See GLM-5.1 MLX in action - demonstration video
Tested on a M3 Ultra 512GB RAM using Inferencer app
- Single inference ~14.1 tokens/s @ 1000 tokens (debug build)
- Batched inference ~20.4 total tokens/s across two inferences
- Memory usage: ~419.30 GiB
Q4.8-INF typically achieves near lossless accuracy in our coding test, using the data-agnostic INF method tuned to fit within a 512GB memory budget
| Quantization (bpw) | Perplexity | Token Accuracy | Missed Divergence |
|---|
| Q2.5-INF | 1.46875 | 86.10% | 34.79% |
| Q3.5 | 197.0 | 44.05% | 72.00% |
| Q4.5 | 1.35937 | 89.75% | 28.98% |
| Q4.8 | 1.26562 | 93.50% | 19.57% |
| Q4.8-INF | 1.21093 | 97.70% | 10.65% |
| Q5.5 | 1.24218 | 94.60% | 17.55% |
| Q6.5 | 1.21875 | 96.85% | 16.03% |
| Q8.5 | 1.21875 | 97.65% | 9.92% |
| Base | 1.20312 | 100.0% | 0.000% |
- Perplexity: Measures the confidence for predicting base tokens (lower is better)
- Token Accuracy: The percentage of correctly generated base tokens
- Missed Divergence: Measures severity of misses; how much the token was missed by
Quantized with a modified version of MLX
Disclaimer
We are not the creator, originator, or owner of any model listed. Each model is created and provided by third parties. Models may not always be accurate or contextually appropriate. You are responsible for verifying the information before making important decisions. We are not liable for any damages, losses, or issues arising from its use, including data loss or inaccuracies in AI-generated content.
