The Large Base Model is the most capable model in the Minibase lineup. With 1.7 billion parameters, it provides substantially deeper reasoning, stronger generalization, and more fluent generation than the smaller models.
While it requires more compute and memory resources, it’s designed for situations where quality and expressiveness matter more than efficiency. The Large Base Model is best for developers who want to push beyond short, simple outputs and into tasks that require more nuance, context, and creativity.
Typical Use Cases
Complex Classification: Handles multi-label and hierarchical categories with higher accuracy.
Structured Extraction: More robust at parsing unstructured text into detailed fields.
Conversational Q&A: Provides stronger dialogue capabilities, supporting multiple turns with greater consistency.
Instruction Following (advanced): Executes more complex instructions reliably.
Summarization (detailed): Can generate longer, more nuanced summaries.
Content Generation: Produces higher-quality free-form text compared to smaller models.
This model is best when response quality, reasoning depth, and flexibility are more important than footprint size.
Model Specs
# of Parameters: ~1.7B
Format Type: causal_lm
Download Size: ~1.7 GB (estimate)
Max Sequence Length: 4,096 tokens
Context Window: 4,096 tokens
The larger parameter count allows for better accuracy and richer outputs, but also means more VRAM and disk space are required for training and deployment.
How to Use
Step 1: Choose the model
From the Models tab, select the Large Base Model (1.7B) when creating a new fine-tuned model.
Step 2: Prepare your dataset
Use Instruction, Input, and Response fields. This model can handle longer, more detailed instructions and outputs, but clarity and consistency are still essential.
Example:
Instruction: “Summarize this paragraph into 2–3 sentences.”
Input: “Artificial intelligence has advanced rapidly in recent years, with new breakthroughs in natural language processing, image recognition, and generative models. These advances have sparked both excitement and concern about the future of work, creativity, and ethics in technology.”
Response: “AI has recently advanced in language, vision, and generative models. These breakthroughs bring new opportunities but also raise questions about jobs and ethics.”
Step 3: Train your model
Upload datasets in Excel, JSON, or JSON-L format (converted internally to JSON-L).
Training uses LoRA for efficient fine-tuning, though larger datasets (10,000+ examples) are recommended to take full advantage of the model’s size.
Step 4: Test in the browser
On the Model Details page, interact with your model. Try adjusting Temperature for more or less creativity and Max Tokens for longer or shorter answers.
Step 5: Deploy or download
Download in GGUF format for local or server-based use.
Deploy instantly with Minibase Cloud and call it via APIs.
Tips & Best Practices
Leverage the scale: This model benefits from larger, richer datasets (10k–20k examples).
Allow longer outputs: The Large Base Model can handle paragraphs and multi-turn responses.
Compute resources: Training and inference require significantly more GPU memory; quantization (Q4_K_M or Q4_K_S) is recommended for deployment.
Quantization options:
High (Q8_0): Maximum quality, larger footprint.
Medium (Q4_K_M): Balanced; recommended for most deployments.
Low (Q4_K_S): Compact; enables edge or lower-resource use cases.
Troubleshooting
The GGUF model download is too large
The GGUF model download is too large
The Large Base Model is resource-heavy. Use quantized formats that can reduce the footprint to optimize for resource consumption.
Responses are inconsistent across long tasks
Responses are inconsistent across long tasks
Provide clear, structured training data and consider instruction fine-tunes with multi-turn examples.
Training seems slow
Training seems slow
The Large Base Model is larger so takes more time to train. Try keeping your use cases narrowly focused and examples clear. Don't try to do too much with a single model.
Need More Help?
Need More Help?
Join our Discord support server to chat with our team and get real-time assistance.