SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2102.02779 · ICML · 2021

Unifying Vision-and-Language Tasks via Text Generation

Jaemin Cho, Jie Hao, Tan Bansal

arXiv · PDF · Open in the Atlas

Code that ran

2 of 12 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (2 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 2 have no contract verdict; 0 failures of our instrument; 10 no recorded run), measured 2026-11-22; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; 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/2102.02779.json · badge: /api/ran/2102.02779.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
j-min/VL-T5 marked official in Papers with Code
Papers with Code archive
2 ran · 10 no recorded run
FunctionStatusWhere it lives
VLSeq2SeqLMOutput Ran
ran on a synthesized input · constructed an object; no result was computed
j-min/VL-T5/VL-T5/src/multitask_model.py
code served (permissive licence) · get_code("b5a6c04b009cca88")
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 an instance of its own class, VLSeq2SeqLMOutput: the run constructed the object and did not call it, so no result was computed. 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

torch.manual_seed(0)

# Create small tensors with distinct dimensions
batch_size = 2
seq_length = 3
vocab_size = 4
hidden_size = 5
num_heads = 2
num_layers = 2
embed_size_per_head = 2

# Loss tensor
loss = torch.tensor([0.5], dtype=torch.float32)

# Logits tensor
logits = torch.randn(batch_size, seq_length, vocab_size, dtype=torch.float32)

# Past key values - list of tensors for each layer
past_key_values = [
    torch.randn(2, batch_size, num_heads, seq_length, embed_size_per_head, dtype=torch.float32)
    for _ in range(num_layers)
]

# Decoder last hidden state
decoder_last_hidden_state = torch.randn(batch_size, seq_length, hidden_size, dtype=torch.float32)

# Decoder hidden states - tuple of tensors (one per layer + embedding)
decoder_hidden_states = tuple(
    torch.randn(batch_size, seq_length, hidden_size, dtype=torch.float32)
    for _ in range(num_layers + 1)
)

# Decoder attentions - tuple of tensors (one per layer)
decoder_attentions = tuple(
    torch.randn(batch_size, num_heads, seq_length, seq_length, dtype=torch.float32)
    for _ in range(num_layers)
)

# Encoder last hidden state
encoder_last_hidden_state = torch.randn(batch_size, seq_length, hidden_size, dtype=torch.float32)

# Encoder hidden states
encoder_hidden_states = tuple(
    torch.randn(batch_size, seq_length, hidden_size, dtype=torch.float32)
    for _ in range(num_layers + 1)
)

# Encoder attentions
encoder_attentions = tuple(
    torch.randn(batch_size, num_heads, seq_length, seq_length, dtype=torch.float32)
    for _ in range(num_layers)
)

# Visual encoder last hidden state
vis_encoder_last_hidden_state = torch.randn(batch_size, 6, hidden_size, dtype=torch.float32)

# Visual encoder hidden states
vis_encoder_hidden_states = tuple(
    torch.randn(batch_size, 6, hidden_size, dtype=torch.float32)
    for _ in range(num_layers + 1)
)

# Visual encoder attentions
vis_encoder_attentions = tuple(
    torch.randn(batch_size, num_heads, 6, 6, dtype=torch.float32)
    for _ in range(num_layers)
)

ARGS = {
    "loss": loss,
    "logits": logits,
    "past_key_values": past_key_values,
    "decoder_last_hidden_state": decoder_last_hidden_state,
    "decoder_hidden_states": decoder_hidden_states,
    "decoder_attentions": decoder_attentions,
    "encoder_last_hidden_state": encoder_last_hidden_state,
    "encoder_hidden_states": encoder_hidden_states,
    "encoder_attentions": encoder_attentions,
    "vis_encoder_last_hidden_state": vis_encoder_last_hidden_state,
    "vis_encoder_hidden_states": vis_encoder_hidden_states,
    "vis_encoder_attentions": vis_encoder_attentions,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, transformers — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type VLSeq2SeqLMOutput — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a VLSeq2SeqLMOutput 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
VisualEmbedding Ran
ran on a synthesized input · constructed an object; no result was computed
j-min/VL-T5/VL-T5/src/multitask_model.py
code served (permissive licence) · get_code("e44fd46c867cf8be")
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 an instance of its own class, VisualEmbedding: the run constructed the object and did not call it, so no result was computed. 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
from transformers.models.t5.modeling_t5 import T5LayerNorm

torch.manual_seed(0)

class Config:
    def __init__(self):
        self.feat_dim = 5
        self.pos_dim = 4
        self.n_images = 3
        self.d_model = 8
        self.layer_norm_epsilon = 1e-6
        self.individual_vis_layer_norm = True
        self.use_vis_layer_norm = True
        self.use_vis_order_embedding = True

class ObjOrderEmbedding(nn.Module):
    def __init__(self, num_embeddings, embedding_dim):
        super().__init__()
        self.num_embeddings = num_embeddings
        self.embedding = nn.Embedding(num_embeddings, embedding_dim)
    
    def forward(self, input_ids):
        return self.embedding(input_ids)

config = Config()
obj_order_embedding = ObjOrderEmbedding(num_embeddings=10, embedding_dim=config.d_model)

ARGS = {
    'config': config,
    'obj_order_embedding': obj_order_embedding,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, transformers — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type VisualEmbedding — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a VisualEmbedding 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
JointEncoder Not yet run
unverified (harvested, no recorded run)
j-min/VL-T5/VL-T5/src/multitask_model.py
code served (permissive licence) · get_code("9ae4551ab4057d4f")
How it ran

no invocation recorded.

VLT5 Not yet run
unverified (harvested, no recorded run)
j-min/VL-T5/VL-T5/src/multitask_model.py
code served (permissive licence) · get_code("5d69b18e68c1e195")
How it ran

no invocation recorded.

VLT5COCOCaption Not yet run
unverified (harvested, no recorded run)
j-min/VL-T5/VL-T5/src/multitask_model.py
code served (permissive licence) · get_code("943c42cc626cc9a5")
How it ran

no invocation recorded.

VLT5GQA Not yet run
unverified (harvested, no recorded run)
j-min/VL-T5/VL-T5/src/multitask_model.py
code served (permissive licence) · get_code("c6c8a88fa5a29395")
How it ran

no invocation recorded.

VLT5MMT Not yet run
unverified (harvested, no recorded run)
j-min/VL-T5/VL-T5/src/multitask_model.py
code served (permissive licence) · get_code("8f8e3dcb1533f514")
How it ran

no invocation recorded.

VLT5MultiTask Not yet run
unverified (harvested, no recorded run)
j-min/VL-T5/VL-T5/src/multitask_model.py
code served (permissive licence) · get_code("615b17a71b757c72")
How it ran

no invocation recorded.

VLT5NLVR Not yet run
unverified (harvested, no recorded run)
j-min/VL-T5/VL-T5/src/multitask_model.py
code served (permissive licence) · get_code("a991fbc2a12b6068")
How it ran

no invocation recorded.

VLT5RefCOCO Not yet run
unverified (harvested, no recorded run)
j-min/VL-T5/VL-T5/src/multitask_model.py
code served (permissive licence) · get_code("99a4eac0e06e3904")
How it ran

no invocation recorded.

VLT5VCR Not yet run
unverified (harvested, no recorded run)
j-min/VL-T5/VL-T5/src/multitask_model.py
code served (permissive licence) · get_code("b4349d2ccfd035b9")
How it ran

no invocation recorded.

VLT5VQA Not yet run
unverified (harvested, no recorded run)
j-min/VL-T5/VL-T5/src/multitask_model.py
code served (permissive licence) · get_code("79580be4d7e75e1a")
How it ran

no invocation recorded.

Also present in other papers' code

Of 12 samples lifted from this paper's linked repositories, 12 carry a function identity and 0 do not. Of the 12 distinct functions with an identity, 4 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 4 later) and 8 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
JointEncoder
brick of this paper's assembly: helper
2 other papers
2 later
2212.02746 (2022-12)
UniGeo: Unifying Geometry Logical Reasoning via Reformulating Mathematical Expression
2305.11768 (2023-05)
Generating Visual Spatial Description via Holistic 3D Scene Understanding
VLSeq2SeqLMOutput
brick of this paper's assembly: helper
2 other papers
2 later
2212.02746 (2022-12)
UniGeo: Unifying Geometry Logical Reasoning via Reformulating Mathematical Expression
2305.11768 (2023-05)
Generating Visual Spatial Description via Holistic 3D Scene Understanding
VLT5
brick of this paper's assembly: model
1 other paper
1 later
2212.02746 (2022-12)
UniGeo: Unifying Geometry Logical Reasoning via Reformulating Mathematical Expression
VisualEmbedding
brick of this paper's assembly: helper
1 other paper
1 later
2305.11768 (2023-05)
Generating Visual Spatial Description via Holistic 3D Scene Understanding

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

Existing methods for vision-and-language learning typically require designing task-specific architectures and objectives for each task. For example, a multi-label answer classifier for visual question answering, a region scorer for referring expression comprehension, and a language decoder for image captioning, etc. To alleviate these hassles, in this work, we propose a unified framework that learns different tasks in a single architecture with the same language modeling objective, i.e., multimodal conditional text generation, where our models learn to generate labels in text based on the visual and textual inputs. On 7 popular vision-and-language benchmarks, including visual question answering, referring expression comprehension, visual commonsense reasoning, most of which have been previously modeled as discriminative tasks, our generative approach (with a single unified architecture) reaches comparable performance to recent taskspecific state-of-the-art vision-and-language models. Moreover, our generative approach shows better generalization ability on questions that have rare answers. Also, we show that our framework allows multi-task learning in a single architecture with a single set of parameters, achieving similar performance to separately optimized singletask models.

For agents

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

get_harvested_code_for_paper("2102.02779")
get_code_for_paper("2102.02779")
have("2102.02779")

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

curl https://syntology.ai/api/ran/2102.02779.json

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

[![Syntology run record](https://syntology.ai/api/ran/2102.02779.svg)](https://syntology.ai/paper/2102.02779)

Connect an agent — have() is free.