Repository navigation
Conversation
Image source for the Day-0 Beam release. SGLang is pinned to 9ac1c65b, the tip of the public beam-miles branch on sgl-project/sglang.
Restructure docs/models/reflection/beam.md into the layout of docs/models/index.md (as the Step-5-Preview page in #4104 does): model introduction from the released HF config, both launchers (full parameter and LoRA), the Beam SGLang pin, launch and resume for all three topologies, and per-recipe parallelism, algorithm, optimizer and rollout settings. Register the page after MiMo in docs.json and add a Reflection row to the models table. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Both Beam launchers used prepare/execute/execute-colocate and hardcoded an external Ray cluster. Match the convention #2431 gave every other launcher: - execute -> train; add full-train (prepare + train in one command) - execute-colocate -> a --colocate flag, as run_deepseek_v4.py does - external Ray is now a mode, not an assumption: --join-ray-workers makes the launcher build the cluster from --ray-hostfile, and only without it do we set MILES_SCRIPT_EXTERNAL_RAY The emitted command is unchanged: the train snapshot is byte-identical to the old execute one. run_beam_lora.py had no committed snapshot at all; add it.
Contributor
|
Hi @zhihengy, thanks for putting this up. I'm wondering if you have a rough timeline on merging this into Miles upstream? |
|
Thanks for this PR. I'm following it to add support for Beam in modal-labs/spindle#252 |
This was referenced Nov 12, 2026
This branch has not been deployed
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
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.
ci-sglang-pr: beam-miles
Day-0 RL support for Reflection Beam 1.0 (501B MoE, 23B active): GRPO with SGLang rollout and Megatron training in BF16 — full parameter on 96 H200 (64 colocated), and LoRA on 32 H200. Image:
docker/Dockerfile.beam, i.e.radixark/miles:v0.1.2plus SGLangbeam-milesat9ac1c65b. Recipe page: #4138.miles_plugins/models/beam.py): Beam layer spec on Megatron — sliding-window layers with RoPE and full-attention layers without it (SSMax query scaling), a sigmoid output gate with per-head bias, sandwich norms with depth-scaled residuals, and an FP32 router weight.miles_plugins/megatron_bridge/beam.py): a Megatron-BridgeBeamForCausalLMbridge that loads the BF16 HF checkpoint directly (notorch_distconversion), with the FP32 logit softcap and the fused[q, g, k, v]mapping, including LoRA adapter export.scripts/run_beam.py(executeon 96 H200, async;execute-colocateon 64 H200),scripts/run_beam_lora.py(32 H200), rewardexamples/beam/reward.py, model dimensionsscripts/models/beam.py.Shared changes that touch other models
expert_biasunder any LoRA (previously multi-LoRA only), and passesdata_parallel_random_init.--lora-target-modulesreaches SGLang as module names; the attentionself_attn.gate_projstays qualified and is not counted as an expert projection.register_hf_config_aliasesregisters SGLang'sBeamConfigformodel_type: beam.Branch state: branched from
beam-image(the Day-0 image source, on v0.1.29926b1ca). Before this leaves draft it needs main merged in:miles/backends/megatron_utils/lora/utils.pyand its test conflict, andrelease-lock.jsoncomes from the v0.1.2 release line.🤖 Generated with Claude Code