Ollama model feed-back
From Notes_Wiki
Revision as of 05:31, 8 January 2025 by Saurabh (talk | contribs) (Created page with "Home > Local system based AI tools > Ollama > Ollama model feed-back We can go to https://ollama.com/ and get information on large no. of models available include their parameter options 9B or 70B etc. and for each type of parameter option we can also see the size of model on the page. Based on the available memory on the system and advantages of particular model we can download it. The name through which can download the model is also writte...")
Home > Local system based AI tools > Ollama > Ollama model feed-back
We can go to https://ollama.com/ and get information on large no. of models available include their parameter options 9B or 70B etc. and for each type of parameter option we can also see the size of model on the page.
Based on the available memory on the system and advantages of particular model we can download it. The name through which can download the model is also written before the size. For example at https://ollama.com/library/phi3 we can see that phi3 is available for 3.8b billion parameters for 2.2GB size and can be downloaded via:
ollama run phi3
Overall for following models I can share feed-back:
- phi3
- medium-128k (14B - 7.9GB) : We get good large context window and decent accuracy
- deepseek-v2
- 16b (16B - 8.9GB) : Based on mixture of experts so it is very fast for its parameeter / memory size.
- gemma2
- 27b (27B - 15 GB) : This is the high parameter count model that can be run on laptop with 32GB RAM. This is very slow compared to others.
- llama3.2 (3B - 2.0 GB)
- : Very small and efficient model that can be run with limited resources
- llama3 (8B - 4.7GB)
- : Decently sized to run on local laptop, be fast, memory efficient and still give reasonable output.
Home > Local system based AI tools > Ollama > Ollama model feed-back