Orthrus-Qwen3: up to 7.8×tokens/forward on Qwen3, identical output distribution

github.com

207 points by FranckDernoncou 21 hours ago


boredatoms - an hour ago

So will this help openai/anthropic have lower congestion in the afternoons if they implement something similar?

xiphias2 - 15 hours ago

The most interesting part of this idea for me is how it wasn't tried / implemented before, as it makes sense.

I haven't read the paper but of course DTree tricks work here as well

bertili - 12 hours ago

Does this translate into a similar reduction in compute?

What's the catch?

DeathArrow - 10 hours ago

If someone can make this work with GGUF and Quantized Qwen 3.6 or Deepseek 4 it would greatly help running local models.

dnlserrano - 6 hours ago

I wonder what our man @antirez will make of this

spwa4 - 9 hours ago

I don't understand. This distills a diffusion transformer out of Qwen3. And while the provably identical is nice, a full diffusion transformer would be a lot faster still.

FranckDernoncou - 21 hours ago

Paper: https://arxiv.org/abs/2605.12825 ; Code+models: https://github.com/chiennv2000/orthrus ; Disclosure: co-author.

Idea: Inject a trainable diffusion attention module into each layer of a frozen AR Transformer. Both heads share one KV cache. Diffusion head projects K=32 tokens in parallel; AR head verifies in a second pass and accepts the longest matching prefix. Output distribution is provably identical to the base model.

Results:

- Up to 7.8x TPF, ~6x wall-clock on MATH-500.

- 16% of params trained, <1B tokens, 24h on 8xH200.

- vs. diffusion LMs (Dream, Fast-dLLM-v2, SDAR, Mercury, Gemini Diffusion): they modify base weights and lose accuracy (Fast-dLLM-v2: -11 pts on MATH-500). Orthrus freezes the backbone; accuracy matches Qwen3-8B exactly.

- vs. Speculative Decoding (EAGLE-3, DFlash): no external drafter, no separate cache, zero TTFT penalty (no drafter to init/sync). KV overhead is O(1) (~4.5 MiB flat). Acceptance length on MATH-500: 11.7 vs. 7.9 (DFlash) vs. 3.5 (EAGLE-3).

- Single-step denoising beats multi-step (6.35 vs. 3.53 TPF). KL distillation beats CE on acceptance rate.

Limitations: strictly bounded by the frozen base model (inherits its biases, hallucinations, knowledge gaps); Qwen3-only evaluation; greedy + rejection sampling only.

GeorgeToresco - 3 hours ago

[flagged]

holotherapper - 7 hours ago

[dead]