API Reference

API Reference

Top-level functions

multi_turboquant.compress(x, config, which="k", layer_idx=0)
multi_turboquant.decompress(compressed, dtype=torch.float16)
multi_turboquant.get_method(CacheMethod.TURBO3)
multi_turboquant.get_preset("balanced")
multi_turboquant.list_presets()
multi_turboquant.recommend_preset(vram_gb, model_size_b, context_length)
multi_turboquant.registered_methods()
multi_turboquant.plan_agents(gpus, model_params_b, desired_agents, desired_context)
multi_turboquant.plan_scenarios(gpus, model_params_b, model_quant)

Integration functions

multi_turboquant.integration.get_llamacpp_args(config)
multi_turboquant.integration.get_llamacpp_command(config, model_path, port, ...)
multi_turboquant.integration.patch_vllm(config)
multi_turboquant.integration.is_vllm_patched()
multi_turboquant.integration.BridgeAdapter(config)

Hardware functions

multi_turboquant.hardware.detect_gpus()
multi_turboquant.hardware.detect_platform()
multi_turboquant.hardware.get_cuda_visible_devices_for_primary(gpus)
multi_turboquant.hardware.get_build_instructions(platform)
multi_turboquant.hardware.get_recommended_engine(platform)

Calibration functions

multi_turboquant.calibration.generate_turboquant_metadata(model_path, recipe)
multi_turboquant.calibration.generate_triattention_stats(model, tokenizer, prompts)
multi_turboquant.calibration.auto_calibrate(config, model_path)

Compatibility functions

multi_turboquant.compatibility.check_config(config, platform)
multi_turboquant.compatibility.get_available_methods(platform)
multi_turboquant.compatibility.get_recommended_config(platform)
multi_turboquant.compatibility.get_cmake_flags(platform)