Skip to content

Add Reflection Beam 256k Miles LoRA recipe - #762

Merged
zhouhelena1 merged 16 commits into
mainfrom
codex/reflection-beam
Oct 27, 2026
Merged

zhouhelena1 merged 16 commits into
mainfrom
codex/reflection-beam

Conversation

@zhouhelena1

@zhouhelena1 zhouhelena1 commented Oct 26, 2026 •

Copy link
Copy Markdown
Contributor

Adds Reflection Beam 1.0 Miles LoRA recipe, following Miles #4149.

Uses 32 H200s across 4 nodes: 16 for training and 16 for rollout, TP8/PP1/CP2/EP16, expert TP1, rank 16 / alpha 32, and a 256k context length (224k prompt + 32k response). Training uses upstream's 32 prompts × 8 completions (global batch size 256), microbatch size 1, FlashAttention, full activation recomputation, and asynchronous trainer/rollout execution.

Upstream LoRA defaults to async and supports synchronous execution with --no-async-training. This Dojo recipe likewise supports async_mode=False with the same separate 16-trainer/16-rollout GPU allocation.

Autoconfig: Beam 256k DAPO-math

Batch-256 validation generated all 256 responses from 32 unpadded DAPO-math prompts with normal EOS stopping (2,286,539 output tokens; 240 natural completions, 16 at the 32k cap). Training then exposed a CP/TP padding-alignment assertion. Follow-up fix #765 preserves batch 256 and validates training against that saved batch. This tests the 256k context cap without forcing every sample to fill the window; the earlier batch-eight runs below validate full-window capacity.

Validation runs:

Retained recipe-specific patches:

  • Context parallelism: propagate cp_comm_type through the Miles bridge and normalize singleton CLI lists for Beam's sliding-window attention.
  • Expert LoRA memory: reuse unpadded inputs and accumulate adapter outputs without redundant full-width buffers.
  • Expert activation memory: chunk unfused SwiGLU temporaries while preserving BF16 rounding.
  • LoRA timing: adapt the shared step/substep reporting patch to the pinned Miles executor-based driver.

@github-actions

Copy link
Copy Markdown

@andrewhinh Andrew Hinh (andrewhinh) left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

i left non-blocking comments re hygiene. and plz clean up the docstrings/comments.

Comment thread modal_dojo/common/models/validation.py Resolved

This branch was successfully deployed

1 active deployment
modal-preview — d70fe3f6 Deployed Oct 27, 2026 by zhouhelena1 via preview #7317
modal — d70fe3f6 Deployed Oct 27, 2026 by zhouhelena1 via validate (Qwen3.5-9B) #3212
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants