AI Models
Download
|
|
Running Imagix 4D's AI-Powered Code Analysis on a Local AI Server requires downloading and installing GGUF-format AI models that have been trained on software design and analysis. The number of such models is large and growing, you may find models you prefer. Among those we have tested for this application, we have found the following models to achieve reasonable performance on modern 64 bit CPUs.
The times listed below are for comparing the models' relative performance. They show the results when running on an Intel CPU with 32 GB system RAM. By installing the Imagix Local AI Server on a system optimized for running AI due to having either an appropriate graphics card (GPU with sufficient VRAM) or unified memory, noticably higher performance can be achieved.
Hugging Face has the largest storage of open-source GGUF models. The links in this table are to each model's extended description on huggingface.co. From there, you can download the actual .gguf model through the linked webpage's Files tab.
Model
(download) |
Summary |
|---|
| omnicoder-9b-q4_k_m |
Produces the most elaborate and useful answers.Typically runs 4-5 minutes. Needs 10 GB to 12 GB VRAM/RAM. The model file size is roughly 5.5 GB. |
| Qwen3-4B-Instruct-2507-Q4_K_M |
Produces reasonable results most of the time. Usually needs 4-5 minutes. Needs 6 GB to 8 GB VRAM/RAM. The model file size is roughly 2.5 GB. |
| Llama-3.2-3B-Instruct-IQ4_NL |
Produces reasonable results, typically in 2-3 minutes. Needs 6 GB to 8 GB VRAM/RAM. The model file size is roughly 2.0 GB. |
| LFM2.5-1.2B-Instruct-Q8_0 |
Select the 8-bit variant Q8_0 for best results. Runs fast, typically in 1-2 minutes. Results are short but still helpful for explaining Imagix 4D screens. Needs 6 GB VRAM/RAM. The model file size is roughly 1.3 GB. |
|
