diff --git a/experiments/nf_evaluation/README.md b/experiments/nf_evaluation/README.md
new file mode 100644
index 000000000..52273a52a
--- /dev/null
+++ b/experiments/nf_evaluation/README.md
@@ -0,0 +1,56 @@
+# Comprehensive Evaluation of Neuralforecast models
+
+In this experiment, we tested all available models in Neuralforecast on benchmark datasets to evaluate their speed and forecasting performance.
+
+The datasets used for this benchmark are:
+- M4 (yearly)
+- M4 (quarterly)
+- M4 (monthly)
+- M4 (daily)
+- Ettm2 (15 min)
+- Electricity (hourly)
+- Weather (10 min)
+- Traffic (hourly)
+- ILI (weekly)
+
+Each model went through hyperparameter optimization. The test was completed locally on CPU.
+
+The table below summarizes the results
+*Table will be updated as results are obtained*
+
+<br>
+
+## Reproducibility
+
+1. Create a conda environment `nf_evaluation` using the `environment.yml` file.
+  ```shell
+  conda env create -f environment.yml
+  ```
+
+3. Activate the conda environment using 
+  ```shell
+  conda activate nf_evaluation
+  ```
+
+Alternatively simply installing neuralforecast and datasetsforecast with pip may suffice:
+```
+pip install git+https://github.com/Nixtla/datasetsforecast.git
+pip install git+https://github.com/Nixtla/neuralforecast.git
+```
+
+4. Run the experiments for each dataset and each model using the 
+- `--dataset` parameter in `[M4-yearly, M4-quarterly, M4-monthly, M4-daily, Ettm2, Electricity, Weather, Traffic, ILI]`
+- `--model` parameter in `['AutoLSTM', 'AutoRNN', 'AutoGRU', 'AutoDilatedRNN', 'AutoDeepAR', 'AutoTCN', 'AutoMLP', 'AutoNBEATS', 'AutoNHITS', 'AutoDLinear', 'AutoTFT', 'AutoVanillaTransformer', 'AutoInformer', 'AutoAutoformer', 'AutoFEDformer', 'AutoTimesNet', 'AutoPatchTST']`
+<br>
+
+```shell
+python run_experiments.py --dataset M4-yearly --model AutoMLP
+```
+
+You can also run all experiments in a single command using
+<br>
+```shell
+run.sh
+```
+
+5. The script creates a folder `results/<dataset>` which contains a CSV file with the metrics for the specified model
\ No newline at end of file
diff --git a/experiments/nf_evaluation/datasets.py b/experiments/nf_evaluation/datasets.py
new file mode 100644
index 000000000..d95b09e41
--- /dev/null
+++ b/experiments/nf_evaluation/datasets.py
@@ -0,0 +1,133 @@
+import pandas as pd
+
+from datasetsforecast.m4 import M4
+from datasetsforecast.m3 import M3
+from datasetsforecast.long_horizon import LongHorizon
+
+
+def get_dataset(name):
+
+    # Read data and parameters
+    if name == 'M3-yearly':
+        Y_df, *_ = M3.load(directory='./', group='Yearly')
+        Y_df['ds'] = pd.to_datetime(Y_df['ds'])
+        freq = 'Y'
+        h = 6
+        val_size = 6
+        test_size = 6
+    elif name == 'M3-quarterly':
+        Y_df, *_ = M3.load(directory='./', group='Quarterly')
+        Y_df['ds'] = pd.to_datetime(Y_df['ds'])
+        freq = 'Q'
+        h = 8
+        val_size = 8
+        test_size = 8
+    elif name == 'M3-monthly':
+        Y_df, *_ = M3.load(directory='./', group='Monthly')
+        Y_df['ds'] = pd.to_datetime(Y_df['ds'])
+        freq = 'M'
+        h = 12
+        val_size = 12
+        test_size = 12
+    elif name == 'M4-yearly':
+        Y_df, *_ = M4.load(directory='./', group='Yearly')
+        Y_df['ds'] = Y_df['ds'].astype(int)
+        freq = 1
+        h = 6
+        val_size = 6
+        test_size = 6
+    elif name == 'M4-quarterly':
+        Y_df, *_ = M4.load(directory='./', group='Quarterly')
+        Y_df['ds'] = Y_df['ds'].astype(int)
+        freq = 4
+        h = 8
+        val_size = 8
+        test_size = 8
+    elif name == 'M4-monthly':
+        Y_df, *_ = M4.load(directory='./', group='Monthly')
+        Y_df['ds'] = Y_df['ds'].astype(int)
+        freq = 12
+        h = 18
+        val_size = 18
+        test_size = 18
+    elif name == 'M4-daily':
+        Y_df, *_ = M4.load(directory='./', group='Daily')
+        Y_df['ds'] = Y_df['ds'].astype(int)
+        freq = 365
+        h = 14
+        val_size = 14
+        test_size = 14
+    elif name == 'M4-hourly':
+        Y_df, *_ = M4.load(directory='./', group='Hourly')
+        Y_df['ds'] = Y_df['ds'].astype(int)
+        freq = 24
+        h = 48
+        val_size = 48
+        test_size = 48
+    elif name == 'Ettm2':
+        Y_df, *_ = LongHorizon.load(directory='./', group='ETTm2')
+        Y_df['ds'] = pd.to_datetime(Y_df['ds'])
+        freq = '15T'
+        h = 720
+        val_size = 11520
+        test_size = 11520
+    elif name == 'Ettm1':
+        Y_df, *_ = LongHorizon.load(directory='./', group='ETTm1')
+        Y_df['ds'] = pd.to_datetime(Y_df['ds'])
+        freq = '15T'
+        h = 720
+        val_size = 11520
+        test_size = 11520
+    elif name == 'Etth1':
+        Y_df, *_ = LongHorizon.load(directory='./', group='ETTh1')
+        Y_df['ds'] = pd.to_datetime(Y_df['ds'])
+        freq = 'H'
+        h = 720
+        val_size = 2880
+        test_size = 2880
+    elif name == 'Etth2':
+        Y_df, *_ = LongHorizon.load(directory='./', group='ETTh2')
+        Y_df['ds'] = pd.to_datetime(Y_df['ds'])
+        freq = 'H'
+        h = 720
+        val_size = 2880
+        test_size = 2880
+    elif name == 'Electricity':
+        Y_df, *_ = LongHorizon.load(directory='./', group='ECL')
+        Y_df['ds'] = pd.to_datetime(Y_df['ds'])
+        freq = 'H'
+        h = 720
+        val_size = 2632
+        test_size = 5260
+    elif name == 'Exchange':
+        Y_df, *_ = LongHorizon.load(directory='./', group='Exchange')
+        Y_df['ds'] = pd.to_datetime(Y_df['ds'])
+        freq = 'D'
+        h = 720
+        val_size = 760
+        test_size = 1517
+    elif name == 'Weather':
+        Y_df, *_ = LongHorizon.load(directory='./', group='Weather')
+        Y_df['ds'] = pd.to_datetime(Y_df['ds'])
+        freq = '10T'
+        h = 720
+        val_size = 5270
+        test_size = 10539
+    elif name == 'Traffic':
+        Y_df, *_ = LongHorizon.load(directory='./', group='TrafficL')
+        Y_df['ds'] = pd.to_datetime(Y_df['ds'])
+        freq = 'H'
+        h = 720
+        val_size = 1756
+        test_size = 3508
+    elif name == 'ILI':
+        Y_df, *_ = LongHorizon.load(directory='./', group='ILI')
+        Y_df['ds'] = pd.to_datetime(Y_df['ds'])
+        freq = 'W'
+        h = 60
+        val_size = 97
+        test_size = 193
+    else:
+        raise Exception("Frequency not defined")
+
+    return Y_df, h, freq, val_size, test_size
\ No newline at end of file
diff --git a/experiments/nf_evaluation/environment.yml b/experiments/nf_evaluation/environment.yml
new file mode 100644
index 000000000..744060d0f
--- /dev/null
+++ b/experiments/nf_evaluation/environment.yml
@@ -0,0 +1,9 @@
+name: nf_evaluation
+channels:
+  - conda-forge
+dependencies:
+  - numpy<1.24
+  - pip
+  - pip:
+    - "git+https://github.com/Nixtla/datasetsforecast.git"
+    - "git+https://github.com/Nixtla/neuralforecast.git"
\ No newline at end of file
diff --git a/experiments/nf_evaluation/gpu_1.sh b/experiments/nf_evaluation/gpu_1.sh
new file mode 100755
index 000000000..59cbf1ecf
--- /dev/null
+++ b/experiments/nf_evaluation/gpu_1.sh
@@ -0,0 +1,11 @@
+#!/bin/bash
+
+datasets=("M3-yearly" "M3-quarterly" "M3-monthly" "M4-yearly")
+
+models=("AutoLSTM" "AutoRNN" "AutoGRU" "AutoDilatedRNN" "AutoDeepAR" "AutoTCN" "AutoMLP" "AutoNBEATS" "AutoNHITS" "AutoDLinear" "AutoTFT" "AutoVanillaTransformer" "AutoInformer" "AutoAutoformer" "AutoFEDformer" "AutoTimesNet" "AutoPatchTST", "AutoTSMixer", "AutoiTransformer")
+
+for dataset in "${datasets[@]}"; do
+    for model in "${models[@]}"; do
+        CUDA_VISIBLE_DEVICES=0 python run_experiments.py --dataset "$dataset" --model "$model"
+    done
+done
\ No newline at end of file
diff --git a/experiments/nf_evaluation/gpu_2.sh b/experiments/nf_evaluation/gpu_2.sh
new file mode 100755
index 000000000..301914c31
--- /dev/null
+++ b/experiments/nf_evaluation/gpu_2.sh
@@ -0,0 +1,11 @@
+#!/bin/bash
+
+datasets=("M4-quarterly" "M4-monthly" "M4-daily" "M4-hourly")
+
+models=("AutoLSTM" "AutoRNN" "AutoGRU" "AutoDilatedRNN" "AutoDeepAR" "AutoTCN" "AutoMLP" "AutoNBEATS" "AutoNHITS" "AutoDLinear" "AutoTFT" "AutoVanillaTransformer" "AutoInformer" "AutoAutoformer" "AutoFEDformer" "AutoTimesNet" "AutoPatchTST" "AutoTSMixer", "AutoiTransformer")
+
+for dataset in "${datasets[@]}"; do
+    for model in "${models[@]}"; do
+        CUDA_VISIBLE_DEVICES=1 python run_experiments.py --dataset "$dataset" --model "$model"
+    done
+done
\ No newline at end of file
diff --git a/experiments/nf_evaluation/gpu_3.sh b/experiments/nf_evaluation/gpu_3.sh
new file mode 100755
index 000000000..358dc38f1
--- /dev/null
+++ b/experiments/nf_evaluation/gpu_3.sh
@@ -0,0 +1,11 @@
+#!/bin/bash
+
+datasets=("Ettm2" "Ettm1" "Etth1" "Etth2")
+
+models=("AutoLSTM" "AutoRNN" "AutoGRU" "AutoDilatedRNN" "AutoDeepAR" "AutoTCN" "AutoMLP" "AutoNBEATS" "AutoNHITS" "AutoDLinear" "AutoTFT" "AutoVanillaTransformer" "AutoInformer" "AutoAutoformer" "AutoFEDformer" "AutoTimesNet" "AutoPatchTST" "AutoTSMixer", "AutoiTransformer")
+
+for dataset in "${datasets[@]}"; do
+    for model in "${models[@]}"; do
+        CUDA_VISIBLE_DEVICES=2 python run_experiments.py --dataset "$dataset" --model "$model"
+    done
+done
\ No newline at end of file
diff --git a/experiments/nf_evaluation/gpu_4.sh b/experiments/nf_evaluation/gpu_4.sh
new file mode 100755
index 000000000..562cc8a3c
--- /dev/null
+++ b/experiments/nf_evaluation/gpu_4.sh
@@ -0,0 +1,11 @@
+#!/bin/bash
+
+datasets=("Electricity" "Exchange" "Weather" "Traffic" "ILI")
+
+models=("AutoLSTM" "AutoRNN" "AutoGRU" "AutoDilatedRNN" "AutoDeepAR" "AutoTCN" "AutoMLP" "AutoNBEATS" "AutoNHITS" "AutoDLinear" "AutoTFT" "AutoVanillaTransformer" "AutoInformer" "AutoAutoformer" "AutoFEDformer" "AutoTimesNet" "AutoPatchTST" "AutoTSMixer", "AutoiTransformer")
+
+for dataset in "${datasets[@]}"; do
+    for model in "${models[@]}"; do
+        CUDA_VISIBLE_DEVICES=3 python run_experiments.py --dataset "$dataset" --model "$model"
+    done
+done
\ No newline at end of file
diff --git a/experiments/nf_evaluation/models.py b/experiments/nf_evaluation/models.py
new file mode 100644
index 000000000..bd6e228c7
--- /dev/null
+++ b/experiments/nf_evaluation/models.py
@@ -0,0 +1,32 @@
+from neuralforecast.auto import *
+from neuralforecast.losses.pytorch import HuberLoss, DistributionLoss
+
+
+def get_model(model_name, horizon, num_samples):
+    """Returns the model class given the model name.
+    """
+    model_dict = {
+        'AutoLSTM': AutoLSTM(config=None, h=horizon, loss=HuberLoss(), num_samples=num_samples),
+        'AutoRNN': AutoRNN(config=None, h=horizon, loss=HuberLoss(), num_samples=num_samples),
+        'AutoGRU': AutoGRU(config=None, h=horizon, loss=HuberLoss(), num_samples=num_samples),
+        'AutoDilatedRNN': AutoDilatedRNN(config=None, h=horizon, loss=HuberLoss(), num_samples=num_samples),
+        'AutoDeepAR': AutoDeepAR(config=None, h=horizon,
+                             loss=DistributionLoss(distribution='StudentT', level=[80, 90], return_params=False),
+                             num_samples=num_samples),
+        'AutoTCN': AutoTCN(config=None, h=horizon, loss=HuberLoss(), num_samples=num_samples),
+        'AutoMLP': AutoMLP(config=None, h=horizon, loss=HuberLoss(), num_samples=num_samples),
+        'AutoNBEATS': AutoNBEATS(config=None, h=horizon, loss=HuberLoss(), num_samples=num_samples),
+        'AutoNHITS': AutoNHITS(config=None, h=horizon, loss=HuberLoss(), num_samples=num_samples),
+        'AutoDLinear': AutoDLinear(config=None, h=horizon, loss=HuberLoss(), num_samples=num_samples),
+        'AutoTFT': AutoTFT(config=None, h=horizon, loss=HuberLoss(), num_samples=num_samples),
+        'AutoVanillaTransformer': AutoVanillaTransformer(config=None, h=horizon, loss=HuberLoss(), num_samples=num_samples),
+        'AutoInformer': AutoInformer(config=None, h=horizon, loss=HuberLoss(), num_samples=num_samples),
+        'AutoAutoformer': AutoAutoformer(config=None, h=horizon, loss=HuberLoss(), num_samples=num_samples),
+        'AutoFEDformer': AutoFEDformer(config=None, h=horizon, loss=HuberLoss(), num_samples=num_samples),
+        'AutoTimesNet': AutoTimesNet(config=None, h=horizon, loss=HuberLoss(), num_samples=num_samples),
+        'AutoPatchTST': AutoPatchTST(config=None, h=horizon, loss=HuberLoss(), num_samples=num_samples),
+        'AutoTSMixer': AutoTSMixer(config=None, h=horizon, loss=HuberLoss(), num_samples=num_samples, n_series=1),
+        'AutoiTransformer': AutoiTransformer(config=None, h=horizon, loss=HuberLoss(), num_samples=num_samples, n_series=1)
+    }
+
+    return model_dict[model_name]
\ No newline at end of file
diff --git a/experiments/nf_evaluation/run.sh b/experiments/nf_evaluation/run.sh
new file mode 100644
index 000000000..1e75bd3ae
--- /dev/null
+++ b/experiments/nf_evaluation/run.sh
@@ -0,0 +1,11 @@
+#!/bin/bash
+
+datasets=("M3-yearly" "M3-quarterly" "M3-monthly" "M4-yearly" "M4-quarterly" "M4-monthly" "M4-daily" "M4-hourly" "Ettm2" "Ettm1" "Etth1" "Etth2" "Electricity" "Exchange" "Weather" "Traffic" "ILI")
+
+models=("AutoLSTM" "AutoRNN" "AutoGRU" "AutoDilatedRNN" "AutoDeepAR" "AutoTCN" "AutoMLP" "AutoNBEATS" "AutoNHITS" "AutoDLinear" "AutoTFT" "AutoVanillaTransformer" "AutoInformer" "AutoAutoformer" "AutoFEDformer" "AutoTimesNet" "AutoPatchTST")
+
+for dataset in "${datasets[@]}"; do
+    for model in "${models[@]}"; do
+        python run_experiments.py --dataset "$dataset" --model "$model"
+    done
+done
\ No newline at end of file
diff --git a/experiments/nf_evaluation/run_experiments.py b/experiments/nf_evaluation/run_experiments.py
new file mode 100644
index 000000000..9bdfc7af9
--- /dev/null
+++ b/experiments/nf_evaluation/run_experiments.py
@@ -0,0 +1,68 @@
+import os
+import argparse
+import time
+
+import pandas as pd
+from neuralforecast.core import NeuralForecast
+from neuralforecast.losses.numpy import mae, mse, smape, mape
+
+from models import get_model
+from datasets import get_dataset
+
+# For compatibitlity with Mac with M chip
+os.environ["PYTORCH_ENABLE_MPS_FALLBACK"] = "1"
+
+# For memory allocation
+os.environ["PYTORCH_CUDA_ALLOC_CONF"] = "expandable_segments:True"
+
+def main(args):
+
+    # Load dataset and model
+    Y_df, h, freq, val_size, test_size = get_dataset(args.dataset)
+    model = get_model(model_name=args.model, horizon=h, num_samples=20)
+
+    # Start time
+    start_time = time.time()
+
+    # Train model
+    nf = NeuralForecast(models=[model], freq=freq)
+    forecasts_df = nf.cross_validation(df=Y_df, val_size=val_size, test_size=test_size, n_windows=None, verbose=True)
+
+    # Calculate elapsed time
+    elapsed_time = time.time() - start_time
+
+    # Evaluation
+    model_mae = mae(y=forecasts_df['y'], y_hat=forecasts_df[f'{args.model}'])
+    model_mse = mse(y=forecasts_df['y'], y_hat=forecasts_df[f'{args.model}'])
+    model_smape = smape(y=forecasts_df['y'], y_hat=forecasts_df[f'{args.model}'])
+    model_mape = mape(y=forecasts_df['y'], y_hat=forecasts_df[f'{args.model}'])
+
+    metrics = {
+            'Model': [args.model],
+            'MAE': [model_mae],
+            'MSE': [model_mse],
+            'sMAPE': [model_smape],
+            'MAPE': [model_mape],
+            'time': [elapsed_time]
+    }
+        
+    
+    # Save results
+    results_path = f'./results/{args.dataset}'
+    os.makedirs(results_path, exist_ok=True)
+
+    metrics_df = pd.DataFrame(metrics)
+    metrics_df.to_csv(f'{results_path}/{args.model}_metrics.csv', header=True, index=False)
+	
+def parse_args():
+	parser = argparse.ArgumentParser(description="script arguments")
+	parser.add_argument('--dataset', type=str, help='dataset to train models on')
+	parser.add_argument('--model', type=str, help='name of the model')
+	return parser.parse_args()
+
+if __name__ == '__main__':
+	# parse arguments
+	args = parse_args()
+	
+    # Run experiment
+	main(args)
\ No newline at end of file
diff --git a/nbs/models.ipynb b/nbs/models.ipynb
index 82331c3b2..6895a8434 100644
--- a/nbs/models.ipynb
+++ b/nbs/models.ipynb
@@ -226,7 +226,7 @@
     "        \"context_size\": tune.choice([5, 10, 50]),\n",
     "        \"decoder_hidden_size\": tune.choice([64, 128, 256, 512]),\n",
     "        \"learning_rate\": tune.loguniform(1e-4, 1e-1),\n",
-    "        \"max_steps\": tune.choice([500, 1000]),\n",
+    "        \"max_steps\": tune.quniform(lower=500, upper=5000, q=500),\n",
     "        \"batch_size\": tune.choice([16, 32]),\n",
     "        \"loss\": None,\n",
     "        \"random_seed\": tune.randint(1, 20)\n",
@@ -369,7 +369,7 @@
     "        \"context_size\": tune.choice([5, 10, 50]),\n",
     "        \"decoder_hidden_size\": tune.choice([64, 128, 256, 512]),\n",
     "        \"learning_rate\": tune.loguniform(1e-4, 1e-1),\n",
-    "        \"max_steps\": tune.choice([500, 1000]),\n",
+    "        \"max_steps\": tune.quniform(lower=500, upper=5000, q=500),\n",
     "        \"batch_size\": tune.choice([16, 32]),\n",
     "        \"loss\": None,\n",
     "        \"random_seed\": tune.randint(1, 20)\n",
@@ -508,7 +508,7 @@
     "        \"context_size\": tune.choice([5, 10, 50]),\n",
     "        \"decoder_hidden_size\": tune.choice([64, 128, 256, 512]),\n",
     "        \"learning_rate\": tune.loguniform(1e-4, 1e-1),\n",
-    "        \"max_steps\": tune.choice([500, 1000]),\n",
+    "        \"max_steps\": tune.quniform(lower=500, upper=5000, q=500),\n",
     "        \"batch_size\": tune.choice([16, 32]),\n",
     "        \"loss\": None,\n",
     "        \"random_seed\": tune.randint(1, 20)\n",
@@ -646,7 +646,7 @@
     "        \"context_size\": tune.choice([5, 10, 50]),\n",
     "        \"decoder_hidden_size\": tune.choice([64, 128]),\n",
     "        \"learning_rate\": tune.loguniform(1e-4, 1e-1),\n",
-    "        \"max_steps\": tune.choice([500, 1000]),\n",
+    "        \"max_steps\": tune.quniform(lower=500, upper=5000, q=500),\n",
     "        \"batch_size\": tune.choice([16, 32]),\n",
     "        \"loss\": None,\n",
     "        \"random_seed\": tune.randint(1, 20)\n",
@@ -784,7 +784,7 @@
     "        \"lstm_dropout\": tune.uniform(0.0, 0.5),\n",
     "        \"learning_rate\": tune.loguniform(1e-4, 1e-1),\n",
     "        \"scaler_type\": tune.choice(['robust', 'minmax1']),\n",
-    "        \"max_steps\": tune.choice([500, 1000, 2000]),\n",
+    "        \"max_steps\": tune.quniform(lower=500, upper=5000, q=500),\n",
     "        \"batch_size\": tune.choice([32, 64, 128, 256]),\n",
     "        \"windows_batch_size\": tune.choice([128, 256, 512, 1024]),\n",
     "        \"loss\": None,\n",
@@ -924,7 +924,7 @@
     "        \"context_size\": tune.choice([5, 10, 50]),\n",
     "        \"decoder_hidden_size\": tune.choice([64, 128, 256, 512]),\n",
     "        \"learning_rate\": tune.loguniform(1e-4, 1e-1),\n",
-    "        \"max_steps\": tune.choice([500, 1000]),\n",
+    "        \"max_steps\": tune.quniform(lower=500, upper=5000, q=500),\n",
     "        \"batch_size\": tune.choice([16, 32]),\n",
     "        \"loss\": None,\n",
     "        \"random_seed\": tune.randint(1, 20)\n",
@@ -1070,7 +1070,7 @@
     "        \"num_layers\": tune.randint(2, 6),\n",
     "        \"learning_rate\": tune.loguniform(1e-4, 1e-1),\n",
     "        \"scaler_type\": tune.choice([None, 'robust', 'standard']),\n",
-    "        \"max_steps\": tune.choice([500, 1000]),\n",
+    "        \"max_steps\": tune.quniform(lower=500, upper=5000, q=500),\n",
     "        \"batch_size\": tune.choice([32, 64, 128, 256]),\n",
     "        \"windows_batch_size\": tune.choice([128, 256, 512, 1024]),\n",
     "        \"loss\": None,\n",
@@ -1205,7 +1205,7 @@
     "        \"h\": None,\n",
     "        \"learning_rate\": tune.loguniform(1e-4, 1e-1),\n",
     "        \"scaler_type\": tune.choice([None, 'robust', 'standard']),\n",
-    "        \"max_steps\": tune.choice([500, 1000]),\n",
+    "        \"max_steps\": tune.quniform(lower=500, upper=5000, q=500),\n",
     "        \"batch_size\": tune.choice([32, 64, 128, 256]),\n",
     "        \"windows_batch_size\": tune.choice([128, 256, 512, 1024]),\n",
     "        \"loss\": None,\n",
@@ -1341,7 +1341,7 @@
     "        \"h\": None,\n",
     "        \"learning_rate\": tune.loguniform(1e-4, 1e-1),\n",
     "        \"scaler_type\": tune.choice([None, 'robust', 'standard']),\n",
-    "        \"max_steps\": tune.choice([500, 1000]),\n",
+    "        \"max_steps\": tune.quniform(lower=500, upper=5000, q=500),\n",
     "        \"batch_size\": tune.choice([32, 64, 128, 256]),\n",
     "        \"windows_batch_size\": tune.choice([128, 256, 512, 1024]),\n",
     "        \"loss\": None,\n",
@@ -1482,7 +1482,7 @@
     "                                         [40, 20, 1], [1, 1, 1]]),\n",
     "       \"learning_rate\": tune.loguniform(1e-4, 1e-1),\n",
     "       \"scaler_type\": tune.choice([None, 'robust', 'standard']),\n",
-    "       \"max_steps\": tune.quniform(lower=500, upper=1500, q=100),\n",
+    "       \"max_steps\": tune.quniform(lower=500, upper=5000, q=500),\n",
     "       \"batch_size\": tune.choice([32, 64, 128, 256]),\n",
     "       \"windows_batch_size\": tune.choice([128, 256, 512, 1024]),\n",
     "       \"loss\": None,\n",
@@ -1620,7 +1620,7 @@
     "       \"moving_avg_window\": tune.choice([11, 25, 51]),\n",
     "       \"learning_rate\": tune.loguniform(1e-4, 1e-1),\n",
     "       \"scaler_type\": tune.choice([None, 'robust', 'standard']),\n",
-    "       \"max_steps\": tune.quniform(lower=500, upper=1500, q=100),\n",
+    "       \"max_steps\": tune.quniform(lower=500, upper=5000, q=500),\n",
     "       \"batch_size\": tune.choice([32, 64, 128, 256]),\n",
     "       \"windows_batch_size\": tune.choice([128, 256, 512, 1024]),\n",
     "       \"loss\": None,\n",
@@ -1901,7 +1901,7 @@
     "        \"n_head\": tune.choice([4, 8]),\n",
     "        \"learning_rate\": tune.loguniform(1e-4, 1e-1),\n",
     "        \"scaler_type\": tune.choice([None, 'robust', 'standard']),\n",
-    "        \"max_steps\": tune.choice([500, 1000, 2000]),\n",
+    "        \"max_steps\": tune.quniform(lower=500, upper=5000, q=500),\n",
     "        \"batch_size\": tune.choice([32, 64, 128, 256]),\n",
     "        \"windows_batch_size\": tune.choice([128, 256, 512, 1024]),\n",
     "        \"loss\": None,\n",
@@ -2038,7 +2038,7 @@
     "        \"n_head\": tune.choice([4, 8]),\n",
     "        \"learning_rate\": tune.loguniform(1e-4, 1e-1),\n",
     "        \"scaler_type\": tune.choice([None, 'robust', 'standard']),\n",
-    "        \"max_steps\": tune.choice([500, 1000, 2000]),\n",
+    "        \"max_steps\": tune.quniform(lower=500, upper=5000, q=500),\n",
     "        \"batch_size\": tune.choice([32, 64, 128, 256]),\n",
     "        \"windows_batch_size\": tune.choice([128, 256, 512, 1024]),\n",
     "        \"loss\": None,\n",
@@ -2175,7 +2175,7 @@
     "        \"n_head\": tune.choice([4, 8]),\n",
     "        \"learning_rate\": tune.loguniform(1e-4, 1e-1),\n",
     "        \"scaler_type\": tune.choice([None, 'robust', 'standard']),\n",
-    "        \"max_steps\": tune.choice([500, 1000, 2000]),\n",
+    "        \"max_steps\": tune.quniform(lower=500, upper=5000, q=500),\n",
     "        \"batch_size\": tune.choice([32, 64, 128, 256]),\n",
     "        \"windows_batch_size\": tune.choice([128, 256, 512, 1024]),\n",
     "        \"loss\": None,\n",
@@ -2312,7 +2312,7 @@
     "        \"n_head\": tune.choice([4, 8]),\n",
     "        \"learning_rate\": tune.loguniform(1e-4, 1e-1),\n",
     "        \"scaler_type\": tune.choice([None, 'robust', 'standard']),\n",
-    "        \"max_steps\": tune.choice([500, 1000, 2000]),\n",
+    "        \"max_steps\": tune.quniform(lower=500, upper=5000, q=500),\n",
     "        \"batch_size\": tune.choice([32, 64, 128, 256]),\n",
     "        \"windows_batch_size\": tune.choice([128, 256, 512, 1024]),\n",
     "        \"loss\": None,\n",
@@ -2448,7 +2448,7 @@
     "        \"hidden_size\": tune.choice([64, 128, 256]),\n",
     "        \"learning_rate\": tune.loguniform(1e-4, 1e-1),\n",
     "        \"scaler_type\": tune.choice([None, 'robust', 'standard']),\n",
-    "        \"max_steps\": tune.choice([500, 1000, 2000]),\n",
+    "        \"max_steps\": tune.quniform(lower=500, upper=5000, q=500),\n",
     "        \"batch_size\": tune.choice([32, 64, 128, 256]),\n",
     "        \"windows_batch_size\": tune.choice([128, 256, 512, 1024]),\n",
     "        \"loss\": None,\n",
@@ -2587,7 +2587,7 @@
     "        \"learning_rate\": tune.loguniform(1e-4, 1e-1),\n",
     "        \"scaler_type\": tune.choice([None, 'robust', 'standard']),\n",
     "        \"revin\": tune.choice([False, True]),\n",
-    "        \"max_steps\": tune.choice([500, 1000, 5000]),\n",
+    "        \"max_steps\": tune.quniform(lower=500, upper=5000, q=500),\n",
     "        \"batch_size\": tune.choice([32, 64, 128, 256]),\n",
     "        \"windows_batch_size\": tune.choice([128, 256, 512, 1024]),\n",
     "        \"loss\": None,\n",
@@ -3031,7 +3031,7 @@
     "        \"multi_layer\": tune.choice([3, 5, 7]),\n",
     "        \"learning_rate\": tune.loguniform(1e-4, 1e-1),\n",
     "        \"scaler_type\": tune.choice([None, 'robust', 'standard']),\n",
-    "        \"max_steps\": tune.choice([500, 1000, 2000]),\n",
+    "        \"max_steps\": tune.quniform(lower=500, upper=5000, q=500),\n",
     "        \"batch_size\": tune.choice([32, 64, 128, 256]),\n",
     "        \"loss\": None,\n",
     "        \"random_seed\": tune.randint(1, 20),\n",
diff --git a/neuralforecast/auto.py b/neuralforecast/auto.py
index 20932619a..a64676dac 100644
--- a/neuralforecast/auto.py
+++ b/neuralforecast/auto.py
@@ -59,7 +59,7 @@ class AutoRNN(BaseAuto):
         "context_size": tune.choice([5, 10, 50]),
         "decoder_hidden_size": tune.choice([64, 128, 256, 512]),
         "learning_rate": tune.loguniform(1e-4, 1e-1),
-        "max_steps": tune.choice([500, 1000]),
+        "max_steps": tune.quniform(lower=500, upper=5000, q=500),
         "batch_size": tune.choice([16, 32]),
         "loss": None,
         "random_seed": tune.randint(1, 20),
@@ -134,7 +134,7 @@ class AutoLSTM(BaseAuto):
         "context_size": tune.choice([5, 10, 50]),
         "decoder_hidden_size": tune.choice([64, 128, 256, 512]),
         "learning_rate": tune.loguniform(1e-4, 1e-1),
-        "max_steps": tune.choice([500, 1000]),
+        "max_steps": tune.quniform(lower=500, upper=5000, q=500),
         "batch_size": tune.choice([16, 32]),
         "loss": None,
         "random_seed": tune.randint(1, 20),
@@ -205,7 +205,7 @@ class AutoGRU(BaseAuto):
         "context_size": tune.choice([5, 10, 50]),
         "decoder_hidden_size": tune.choice([64, 128, 256, 512]),
         "learning_rate": tune.loguniform(1e-4, 1e-1),
-        "max_steps": tune.choice([500, 1000]),
+        "max_steps": tune.quniform(lower=500, upper=5000, q=500),
         "batch_size": tune.choice([16, 32]),
         "loss": None,
         "random_seed": tune.randint(1, 20),
@@ -275,7 +275,7 @@ class AutoTCN(BaseAuto):
         "context_size": tune.choice([5, 10, 50]),
         "decoder_hidden_size": tune.choice([64, 128]),
         "learning_rate": tune.loguniform(1e-4, 1e-1),
-        "max_steps": tune.choice([500, 1000]),
+        "max_steps": tune.quniform(lower=500, upper=5000, q=500),
         "batch_size": tune.choice([16, 32]),
         "loss": None,
         "random_seed": tune.randint(1, 20),
@@ -345,7 +345,7 @@ class AutoDeepAR(BaseAuto):
         "lstm_dropout": tune.uniform(0.0, 0.5),
         "learning_rate": tune.loguniform(1e-4, 1e-1),
         "scaler_type": tune.choice(["robust", "minmax1"]),
-        "max_steps": tune.choice([500, 1000, 2000]),
+        "max_steps": tune.quniform(lower=500, upper=5000, q=500),
         "batch_size": tune.choice([32, 64, 128, 256]),
         "windows_batch_size": tune.choice([128, 256, 512, 1024]),
         "loss": None,
@@ -418,7 +418,7 @@ class AutoDilatedRNN(BaseAuto):
         "context_size": tune.choice([5, 10, 50]),
         "decoder_hidden_size": tune.choice([64, 128, 256, 512]),
         "learning_rate": tune.loguniform(1e-4, 1e-1),
-        "max_steps": tune.choice([500, 1000]),
+        "max_steps": tune.quniform(lower=500, upper=5000, q=500),
         "batch_size": tune.choice([16, 32]),
         "loss": None,
         "random_seed": tune.randint(1, 20),
@@ -487,7 +487,7 @@ class AutoMLP(BaseAuto):
         "num_layers": tune.randint(2, 6),
         "learning_rate": tune.loguniform(1e-4, 1e-1),
         "scaler_type": tune.choice([None, "robust", "standard"]),
-        "max_steps": tune.choice([500, 1000]),
+        "max_steps": tune.quniform(lower=500, upper=5000, q=500),
         "batch_size": tune.choice([32, 64, 128, 256]),
         "windows_batch_size": tune.choice([128, 256, 512, 1024]),
         "loss": None,
@@ -553,7 +553,7 @@ class AutoNBEATS(BaseAuto):
         "h": None,
         "learning_rate": tune.loguniform(1e-4, 1e-1),
         "scaler_type": tune.choice([None, "robust", "standard"]),
-        "max_steps": tune.choice([500, 1000]),
+        "max_steps": tune.quniform(lower=500, upper=5000, q=500),
         "batch_size": tune.choice([32, 64, 128, 256]),
         "windows_batch_size": tune.choice([128, 256, 512, 1024]),
         "loss": None,
@@ -619,7 +619,7 @@ class AutoNBEATSx(BaseAuto):
         "h": None,
         "learning_rate": tune.loguniform(1e-4, 1e-1),
         "scaler_type": tune.choice([None, "robust", "standard"]),
-        "max_steps": tune.choice([500, 1000]),
+        "max_steps": tune.quniform(lower=500, upper=5000, q=500),
         "batch_size": tune.choice([32, 64, 128, 256]),
         "windows_batch_size": tune.choice([128, 256, 512, 1024]),
         "loss": None,
@@ -698,7 +698,7 @@ class AutoNHITS(BaseAuto):
         ),
         "learning_rate": tune.loguniform(1e-4, 1e-1),
         "scaler_type": tune.choice([None, "robust", "standard"]),
-        "max_steps": tune.quniform(lower=500, upper=1500, q=100),
+        "max_steps": tune.quniform(lower=500, upper=5000, q=500),
         "batch_size": tune.choice([32, 64, 128, 256]),
         "windows_batch_size": tune.choice([128, 256, 512, 1024]),
         "loss": None,
@@ -765,7 +765,7 @@ class AutoDLinear(BaseAuto):
         "moving_avg_window": tune.choice([11, 25, 51]),
         "learning_rate": tune.loguniform(1e-4, 1e-1),
         "scaler_type": tune.choice([None, "robust", "standard"]),
-        "max_steps": tune.quniform(lower=500, upper=1500, q=100),
+        "max_steps": tune.quniform(lower=500, upper=5000, q=500),
         "batch_size": tune.choice([32, 64, 128, 256]),
         "windows_batch_size": tune.choice([128, 256, 512, 1024]),
         "loss": None,
@@ -899,7 +899,7 @@ class AutoTFT(BaseAuto):
         "n_head": tune.choice([4, 8]),
         "learning_rate": tune.loguniform(1e-4, 1e-1),
         "scaler_type": tune.choice([None, "robust", "standard"]),
-        "max_steps": tune.choice([500, 1000, 2000]),
+        "max_steps": tune.quniform(lower=500, upper=5000, q=500),
         "batch_size": tune.choice([32, 64, 128, 256]),
         "windows_batch_size": tune.choice([128, 256, 512, 1024]),
         "loss": None,
@@ -967,7 +967,7 @@ class AutoVanillaTransformer(BaseAuto):
         "n_head": tune.choice([4, 8]),
         "learning_rate": tune.loguniform(1e-4, 1e-1),
         "scaler_type": tune.choice([None, "robust", "standard"]),
-        "max_steps": tune.choice([500, 1000, 2000]),
+        "max_steps": tune.quniform(lower=500, upper=5000, q=500),
         "batch_size": tune.choice([32, 64, 128, 256]),
         "windows_batch_size": tune.choice([128, 256, 512, 1024]),
         "loss": None,
@@ -1035,7 +1035,7 @@ class AutoInformer(BaseAuto):
         "n_head": tune.choice([4, 8]),
         "learning_rate": tune.loguniform(1e-4, 1e-1),
         "scaler_type": tune.choice([None, "robust", "standard"]),
-        "max_steps": tune.choice([500, 1000, 2000]),
+        "max_steps": tune.quniform(lower=500, upper=5000, q=500),
         "batch_size": tune.choice([32, 64, 128, 256]),
         "windows_batch_size": tune.choice([128, 256, 512, 1024]),
         "loss": None,
@@ -1103,7 +1103,7 @@ class AutoAutoformer(BaseAuto):
         "n_head": tune.choice([4, 8]),
         "learning_rate": tune.loguniform(1e-4, 1e-1),
         "scaler_type": tune.choice([None, "robust", "standard"]),
-        "max_steps": tune.choice([500, 1000, 2000]),
+        "max_steps": tune.quniform(lower=500, upper=5000, q=500),
         "batch_size": tune.choice([32, 64, 128, 256]),
         "windows_batch_size": tune.choice([128, 256, 512, 1024]),
         "loss": None,
@@ -1170,7 +1170,7 @@ class AutoFEDformer(BaseAuto):
         "hidden_size": tune.choice([64, 128, 256]),
         "learning_rate": tune.loguniform(1e-4, 1e-1),
         "scaler_type": tune.choice([None, "robust", "standard"]),
-        "max_steps": tune.choice([500, 1000, 2000]),
+        "max_steps": tune.quniform(lower=500, upper=5000, q=500),
         "batch_size": tune.choice([32, 64, 128, 256]),
         "windows_batch_size": tune.choice([128, 256, 512, 1024]),
         "loss": None,
@@ -1240,7 +1240,7 @@ class AutoPatchTST(BaseAuto):
         "learning_rate": tune.loguniform(1e-4, 1e-1),
         "scaler_type": tune.choice([None, "robust", "standard"]),
         "revin": tune.choice([False, True]),
-        "max_steps": tune.choice([500, 1000, 5000]),
+        "max_steps": tune.quniform(lower=500, upper=5000, q=500),
         "batch_size": tune.choice([32, 64, 128, 256]),
         "windows_batch_size": tune.choice([128, 256, 512, 1024]),
         "loss": None,
@@ -1462,7 +1462,7 @@ class AutoStemGNN(BaseAuto):
         "multi_layer": tune.choice([3, 5, 7]),
         "learning_rate": tune.loguniform(1e-4, 1e-1),
         "scaler_type": tune.choice([None, "robust", "standard"]),
-        "max_steps": tune.choice([500, 1000, 2000]),
+        "max_steps": tune.quniform(lower=500, upper=5000, q=500),
         "batch_size": tune.choice([32, 64, 128, 256]),
         "loss": None,
         "random_seed": tune.randint(1, 20),