fix: define dataset configurations and remove duplicates

README.md

@@ -1,39 +1,190 @@
 ---
 license: cc-by-4.0
-
 task_categories:
-- tabular-regression
-- tabular-classification
+  - tabular-regression
+  - tabular-classification
 language:
-- en
+  - en
 tags:
-- gpu
-- memory-estimation
-- utilization-estimation
-- deep-learning
-- resource-management
-- mlp
-- cnn
-- transformer
-pretty_name: GPUMemNet, GPUUtilNet Dataset
-
-paper: https://arxiv.org/abs/2602.17817
+  - gpu
+  - memory-estimation
+  - utilization-estimation
+  - deep-learning
+  - resource-management
+  - mlp
+  - cnn
+  - transformer
+pretty_name: GPUMemNet and GPUUtilNet Dataset
+paper: https://doi.org/10.1145/3805621.3807621
 arxiv: 2602.17817
 repo: https://github.com/itu-rad/GPUMemNet
+configs:
+  - config_name: mlp-memory-step1
+    data_files:
+      - split: train
+        path: MLP/mlp_data1.csv
+
+  - config_name: mlp-memory-step2
+    data_files:
+      - split: train
+        path: MLP/mlp_data2.csv
+
+  - config_name: mlp-utilization
+    data_files:
+      - split: train
+        path: MLP/mlp_data_for_util.csv
+
+  - config_name: mlp-memory-legacy
+    data_files:
+      - split: train
+        path: MLP/fc_data_GPU_memory_extensive.csv
+
+  - config_name: cnn-memory-step1
+    data_files:
+      - split: train
+        path: CNN/cnn_data1.csv
+
+  - config_name: cnn-memory-revised
+    data_files:
+      - split: train
+        path: CNN/cnn_data_new_approach.csv
+
+  - config_name: cnn-utilization
+    data_files:
+      - split: train
+        path: CNN/cnn_data_for_util.csv
+
+  - config_name: transformer-memory
+    data_files:
+      - split: train
+        path: Transformers/transformer_data1.csv
+
+  - config_name: transformer-utilization
+    data_files:
+      - split: train
+        path: Transformers/transformer_data_for_util.csv
 ---
 
-# GPUMemNet Dataset
+# GPUMemNet and GPUUtilNet Dataset
+
+This dataset accompanies the paper
+**“GPU Memory and Utilization Estimation for Training-Aware Resource
+Management: Opportunities and Limitations.”**
+
+It contains synthetic deep learning training configurations and their measured
+GPU memory consumption and utilization characteristics.
+
+## Dataset configurations
+
+The dataset is divided into separate configurations because MLP, CNN, and
+Transformer workloads use different feature schemas.
+
+| Configuration | Rows | Description |
+|---|---:|---|
+| `mlp-memory-step1` | 3,000 | Initial MLP memory and average-utilization measurements |
+| `mlp-memory-step2` | 3,000 | MLP measurements with batch-normalization and dropout features |
+| `mlp-utilization` | 3,000 | MLP average and maximum utilization measurements |
+| `mlp-memory-legacy` | 1,091 | Earlier fully connected network memory dataset |
+| `cnn-memory-step1` | 9,000 | CNN measurements including architecture identifiers |
+| `cnn-memory-revised` | 9,000 | Revised CNN representation |
+| `cnn-utilization` | 9,000 | CNN average and maximum utilization measurements |
+| `transformer-memory` | 5,011 | Transformer memory measurements |
+| `transformer-utilization` | 5,011 | Transformer average and maximum utilization measurements |
+
+## Prediction targets
+
+The primary GPU-memory prediction target is:
+
+- `Max GPU Memory (MiB)`
+
+The legacy MLP configuration uses:
 
-This dataset accompanies the paper **"GPU Memory and Utilization Estimation for Training-Aware Resource Management: Opportunities and Limitations"** and is released as part of the [GPUMemNet repository](https://github.com/itu-rad/GPUMemNet).
+- `gpumemory_max`
 
+The utilization configurations contain average and maximum values for:
 
-## Description
+- `GPUTL`
+- `GRACT`
+- `SMACT`
+- `SMOCC`
+- `FP32A`
+- `DRAMA`
 
-A synthetic, extensible dataset for GPU memory and utilization estimation across three neural network architecture families: MLPs, CNNs, and Transformers. Each sample captures architectural properties (layer counts, depth, batch size, number of parameters, number of activations) alongside measured GPU memory consumption and hardware utilization metrics (SMACT, SMOCC, DRAMA), collected under controlled training conditions.
+## Loading
 
-The dataset is designed to support the development and evaluation of training-aware GPU resource management systems, with a focus on pre-execution memory estimation and interference-aware scheduling through utilization prediction.
+Install the Hugging Face datasets package:
+
+```bash
+pip install datasets
+```
+
+Load a specific configuration:
+
+```python
+from datasets import load_dataset
+
+dataset = load_dataset(
+    "ehyo/GPU-Resources-Estimation-for-Deep-Learning-Training-Tasks",
+    "cnn-utilization",
+)
+```
+
+Each configuration currently provides a `train` split containing the complete
+corresponding table.
+
+## Dataset characteristics
+
+The workload families contain different features:
+
+- **MLP:** depth, activation function, activation counts, parameter counts,
+  batch size, batch-normalization layers, and dropout layers.
+- **CNN:** depth, activation function, layer-type counts, batch size, parameter
+  counts, activation counts, and analytical memory estimates.
+- **Transformer:** sequence length, embedding size, number of layers, number
+  of attention heads, parameter counts, activation counts, and layer-type
+  counts.
+
+The configurations should be loaded independently because their schemas are
+workload-family specific.
+
+## Data collection
+
+The measurements were collected from generated deep learning training
+workloads under controlled execution conditions. The original column names
+and units are preserved for compatibility with the accompanying code and
+published experiments.
+
+Further implementation and experimental details are available in the paper
+and GitHub repository.
 
 ## Repository
 
 Code, models, and reproducibility artifacts are available at:
-[https://github.com/itu-rad/GPUMemNet](https://github.com/itu-rad/GPUMemNet)
+
+https://github.com/itu-rad/GPUMemNet
+
+## Citation
+
+```bibtex
+@inproceedings{yousefzadehaslmiandoab2026gpumemory,
+  author    = {Ehsan Yousefzadeh-Asl-Miandoab and
+               Reza Karimzadeh and
+               Danyal Yorulmaz and
+               Bulat Ibragimov and
+               Pınar Tözün},
+  title     = {GPU Memory and Utilization Estimation for Training-Aware
+               Resource Management: Opportunities and Limitations},
+  booktitle = {Proceedings of the Sixth European Workshop on Machine
+               Learning and Systems},
+  series    = {EuroMLSys '26},
+  pages     = {127--138},
+  publisher = {Association for Computing Machinery},
+  year      = {2026},
+  doi       = {10.1145/3805621.3807621}
+}
+```
+
+## License
+
+This dataset is licensed under the
+[Creative Commons Attribution 4.0 International License](https://creativecommons.org/licenses/by/4.0/).

cnn_data_for_util.csv

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:047ef9d65620a70d76e0f04f2a7d45f5820059da5c1ecd5d564dd2571e78bcfa
-size 20741975

transformer_data_for_util.csv

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:1433a6d47f7c769147dbc8629e8e552a234a24605b6fd0e26db471cf38e37529
-size 15886743