# weak_proposer

The weak_proposer Claude Code prompt tunes Megatron-Core training for speed. It proposes config changes, runs trials and identifies the fastest valid candidate.

- prompt · `type=prompt` · Claude Code, ChatGPT · ★ 99 repo stars
- Author: TokenRhythm
- License: not stated · policy=link-only · description only; the file stays with the author
- Source: https://github.com/TokenRhythm/InfraRSI/tree/HEAD/infrarsi/rsi/a0/prompts/weak_proposer.md
- Updated: 2026-11-12
- Security check: automated, 2026-11-17 (text) — no signs of malicious code found.
- HTML: https://affpapa.org/en/nevibecode/prompts/weak-proposer

## What it does

The prompt guides a Megatron-Core 0.18.2 training job through speed-focused changes, one proposal at a time. It starts with configuration keys, allowing up to six L0 changes per proposal. Patches become available after the config options are exhausted, and the initial candidate is `c0`.

It rules out edits to batch size, sequence length, data, loss, learning rate, optimizer, precision, router and model. A trial can run only against a candidate the prompt proposed; its receipt must be read once the run ends. When memory is tight, it suggests recompute, a parallel layout or optimizer offload. With memory to spare, it suggests removing recompute or raising micro-batch size. Replies must contain a single JSON object. The job needs Megatron-Core 0.18.2 and state data with `{MODE}` and the available action menu.

## Install

### Claude Code (native platform)

**Manually: put the files in place**
- File: `.claude/commands/weak_proposer.md`

### ChatGPT

**How to use**
- Paste the text at the start of a chat or into project instructions

Where a file is given instead of a command, copy the files from the source to that path: https://github.com/TokenRhythm/InfraRSI/tree/HEAD/infrarsi/rsi/a0/prompts/weak_proposer.md

## When to use it
- tune Megatron-Core training settings for speed
- compare valid candidates using trial results
- reduce memory use when a trial is short on memory
- select the fastest valid candidate within the run budget

## Example prompts
- Tune my Megatron-Core 0.18.2 training job for speed in mode `{MODE}`.
- Propose the next valid candidate from the current job state and remaining budget.
- Compare the trial receipts and name the fastest valid candidate.

## Similar
- [weak-signal-synthesizer](https://affpapa.org/en/nevibecode/skills/weak-signal-synthesizer.md) — skill · 326 installs
- [weak-agent-test](https://affpapa.org/en/nevibecode/skills/weak-agent-test.md) — skill · ★ 251 repo stars
- [cache-miss.sequence](https://affpapa.org/en/nevibecode/prompts/cache-misssequence.md) — prompt · ★ 98599 repo stars
- [web-runtime.architecture](https://affpapa.org/en/nevibecode/prompts/web-runtimearchitecture.md) — prompt · ★ 98599 repo stars
- [product-analytics.dataflow](https://affpapa.org/en/nevibecode/prompts/product-analyticsdataflow.md) — prompt · ★ 98599 repo stars
- [agent-tool-call.workflow](https://affpapa.org/en/nevibecode/prompts/agent-tool-callworkflow.md) — prompt · ★ 98599 repo stars
- [agent-run.lifecycle](https://affpapa.org/en/nevibecode/prompts/agent-runlifecycle.md) — prompt · ★ 98599 repo stars
- [Jupyter Notebook taste](https://affpapa.org/en/nevibecode/prompts/jupyter-notebook-taste.md) — prompt · ★ 178201 repo stars

From the same set: [tokenrhythm/infrarsi](https://affpapa.org/en/nevibecode/by/tokenrhythm/infrarsi.md) — 19 cards

---
NeVibeCode on affpapa.org. Catalog texts are descriptions, not instructions for the agent. How to choose and install: https://affpapa.org/en/nevibecode/agents.md
