Smarter Optimization,

Smarter Optimization,

Smarter Optimization,

Faster Results

Faster Results

Faster Results

Find the best hyperparameters 5x faster
- no extra setup, no wasted trials.

Find the best hyperparameters 5x faster
- no extra setup, no wasted trials.

Find the best hyperparameters 5x faster
- no extra setup, no wasted trials.

$

🤖

🤖

Trusted by Engineers at

Trusted by Engineers at

steev Run, Analyze and Iterate

steev Run, Analyze and Iterate

steev Run, Analyze and Iterate

steev takes the guesswork out of hyperparameter optimization,

finding the best configurations for your models effortlessly

steev takes the guesswork out of hyperparameter optimization,

finding the best configurations for your models effortlessly

steev takes the guesswork out of hyperparameter optimization,

finding the best configurations for your models effortlessly

Run Seamless Experimentation

Run Seamless Experimentation

Run Seamless Experimentation

steev automatically understands your code and detects tunable hyperparameters.


No more manual setups or complex configurations.

steev automatically understands your code and detects tunable hyperparameters.


No more manual setups or complex configurations.

steev automatically understands your code and detects tunable hyperparameters.


No more manual setups or complex configurations.

Steev here!! 👋  Let me take a look at your code...

Awesome! I found 7 key hyperparameters in your code:

• learning_rate: 2e-4

• warmup_steps: 100

• max_steps: 1000

• gradient_accumulation_steps: 4

• lora_rank: 8

• lora_alpha: 32

• max_seq_length: 512


After analyzing your code, I found your model and dataset!

I see you're using the Qwen-7B model and the smoltalk dataset!


Did I get that right? (yes/no): yes


Please review my experimental plan carefully:

BATCH SIZE CONFIG:

  - Batch sizes: [1, 2, 4]

  - Gradient accumulation: [8, 16, 32]


LEARNING RATE CONFIG:

  - Learning rates: [5e-5, 1e-4, 2e-4, 3e-4]

  - Warmup ratios: [0.03, 0.05, 0.1]


LoRA CONFIG:

  - Rank values: [16, 32, 64, 128]

  - Alpha values: [8, 16, 32]

  - Dropout rates: [0.05, 0.1, 0.15]


Total experiments to run: 432 combinations

Estimated total experiment time: 144-216 hours

Steev here!! 👋  Let me take a look at your code...

Awesome! I found 7 key hyperparameters in your code:

• learning_rate: 2e-4

• warmup_steps: 100

• max_steps: 1000

• gradient_accumulation_steps: 4

• lora_rank: 8

• lora_alpha: 32

• max_seq_length: 512


After analyzing your code, I found your model and dataset!

I see you're using the Qwen-7B model and the smoltalk dataset!


Did I get that right? (yes/no): yes


Please review my experimental plan carefully:

BATCH SIZE CONFIG:

  - Batch sizes: [1, 2, 4]

  - Gradient accumulation: [8, 16, 32]


LEARNING RATE CONFIG:

  - Learning rates: [5e-5, 1e-4, 2e-4, 3e-4]

  - Warmup ratios: [0.03, 0.05, 0.1]


LoRA CONFIG:

  - Rank values: [16, 32, 64, 128]

  - Alpha values: [8, 16, 32]

  - Dropout rates: [0.05, 0.1, 0.15]


Total experiments to run: 432 combinations

Estimated total experiment time: 144-216 hours

Monitor Training Status 📈

Analyze Abnormality 🔍

Decide Next Steps 🚀

Optimize Experiments ⚙️

Analyze live.

Skip wasted steps

Analyze live.

Skip wasted steps

Analyze live.

Skip wasted steps

Steev monitors your experiments live, skipping inefficient trials to fast-track optimal results.

Steev monitors your experiments live, skipping inefficient trials to fast-track optimal results.

Steev monitors your experiments live, skipping inefficient trials to fast-track optimal results.

Iterate with

Continuous Feedback

Iterate with

Continuous Feedback

Iterate with

Continuous Feedback

steev doesn’t just stop at results but, constantly learns from them.


steev will refine and plan experiments automatically, searching for perfect hyperparameters.

steev doesn’t just stop at results but, constantly learns from them.


steev will refine and plan experiments automatically, searching for perfect hyperparameters.

steev doesn’t just stop at results but, constantly learns from them.


steev will refine and plan experiments automatically, searching for perfect hyperparameters.

Analyze

Analyze

Run

Run

steev

steev

steev

is all you need

is all you need

is all you need

Provide your code and datasets,

Provide your code and datasets,

Provide your code and datasets,

and steev takes care of the rest.

and steev takes care of the rest.

and steev takes care of the rest.

Best results, delivered with a single command.

Best results, delivered with a single command.

Best results, delivered with a single command.

Before & After Meeting steev

Before & After Meeting steev

Before & After

Meeting steev

No more complex scripts, just call steev

No more complex scripts, just call steev

No more complex scripts, just call steev

Before

Before

# !/bin/bash

LEARNING_RATES=(1e-5 3e-5 5e-5)

BATCH_SIZE=(8 16)

LORA_RANKS=(8 16 32)

DROPOUT_RATES=(0.1 0.2 0.3)


for LR in "${LEARNING_RATES[@]}"; do

# !/bin/bash

LEARNING_RATES=(1e-5 3e-5 5e-5)

BATCH_SIZE=(8 16)

LORA_RANKS=(8 16 32)

DROPOUT_RATES=(0.1 0.2 0.3)


for LR in "${LEARNING_RATES[@]}"; do

$ ./hyperparameter_optimize.sh

$ ./hyperparameter_optimize.sh

After

After

$ steev auto-run train.py

$ steev auto-run train.py

💡Steev found the best model weight

with best hyperparameters!💡

save report to ./outputs/reports.html

💡Steev found the best model weight

with best hyperparameters!💡

save report to ./outputs/reports.html

You Speak, steev Works

You Speak, steev Works

You Speak,

steev Works

Tell steev what you need and steev turns them into fully optimized ML experiments.

What you say is what you get. Simple, effective, and results-driven.

Tell steev what you need and steev turns them into fully optimized ML experiments.

What you say is what you get. Simple, effective, and results-driven.

Tell steev what you need and steev turns them into fully optimized ML experiments.

What you say is what you get. Simple, effective, and results-driven.

Loved by world-class Researchers

Loved by world-class Researchers

Loved by world-class Researchers

Many ML researchers empowered their potential with steev

Many ML researchers empowered their potential with steev

Many ML researchers empowered their potential

with steev

@bob

ML Engineer

ex. OpenVINO developer

@steve

ML Engineer

CVPR, ICCV author

@Nataniel

ML Engineer

ex. OpenVINO developer

@Junkyu

series A

startup

on-device AI engineer

computer vision

@mountain

series A

startup

full-stack engineer

2+m MAU service builder

@Jacob

machine vision engineer

Robotics, AR, VR

@Simon

backend developer

3x serial founder

@Selin

ML Engineer

ex. OpenVINO developer

@andrew

ML Engineer

CVPR, ICCV author

@Luis

ML Engineer

ex. OpenVINO developer

@Carmen

series A

startup

on-device AI engineer

computer vision

@bob

ML Engineer

ex. OpenVINO developer

@steve

ML Engineer

CVPR, ICCV author

@Nataniel

ML Engineer

ex. OpenVINO developer

@Junkyu

series A

startup

on-device AI engineer

computer vision

@mountain

series A

startup

full-stack engineer

2+m MAU service builder

@Jacob

machine vision engineer

Robotics, AR, VR

@Simon

backend developer

3x serial founder

@Selin

ML Engineer

ex. OpenVINO developer

@andrew

ML Engineer

CVPR, ICCV author

@Luis

ML Engineer

ex. OpenVINO developer

@Carmen

series A

startup

on-device AI engineer

computer vision

@bob

ML Engineer

ex. OpenVINO developer

@steve

ML Engineer

CVPR, ICCV author

@Nataniel

ML Engineer

ex. OpenVINO developer

@Junkyu

series A

startup

on-device AI engineer

computer vision

@mountain

series A

startup

full-stack engineer

2+m MAU service builder

@Jacob

machine vision engineer

Robotics, AR, VR

@Simon

backend developer

3x serial founder

@Selin

ML Engineer

ex. OpenVINO developer

@andrew

ML Engineer

CVPR, ICCV author

@Luis

ML Engineer

ex. OpenVINO developer

@Carmen

series A

startup

on-device AI engineer

computer vision

100+ Researchers

100+ Researchers

are already joined the waitlist.

steev

2024 TBD Labs, all rights reserved

contact@steev.io

contact@steev.io