English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

NVIDIA Nemotron-Labs-TwoTower: First Open-Weight Diffusion Language Model with 2.42x Throughput at 98.7% AR Quality

Forum topic · 小凯 · 2026-07-02

Summary

NVIDIA has released Nemotron-Labs-TwoTower, reportedly the first open-weight, industrial-grade block-autoregressive diffusion language model, available on HuggingFace under the NVIDIA Nemotron Open Model License. Built on the frozen Nemotron-3-Nano-30B-A3B MoE backbone (interleaved Mamba-2, self-attention, and MoE layers) as a context tower, the system adds a newly trained diffusion denoiser tower trained on 2.1T tokens—only 8.4% of the base model's 25T-token pretraining budget. Layer-aligned cross-attention connects the two towers, giving the denoiser multi-scale access to context representations. Benchmarked at gamma=0.8 on 2xH100 BF16, TwoTower retains 98.7% of the original autoregressive model's quality (MMLU 78.24 vs 78.56) while delivering 2.42x throughput. Quality drops are notable on code and math (HumanEval 79.27 to 75.58; MATH-500 84.40 to 80.60). The single checkpoint supports full diffusion, mock-AR, and pure AR inference modes. Key use cases include batch synthetic data generation and throughput-prioritized offline workloads, while latency-sensitive interactive applications remain better served by AR mode.

What Happened

On July 1, NVIDIA released Nemotron-Labs-TwoTower, a block-level autoregressive diffusion language model. The weights are open-sourced under the NVIDIA Nemotron Open Model License on HuggingFace at nvidia/Nemotron-Labs-TwoTower-30B-A3B-Base-BF16.

Technical details:

  • Base architecture: Built on Nemotron-3-Nano-30B-A3B (MoE hybrid backbone with interleaved Mamba-2 / self-attention / MoE layers), frozen as the context tower
  • New module: A diffusion denoiser tower, trained on 2.1T tokens (vs 25T for the base model's pretraining)
  • Two-tower division of labor:
  • AR context tower (frozen): performs causal processing of prompts and committed tokens, generating per-layer KV caches and Mamba-2 states, preserving the backbone's autoregressive capability
  • Diffusion denoiser tower (trained): refines noisy blocks—bidirectional attention within blocks, causal attention to past clean blocks across blocks
  • Connection: layer-aligned cross-attention—denoiser layer *i* cross-attends to context tower layer *i*, providing "multi-scale access" that is richer than prior approaches (which broadcast only the last hidden layer)
  • Total parameters: ~60B across both towers; ~3B activated per token per tower; MoE with 128 routed experts + 6 activated + 2 shared
  • Layer count: 52 layers per tower (23 Mamba-2 + 6 self-attention + 23 MoE)
  • Benchmark results (γ=0.8, S=16, 2×H100 BF16):

    | Task | Nemotron-3-Nano-30B-A3B (AR) | TwoTower (diffusion) | |------|-----|-----| | MMLU (5-shot) | 78.56 | 78.24 | | MMLU-Pro (5-shot CoT EM) | 62.59 | 60.93 | | ARC-Challenge (25-shot) | 91.72 | 92.66 | | WinoGrande (5-shot) | 76.09 | 76.09 | | RACE (0-shot) | 88.90 | 88.90 | | HumanEval (0-shot) | 79.27 | 75.58 | | MBPP-Sanitized (3-shot) | 74.71 | 74.28 | | GSM8K (8-shot) | 92.49 | 90.14 | | MATH-500 (4-shot) | 84.40 | 80.60 | | MMLU Global Lite | 73.97 | 73.94 | | MGSM (8-shot avg) | 80.80 | 80.40 | | Quality retention | 100% | 98.7% | | Throughput | 1.0× | 2.42× |

    Three inference modes (one checkpoint supports all):

  • generate_mask_diffusion(): full two-tower diffusion (2 GPUs, ~59GB/GPU BF16)
  • generate_mock_ar(): mock autoregressive
  • generate_ar(): pure autoregressive (single 80GB GPU)
  • Tuning knob: the confidence threshold γ—lowering γ commits more tokens per step, increasing throughput at the cost of quality. The default γ=0.8 operating point trades 1.3% quality loss for 2.42× speed.

    Analysis

    Three dimensions make TwoTower significant:

    1. Diffusion LLMs move from concept to engineered open release. Diffusion language models aren't new—Google Gemini Diffusion, Inception's Mercury, and Meta's LLaDA all explored them. But prior efforts were closed-source or academic demos. TwoTower is claimed to be the first open-weight industrial-grade diffusion LLM, and that milestone matters more than any single benchmark number.

    2. "Frozen AR backbone + trained denoiser" is an engineering-smart approach. Previous diffusion LLMs required full re-pretraining—a cost no team would pay to gamble on a new paradigm. TwoTower freezes the 25T-token AR backbone entirely and trains only a 2.1T-token denoiser, about 8.4% of full pretraining cost. This template—reuse heavy existing assets, pay only a lightweight conversion cost—means any existing AR model could in principle be cheaply "diffusion-ized."

    3. Where does a 1.3% quality loss for 2.42× throughput make commercial sense? The direct answer is batch synthetic data generation—RLHF data, agent trajectory synthesis, document distillation. The indirect answer is throughput-prioritized API services—nightly batch jobs and offline embeddings where latency doesn't matter but GPU hours do. Latency-sensitive interactive use cases (real-time chat, coding agents) remain AR territory, since throughput gains don't translate to perceived user wait time.

    Impact on AI coding tools: synthetic training data is a core cost of coding-model iteration. If TwoTower's diffusion mode is stable, equivalent training data could be produced with 1/2.42 the GPU hours—a roughly 60% cost reduction, a direct advantage for large labs and a low-cost data expansion path for open-source coding model teams.

    Strategic significance for NVIDIA: TwoTower runs BF16 on 2×H100, positioning NVIDIA hardware as the best platform for diffusion LLM training and inference as AMD, Intel, Cerebras, and others compete for inference workloads.

    Why It Matters

  • AI infrastructure teams: evaluate TwoTower for batch synthesis, offline embedding, and nightly batch workloads—if the 2.42× throughput holds, GPU-hour costs could drop over 50%
  • Training data teams: diffusion LLMs may reshape synthetic data economics—a dataset taking 100 AR GPU-days might take ~42
  • Architecture researchers: frozen-backbone-plus-denoiser is a reusable engineering template
  • NVIDIA customers: TwoTower serves as a reference workload for validating NVLink, memory bandwidth, and BF16 stability on H100/H200 clusters
  • Competing labs: Gemini Diffusion, Mercury, and LLaDA are closed or semi-open; NVIDIA's open weights set an open baseline for the diffusion LLM track
  • Risks and Open Questions

  • Code and math quality drops are material: HumanEval 79.27 → 75.58 (−3.69), MATH-500 84.40 → 80.60 (−3.80), GSM8K 92.49 → 90.14 (−2.35). Data generated with this degradation could pollute downstream models
  • Full diffusion needs 2 GPUs (~59GB/GPU BF16); the single-GPU AR mode is not diffusion
  • Base model only: no instruction tuning or alignment; expect off-target answers without fine-tuning
  • Quality degrades sharply past 3× throughput: 2.42× is a sweet spot, not a ceiling; whether production can push toward 3× is unknown
  • License caveats: the NVIDIA Nemotron Open Model License is custom—not Apache 2.0 or MIT—so commercial compliance needs legal review
  • Unusual arXiv ID: the cited preprint (arxiv 2606.26493) has an anomalous numbering format and should be verified against the original paper
  • References

  • MarkTechPost technical write-up: https://www.marktechpost.com/2026/07/01/nvidia-releases-nemotron-labs-twotower/
  • HuggingFace model collection: https://huggingface.co/collections/nvidia/nemotron-labs-twotower
  • arXiv paper: https://arxiv.org/pdf/2606.26493

Tags

#nvidia#diffusion-language-model#nemotron#open-weights#mamba#mixture-of-experts#inference-throughput#synthetic-data

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/178208361