Following this reddit post: https://www.reddit.com/r/StrixHalo/comments/1w12gs2/qwen38flashnext_125ba6b_finally_running_on_my/
Step-by-step guide to run Qwen3.8-Flash-Next (125B MoE, ~6B active) locally on a Strix Halo machine (gfx1151) via ROCm and the EngramHalo llama.cpp fork, with MTP speculative decoding and 131K context.
Verified on: Framework Desktop, Ryzen AI MAX+ 395 (Radeon 8060S / gfx1151),
Running apt update produces the following warning:
W: https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2404/x86_64/InRelease:
Key is stored in legacy trusted.gpg keyring (/etc/apt/trusted.gpg), see the
DEPRECATION section in apt-key(8) for details.
OpenCV does a reasonable job of reading videos from file or webcams. It's simple and mostly works. When it comes to writing videos, it however leaves a lot to be desired. There is little control over the codecs and it is almost impossible to know which codecs are installed. It also wants to know things like the frame size at intailisation. This isn't always a problem, but if you don't know it yet it means you have to set up the video writer inside your main processing loop.
To make something as cross-platform compatible as possible it would be nice to use FFmpeg. There are a few python wrappers around, but as far as I can tell they are mainly used for transcoding type applications. One solution is run FFmpeg as a subprocess and set its input to accept a pipe. Then every video frame is passed through the pipe. You write this yourself, in fact it's only a few lines of code. However, the scikit-video package will do this for us, with some nice boilerplate to ma
(Codecs are extracted from https://web.archive.org/web/20120722124832/http://opencv.willowgarage.com/wiki/QuickTimeCodecs )
| brew update && brew install pyenv pyenv-virtualenv | |
| echo 'eval "$(pyenv init --path)"' >> ~/.zprofile | |
| echo 'eval "$(pyenv init -)"' >> ~/.zshrc | |
| echo 'eval "$(pyenv virtualenv-init -)"' >> ~/.zshrc | |
| brew install zlib sqlite bzip2 libiconv libzip openssl rust | |
| CFLAGS="-I$(brew --prefix openssl)/include -I$(brew --prefix bzip2)/include -I$(brew --prefix readline)/include -I$(brew --prefix zlib)/include -I$(brew --prefix sqlite)/include -I$(xcrun --show-sdk-path)/usr/include" LDFLAGS="-L$(brew --prefix openssl)/lib -L$(brew --prefix bzip2)/lib -L$(brew --prefix readline)/lib -L$(brew --prefix zlib)/lib -L$(brew --prefix sqlite)/lib" pyenv install miniforge3-4.10.3-10 |
| """https://stackoverflow.com/questions/14906764/how-to-redirect-stdout-to-both-file-and-console-with-scripting""" | |
| import sys | |
| import traceback | |
| class Tee(object): | |
| """Context manager that copies stdout and any exceptions to a log file. | |
| Example usage: | |
| [style] | |
| based_on_style = pep8 | |
| column_limit = 120 | |
| coalesce_brackets = True | |
| dedent_closing_brackets = False | |
| each_dict_entry_on_separate_line = True | |
| spaces_before_comment = 2 | |
| split_before_first_argument = False | |
| split_before_named_assigns = False | |
| split_all_top_level_comma_separated_values = True |
| # Mac | |
| .DS_Store? | |
| ehthumbs.db | |
| Icon? | |
| Thumbs.db | |
| *.DS_Store | |
| # Windows | |
| Thumbs.db |
NOTE: This is a question I found on StackOverflow which I’ve archived here, because the answer is so effing phenomenal.
If you are not into long explanations, see [Paolo Bergantino’s answer][2].
| " Use Vim settings, rather then Vi settings (much better!). | |
| " This must be first, because it changes other options as a side effect. | |
| set nocompatible | |
| " ================ General Config =================== | |
| set number " Display line number | |
| set ruler " Display column number | |
| set showmode " Show current mode at bottom | |
| set visualbell " No Sounds | |
| set autoread " Reload file changed outside vim |