Skip to content

Models and benchmarks

Sample .tic models

Lossless .tic sample models

Compile your own models

Sample models

A few sample models with varying architectures and sizes. Smaller .tic leaves more memory for KV cache. BF16 is supported today, with more precisions in progress.

ModelPrecisionTypeRaw.tic (savings)
Qwen2.5 7B Instruct on Hugging FaceBF16Text15.23 GB10.86 GB (28.7%)
Mistral 7B Instruct v0.3 on Hugging FaceBF16Text14.50 GB10.33 GB (28.8%)
Llama 3.1 8B Instruct on Hugging FaceBF16Text16.06 GB11.44 GB (28.8%)
Gemma 4 12B IT on Hugging FaceBF16Multimodal23.92 GB17.05 GB (28.7%)
Qwen3.5 27B on Hugging FaceBF16Multimodal55.56 GB39.57 GB (28.8%)
Qwen3.5 35B A3B on Hugging FaceBF16Multimodal, MoE71.90 GB51.27 GB (28.7%)

Next steps

Each compiled model ships with a hash manifest for verification.

Benchmarks include memory, latency, and throughput.