inferencerlabs/GLM-4.7-Flash-MLX-6.5bit

NOTICE

No longer available on HF due to storage restrictions - archived here

INFORMATION

See GLM-4.7-Flash MLX in action - demonstration video

Tested on a M3 Ultra 512GB RAM using Inferencer app v1.9.3

• Single inference ~61 tokens/s @ 1000 tokens
• Batched inference ~120 total tokens/s across three inferences
• Memory usage: ~26 GB

q6.5bit quant achieved 1.2265 perplexity in our testing

Quantization	Perplexity	Token Accuracy	Effective Divergence	Missed Divergence
q4.5	1.36718	90.30%	2.8175%	29.04%
q5.5	1.25000	94.45%	0.9635%	17.36%
q6.5	1.22656	96.65%	0.3764%	11.23%
q8.5	1.22656	97.80%	0.2158%	9.808%
Base	1.21875	100.0%	0.0000%	0.000%

• Perplexity: Measures the confidence for predicting base tokens (lower is better)
• Token Accuracy: The percentage of correctly generated base tokens
• Effective Divergence: How much the deviation affected the final output
• 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 GLM-4.7-Flash.

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.