Base Models
Available models for fine-tuning
ModelStudio offers multiple pre-trained foundation models for fine-tuning. Each model is optimized for different use cases and capabilities.
Available Models
Qwen3 0.6B
Ultra-efficient text generation across multiple languages.
| Property | Value |
|---|---|
| Parameters | 600 million (0.6B) |
| Context Window | 32,768 tokens |
| Languages | 100+ languages |
| Data Types | Text |
| License | Apache 2.0 |
Use Cases:
- Text generation, translation, and conversational AI across 100+ languages including English, Chinese, French, Spanish, and German
- Quick text classification
- Content moderation filtering
- High-volume data tagging
- Chatbots and content creation
Qwen3 1.7B
Powerful model for complex reasoning and code generation.
| Property | Value |
|---|---|
| Parameters | 1.7 billion |
| Context Window | 32,768 tokens |
| Languages | 100+ languages |
| Data Types | Text |
| License | Apache 2.0 |
Use Cases:
- Complex reasoning, mathematical computations, code generation, and multi-turn dialogues
- Perfect for role-playing scenarios
- Logical problem-solving
- Instruction following
- Automated code and formula generation
- Troubleshooting agent reasoning
Qwen2.5-VL 3B Instruct
Multimodal vision-language model for visual understanding.
| Property | Value |
|---|---|
| Parameters | 3 billion |
| Context Window | 32,768 tokens |
| Languages | 100+ languages (including Japanese, Korean, Arabic, Vietnamese) |
| Data Types | Image, text, video |
| Input Types | Local files, URLs, base64 images |
| License | Apache 2.0 |
Use Cases:
- Image understanding, video analysis, and visual question answering
- Handles images of various resolutions and ratios
- Supports multilingual text in images
- Visual data extraction (OCR)
- Manufacturing quality inspection
- Gauge and meter reading
- Document processing
Qwen2.5-VL is the only vision-capable model in ModelStudio, supporting multimodal inputs including images and videos.
Granite 3.3 2B Instruct
Long-context document processing specialist.
| Property | Value |
|---|---|
| Parameters | 2 billion |
| Context Window | 128,768 tokens |
| Languages | English, German, Spanish, French, Japanese, Portuguese, Arabic, Czech, Italian, Korean, Dutch, Chinese |
| Data Types | Text |
| License | Apache 2.0 |
Use Cases:
- Long-form document summarization
- Legal and compliance document Q&A
- Technical documentation categorization
- Document QA and meeting summarization
- Code completion with Fill-in-the-Middle (FIM)
- Structured reasoning tasks
Granite 3.3 features an exceptionally large 128k token context window, making it ideal for processing lengthy documents.
Model Selection Guide
| Task Type | Recommended Model |
|---|---|
| Quick classification/tagging | Qwen3 0.6B |
| Complex reasoning/code | Qwen3 1.7B |
| Image/video processing | Qwen2.5-VL 3B |
| Long documents | Granite 3.3 2B |
| Multi-language (EU) | Granite 3.3 2B |
| General purpose | Qwen3 1.7B |
What is LoRA Fine-Tuning?
ModelStudio uses LoRA (Low-Rank Adaptation) for efficient fine-tuning. This approach:
- Freezes the pre-trained model weights
- Introduces trainable rank decomposition matrices
- Reduces the number of trainable parameters
- Maintains quality while being efficient
This enables fast training (minutes instead of hours) with minimal resource requirements.
Next Steps
- Create a Dataset - Prepare your training data
- Configure Fine-Tuning - Start training your model