gemma-4-31B-9bit-MLX

See gemma-4-31B-30b MLX in action - demonstration video

Tested on a M3 Ultra 512GB RAM using Inferencer app v1.10.10

• Single inference ~17.1 tokens/s @ 1000 tokens (measured in debug mode)
• Vision inference ~ tokens/s (available from v1.11.0)
• Batched inference ~ total tokens/s across five inferences
• Memory usage: ~33.1 GiB

<p style="margin-bottom:0px;"> <strong>9bpw quant typically achieves near lossless accuracy in our coding test</strong> </p> <table style="border-collapse: collapse; text-align:center; margin-top:10px; margin-bottom:0px;"> <thead> <tr><th>Quantization (bpw)</th><th>Perplexity</th><th>Token Accuracy</th><th>Missed Divergence</th></tr> </thead> <tbody> <tr><td><strong>q4.5</strong></td><td>1.32812</td><td>90.5%</td><td>26.44%</td></tr> <tr><td><strong>q5.5</strong></td><td>1.23437</td><td>95.4%</td><td>16.03%</td></tr> <tr><td><strong>q6.5</strong></td><td>1.21875</td><td>96.85%</td><td>12.55%</td></tr> <tr><td><strong>q8.5</strong></td><td>1.21875</td><td>97.65%</td><td>9.92%</td></tr> <tr><td><strong>q9</strong></td><td>1.21093</td><td>97.95%</td><td>9.61%</td></tr> <tr><td><strong>Base</strong></td><td>1.20312</td><td>100.0%</td><td>0.000%</td></tr> </tbody> </table>

• 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

For more details see demonstration video or visit google/gemma-4-31B-it.

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.