SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2604.12928 · 2026

G : Asynchronous Knowledge Retrieval for Full-Duplex Speech Language Models

Karen Livescu, Neil Zeghidour, Manu Orsini, Eugene Kharitonov, Alexandre Défossez, Chung-Ming Chien

arXiv · PDF · Open in the Atlas

Code that ran

6 of 6 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; 6 failures 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/2604.12928.json · badge: /api/ran/2604.12928.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
kyutai-labs/moshi-rag canonical (classifier label)
classifier basis: census
6 ran · 0 no recorded run
FunctionStatusWhere it lives
apply_parametrization_norm Ran
ran; our contract draft was wrong, not the code
kyutai-labs/moshi-rag/moshi/moshi/modules/conv.py
code served (permissive licence) · get_code("61632108e112746f")
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 Linear. 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
from torch import nn
from torch.nn.utils import weight_norm

torch.manual_seed(0)

# Create a simple module to test with
module = nn.Linear(4, 3)

# Define ARGS with all parameters of apply_parametrization_norm
ARGS = {
    'module': module,
    'norm': 'weight_norm'
}
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 Linear — a Linear output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Linear output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not 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
extract_answer_variants Ran
ran; our contract draft was wrong, not the code
kyutai-labs/moshi-rag/moshi/moshi/evaluate/score.py
code served (permissive licence) · get_code("dfeda1e8168320ae")
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 list, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim 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 logging
from typing import Any, Dict, List, Optional

# Test case 1: string input
answer_field_1 = "Paris"

# Test case 2: list input
answer_field_2 = ["London", "UK Capital", ""]

# Test case 3: dict with normalized_aliases and value
answer_field_3 = {
    "normalized_aliases": ["New York", "NYC"],
    "aliases": ["New York City"],
    "normalized_value": "New York",
    "value": "NYC"
}

# Test case 4: dict with only some keys
answer_field_4 = {
    "value": "Tokyo"
}

# Test case 5: dict with aliases as list
answer_field_5 = {
    "aliases": ["Berlin", "German Capital"],
    "normalized_value": "Berlin"
}

# Use the first test case as the primary argument
ARGS = {
    "answer_field": answer_field_1
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 1, dtype family float — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
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
gating_forward_generic Ran
ran; our contract draft was wrong, not the code
kyutai-labs/moshi-rag/moshi/moshi/modules/gating.py
code served (permissive licence) · get_code("26ecc97e6c0629ee")
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 Tensor, ndim 3, dtype family float, shape (2, 5, 3) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape 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
from torch import nn
import torch.nn.functional as F

torch.manual_seed(0)

# Create linear_in: takes input and outputs 2x hidden_dim (for gating)
# Input will be (B, T, input_dim) -> output (B, T, 2*hidden_dim)
linear_in = nn.Linear(in_features=4, out_features=8)

# Create linear_out: takes gated output and projects to output_dim
# Input will be (B, T, hidden_dim) -> output (B, T, output_dim)
linear_out = nn.Linear(in_features=4, out_features=3)

# Create a simple activation function (e.g., ReLU)
activation = nn.ReLU()

# Create input tensor: (batch=2, time=5, features=4)
x = torch.randn(2, 5, 4)

ARGS = {
    "linear_in": linear_in,
    "linear_out": linear_out,
    "activation": activation,
    "x": x
}
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 Tensor, ndim 3, dtype family float, shape (2, 5, 3) — shape (2, 5, 3) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
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
get_extra_padding_for_conv1d Ran
ran; our contract draft was wrong, not the code
kyutai-labs/moshi-rag/moshi/moshi/modules/conv.py
code served (permissive licence) · get_code("5a9465546cfaa0f0")
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 int, ndim 0, dtype family int, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape 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 math

torch.manual_seed(0)

# Create a small tensor with shape (batch, channels, length)
# The function accesses x.shape[-1] to get the length dimension
x = torch.randn(2, 4, 16)  # batch=2, channels=4, length=16

kernel_size = 3
stride = 2
padding_total = 1

ARGS = {
    "x": x,
    "kernel_size": kernel_size,
    "stride": stride,
    "padding_total": padding_total,
}
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 int, ndim 0, dtype family int, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
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
hf_get Ran
ran; our contract draft was wrong, not the code
kyutai-labs/moshi-rag/moshi/moshi/models/loaders.py
code served (permissive licence) · get_code("625b33a3b1ad0220")
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 PosixPath. 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.

from pathlib import Path

ARGS = {
    "filename": "model.safetensors",
    "hf_repo": None,
    "check_local_file_exists": False,
    "revision": None,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
huggingface_hub — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type PosixPath — a PosixPath output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a PosixPath output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not 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
pad_for_conv1d Ran
ran; our contract draft was wrong, not the code
kyutai-labs/moshi-rag/moshi/moshi/modules/conv.py
code served (permissive licence) · get_code("44c94244e949b9fb")
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 Tensor, ndim 3, dtype family float, shape (2, 4, 10) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape 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 math
import torch
from torch.nn import functional as F

torch.manual_seed(0)

# Create a small tensor for testing
# Shape: (batch_size=2, channels=4, length=10)
x = torch.randn(2, 4, 10)

# Define parameters for the convolution
kernel_size = 3
stride = 2
padding_total = 1

ARGS = {
    'x': x,
    'kernel_size': kernel_size,
    'stride': stride,
    'padding_total': padding_total
}
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 Tensor, ndim 3, dtype family float, shape (2, 4, 10) — shape (2, 4, 10) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
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 6 samples lifted from this paper's linked repositories, 6 carry a function identity and 0 do not. Of the 6 distinct functions with an identity, 5 are also present in other papers whose code we hold (5 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.

FunctionAlso present inEarliest-datedOthers named
get_extra_padding_for_conv1d 9 other papers
9 dated earlier · 1 only as an assembly brick
2306.06546 (2023-06)
High-Fidelity Audio Compression with Improved RVQGAN
2308.16692 (2023-08)
SPEECHTOKENIZER: UNIFIED SPEECH TOKENIZER FOR SPEECH LANGUAGE MODELS
2310.00014 (2023-10)
Fewer-token Neural Speech Codec with Time-invariant Codes
and 6 more not named here
apply_parametrization_norm 1 other paper
1 dated earlier
2410.00037 (2024-10)
Moshi: a speech-text foundation model for real-time dialogue
gating_forward_generic 1 other paper
1 dated earlier
2410.00037 (2024-10)
Moshi: a speech-text foundation model for real-time dialogue
hf_get 1 other paper
1 dated earlier
2410.00037 (2024-10)
Moshi: a speech-text foundation model for real-time dialogue
pad_for_conv1d 1 other paper
1 dated earlier
2410.00037 (2024-10)
Moshi: a speech-text foundation model for real-time dialogue

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

Speech-to-speech language models have recently emerged to enhance the naturalness of conversational AI. In particular, full-duplex models are distinguished by their real-time interactivity, including handling of pauses, interruptions, and backchannels. However, improving their factuality remains an open challenge. While scaling the model size could address this gap, it would make real-time inference prohibitively expensive. In this work, we propose Moshi R A G , a modular approach that combines a compact full-duplex interface with selective retrieval to access more powerful knowledge sources. Our asynchronous framework enables the model to identify knowledgedemanding queries and ground its responses in external information. By leveraging the natural temporal gap between response onset and the delivery of core information, the retrieval process can be completed while maintaining a natural conversation flow. With this approach, Moshi R A G achieves factuality comparable to the best publicly released non-duplex speech language models while preserving the interactivity inherent to fullduplex systems. Moreover, our flexible design supports plug-and-play retrieval methods without retraining and demonstrates strong performance on out-of-domain mathematical reasoning tasks.

For agents

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

get_harvested_code_for_paper("2604.12928")
get_code_for_paper("2604.12928")
have("2604.12928")

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

curl https://syntology.ai/api/ran/2604.12928.json

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

[![Syntology run record](https://syntology.ai/api/ran/2604.12928.svg)](https://syntology.ai/paper/2604.12928)

Connect an agent — have() is free.