OllaMan DocsInstall Online Models
Browse and download models from the online library
Overview
The Discover page provides access to Ollama's extensive online model library with 1000+ pre-trained AI models. You can browse, search, and install any model with just a few clicks.
Browsing the Model Library
Open Discover Page
Click Discover in the sidebar to access the online models library.
Browse Available Models
You'll see a grid of model cards showing:
- Model Name: The official model identifier
- Description: What the model is designed for
- Tags: Model categories and capabilities
- Downloads: Popularity indicator
- Last Update: How recently the model was updated
Search for Models
Use the search box at the top to filter models:
- Type model names (e.g., "llama", "mistral", "codellama")
- Search by capability (e.g., "coding", "chat", "translation")
- Filter by model family or creator
Installing a Model
Select a Model
Click on any model card to open its detail page.
Review Model Details
On the model detail page, you'll find:
- Full Description: Detailed information about the model's capabilities
- Model Family: Related models and versions
- Performance Specs: Context window size, parameter count
- Use Cases: Recommended applications
Choose a Version
Scroll to the Versions section. Here you'll see all available parameter sizes:
Common version sizes:
- 1B - 3B: Small, fast models for basic tasks
- 7B - 13B: Medium models, good balance of speed and quality
- 20B - 70B: Large models, highest quality but require more resources
Each version shows:
- Parameter Size: e.g., 1B, 7B, 70B
- Context Window: Maximum input length
- Model Size: Disk space required
- Quantization: Compression method (Q4_0, Q8_0, etc.)
Click Pull Button
Find the version you want and click the Pull button on its right side.
Monitor Download Progress
The download manager appears at the bottom-right corner, showing:
- Download Progress: Percentage completed
- Download Speed: Current transfer rate (MB/s)
- Size Information: Downloaded / Total size
- Estimated Time: Time remaining
Understanding Model Versions
Parameter Sizes
Small (1B-3B)
- Fastest inference speed
- Lowest memory/disk usage
- Good for simple tasks
- Can run on any hardware
Medium (7B-13B)
- Balanced performance
- Reasonable resource use
- Suitable for most tasks
- Recommended for typical users
Large (20B-70B+)
- Best quality output
- Highest resource requirements
- Complex reasoning tasks
- Needs powerful hardware
Quantization Levels
Quantization reduces model size while maintaining quality:
- Q8_0: Highest quality, larger file size
- Q6_K: Excellent quality, good compression
- Q5_K: Great balance of size and quality
- Q4_0: Smaller size, slightly lower quality
- Q3_K: Most compressed, reduced accuracy
Which Quantization to Choose?
For most users, Q4_0 or Q5_K provides the best balance. Try Q8_0 if you have plenty of disk space and want maximum quality.
Download Management
Multiple Downloads
You can download multiple models simultaneously:
- Each download shows in the download manager
- Downloads run in parallel
- Hover over the manager to see all active downloads
Download States
After Installation
Once download completes:
- Find Your Model: Go to Installed (Local Models) to see the new model
- Start Chatting: Click the Chat button to begin using it
- Check Dashboard: View the model in your statistics
Recommended Models
For General Chat
- llama3.1:8b - Meta's latest, excellent for conversation
- mistral:7b - Fast and capable general-purpose model
- gemma2:9b - Google's efficient chat model
For Coding
- codellama:13b - Specialized for code generation
- deepseek-coder:6.7b - Excellent code understanding
- starcoder2:7b - Multi-language code expert
For Creative Writing
- llama3.1:70b - Top-tier creative output (requires powerful hardware)
- mistral-nemo:12b - Great storytelling capabilities
- nous-hermes2:34b - Creative and coherent narratives
Troubleshooting
Best Practices
Download Tips
- Start small: Try 7B or 13B versions before downloading 70B models
- Check disk space: Ensure you have 2x the model size available
- Use WiFi: Large downloads can consume significant mobile data
- Download overnight: Big models can take hours on slower connections
- Test before committing: Try smaller versions to evaluate the model first