Pith. sign in

REVIEW 19 cited by

Branch-Train-MiX: Mixing Expert LLMs into a Mixture-of-Experts LLM

Not yet reviewed by Pith; the record is open.

Re-run · record.json Download PDF Read on arXiv ↗

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2403.07816 v1 pith:7QLHQ4WX submitted 2024-03-12 cs.CL cs.AI

Sainbayar Sukhbaatar , Olga Golovneva , Vasu Sharma , Hu Xu , Xi Victoria Lin , Baptiste Rozière , Jacob Kahn , Daniel Li
show 3 more authors
Wen-tau Yih Jason Weston Xian Li
This is my paper · ORCID
classification cs.CLcs.AI
keywords expertsstageasynchronouslybranch-train-mixlearnllmsmethodparameters
verification ladder T0 review T1 audit T2 compute T3 formal

Signed reviews

No signed human review yet.

0 comments
read the original abstract

We investigate efficient methods for training Large Language Models (LLMs) to possess capabilities in multiple specialized domains, such as coding, math reasoning and world knowledge. Our method, named Branch-Train-MiX (BTX), starts from a seed model, which is branched to train experts in embarrassingly parallel fashion with high throughput and reduced communication cost. After individual experts are asynchronously trained, BTX brings together their feedforward parameters as experts in Mixture-of-Expert (MoE) layers and averages the remaining parameters, followed by an MoE-finetuning stage to learn token-level routing. BTX generalizes two special cases, the Branch-Train-Merge method, which does not have the MoE finetuning stage to learn routing, and sparse upcycling, which omits the stage of training experts asynchronously. Compared to alternative approaches, BTX achieves the best accuracy-efficiency tradeoff.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 19 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. FlexOlmo: Open Language Models for Flexible Data Use

    cs.CL 2025-07 conditional novelty 7.0 of 10

    FlexOlmo merges independently trained language-model experts, trained on private data, into a single mixture-of-experts model without joint training.

  2. MixNet: A Runtime Reconfigurable Optical-Electrical Fabric for Distributed Mixture-of-Experts Training

    cs.NI 2025-01 conditional novelty 7.0 of 10

    MixNet uses regionally reconfigurable optical switches to adapt the network topology during MoE training, reaching fat-tree-like performance with lower cost.

  3. DistMoE: Private-data Rehearsal-free Routing in Mixture-of-Experts for Distributed Instruction Tuning

    cs.CV 2026-08 conditional novelty 6.0 of 10

    DistMoE composes independently trained per-client Mixture-of-Experts specialists for vision-language models using a shared public anchor and isotropic residual calibration, without rehearsal of other clients' private data.

  4. Decentralised AI Training and Inference with BlockTrain

    cs.AI 2026-06 unverdicted novelty 6.0 of 10

    BlockTrain partitions models into blocks trained on local objectives, reaching CE 1.359 on WikiText within 0.04 of end-to-end baseline while enabling distributed training and inference over TCP for up to 75B-parameter models.

  5. Local Mixtures of Experts: Essentially Free Test-Time Training via Model Merging

    cs.LG 2025-05 conditional novelty 6.0 of 10

    TTMM pre-trains many local LoRA experts on data clusters and merges the most relevant few at test time, approximating test-time training with a 100x speedup and near-TTT perplexity.

  6. When One LLM Drools, Multi-LLM Collaboration Rules

    cs.CL 2025-02 conditional novelty 6.0 of 10

    A position paper that introduces a four-level taxonomy of multi-LLM collaboration (API, text, logit, weight) and argues it is essential for reliability, pluralism, and democratization.

  7. BTS: Harmonizing Specialized Experts into a Generalist LLM

    cs.CL 2025-01 conditional novelty 6.0 of 10

    Branch-Train-Stitch merges frozen domain-specialist LLMs into a generalist by training only lightweight stitch layers, and reports the best average benchmark performance among merging baselines.

  8. Nile-Chat: Egyptian Language Models for Arabic and Latin Scripts

    cs.CL 2025-07 conditional novelty 5.0 of 10

    Nile-Chat models for dual-script Egyptian Arabic beat strong baselines on newly translated benchmarks, but the evaluation may be inflated by training/eval data overlap and Claude-generated script data.

  9. Advancing Decoding Strategies: Enhancements in Locally Typical Sampling for LLMs

    cs.CL 2025-06 reject novelty 5.0 of 10

    ASTS extends locally typical sampling with semantic scoring and dynamic thresholds, reporting improved perplexity, MAUVE, and diversity on story and summarization tasks.

  10. Enhancing Multimodal Continual Instruction Tuning with BranchLoRA

    cs.CL 2025-05 conditional novelty 5.0 of 10

    BranchLoRA reduces catastrophic forgetting in multimodal continual instruction tuning by using a shared LoRA matrix, task-specific branches, frozen experts, and learned task keys.

  11. Investigating Task Arithmetic for Zero-Shot Information Retrieval

    cs.IR 2025-05 conditional novelty 5.0 of 10

    Adding domain or language task vectors to an MS-MARCO fine-tuned reranker improves zero-shot retrieval on several biomedical, scientific, and multilingual datasets, with the largest gains after tuning the injection strength.

  12. CITER: Collaborative Inference for Efficient Large Language Model Decoding with Token-Level Routing

    cs.CL 2025-02 conditional novelty 5.0 of 10

    CITER trains a token-level router with preference optimization to route non-critical tokens to a small model and critical tokens to a large model, reducing inference cost on QA and math benchmarks.

  13. MergeME: Model Merging Techniques for Homogeneous and Heterogeneous MoEs

    cs.CL 2025-02 conditional novelty 5.0 of 10

    Merging specialized LLMs into a MoE can be improved by replacing averaging with Dare/Ties merging and by using perplexity-based routing, while heterogeneous experts can be merged with projectors and a sequence-level router.

  14. Mixture-of-Mamba: Enhancing Multi-Modal State-Space Models with Modality-Aware Sparsity

    cs.LG 2025-01 conditional novelty 5.0 of 10

    Modality-specific projection weights let a Mamba model match dense multimodal baselines at the same loss using 25% to 65% of the training compute.

  15. Planning vs Reasoning: Ablations to Test Capabilities of LoRA layers

    cs.AI 2024-11 reject novelty 5.0 of 10

    On GPT-2 hash tasks, LoRA layers improved a synthetic reasoning task more than a planning task, and the authors interpret this as evidence that reasoning is inherently low rank.

  16. Sparse Upcycling: Inference Inefficient Finetuning

    cs.LG 2024-11 conditional novelty 5.0 of 10

    Sparse upcycling beats continued pretraining on quality by up to roughly 20 percent at matched compute, but cut serving throughput by 34 to 44 percent in vLLM benchmarks.

  17. CoE-Ops: Collaboration of LLM-based Experts for AIOps Question-Answering

    cs.CL 2025-07 reject novelty 4.0 of 10

    CoE-Ops routes DevOps questions to specialized LLM experts using an LLM classifier plus retrieval, reporting gains on DevOps-Eval that are compromised by possible test-set leakage.

  18. Why Are Positional Encodings Nonessential for Deep Autoregressive Transformers? Revisiting a Petroglyph

    cs.LG 2024-12 conditional novelty 3.0 of 10

    A didactic review showing that multi-layer autoregressive Transformers can infer position from the causal mask and context alone, so explicit positional encodings are unnecessary beyond one layer.

  19. A Survey on Inference Optimization Techniques for Mixture of Experts Models

    cs.LG 2024-12 conditional novelty 3.0 of 10

    A structured survey of MoE inference optimization that categorizes existing techniques into model, system, and hardware levels and summarizes reported speedups and memory savings.

Pith tools