feat(megatron): enable Run Deck for training sweeps - #427
Draft
atnair-amd wants to merge 2 commits into
Draft
atnair-amd wants to merge 2 commits into
atnair-amd wants to merge 2 commits into
Conversation
A GPU or topology mismatch in the legacy Llama lookup now returns None
instead of erroring every test during module setup. Named-cell charts
accept a single point, bar_pct reuses the numeric guard, crashed combos
stay visible as na, and min gates use the worst sample. Chart x labels
use the same {x} template as other decks.
cursor
Bot
force-pushed
the
atnair/megatron-run-deck-a3e1
branch
from
September 15, 2026 21:11
a722552 to
76e481c
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Adds a render-only sweep Run Deck for Megatron training stems (
megatron_single,megatron_distributed, and the Llama 8B/70B variants via profile aliases).Plan / code surfaces
profiles/megatron.jsonwithdataset_builder: sweepandlayout: named_cellstrain_res_dict,rundeck_variant) in megatron conftest plus the four legacy Llama suite filesGraphs / tables
Review follow-up
rundeck_variantGPU/topology mismatches return None instead of erroring the modulebar_pctuses the same numeric/valueguard as cell records{x}(compatible with the RCCL deck)dimension_keys/metric_prefixnarowsQuality
make fmt-check/make lintclean. Hardware not run. Draft — do not merge yet.Note: this PR still touches shared sweep chart/gate-matrix code; review that path against vLLM/SGLang decks.
Supercedes #421.