unsloth
GLM 4.5 Air
GLM 4.5 Air is a 110.5B mixture-of-experts glm4moe model from unsloth, with about 23.5B 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.
Parameters110B / 23.5B active
Context128K
Attentiongqa
Licensemit
46 layers · 8 KV heads · head dim 128 · GGUF repo · config
Quant file sizes
| Quant | File size | Source |
|---|---|---|
| BF16 | 205.8 GB | Hugging Face file bytes |
| UD-Q8_K_XL | 119.0 GB | Hugging Face file bytes |
| Q8_0 | 109.4 GB | Hugging Face file bytes |
| UD-Q6_K_XL | 94.6 GB | Hugging Face file bytes |
| Q6_K | 92.2 GB | Hugging Face file bytes |
| UD-Q5_K_XL | 77.2 GB | Hugging Face file bytes |
| Q5_K_M | 77.8 GB | Hugging Face file bytes |
| Q5_K_S | 73.0 GB | Hugging Face file bytes |
| UD-Q4_K_XL | 63.1 GB | Hugging Face file bytes |
| Q4_K_M | 68.0 GB | Hugging Face file bytes |
| Q4_K_S | 62.4 GB | Hugging Face file bytes |
| IQ4_NL | 58.4 GB | Hugging Face file bytes |
| IQ4_XS | 56.3 GB | Hugging Face file bytes |
| Q4_1 | 64.6 GB | Hugging Face file bytes |
| Q4_0 | 58.5 GB | Hugging Face file bytes |
| UD-Q3_K_XL | 51.0 GB | Hugging Face file bytes |
| Q3_K_M | 53.3 GB | Hugging Face file bytes |
| Q3_K_S | 49.0 GB | Hugging Face file bytes |
| UD-IQ3_XXS | 47.9 GB | Hugging Face file bytes |
| UD-Q2_K_XL | 44.2 GB | Hugging Face file bytes |
| Q2_K | 84.6 GB | Hugging Face file bytes |
| UD-IQ2_XXS | 39.8 GB | Hugging Face file bytes |
| UD-IQ1_M | 37.3 GB | Hugging Face file bytes |
| UD-IQ1_S | 35.9 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 | Won't fit | UD-IQ1_S | — |
| GeForce RTX 4060 Ti 16GB | Partial offload | UD-IQ1_M | 7.4–10 |
| GeForce RTX 4090 | Expert offload | UD-Q2_K_XL | 48–65 |
| GeForce RTX 5090 | Expert offload | UD-Q2_K_XL | 85–116 |
| 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, 96–130 tok/s
- NVIDIA L40S — Tight fit, UD-IQ2_XXS, 45–61 tok/s
- NVIDIA A40 — Tight fit, UD-IQ2_XXS, 36–49 tok/s
- RTX 6000 Ada — Tight fit, UD-IQ2_XXS, 50–68 tok/s
- RTX A6000 — Tight fit, UD-IQ2_XXS, 40–54 tok/s
- Apple M1 Ultra — Tight fit, UD-Q2_K_XL, 38–52 tok/s
Run it
Commands for the GeForce RTX 4060 Ti 16GB estimate (UD-IQ1_M).
llama.cpp
llama-server -hf unsloth/GLM-4.5-Air-GGUF:UD-IQ1_M -c 4096 -ngl 16
No matching name in the Ollama library snapshot.
LM Studio
lms get unsloth/GLM-4.5-Air-GGUF # Then load quant UD-IQ1_M