refactor(hybrid): rename configs — zebra_ for hybrids, FLA names for … - #1007
Open
clairesonglee wants to merge 13 commits into
Open
refactor(hybrid): rename configs — zebra_ for hybrids, FLA names for …#1007clairesonglee wants to merge 13 commits into
clairesonglee wants to merge 13 commits into
Conversation
…pure Reserve zebra_* for recurrent+MLA hybrids; rename pure KDA/GDN to kda_*/gdn_*. Add MI355X hybrid pretrain configs, deprecation symlinks, bridge recipe shim, extends: fixes, and preset validation tests. Co-authored-by: Cursor <cursoragent@cursor.com>
![github-code-quality[bot]](https://avatars.githubusercontent.com/u/9919?s=60&v=4){kind=small}
Resolve hybrid-model doc conflicts: adopt primus-cli launch examples from main with renamed zebra_* / kda_* / gdn_* config paths. Co-authored-by: Cursor <cursoragent@cursor.com>
Remove unused yaml import and apply black formatting for code-lint CI. Co-authored-by: Cursor <cursoragent@cursor.com>
Use kda_1B.yaml / gdn_1B.yaml model presets and
{arch}_{size}-precision-pretrain.yaml experiment configs (aligned with
llama3.2_1B-BF16-pretrain.yaml). Keep *_pure* paths as deprecated symlinks.
Co-authored-by: Cursor <cursoragent@cursor.com>
Adopt Megatron-style experiment names (zebra_llama_kda_1B_BF16-pretrain.yaml) across MI300X/MI325X/MI355X. Keep zebra_*_hybrid-pretrain paths as symlinks. Co-authored-by: Cursor <cursoragent@cursor.com>
Primus pure configs use kda_1B.yaml / kda_1B-precision-pretrain.yaml; update docs and HF/init converters to prefer kda_1B.json (and gdn_1B.json) with fallback to legacy *_pure.json in older FLA checkouts. Co-authored-by: Cursor <cursoragent@cursor.com>
Drop deprecated symlinks and recipe shims, rename pure pretrain configs from -precision- to _BF16-, and update docs, tools, and tests to canonical paths only. Co-authored-by: Cursor <cursoragent@cursor.com>
clairesonglee
marked this pull request as ready for review
August 21, 2026 22:30
clairesonglee
requested review from
Xiaoming-AMD,
limou102 and
wenxie-amd
as code owners
August 21, 2026 22:30
clairesonglee
marked this pull request as draft
August 21, 2026 22:41
Rename configs, Bridge recipes, HF tooling, and docs from zebra/hylo_llama to hylo/hylo_llama so hybrid model naming is consistent end to end. Co-authored-by: Cursor <cursoragent@cursor.com>
clairesonglee
marked this pull request as ready for review
August 21, 2026 23:31
Raise kda_1B and hylo mamba 1B/3B global batch sizes to benchmarked limits on 8× MI300X and annotate all eight configs with sweep TFLOP/s and mem notes. Co-authored-by: Cursor <cursoragent@cursor.com>
Keep GBS, memory, and OOM notes only; tune hylo_llama_mamba_3B to GBS=48. Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
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.
Replace zebra_* model names with new name hylo_*
Reserve hylo_* for recurrent+MLA hybrids; rename pure KDA/GDN to kda_/gdn_.
Add MI355X hybrid pretrain configs, deprecation symlinks, bridge recipe shim, extends: fixes, and preset validation tests.
Keep batch sizes used for convergence experiments in documentation