SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2602.04804 · ICML · 2026

OmniSIFT: Modality-Asymmetric Token Compression for Efficient Omni-modal Large Language Models

Qiang Liu, Liang Wang, Pengfei Wan, Jiaheng Liu, Junfei Wu, Xuyang Liu, Yue Ding, Jungang Li, Yang Shi, Xinlong Chen, Bohan Zeng, Yushuo Guan, and 4 more

arXiv · PDF · Open in the Atlas

Code that ran

1 of 1 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 1 failure of our instrument; 0 no recorded run), measured 2026-11-22; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2602.04804.json · badge: /api/ran/2602.04804.svg

Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.

RepositoryRoleRan
dingyue772/OmniSIFT community
Papers with Code archive
1 ran · 0 no recorded run
FunctionStatusWhere it lives
similarity_pruning Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
dingyue772/OmniSIFT/omnisift/compression_units.py
code served (permissive licence) · get_code("9ce0b2117b32b5bf")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

# Create a tiny compressioner model
class Compressioner(nn.Module):
    def __init__(self, embed_dim=4):
        super().__init__()
        self.linear = nn.Linear(embed_dim * 2, 1)
    
    def forward(self, v_embeds, a_embeds):
        # v_embeds: (num_v, embed_dim)
        # a_embeds: (num_a, embed_dim)
        # Output scores for each audio token based on video context
        batch_scores = []
        for a_emb in a_embeds:
            for v_emb in v_embeds:
                combined = torch.cat([v_emb, a_emb], dim=-1)
                score = self.linear(combined)
                batch_scores.append(score)
        if batch_scores:
            return torch.cat(batch_scores, dim=0)[:len(a_embeds)]
        else:
            return torch.zeros(len(a_embeds), device=a_embeds.device)

compressioner = Compressioner(embed_dim=4)

# Create tiny input tensors
batch_size = 2
seq_len = 12
embed_dim = 4

inputs_embeds = torch.randn(batch_size, seq_len, embed_dim)

# Create input_ids with some audio and video tokens
input_ids = torch.tensor([
    [1, 2, 2, 3, 3, 1, 2, 2, 3, 3, 1, 1],  # batch 0
    [1, 2, 2, 3, 3, 1, 2, 2, 3, 3, 1, 1],  # batch 1
])

audio_token_id = 2
video_token_id = 3

# Create position_ids: shape (3, batch_size, seq_len) for [time, height, width]
position_ids = torch.zeros((3, batch_size, seq_len), dtype=torch.long)
# time dimension - assign frame ids
position_ids[0, :, :] = torch.tensor([[0, 0, 0, 1, 1, 1, 2, 2, 2, 3, 3, 3],
                                       [0, 0, 0, 1, 1, 1, 2, 2, 2, 3, 3, 3]])
# height dimension
position_ids[1, :, :] = torch.randint(0, 2, (batch_size, seq_len))
# width dimension
position_ids[2, :, :] = torch.randint(0, 2, (batch_size, seq_len))

merging_ratio_a = 0.3
merging_ratio_v = 0.5

ARGS = {
    "compressioner": compressioner,
    "inputs_embeds": inputs_embeds,
    "input_ids": input_ids,
    "audio_token_id": audio_token_id,
    "video_token_id": video_token_id,
    "position_ids": position_ids,
    "merging_ratio_a": merging_ratio_a,
    "merging_ratio_v": merging_ratio_v,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block

Also present in other papers' code

Of 1 samples lifted from this paper's linked repositories, 1 carry a function identity and 0 do not. Of the 1 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 1 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Omni-modal Large Language Models (Omni-LLMs) have demonstrated strong capabilities in audio-video understanding tasks. However, their reliance on long multimodal token sequences leads to substantial computational overhead. Despite this challenge, token compression methods designed for Omni-LLMs remain limited. To bridge this gap, we propose OmniSIFT (Omnimodal Spatio-temporal Informed Fine-grained Token compression), a modality-asymmetric token compression framework tailored for Omni-LLMs. Specifically, OmniSIFT adopts a twostage compression strategy: (i) a spatio-temporal video pruning module that removes video redundancy arising from both intra-frame structure and inter-frame overlap, and (ii) a vision-guided audio selection module that filters audio tokens. The entire framework is optimized end-to-end via a differentiable straight-through estimator. Extensive experiments on five representative benchmarks demonstrate the efficacy and robustness of OmniSIFT. Notably, for Qwen2.5-Omni-7B, Om-niSIFT introduces only 4.85M parameters while maintaining lower latency than training-free baselines such as OmniZip. With merely 25% of the original token context, OmniSIFT consistently outperforms all compression baselines and even surpasses the performance of the full-token model on several tasks. Code is available at https: //github.com/dingyue772/OmniSIFT.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2602.04804")
get_code_for_paper("2602.04804")
have("2602.04804")

The run record, dated, one paper per request, free:

curl https://syntology.ai/api/ran/2602.04804.json

A badge for a README (the split and the date, never a ratio):

[![Syntology run record](https://syntology.ai/api/ran/2602.04804.svg)](https://syntology.ai/paper/2602.04804)

Connect an agent — have() is free.