Qwen3-30B-A3B-Instruct-2507 GGUF (ShapeLearn Quantized)
This is a GGUF-quantized version of Qwen3-30B-A3B-Instruct-2507 produced with ByteShape's ShapeLearn, which learns the optimal datatype per tensor to maintain high quality even at very low bitlengths.
To learn more about ShapeLearn and to see detailed benchmarks across GPUs, CPUs, and even the Raspberry Pi, please visit our blog.
If you have questions or want to share feedback, reach us on Reddit.
How to Pick a Model
We provide CPU and GPU optimized variants for llama.cpp:
- CPUs: Models labeled as KQ, optimized for CPU inference with predominantly KQ quantization.
- GPUs: Models labeled as IQ, optimized for GPU inference with a hybrid approach combining KQ and IQ quantization for better throughput.
Each hardware target includes a range of models covering different size and quality tradeoffs.
The charts below show quality vs tokens per second for each device, comparing ShapeLearn models with Unsloth or MagicQuant baselines.
Selection rule: Choose the model with the highest quality at your target throughput or the fastest model that still meets your required quality.
CPU Models
Table sorted by model size (match the chart numbers to model IDs):
| Model ID | Bits/Weight | Model Size | Normalized Quality |
|---|---|---|---|
| KQ-1 | 2.66 | 10.2 GB | 92.84% |
| KQ-2 | 2.70 | 10.3 GB | 94.18% |
| KQ-3 | 2.85 | 10.9 GB | 95.49% |
| KQ-4 | 3.18 | 12.1 GB | 96.97% |
| KQ-5 | 3.25 | 12.4 GB | 97.97% |
| KQ-6 | 3.61 | 13.8 GB | 98.75% |
| KQ-7 | 3.92 | 14.9 GB | 98.86% |
| KQ-8 | 4.41 | 16.9 GB | 99.34% |
| KQ-9 | 4.67 | 17.8 GB | 99.75% |
GPU Models
Table sorted by model size (match the chart numbers to model IDs):
| Model ID | Bits/Weight | Model Size | Normalized Score |
|---|---|---|---|
| IQ-1 | 2.69 | 10.3 GB | 94.24% |
| IQ-2 | 2.75 | 10.5 GB | 95.48% |
| IQ-3 | 3.02 | 11.5 GB | 95.83% |
| IQ-4 | 3.29 | 12.5 GB | 97.35% |
| IQ-5 | 3.63 | 13.9 GB | 97.74% |
| IQ-6 | 3.87 | 14.8 GB | 98.66% |
| IQ-7 | 4.41 | 16.9 GB | 99.34% |
| IQ-8 | 4.67 | 17.8 GB | 99.75% |
Notes on quantization labels
The labels you see (for example IQ4_XS) are only there to make Hugging Face show our models in the GGUF table. We do not use the conventional quantization profiles as defined in llama.cpp. In our case, these labels indicate the primary quantization approach and average bit length. Note that both KQ and IQ models may use a mix of quantization techniques optimized for their target hardware, which is why several models can share the same tag.
Running these models with Ollama
All GGUF files in this repo can be used directly with Ollama.
To run a model with Ollama, use:
ollama run hf.co/byteshape/Qwen3-30B-A3B-Instruct-2507-GGUF:FILE_NAME.gguf
Replace FILE_NAME.gguf with the GGUF filename you want. For example:
ollama run hf.co/byteshape/Qwen3-30B-A3B-Instruct-2507-GGUF:Qwen3-30B-A3B-Instruct-2507-IQ4_XS-3.63bpw.gguf