Quantized builds of the UIGEN-X 30B Mixture-of-Experts coding assistant for local inference with Ollama / llama.cpp runtimes. Each variant ships with the Modelfile exported from the Ollama registry plus the corresponding GGUF binary.

Base model: smirki/UIGEN-X-30B-MoE-merged-checkpoint-200
Architecture: 30B parameter MoE tuned for software engineering tasks.

Variants

Variant	Size	Blob
`q2_k`	10.49 GB	`sha256-89dc7fdb0a4a30a6bd4e8a611db45fd821ccce4748d5bda1fc5151b5be0fc0fd`
`q3_k_s`	12.38 GB	`sha256-5dc18116ed7f2c98b96361b4a12e8b43fb4e75ee3dc162ba73a74226a3621a3c`
`Q4_K_M`	17.28 GB	`sha256-4c641495ea35d559011305746eeda3b9cc1c3ef6cabce16c2c260962715957e5`
`Q5_K_M`	20.23 GB	`sha256-3848f1b66aeee5b454ad3c3d6133da2723e84e7cfa0bee9e67d81639cdbd9b4d`
`Q6_K`	23.37 GB	`sha256-4a08030b66cb100eff2c183500b7c96ef44e3f4816911d32fe55a1e2a4aa1d47`
`Q8_0`	30.25 GB	`sha256-fc84a665b1889d7c6bed55b2b5dedee464ba80400bf1fbd43164b48eb219e2a7`

Usage with Ollama

Example with the Q5_K_M quantization:

bash

ollama create uigen-x-30b-moe-q5-k-m -f modelfiles/uigen-x-30b-moe--Q5_K_M.Modelfile
ollama run uigen-x-30b-moe-q5-k-m

Swap Q5_K_M for any other variant listed above.

Source

Originally published on my Ollama profile: https://ollama.com/richardyoung/uigen-x-30b-moe