Welcome back, User!

Manage your GPU workloads and monitor cluster activity.

Active Jobs
0
GPU Hours Used
0.0
Available GPUs
4x RTX 4090
Estimated Cost
$0.00
Quick Actions
🚀
Submit Job
Run your first GPU workload
📊
Ray Dashboard
Monitor cluster & jobs
📋
View Jobs
Check job history & logs
🔑 API Key
••••••••••••••••••••••••••••••••
⚠️ Keep this secret! Don't share or commit to version control.
👤 Your AccountSynced
EmailLoading...
Created--
StoragePostgreSQL
⚡ Active GPU WorkersVast.ai
🔍
Checking for active workers...
Recent Jobs
📭
No jobs yet
Submit your first job to see it here
🚀 Submit a Job

Choose how you want to submit jobs to OptimL:

✨
Easy Mode
Form-based fine-tuning
RECOMMENDED
📦
Advanced
Upload custom code & data
BRING YOUR OWN
💻
CLI
Submit via terminal
POWER USERS
🐍
Python SDK
Programmatic access
COMING SOON
✨ Easy Mode: LLM Fine-TuningNo Code Required

Select the pre-trained model to fine-tune

Training dataset for fine-tuning

URL to your JSON/JSONL dataset file

Model compression for smaller, faster inference

Higher = smaller model, lower accuracy

Training Parameters
✨
Easy Mode: Just select your options and click submit. OptimL will generate the configuration and submit the job automatically. No coding required!
📦 Advanced Mode: Custom Code & DatasetBring Your Own
💡
Upload your own code and dataset to run alongside OptimL's compression. Your code will be uploaded to the cluster and executed on GPUs. OptimL libraries are always available.
📄
Drop Python files here or click to browse
Supports: .py, .yaml, .json, .txt, .sh (max 10MB total)
📊
Drop dataset file here or click to browse
Supports: .json, .jsonl, .csv, .parquet (max 100MB)

Command to run your code (e.g., python train.py, torchrun --nproc_per_node=2 train.py)

Number of GPUs for your job

Additional Dependencies (Optional)

One package per line. OptimL, PyTorch, and common ML libraries are pre-installed.

📝
How it works: Your files are securely uploaded to our storage, then passed to Ray workers via working_dir. Your code runs with full access to GPUs and OptimL's compression algorithms.
🎬 Live Demo: Job Submission
1
Submit
2
Upload
3
Run
4
Done
Terminal
user@optiml:~$ ▊
──────────────────────────────
Server: ray.optiml.org:8265
──────────────────────────────
Uploading working directory...
✓ train.py
✓ data/train.jsonl
✓ config.yaml
Uploaded 3 files
──────────────────────────────
✓ Job 'raysubmit_7f3kL9' submitted
──────────────────────────────
Next:
ray job logs/status/stop <id>
Tailing logs...
[OptimL] Loading model: Llama-2-7b-hf
[OptimL] GPU: RTX 4090 (24GB)
[OptimL] QLoRA adapters applied
[Train] Epoch 1/3 | Loss: 2.34→1.87
[Train] Epoch 2/3 | Loss: 1.87→1.12
[Train] Epoch 3/3 | Loss: 1.12→0.89
[OptimL] Model saved
[OptimL] Compressing to INT4...
[OptimL] Export complete
──────────────────────────────
Job succeeded
──────────────────────────────
user@optiml:~$ ▊
💻 Submit via CLI
1Install
Terminal
pip install optiml 'ray[default]'
2Code
train.py

          

          

            
3Submit

            
Terminal

                      

          

            
4Monitor

            
Terminal

            Open Ray Dashboard →
          

        

      

      

        
🐍 Python SDK

        
🚧
Coming Soon
SDK under development.