Combination Engine

The Combination Engine is the heart of Pruna AI’s optimization framework. Forget manual trial-and-error or juggling a patchwork of tools. Pruna gives you a unified library of 50+ best-in-class algorithms, organized into ten powerful methods, all designed to squeeze maximum efficiency from your models.

To give you an example, one of the base Smash_Config for the Diffusers model is combining 5 algorithms! Disclaimer: the "best" config always depends on your use case ;)

smash_config = SmashConfig()
smash_config["compiler"] = "torch_compile"
smash_config["torch_compile_target"] = "module_list"
smash_config["quantizer"] = "fp8"
smash_config["factorizer"] = "qkv_diffusers"
smash_config["cacher"] = "auto"
smash_config["auto_cache_mode"] = "midpoint"
smash_config["auto_objective"] = "quality"
smash_config["auto_speed_factor"] = 0.46
smash_config["pruner"] = "padding_pruning"
smash_config["padding_pruning_min_sequence_length"] = 128
smash_config._prepare_saving = False
smash_config.add_tokenizer(pipe_t2i.tokenizer_2)

While Pruna is open-source at its core, as a customer you get exclusive access to all state-of-the-art techniques, proprietary methods, and every future release, ensuring your models always stay ahead of the curve.