MUSA GPU Training Optimizer
open-infra-skills/infra-skills
Profiles, benchmarks and tunes AI training workloads on Moore Threads MUSA GPUs with a measurement-first process that keeps model behavior unchanged.
pyqula's CPU/GPU switch (src/pyqula/gpu.py), how a routine is routed onto the device, per-call precision, and the tiered porting plan in documentation/gpuportingplan.md.
Install GPU Backend
$ npx skills add joselado/pyqula --skill gpu-backend -a claude-codeProject install by default; add -g for ~/.claude/skills/.
$ gh skill install joselado/pyqula gpu-backend --agent claude-codeProject scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).
$ git clone --depth 1 https://github.com/joselado/pyqula.git skills-src && mkdir -p .claude/skills && cp -r skills-src/.claude/skills/gpu-backend .claude/skills/gpu-backend && rm -rf skills-srcUse ~/.claude/skills/ instead of .claude/skills for a personal install. The folder must contain SKILL.md.
Claude Code skills documentation · loads skills from .claude/skills/
Install the "gpu-backend" agent skill from https://github.com/joselado/pyqula/tree/master/.claude/skills/gpu-backend into .claude/skills/gpu-backend/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "gpu-backend", then confirm the skill loads.Claude Code copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$skill-installer install https://github.com/joselado/pyqula/tree/master/.claude/skills/gpu-backendType this inside Codex. $skill-installer <name> installs a curated skill from openai/skills. The installer writes to $CODEX_HOME/skills (default ~/.codex/skills). Restart Codex if the skill does not show up.
$ npx skills add joselado/pyqula --skill gpu-backend -a codexProject install goes to .agents/skills/; add -g for ~/.codex/skills/.
$ gh skill install joselado/pyqula gpu-backend --agent codexProject scope by default (.agents/skills/); add --scope user for a personal install.
$ git clone --depth 1 https://github.com/joselado/pyqula.git skills-src && mkdir -p .agents/skills && cp -r skills-src/.claude/skills/gpu-backend .agents/skills/gpu-backend && rm -rf skills-srcUse ~/.agents/skills/ instead of .agents/skills for a personal install.
Codex skills documentation · loads skills from .agents/skills/
Install the "gpu-backend" agent skill from https://github.com/joselado/pyqula/tree/master/.claude/skills/gpu-backend into .agents/skills/gpu-backend/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "gpu-backend", then confirm the skill loads.Codex copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ npx skills add joselado/pyqula --skill gpu-backend -a cursorProject install goes to .agents/skills/; add -g for ~/.cursor/skills/.
$ gh skill install joselado/pyqula gpu-backend --agent cursorProject scope by default (.agents/skills/); add --scope user for a personal install.
$ git clone --depth 1 https://github.com/joselado/pyqula.git skills-src && mkdir -p .cursor/skills && cp -r skills-src/.claude/skills/gpu-backend .cursor/skills/gpu-backend && rm -rf skills-srcUse ~/.cursor/skills/ instead of .cursor/skills for a personal install.
Cursor skills documentation · loads skills from .cursor/skills/, .agents/skills/, .claude/skills/, .codex/skills/
Install the "gpu-backend" agent skill from https://github.com/joselado/pyqula/tree/master/.claude/skills/gpu-backend into .cursor/skills/gpu-backend/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "gpu-backend", then confirm the skill loads.Cursor copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ gemini skills install https://github.com/joselado/pyqula.git --path .claude/skills/gpu-backend--scope user (default) or --scope workspace; --path is the subfolder of the repo that holds the skill; --consent skips the security confirmation prompt.
$ npx skills add joselado/pyqula --skill gpu-backend -a gemini-cliProject install goes to .agents/skills/; add -g for ~/.gemini/skills/.
$ gh skill install joselado/pyqula gpu-backend --agent gemini-cliProject scope by default (.agents/skills/); add --scope user for a personal install.
$ git clone --depth 1 https://github.com/joselado/pyqula.git skills-src && mkdir -p .gemini/skills && cp -r skills-src/.claude/skills/gpu-backend .gemini/skills/gpu-backend && rm -rf skills-srcUse ~/.gemini/skills/ instead of .gemini/skills for a personal install, then run /skills reload.
Gemini CLI skills documentation · loads skills from .gemini/skills/, .agents/skills/
Install the "gpu-backend" agent skill from https://github.com/joselado/pyqula/tree/master/.claude/skills/gpu-backend into .gemini/skills/gpu-backend/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "gpu-backend", then confirm the skill loads.Gemini CLI copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ gh skill install joselado/pyqula gpu-backendInstalls for Copilot at project scope by default; add --scope user for a personal install. Preview a skill first with gh skill preview. Needs GitHub CLI 2.90.0 or later (public preview).
$ npx skills add joselado/pyqula --skill gpu-backend -a github-copilotProject install goes to .agents/skills/; add -g for ~/.copilot/skills/.
$ git clone --depth 1 https://github.com/joselado/pyqula.git skills-src && mkdir -p .github/skills && cp -r skills-src/.claude/skills/gpu-backend .github/skills/gpu-backend && rm -rf skills-srcUse ~/.copilot/skills/ instead of .github/skills for a personal install. Commit .github/skills so cloud agent and code review can use it.
GitHub Copilot skills documentation · loads skills from .github/skills/, .claude/skills/, .agents/skills/
Install the "gpu-backend" agent skill from https://github.com/joselado/pyqula/tree/master/.claude/skills/gpu-backend into .github/skills/gpu-backend/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "gpu-backend", then confirm the skill loads.GitHub Copilot copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ npx skills add joselado/pyqula --skill gpu-backend -a opencodeOpenCode documents no install command of its own. Project install goes to .agents/skills/; add -g for ~/.config/opencode/skills/.
$ gh skill install joselado/pyqula gpu-backend --agent opencodeProject scope by default (.agents/skills/); add --scope user for a personal install.
$ git clone --depth 1 https://github.com/joselado/pyqula.git skills-src && mkdir -p .opencode/skills && cp -r skills-src/.claude/skills/gpu-backend .opencode/skills/gpu-backend && rm -rf skills-srcUse ~/.config/opencode/skills/ instead of .opencode/skills for a personal install.
OpenCode skills documentation · loads skills from .opencode/skills/, .claude/skills/, .agents/skills/
Install the "gpu-backend" agent skill from https://github.com/joselado/pyqula/tree/master/.claude/skills/gpu-backend into .opencode/skills/gpu-backend/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "gpu-backend", then confirm the skill loads.OpenCode copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
Facts
- Skill name
gpu-backend- GitHub stars
- 145
- Token cost
- ~679 tokens
- SKILL.md length
- 309 words
- Files
- 1
- Skills in repo
- 6
- Repo updated
- First seen
- Licence
- GPL-3.0
At a glance
pyqula's CPU/GPU switch (src/pyqula/gpu.py), how a routine is routed onto the device, per-call precision, and the tiered porting plan in documentation/gpuportingplan.md.
- Tasks that involve Deep learning
- SKILL.md covers One package-wide switch, The porting plan, and the… and What the GPU is actually good…
- Instructions only: no scripts, shell commands, URLs or credentials in SKILL.md
What it does
GPU Backend is an agent skill from joselado/pyqula. pyqula's CPU/GPU switch (src/pyqula/gpu.py), how a routine is routed onto the device, per-call precision, and the tiered porting plan in documentation/gpuportingplan.md. Load this before moving any compute onto the GPU, before adding a jax code path, before touching kpmtk/kpmjax.py, htk/eigenvectorsjax.py or any other jax.py module, and whenever a task mentions the GPU, CUDA, the device, jax, or making something faster by moving it off the CPU. Each tier of the porting plan wants the maintainer's sign-off before…
Its SKILL.md is about 680 tokens, which your agent loads only when the skill is triggered. It is a single SKILL.md file with no bundled scripts.
It sits in AI & LLM Engineering, covering Deep learning. It works with CUDA. The repository describes itself as: Python library to compute properties of quantum tight binding models, including topological, electronic and magnetic properties and including the effect of many-body interactions. The licence is GPL-3.0.
When your agent uses it
- Tasks that involve Deep learning
Example prompts
- “/gpu-backend”
What it can do on your machine
Read from SKILL.md and the folder at commit a61709a. It shows what the files ask for, not the result of running them.
Tool permissions
Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.
From allowed-tools in the SKILL.md frontmatter.
Runs code
No scripts in the folder and no shell commands in SKILL.md.
From the folder's file list and the shell code blocks in SKILL.md.
Network
No URLs in SKILL.md.
From URLs in SKILL.md, links to its own repository left out.
Credentials
Names no API keys, tokens, secrets or passwords.
From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.
Context cost
GPU Backend loads about 679 tokens when it runs. Until then it costs about 151 tokens; SKILL.md has 309 words of instructions outside code blocks.
Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.
Safety
Auto-check passed
The automated check found no risky patterns in SKILL.md.
Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); files beside SKILL.md are not scanned.
SKILL.md
The full file from joselado/pyqula at commit a61709a, republished under its GPL-3.0 licence (© joselado). 309 words, ~679 tokens.
Download SKILL.mdSave it as
.claude/skills/gpu-backend/SKILL.md (or your agent's skills folder).- name
- gpu-backend
- description
- pyqula's CPU/GPU switch (src/pyqula/gpu.py), how a routine is routed onto the device, per-call precision, and the tiered porting plan in documentation/gpu_porting_plan.md. Load this before moving any compute onto the GPU, before adding a jax code path, before touching kpmtk/kpmjax.py, htk/eigenvectorsjax.py or any other *jax.py module, and whenever a task mentions the GPU, CUDA, the device, jax, or making something faster by moving it off the CPU. Each tier of the porting plan wants the maintainer's sign-off before it starts, so check here before proposing GPU work rather than after.
The GPU backend
One package-wide switch
The CPU/GPU backend is one switch, src/pyqula/gpu.py. gpu.set_gpu(True) puts every
GPU-capable routine on the device and points jax's default device there, so the jax modules
with no backend branch of their own follow it too. The default is the CPU, on every machine.
A new GPU path routes on gpu.get_gpu() rather than growing a switch of its own. Precision
stays per-call: kpm_prec, chi_prec, eigh_prec. The old per-call kpm_cpugpu and
chi_cpugpu arguments were removed and now raise.
Importing gpu.py switches jax to double precision (jax_enable_x64) for the whole
process, and that is the only place it is set. A new jax module imports gpu before it
creates any jax array, rather than setting x64 itself: when eleven modules each set it at
import, the jax code that did not ran in single or double precision depending on import
order.
The porting plan, and the sign-off rule
documentation/gpu_porting_plan.md is the maintainer-facing roadmap for moving
compute-heavy paths onto GPU via jax, which is already a hard dependency. Read it before
starting any GPU-related work in this repo.
| Tier | What | State |
|---|---|---|
| 1 | The batched KPM GPU path (kpmtk/kpmjax.py, kpmtk/kpmnumba.py) | done |
| 2 | Batched dense diagonalization (htk/eigenvectorsjax.py) | done |
| 3 | Scoping the forced-CPU jax modules | done |
| 4 | Why sparse/ARPACK-based Green's-function work is harder and lower priority | not started |
Each tier wants explicit sign-off before it starts. Propose, do not begin.
What the GPU is actually good for here
The shape that pays on the device is a batched dense solve: many independent matrices diagonalized or multiplied at once, large enough that the transfer is amortized. Single sparse solves and ARPACK-style iterative work are the unfavourable case, which is why tier 4 sits where it does.
Performance conclusions belong in the tracked roadmaps. The hostnames, scratch paths and job IDs that produced them do not -- see the HPC rule in CLAUDE.md.
© joselado, GPL-3.0. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file
Files
Just SKILL.md in .claude/skills/gpu-backend of joselado/pyqula.
Open the folder on GitHubat commit a61709a
Compare with similar skills
GPU Backend next to the 5 skills that share the most tags, products or categories with it. Stars are the repository's; “used in” counts other GitHub owners with a copy.
| Skill | Stars | Used in | Tokens | Auto-check | Licence | Repo updated |
|---|---|---|---|---|---|---|
| GPU Backend this skilljoselado/pyqula | 145 | — | ~679 | Automated safety check: Pass | GPL-3.0 | |
| MUSA GPU Training Optimizeropen-infra-skills/infra-skills | 141 | — | ~1.7k | Automated safety check: Pass | Apache-2.0 | |
| DGX Spark Training Gotchaswshobson/agents | 40k | 1 repos | ~2k | Automated safety check: Pass | MIT | |
| Megatron-LM on SLURMNVIDIA/Megatron-LM | 18k | — | ~1.8k | Automated safety check: Pass | Apache-2.0 | |
| Extending Ocannlahrefs/ocannl | 118 | — | ~728 | Automated safety check: Pass | BSD-2-Clause | |
| Mamba State-Space ModelsOrchestra-Research/AI-Research-SKILLs | 13k | 3 repos | ~1.8k | Automated safety check: Pass | MIT |
Similar skills
DGX Spark Training Gotchas
wshobson/agents
Preflight checks and diagnosis for ten known failure modes of ML training on NVIDIA DGX Spark's GB10, spanning launch errors, memory, thermals, bandwidth and precision.
- Official
Megatron-LM on SLURM
NVIDIA/Megatron-LM
Shows how to launch distributed Megatron-LM training on a SLURM cluster: sbatch skeleton, torch.distributed.run setup, CUDA_DEVICE_MAX_CONNECTIONS rules and failure diagnosis.
Extending Ocannl
ahrefs/ocannl
Touch-lists for common OCANNL extension tasks: adding a primitive operation, adding or extending a backend, extending shape inference, and diagnosing output differences between backends.
Mamba State-Space Models
Orchestra-Research/AI-Research-SKILLs
Guide to using Mamba selective state-space models for linear-time sequence modeling, from the Mamba block and pretrained checkpoints to Mamba-2 and speed comparisons.
GPU Optimizer
Mathews-Tom/armory
GPU optimization for consumer NVIDIA GPUs (8-24GB VRAM) covering mixed precision, gradient checkpointing, XGBoost GPU, CuPy/cuDF migration, and torch.compile.
More from joselado/pyqula
Refresh Docs
joselado/pyqula
Refresh pyqula's documentation after a change - recount the test suite, propagate every number that moved, re-run the static user-guide checks, and rebuild documentation/userguide.pdf.
Error Conventions
joselado/pyqula
How pyqula raises errors -- which exception type for which failure, the registries behind string-selected options (mode=, solver=, channel=, operator names), and the shared Hilbert-space guards in…
New Feature
joselado/pyqula
The completeness checklist for adding a user-facing feature to pyqula - where the implementation goes, what kind of test it needs, and the five documentation surfaces that must move with it.
User Guide Voice
joselado/pyqula
The maintainer's writing voice for documentation/userguide.md -- the three registers (chapter prose, section intros, catalogue bullets), the spelling decisions, what not to write, and which chapters…
Wannierization
joselado/pyqula
pyqula's Wannierization (wanniertk/), what h.getwannierhamiltonian() returns, the disentanglement window keywords and which combinations raise NotImplementedError, and the bundled pure-Python…
Related
Works with
Categories
Questions about GPU Backend
What does GPU Backend do?
pyqula's CPU/GPU switch (src/pyqula/gpu.py), how a routine is routed onto the device, per-call precision, and the tiered porting plan in documentation/gpuportingplan.md. GPU Backend is an agent skill from joselado/pyqula.md.
When should I use GPU Backend?
GPU Backend fits situations like: tasks that involve Deep learning.
How do I install GPU Backend in Claude Code?
Run `npx skills add joselado/pyqula --skill gpu-backend -a claude-code`. Or copy the skill folder (.claude/skills/gpu-backend in joselado/pyqula) into .claude/skills/gpu-backend in your project. Claude Code loads it when a task matches its description.
How do I install GPU Backend in Codex?
Run `npx skills add joselado/pyqula --skill gpu-backend -a codex`. Or copy the skill folder (.claude/skills/gpu-backend in joselado/pyqula) into .agents/skills/gpu-backend in your project. Codex loads it when a task matches its description.
Can I use GPU Backend in Cursor, Gemini CLI or GitHub Copilot?
Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add joselado/pyqula --skill gpu-backend -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/gpu-backend, .gemini/skills/gpu-backend, .github/skills/gpu-backend and .opencode/skills/gpu-backend in your project.
What does GPU Backend need to run?
SKILL.md names no scripts, command-line tools or credentials: GPU Backend is instructions for the agent only.
Does GPU Backend access the network?
SKILL.md contains no URLs. Any network use would come from the scripts or tools the agent runs. This is read from the text; nothing was executed.
Is GPU Backend safe to install?
Our automated static check of SKILL.md found no risky patterns, such as piping downloads into a shell, reading credential files or hidden Unicode. It is not a guarantee. Review the folder before installing.
What licence does GPU Backend use?
GPU Backend is published under the GPL-3.0 licence (the repository's licence). It allows redistribution, so the full SKILL.md is shown on this page.
How many tokens does GPU Backend use?
About 679 tokens (SKILL.md is roughly 2.7k characters). Agents keep only the skill's name and description in context until a task matches; then they load SKILL.md in full.
What are the alternatives to GPU Backend?
Skills that share tags, products or a category with GPU Backend: MUSA GPU Training Optimizer (open-infra-skills/infra-skills, 141 stars), DGX Spark Training Gotchas (wshobson/agents, 40k stars), Megatron-LM on SLURM (NVIDIA/Megatron-LM, 18k stars) and Extending Ocannl (ahrefs/ocannl, 118 stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.
Who maintains GPU Backend?
joselado (a GitHub user) maintains it in joselado/pyqula, which has 145 GitHub stars. The repository holds 6 skills in this directory. The repository was last updated on October 7, 2026.
Source: joselado/pyqula on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.