unsloth
Qwen3.5 122B A10B MTP
Qwen3.5 122B A10B MTP is a 124.6B mixture-of-experts qwen35moe model from unsloth, with about 10.0B parameters active per token. Configured context length is 262,144 tokens. License: apache-2.0. Quant sizes are the GGUF file bytes published on Hugging Face when a file exists.
Parameters125B / 10.0B active
Context256K
Attentionhybrid
Licenseapache-2.0
48 layers · 2 KV heads · head dim 256 · 36 linear-attention layers · GGUF repo · config
Quant file sizes
| Quant | File size | Source |
|---|---|---|
| BF16 | 232.2 GB | Hugging Face file bytes |
| UD-Q8_K_XL | 125.7 GB | Hugging Face file bytes |
| Q8_0 | 123.5 GB | Hugging Face file bytes |
| UD-Q6_K_XL | 106.1 GB | Hugging Face file bytes |
| UD-Q6_K | 97.0 GB | Hugging Face file bytes |
| UD-Q5_K_XL | 87.4 GB | Hugging Face file bytes |
| UD-Q5_K_M | 87.2 GB | Hugging Face file bytes |
| UD-Q5_K_S | 82.2 GB | Hugging Face file bytes |
| UD-Q4_K_XL | 73.2 GB | Hugging Face file bytes |
| UD-Q4_K_M | 72.9 GB | Hugging Face file bytes |
| UD-Q4_K_S | 68.4 GB | Hugging Face file bytes |
| UD-IQ4_XS | 57.7 GB | Hugging Face file bytes |
| IQ4_NL | 58.8 GB | Hugging Face file bytes |
| UD-Q3_K_XL | 54.4 GB | Hugging Face file bytes |
| Q3_K_M | 54.2 GB | Hugging Face file bytes |
| UD-IQ3_S | 48.5 GB | Hugging Face file bytes |
| UD-IQ3_XXS | 44.6 GB | Hugging Face file bytes |
| UD-Q2_K_XL | 39.9 GB | Hugging Face file bytes |
| UD-IQ2_XXS | 37.5 GB | Hugging Face file bytes |
| UD-IQ1_M | 36.0 GB | Hugging Face file bytes |
On common machines
| Device | Verdict | Quant | Speed |
|---|---|---|---|
| GeForce RTX 4060 Ti 8GB | Won't fit | UD-IQ1_M | — |
| GeForce RTX 3060 12GB | Partial offload | UD-IQ1_M | 20–27 |
| GeForce RTX 4060 Ti 16GB | Partial offload | UD-IQ2_XXS | 20–27 |
| GeForce RTX 4090 | Partial offload | UD-IQ3_XXS | 23–31 |
| GeForce RTX 5090 | Partial offload | UD-Q3_K_XL | 21–29 |
| Apple M4 Pro | Won't fit | UD-IQ1_M | — |
| Apple M5 | Won't fit | UD-IQ1_M | — |
| CPU only · DDR5-5600 | Won't fit | UD-IQ1_M | — |
Smallest devices that fit it fully
- NVIDIA A100 40GB — Tight fit, UD-IQ1_M, 259–353 tok/s
- NVIDIA L40S — Tight fit, UD-Q2_K_XL, 131–178 tok/s
- NVIDIA A40 — Tight fit, UD-Q2_K_XL, 105–143 tok/s
- RTX 6000 Ada — Tight fit, UD-Q2_K_XL, 145–198 tok/s
- RTX A6000 — Tight fit, UD-Q2_K_XL, 116–158 tok/s
- Apple M1 Ultra — Tight fit, UD-IQ3_XXS, 109–149 tok/s
Run it
Commands for the GeForce RTX 4060 Ti 16GB estimate (UD-IQ2_XXS).
llama.cpp
llama-server -hf unsloth/Qwen3.5-122B-A10B-MTP-GGUF:UD-IQ2_XXS -c 4096 -ngl 18
Ollama
ollama run qwen3.5
LM Studio
lms get unsloth/Qwen3.5-122B-A10B-MTP-GGUF # Then load quant UD-IQ2_XXS