unsloth
GLM 4.6V
GLM 4.6V is a 106.9B mixture-of-experts glm4moe model from unsloth, with about 22.7B parameters active per token. Configured context length is 131,072 tokens. License: mit. Quant sizes are the GGUF file bytes published on Hugging Face when a file exists.
Parameters107B / 22.7B active
Context128K
Attentiongqa
Licensemit
46 layers · 8 KV heads · head dim 128 · GGUF repo · config
Quant file sizes
| Quant | File size | Source |
|---|---|---|
| BF16 | 199.1 GB | Hugging Face file bytes |
| UD-Q8_K_XL | 115.4 GB | Hugging Face file bytes |
| Q8_0 | 105.8 GB | Hugging Face file bytes |
| UD-Q6_K_XL | 91.6 GB | Hugging Face file bytes |
| Q6_K | 89.3 GB | Hugging Face file bytes |
| UD-Q5_K_XL | 74.8 GB | Hugging Face file bytes |
| Q5_K_M | 75.2 GB | Hugging Face file bytes |
| Q5_K_S | 70.7 GB | Hugging Face file bytes |
| UD-Q4_K_XL | 61.0 GB | Hugging Face file bytes |
| Q4_K_M | 65.7 GB | Hugging Face file bytes |
| Q4_K_S | 60.4 GB | Hugging Face file bytes |
| IQ4_NL | 56.4 GB | Hugging Face file bytes |
| IQ4_XS | 54.3 GB | Hugging Face file bytes |
| Q4_1 | 62.5 GB | Hugging Face file bytes |
| Q4_0 | 56.6 GB | Hugging Face file bytes |
| UD-Q3_K_XL | 49.1 GB | Hugging Face file bytes |
| Q3_K_M | 51.6 GB | Hugging Face file bytes |
| Q3_K_S | 47.4 GB | Hugging Face file bytes |
| UD-IQ3_XXS | 46.3 GB | Hugging Face file bytes |
| UD-Q2_K_XL | 42.6 GB | Hugging Face file bytes |
| Q2_K | 81.8 GB | Hugging Face file bytes |
| UD-IQ2_XXS | 38.4 GB | Hugging Face file bytes |
| UD-IQ1_M | 35.9 GB | Hugging Face file bytes |
| UD-IQ1_S | 34.5 GB | Hugging Face file bytes |
On common machines
| Device | Verdict | Quant | Speed |
|---|---|---|---|
| GeForce RTX 4060 Ti 8GB | Won't fit | UD-IQ1_S | — |
| GeForce RTX 3060 12GB | Partial offload | UD-IQ1_S | 7.7–10 |
| GeForce RTX 4060 Ti 16GB | Partial offload | UD-IQ2_XXS | 7.1–9.7 |
| GeForce RTX 4090 | Expert offload | UD-Q2_K_XL | 49–67 |
| GeForce RTX 5090 | Expert offload | UD-Q2_K_XL | 87–119 |
| Apple M4 Pro | Won't fit | UD-IQ1_S | — |
| Apple M5 | Won't fit | UD-IQ1_S | — |
| CPU only · DDR5-5600 | Won't fit | UD-IQ1_S | — |
Smallest devices that fit it fully
- NVIDIA A100 40GB — Tight fit, UD-IQ1_S, 91–124 tok/s
- NVIDIA L40S — Tight fit, UD-Q2_K_XL, 45–62 tok/s
- NVIDIA A40 — Tight fit, UD-Q2_K_XL, 36–50 tok/s
- RTX 6000 Ada — Tight fit, UD-Q2_K_XL, 50–69 tok/s
- RTX A6000 — Tight fit, UD-Q2_K_XL, 40–55 tok/s
- Apple M1 Ultra — Tight fit, UD-Q2_K_XL, 39–53 tok/s
Run it
Commands for the GeForce RTX 4060 Ti 16GB estimate (UD-IQ2_XXS).
llama.cpp
llama-server -hf unsloth/GLM-4.6V-GGUF:UD-IQ2_XXS -c 4096 -ngl 16
No matching name in the Ollama library snapshot.
LM Studio
lms get unsloth/GLM-4.6V-GGUF # Then load quant UD-IQ2_XXS