| // in your build.zig, you can specify a custom test runner: | |
| // const tests = b.addTest(.{ | |
| // .target = target, | |
| // .optimize = optimize, | |
| // .test_runner = "test_runner.zig", // add this line | |
| // .root_source_file = .{ .path = "src/main.zig" }, | |
| // }); | |
| const std = @import("std"); | |
| const builtin = @import("builtin"); |
Good question! I am collecting human data on how quantization affects outputs. See here for more information: ggerganov/llama.cpp#5962
In the meantime, use the largest that fully fits in your GPU. If you can comfortably fit Q4_K_S, try using a model with more parameters.
See the wiki upstream: https://github.com/ggerganov/llama.cpp/wiki/Feature-matrix
Given a Parent class with value property, Child can inherit and overload the property while accessing Parent property getter and setter.
Although we could just reimplement the Child.value property logic completely without using Parent.value whatsover, this would violate the DRY principle and, more important, it wouldn't allow for proper multiple inheritance (as show in the example property_inheritance.py bellow).
Two options:
Child redefines value property completely, both getter and setter.
| ;; markdown-notebook mode | |
| ;; Helps you to insert python code blocks and execute their results | |
| ;; and place output back inside the document. | |
| ;; Install | |
| ;; Put this file somewhere and then add the following line to your emacs.el | |
| ;; (load-file "~/path/to/mdnb.el") | |
| ;; Help | |
| ;; Load the minor mode with M-x markdown-notebook-mode |
Thank you to SpecterOps for supporting this research, to Elad for helping draft this blog, and to Sarah, Daniel, and Adam for proofreading and editing! Crossposted on the SpecterOps Blog.
What follows is the culmination of two years of research with funding by SpecterOps and contributions from many of my coworkers.
Special thanks are needed to Elad, Lee, Will, Daniel, and Kai. Elad, Lee, and Will have contributed several ideas to the project, which are documented here, and have each spent multiple days testing the tool. Daniel has answered all of my inevitable questions about AzureAD (whoops, now Ent
| #!/usr/bin/env python | |
| import math | |
| import matplotlib.pyplot as plt | |
| import torch | |
| import torch.nn as nn | |
| from sklearn.datasets import make_moons | |
| from torch import Tensor | |
| from tqdm import tqdm |
| import mmap | |
| import torch | |
| import json | |
| import os | |
| from huggingface_hub import hf_hub_download | |
| def load_file(filename, device): | |
| with open(filename, mode="r", encoding="utf8") as file_obj: | |
| with mmap.mmap(file_obj.fileno(), length=0, access=mmap.ACCESS_READ) as m: |
Starting in 1.20, the Material class has completely disappeared and has been replaced with a series of properties that are chained to BlockBehavior.Properties.of. This list exists so you can easily figure out which propeties to use to replace the old materials.
Every Material here gives its default MapColor (previously known as MaterialColor), regardless of if it had one or not. It also defines which properties to use to replicate the exact behavior used previously.
The very bottom of this file defines the more complicated replacements to certain material properties.
