with RePlay.
-2. [02_models_comparison.ipynb](https://github.com/sb-ai-lab/RePlay/blob/main/examples/02_models_comparison.ipynb) - reproducible models comparison on [MovieLens-1M dataset](https://grouplens.org/datasets/movielens/1m/).
-3. [03_features_preprocessing_and_lightFM.ipynb](https://github.com/sb-ai-lab/RePlay/blob/main/examples/03_features_preprocessing_and_lightFM.ipynb) - LightFM example with pyspark for feature preprocessing.
-4. [04_splitters.ipynb](https://github.com/sb-ai-lab/RePlay/blob/main/examples/04_splitters.ipynb) - An example of using RePlay data splitters.
-5. [05_feature_generators.ipynb](https://github.com/sb-ai-lab/RePlay/blob/main/examples/05_feature_generators.ipynb) - Feature generation with RePlay.
-6. [06_item2item_recommendations.ipynb](https://github.com/sb-ai-lab/RePlay/blob/main/examples/06_item2item_recommendations.ipynb) - Item to Item recommendations example.
-7. [07_filters.ipynb](https://github.com/sb-ai-lab/RePlay/blob/main/examples/07_filters.ipynb) - An example of using filters.
-8. [08_recommending_for_categories.ipynb](https://github.com/sb-ai-lab/RePlay/blob/main/examples/08_recommending_for_categories.ipynb) - An example of recommendation for product categories.
-9. [09_sasrec_example.ipynb](https://github.com/sb-ai-lab/RePlay/blob/main/examples/09_sasrec_example.ipynb) - An example of using transformer-based SASRec model to generate recommendations.
-10. [10_bert4rec_example.ipynb](https://github.com/sb-ai-lab/RePlay/blob/main/examples/10_bert4rec_example.ipynb) - An example of using transformer-based BERT4Rec model to generate recommendations.
-11. [11_sasrec_dataframes_comparison.ipynb](https://github.com/sb-ai-lab/RePlay/blob/main/examples/11_sasrec_dataframes_comparison.ipynb) - speed comparison of using different frameworks (pandas, polars, pyspark) for data processing during SASRec training.
-12. [12_neural_ts_exp.ipynb](https://github.com/sb-ai-lab/RePlay/blob/main/examples/12_neural_ts_exp.ipynb) - An example of using Neural Thompson Sampling bandit model (based on Wide&Deep architecture).
-13. [13_personalized_bandit_comparison.ipynb](https://github.com/sb-ai-lab/RePlay/blob/main/examples/13_personalized_bandit_comparison.ipynb) - A comparison of context-free and contextual bandit models.
-14. [14_hierarchical_recommender.ipynb](https://github.com/sb-ai-lab/RePlay/blob/main/examples/14_hierarchical_recommender.ipynb) - An example of using HierarchicalRecommender with user-disjoint LinUCB.
-
-### Videos and papers
-* **Video guides**:
-	- [Replay for offline recommendations, AI Journey 2021](https://www.youtube.com/watch?v=ejQZKGAG0xs)
+Parameters for SASRec are defined in the sasrec_<dataset_name>.yaml files. 
+To use CCE-, specify the following configuration:
+```
+loss_type: CCE
+loss_sample_count: <number_of_negative_samples>
+```
+If `loss_sample_count: null`, the training will use the standard CCE method.
 
-* **Research papers**:
-    - [RePlay: a Recommendation Framework for Experimentation and Production Use](https://arxiv.org/abs/2409.07272) Alexey Vasilev, Anna Volodkevich, Denis Kulandin, Tatiana Bysheva, Anton Klenitskiy. In The 18th ACM Conference on Recommender Systems (RecSys '24)
-	- [Turning Dross Into Gold Loss: is BERT4Rec really better than SASRec?](https://doi.org/10.1145/3604915.3610644) Anton Klenitskiy, Alexey Vasilev. In The 17th ACM Conference on Recommender Systems (RecSys '23)
-    - [The Long Tail of Context: Does it Exist and Matter?](https://arxiv.org/abs/2210.01023). Konstantin Bauman, Alexey Vasilev, Alexander Tuzhilin. In Workshop on Context-Aware Recommender Systems (CARS) (RecSys '22)
-    - [Multiobjective Evaluation of Reinforcement Learning Based Recommender Systems](https://doi.org/10.1145/3523227.3551485). Alexey Grishanov, Anastasia Ianina, Konstantin Vorontsov. In The 16th ACM Conference on Recommender Systems (RecSys '22)
-    - [Quality Metrics in Recommender Systems: Do We Calculate Metrics Consistently?](https://doi.org/10.1145/3460231.3478848) Yan-Martin Tamm, Rinchin Damdinov, Alexey Vasilev. In The 15th ACM Conference on Recommender Systems (RecSys '21)
+To reproduce CE- grid search results, we provide a special trainer. It is available in `replay_benchmarks/grid_params_search_runner.py`. To set a grid for grid-search, you can modify the `replay_benchmarks/configs/mode/hyperparameter_experiment.yaml` file. Additionally, you need to change the usage mode in the main config (`replay_benchmarks/configs/config.yaml`). There, the parameter `mode: train` should be changed to `mode: hyperparameter_experiment`.
 
-<a name="contributing"></a>
-## 💡 Contributing to RePlay
+The `hyperparameter_experiment.yaml` configuration is used solely to iterate over `batch_size`, `max_seq_len`, and `loss_sample_count`. To change other parameters, you need to modify them in their respective configuration files.
 
-We welcome community contributions. For details please check our [contributing guidelines](CONTRIBUTING.md).
+## Acknowledgements 
+This repository is build upon the [RePlay repository]
+(https://github.com/sb-ai-lab/RePlay/tree/main). Triton kernels is based on the code of [ml-cross-entropy](
+https://github.com/apple/ml-cross-entropy/tree/main).
 
diff --git a/clean.sh b/clean.sh
@@ -0,0 +1,9 @@
+#!/bin/bash
+
+rm -rf output_log.log
+
+rm -rf lightning_logs
+
+rm -rf replay_benchmarks/artifacts
+
+rm -rf replay_benchmarks/__pycache__
diff --git a/kernels/cut_cross_entropy/cce.py b/kernels/cut_cross_entropy/cce.py
@@ -1,4 +1,5 @@
 # Copyright (C) 2024 Apple Inc. All Rights Reserved.
+# This software includes modifications
 from dataclasses import dataclass
 from typing import cast
 

diff --git a/kernels/cut_cross_entropy/cce_backward.py b/kernels/cut_cross_entropy/cce_backward.py
@@ -1,4 +1,5 @@
 # Copyright (C) 2024 Apple Inc. All Rights Reserved.
+# This software includes modifications
 import torch
 import triton
 import triton.language as tl

diff --git a/kernels/cut_cross_entropy/cce_lse_forward.py b/kernels/cut_cross_entropy/cce_lse_forward.py
@@ -1,4 +1,5 @@
 # Copyright (C) 2024 Apple Inc. All Rights Reserved.
+# This software includes modifications
 from typing import Literal, overload
 
 import torch

diff --git a/kernels/cut_cross_entropy/linear_cross_entropy.py b/kernels/cut_cross_entropy/linear_cross_entropy.py
@@ -1,4 +1,5 @@
 # Copyright (C) 2024 Apple Inc. All Rights Reserved.
+# This software includes modifications
 import enum
 import platform
 from enum import auto

diff --git a/kernels/cut_cross_entropy/tl_autotune.py b/kernels/cut_cross_entropy/tl_autotune.py
@@ -1,4 +1,5 @@
 # Copyright (C) 2024 Apple Inc. All Rights Reserved.
+# This software includes modifications
 import functools
 import heapq
 import os

diff --git a/kernels/fused_linear_cross_entropy/fused_linear_ce_loss.py b/kernels/fused_linear_cross_entropy/fused_linear_ce_loss.py
@@ -1,3 +1,4 @@
+#Liger-Kernel/src/liger_kernel/ops/fused_linear_cross_entropy.py
 import torch
 import triton
 import triton.language as tl

diff --git a/kernels/multinomial_sampling/__init__.py b/kernels/multinomial_sampling/__init__.py
diff --git a/kernels/multinomial_sampling/cuda_multinomial.cpp b/kernels/multinomial_sampling/cuda_multinomial.cpp
diff --git a/kernels/multinomial_sampling/cuda_multinomial_kernel.cu b/kernels/multinomial_sampling/cuda_multinomial_kernel.cu