The Standard Base Model is the middle-tier language model available on Minibase. It provides a balance between efficiency and capability—larger than the Small Base Model, but still optimized for fast training and deployment.
While not as resource-light as the Small Base Model, it delivers stronger reasoning and flexibility, making it a better choice for more complex tasks, short-form conversations, and instruction following. It’s well suited for developers who want higher output quality while keeping memory usage and compute costs modest.
Typical Use Cases
Classification (multi-label): Handles nuanced or multi-category labeling tasks.
Extraction (structured data): Reliable for pulling entities, values, or fields from text.
Q&A and Dialogue (short-form): Performs well for factual questions and brief conversations.
Instruction Following (intermediate): Good at handling task-specific fine-tunes where consistency matters.
Summarization (concise): Can generate short, accurate summaries.
The Standard Base Model is best when you need a balance of efficiency and quality, without moving up to billion-parameter scale.
Model Specs
# of Parameters: ~360M
Format Type: causal_lm
Download Size: ~360 MB (estimate)
Max Sequence Length: 4,096 tokens
Context Window: 4,096 tokens
This expanded context and parameter size provide higher fidelity responses and better fine-tune headroom, while still being deployable on consumer GPUs and edge servers.
How to Use
Step 1: Choose the model
From the Models tab, select the Standard Base Model when creating a new fine-tuned model.
Step 2: Prepare your dataset
Use Instruction, Input, and Response fields. Examples can be longer than with the Small Base Model, but still keep them structured and consistent.
Example:
Instruction: “Extract the company name from this sentence.”
Input: “OpenAI released a new model today.”
Response: “OpenAI”
Step 3: Train your model
Upload datasets in Excel, JSON, or JSON-L format (standardized internally to JSON-L). Training uses LoRA for efficient adaptation.
Step 4: Test in the browser
On the Model Details page, chat with your model directly. Adjust Temperature (randomness) and Max Tokens (length) to test outputs.
Step 5: Deploy or download
Download in GGUF format for local or on-device use.
Deploy instantly with Minibase Cloud and access through APIs.
Tips & Best Practices
Balance dataset scope: This model can handle slightly broader tasks than the Small Base Model, but still performs best on focused, well-scoped use cases.
Dataset size: Aim for 5,000–25,000 examples for stable fine-tunes. Smaller sets work, but larger ones let the model’s extra capacity shine.
Keep outputs concise: Works best for labels, keywords, summaries, or structured JSON. Long-form creative text is better left for larger models.
Quantization options:
High (Q8_0): Near-lossless quality.
Medium (Q4_K_M): Balanced; recommended default.
Low (Q4_K_S): Compact, fastest, edge-optimized.
Troubleshooting
Responses feel too simple
Responses feel too simple
The Standard Base Model is more capable than the Small Base Model, but not a replacement for very large models. For complex multi-turn reasoning, move up to the Large Base Model.
The model is inconsistent
The model is inconsistent
Make sure your dataset examples are clear, tightly scoped, and demonstrate exactly the expected response format.
Training seems slow
Training seems slow
The Standard 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.