KALAVAI: Predicting When Independent Specialist Fusion Works -- A Quantitative Model for Post-Hoc Cooperative LLM Training

• URL: http://arxiv.org/abs/2603.22755v2
• Date: Fri, 27 Mar 2026 15:25:04 GMT
• Title: KALAVAI: Predicting When Independent Specialist Fusion Works -- A Quantitative Model for Post-Hoc Cooperative LLM Training
• Abstract summary: Independently trained domain specialists can be fused post-hoc into a single model that outperforms any individual specialist.<n>In the KALAVAI protocol, contributors fine-tune copies of a shared checkpoint independently, then submit for lightweight MoE routing (500 steps)
• Score: 3.151184728006369
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Independently trained domain specialists can be fused post-hoc into a single model that outperforms any individual specialist, and the gain is predictable: gain = 0.82 x divergence - 2.72 (R^2 = 0.856, n=6, 3-26% divergence). This enables practitioners to estimate cooperative value before committing compute. Below ~3.3% divergence, gains approach zero.In the KALAVAI protocol, contributors fine-tune copies of a shared checkpoint independently, then submit for lightweight MoE routing (500 steps). Gains are consistent: +7.72% at 410M (+/-0.02%, 3 seeds), +7.49% at 1B (+/-0.01%, 3 seeds), +6.53% at 6.9B, each over the best specialist. The router matches domain-oracle routing within <10^{-5} nats. Cross-lingual fusion (Tamil/Yoruba/Welsh/Code) achieves +21.76%, with Yoruba perplexity falling 41.9 to 7.7. A 20-contributor federation achieves +16.71% (+/-0.07pp, 3 seeds).Three requirements bound the protocol. Shared initialisation is necessary: checkpoint mismatch degrades routing. Frozen layers are optional below ~10,000 steps and beneficial beyond. Learned routing is essential: uniform averaging degrades by -1.2% vs. best specialist, while any trained router achieves oracle-optimal assignment.

Related papers

• From Production Traffic to Post-Training: Building a Self-Hosted LLM That Covers the Corporate Request Mix [55.08143827481755]
  We consolidate traffic from over 200 internal applications onto a single model.<n>We train a separate GRPO expert per axis and merge them via two-stage SLERP.<n>The model absorbs 50% of platform traffic, 116M requests per month, at a fraction of the serving cost.
  arXiv  Detail & Related papers  (2026-09-01T17:39:26Z)
• Annotations as Rollouts: Efficient and Scalable Reinforcement Learning for Video MLLMs [104.11820606942855]
  Video-ORA-9B decodes in 130 ms instead of 4,780 ms without chain-of-thought.<n>OraRL scales with model size and data, surpassing its backbone from 0.8B to 9B and GRPO up to 100k prompts.
  arXiv  Detail & Related papers  (2026-08-20T18:28:19Z)
• LLMs Can Predict Failure Risk, But Struggle to Predict Which Collaboration Protocol Pays Off: Cost-Aware Protocol Routing Across Reasoning Tasks [3.789882665982407]
  Multi-agent large language model (LLM) systems can improve reasoning by spending more computation.<n>We isolate this decision by running every problem under four protocols while holding the solver fixed within each setting.
  arXiv  Detail & Related papers  (2026-08-14T22:35:37Z)
• Router Sensitivity Under Lightweight Fine-Tuning Identifies Prunable Experts in Mixture-of-Experts Models [3.747647072414422]
  Mixture-of-Experts (MoE) models decouple total parameters from per-token compute, but deployment still requires storing every expert.<n>Recent theory shows that pruning experts with the smallest router-norm changes during fine-tuning can preserve accuracy, but assumes full fine-tuning.<n>We test whether lightweight adaptation can recover this signal.
  arXiv  Detail & Related papers  (2026-08-08T03:36:41Z)
• Efficient Visual Pointing for Embodied AI:Agent-Driven Data Synthesis, Cross-Block Attention, and Iterative Correction [55.11480729304395]
  We describe our PointArena 2026 solution, which achieves 77.2% overall accuracy and ranks second on the benchmark.<n>The ap proach targets three failure modes. First, agent-driven syn thesis builds large semantic and anchor-relative candidate pools.<n>Second, a determinis tic steerable-data pipeline creates a verified 10,000-sample main set, plus reserve samples, using masks, templates, and path verification.
  arXiv  Detail & Related papers  (2026-06-29T06:39:03Z)
• Automated Proving of Shannon-Type Entropy Inequalities via Fine-Tuned Language Models and Guided Tree Search [50.16356451328644]
  Proving Shannon-type entropy inequalities is a fundamental task in information theory.<n>We investigate whether small-scale large language models, fine-tuned on atomic proof steps, can automate this process.<n>We show that GPT-5.5 solves 1.7% samples under zero-shot prompting while Psitip solves 33.3% samples.
  arXiv  Detail & Related papers  (2026-06-04T05:43:12Z)
• Beyond Inference-Only Deployment: Comparing Weight-Based Consolidation Against Cascading Compaction [1.6214121783846343]
  We evaluate a nightly consolidation of interaction knowledge into model weights via synthesis, synthesis, and Low-Rank Adaptation (LoRA) fine-tuning on a single consumer.<n>Across ten realistic software development conversations, cascading compaction retain 36.8 +/- 3.0% of knowledge.<n>LoRA consolidates interaction knowledge into model weights via synthesis, synthesis, and Low-Rank Adaptation (LoRA) fine-tuning on a single consumer.
  arXiv  Detail & Related papers  (2026-05-23T16:57:25Z)
• Step-wise Rubric Rewards for LLM Reasoning [72.17879367869503]
  Reinforcement Learning with Verifiable Rewards (RLVR) is widely used to improve reasoning in large language models.<n>We find 18.2% of steps in correct responses are wrong yet positively rewarded.<n>Across six mathematical reasoning benchmarks, SRaR improves average accuracy over RaR by 3.57 points.
  arXiv  Detail & Related papers  (2026-05-17T07:08:14Z)
• Orchard: An Open-Source Agentic Modeling Framework [124.68499958175111]
  We present Orchard, an open-source framework for scalable agentic modeling.<n>At its core is Orchard Env, a lightweight environment service providing reusable primitives for sandbox lifecycle management.<n>On top of Orchard Env, we build three agentic modeling recipes.
  arXiv  Detail & Related papers  (2026-05-14T16:35:12Z)
• More Is Not Always Better: Cross-Component Interference in LLM Agent Scaffolding [4.738949927143789]
  LLM agent systems are built by stacking components (planning, tools, memory, self-reflection, retrieval)<n>We run a full factorial experiment over all 25=32 subsets of five components on HotpotQA and GSM8K with Llama-3.1-8B/70B conditions.<n>The All-In system is consistently suboptimal: on HotpotQA, a single-tool agent surpasses All-In by 32% (F1 0.233 vs 0.177, p23)<n>Our findings suggest maximally-equipped agent should be replaced by subset selection via interaction-aware analysis.
  arXiv  Detail & Related papers  (2026-05-07T06:01:43Z)
• Equifinality in Mixture of Experts: Routing Topology Does Not Determine Language Modeling Quality [0.0]
  We build a sparse Mixture-of-Experts (MoE) architecture using cosine-similarity routing against learned centroids.<n>We find that routing topology does not determine perplexity (PPL)<n>Expert-level and causal controllability -- which coexist with topology-level equifinality -- are explored in a companion paper.
  arXiv  Detail & Related papers  (2026-04-15T21:02:02Z)
• D2-LoRA: A Synergistic Approach to Differential and Directional Low-Rank Adaptation [0.24636535146231608]
  D2-LoRA achieves 76.4 percent average accuracy across eight question answering and reading comprehension benchmarks.<n>The method combines signed low-rank residual updates with additive and subtractive components, together with a train-time column-wise projection.
  arXiv  Detail & Related papers  (2026-02-16T13:19:42Z)
• PROTEUS: SLA-Aware Routing via Lagrangian RL for Multi-LLM Serving Systems [1.0978496459260902]
  We present PROTEUS, a router that accepts accuracy targets tau as runtime input.<n>A single trained model serves the full accuracy spectrum without retraining.<n>We evaluate on RouterBench (11 models, 405K queries) and SPROUT (14 models, 45K queries)
  arXiv  Detail & Related papers  (2026-01-27T09:38:16Z)
• MatchTIR: Fine-Grained Supervision for Tool-Integrated Reasoning via Bipartite Matching [60.886768806064936]
  Tool-Integrated Reasoning empowers large language models to tackle complex tasks by interleaving reasoning steps with external tool interactions.<n>Existing reinforcement learning methods rely on outcome- or trajectory-level rewards, assigning uniform advantages to all steps within a trajectory.<n>We propose MatchTIR, a framework that introduces fine-grained supervision via bipartite matching-based turn-level reward assignment and dual-level advantage estimation.
  arXiv  Detail & Related papers  (2026-01-15T18:59:23Z)
• Towards a Science of Scaling Agent Systems [79.64446272302287]
  We formalize a definition for agent evaluation and characterize scaling laws as the interplay between agent quantity, coordination structure, modelic, and task properties.<n>We derive a predictive model using coordination metrics, that cross-validated R2=0, enabling prediction on unseen task domains.<n>We identify three effects: (1) a tool-coordination trade-off: under fixed computational budgets, tool-heavy tasks suffer disproportionately from multi-agent overhead, and (2) a capability saturation: coordination yields diminishing or negative returns once single-agent baselines exceed 45%.
  arXiv  Detail & Related papers  (2025-12-09T06:52:21Z)
• Unifying Mixture of Experts and Multi-Head Latent Attention for Efficient Language Models [1.7272658301768147]
  MoE-MLA-RoPE is a novel architecture combination that combines Mixture of Experts (MoE) with Multi-head Latent Attention (MLA) and Rotary Position Embeddings (RoPE) for efficient language modeling.<n>Our approach addresses the fundamental trade-off between model capacity and computational efficiency through three key innovations.
  arXiv  Detail & Related papers  (2025-08-02T08:33:30Z)
• Scalable Reinforcement Post-Training Beyond Static Human Prompts: Evolving Alignment via Asymmetric Self-Play [52.3079697845254]
  eva is the first method that allows language models to adaptively create training prompts in both offline and online RL post-training.<n>We show eva can create effective RL curricula and is robust across ablations.
  arXiv  Detail & Related papers  (2024-10-31T08:15:32Z)
• Advancing LLM Reasoning Generalists with Preference Trees [119.57169648859707]
  We introduce Eurus, a suite of large language models (LLMs) optimized for reasoning. Eurus models achieve state-of-the-art results among open-source models on a diverse set of benchmarks.
  arXiv  Detail & Related papers  (2024-04-02T16:25:30Z)
• End-to-End Sensitivity-Based Filter Pruning [49.61707925611295]
  We present a sensitivity-based filter pruning algorithm (SbF-Pruner) to learn the importance scores of filters of each layer end-to-end. Our method learns the scores from the filter weights, enabling it to account for the correlations between the filters of each layer.
  arXiv  Detail & Related papers  (2022-04-15T10:21:05Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.