SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2405.14567 · 2024

EHRMamba: Towards Generalizable and Scalable Foundation Models for Electronic Health Records

arXiv · PDF · Open in the Atlas

Code that ran

14 of 14 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; 3 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 11 failures of our instrument; 0 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/2405.14567.json · badge: /api/ran/2405.14567.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
VectorInstitute/odyssey canonical (classifier label)
classifier basis: census
14 ran · 0 no recorded run
FunctionStatusWhere it lives
annealed_alpha Ran
ran, honoured the contract we drafted
VectorInstitute/odyssey/odyssey/models/concept_bottleneck.py
code served (permissive licence) · get_code("652cb2ddc9a741b5")
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 float, ndim 0, dtype family float, 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 math

torch_seed = 0

ARGS = {
    "step": 1000,
    "anneal_steps": 10000,
    "start": 0.1,
    "end": 0.9,
    "cosine": False,
}
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 float, ndim 0, dtype family float, 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
best_scorer_per_cell Ran
ran; our contract draft was wrong, not the code
VectorInstitute/odyssey/odyssey/inference/model_comparison.py
code served (permissive licence) · get_code("9cc15076d0a08f28")
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 dict. 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 collections.abc import Sequence
from dataclasses import dataclass

@dataclass(frozen=True)
class ScorerAUROC:
    """One (event, horizon, scorer[, stratum]) AUROC cell."""
    event: str
    horizon_hours: float
    scorer: str
    stratum: str
    n: int
    n_positive: int
    auroc: float | None

# Create test data with various AUROC values and strata
rows = [
    ScorerAUROC(event="mortality", horizon_hours=24.0, scorer="model_a", stratum="all", n=100, n_positive=30, auroc=0.85),
    ScorerAUROC(event="mortality", horizon_hours=24.0, scorer="model_b", stratum="all", n=100, n_positive=30, auroc=0.82),
    ScorerAUROC(event="mortality", horizon_hours=24.0, scorer="model_c", stratum="all", n=100, n_positive=30, auroc=None),
    ScorerAUROC(event="readmission", horizon_hours=48.0, scorer="model_a", stratum="all", n=150, n_positive=45, auroc=0.78),
    ScorerAUROC(event="readmission", horizon_hours=48.0, scorer="model_b", stratum="all", n=150, n_positive=45, auroc=0.81),
    ScorerAUROC(event="mortality", horizon_hours=24.0, scorer="model_a", stratum="icu", n=50, n_positive=20, auroc=0.88),
    ScorerAUROC(event="mortality", horizon_hours=24.0, scorer="model_b", stratum="icu", n=50, n_positive=20, auroc=0.80),
    ScorerAUROC(event="empty_cell", horizon_hours=72.0, scorer="model_a", stratum="all", n=10, n_positive=1, auroc=None),
    ScorerAUROC(event="empty_cell", horizon_hours=72.0, scorer="model_b", stratum="all", n=10, n_positive=1, auroc=None),
]

ARGS = {
    "rows": rows,
    "stratum": "all",
}
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 dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict 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
censored_hazard_log_likelihood Ran
ran; our contract draft was wrong, not the code
VectorInstitute/odyssey/odyssey/models/time_to_event.py
code served (permissive licence) · get_code("2407e51f7377ab2c")
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 1, dtype family float, shape (2,) 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 torch.nn.functional as F

torch.manual_seed(0)

# Create small tensors for testing
batch_size = 2
num_bins = 4

# hazard_logits: shape (batch_size, num_bins)
hazard_logits = torch.randn(batch_size, num_bins)

# target_bin: shape (batch_size,) with values in range [0, num_bins)
target_bin = torch.tensor([1, 2], dtype=torch.long)

# observed: shape (batch_size,) boolean tensor
observed = torch.tensor([True, False], dtype=torch.bool)

ARGS = {
    "hazard_logits": hazard_logits,
    "target_bin": target_bin,
    "observed": observed,
}
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 1, dtype family float, shape (2,) — shape (2,) 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
concept_loss Ran
ran; our contract draft was wrong, not the code
VectorInstitute/odyssey/odyssey/models/concept_bottleneck.py
code served (permissive licence) · get_code("b040d3186985f9cd")
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 0, dtype family float, 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 torch.nn.functional as F

torch.manual_seed(0)

# Define small tensors for testing
batch_size = 2
num_concepts = 4

# concept_logits: shape (batch_size, num_concepts) - raw model outputs
concept_logits = torch.randn(batch_size, num_concepts)

# concept_labels: shape (batch_size, num_concepts) - binary labels (0 or 1)
concept_labels = torch.randint(0, 2, (batch_size, num_concepts))

# concept_mask: shape (batch_size, num_concepts) - 1 = observed, 0 = unobserved
concept_mask = torch.randint(0, 2, (batch_size, num_concepts))

# pos_weight: shape (num_concepts,) - positive class weight per concept
pos_weight = torch.tensor([2.0, 3.0, 1.5, 2.5])

ARGS = {
    "concept_logits": concept_logits,
    "concept_labels": concept_labels,
    "concept_mask": concept_mask,
    "pos_weight": pos_weight,
}
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 0, dtype family float, 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
gap_to_bin Ran
ran; our contract draft was wrong, not the code
VectorInstitute/odyssey/odyssey/models/time_to_event.py
code served (permissive licence) · get_code("20e5b21ec781dd3a")
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 1, dtype family int, shape (5,) 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

torch.manual_seed(0)

# gap_to_bin takes:
# 1. gap_hours: torch.Tensor - a tensor of non-negative gap values in hours
# 2. edges: Sequence[float] - a sequence of bin edge boundaries

# Create a small tensor of gap hours (non-negative values)
gap_hours = torch.tensor([0.0, 1.5, 3.0, 5.5, 10.0], dtype=torch.float32)

# Create a sequence of edges (bin boundaries)
edges = [2.0, 4.0, 8.0]

ARGS = {
    "gap_hours": gap_hours,
    "edges": edges,
}
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 1, dtype family int, shape (5,) — shape (5,) 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
hazard_log_likelihood Ran
ran; our contract draft was wrong, not the code
VectorInstitute/odyssey/odyssey/models/time_to_event.py
code served (permissive licence) · get_code("fa134ddfdde7b9ed")
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 1, dtype family float, shape (2,) 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 torch.nn.functional as F

torch.manual_seed(0)

# Construct arguments for hazard_log_likelihood
# The function expects:
# - hazard_logits: shape (..., num_bins)
# - target_bin: shape (...) long tensor with values < num_bins

batch_size = 2
num_bins = 4

# Create hazard_logits with shape (2, 4)
hazard_logits = torch.randn(batch_size, num_bins)

# Create target_bin with shape (2,) containing valid bin indices
target_bin = torch.tensor([1, 3], dtype=torch.long)

ARGS = {
    "hazard_logits": hazard_logits,
    "target_bin": target_bin,
}
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 1, dtype family float, shape (2,) — shape (2,) 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
intervention_apply_mask Ran
ran; our contract draft was wrong, not the code
VectorInstitute/odyssey/odyssey/models/concept_bottleneck.py
code served (permissive licence) · get_code("1257aeb0e0ca5e64")
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 2, dtype family float, shape (2, 4) 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 dataclasses import dataclass

@dataclass(frozen=True)
class BottleneckIntervention:
    """A do()-style edit applied inside the bottleneck's mixing step."""
    probs: torch.Tensor | None = None
    probs_mask: torch.Tensor | None = None
    uncertain_band: float | None = None
    zero_known: bool = False
    zero_unknown: bool = False
    zero_residual: bool = False

def intervention_apply_mask(intervention: BottleneckIntervention, own_probs: torch.Tensor) -> torch.Tensor | None:
    """Where an intervention's ``probs`` actually replace the model's own."""
    apply: torch.Tensor | None = None
    if intervention.probs_mask is not None:
        apply = intervention.probs_mask.expand_as(own_probs)
    if intervention.uncertain_band is not None:
        band = (own_probs - 0.5).abs() < intervention.uncertain_band
        apply = band if apply is None else apply & band
    return apply

torch.manual_seed(0)

# Create test tensors
own_probs_tensor = torch.rand(2, 4)  # batch=2, num_concepts=4

# Case 1: Both probs_mask and uncertain_band set
probs_mask_tensor = torch.ones(2, 4, dtype=torch.bool)
intervention_1 = BottleneckIntervention(
    probs=torch.rand(2, 4),
    probs_mask=probs_mask_tensor,
    uncertain_band=0.2,
    zero_known=False,
    zero_unknown=False,
    zero_residual=False
)

ARGS = {
    "intervention": intervention_1,
    "own_probs": own_probs_tensor
}
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 2, dtype family float, shape (2, 4) — shape (2, 4) 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
middle_layer Ran
ran, honoured the contract we drafted
VectorInstitute/odyssey/odyssey/models/injection.py
code served (permissive licence) · get_code("37d5d57f4239c77d")
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 torch.nn as nn

torch.manual_seed(0)

class SimpleBackbone(nn.Module):
    def __init__(self, num_layers=4):
        super().__init__()
        self.layers = nn.ModuleList([
            nn.Linear(4, 4) for _ in range(num_layers)
        ])
    
    def forward(self, x):
        for layer in self.layers:
            x = layer(x)
        return x

backbone = SimpleBackbone(num_layers=4)

ARGS = {
    'backbone': backbone
}
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
pinball_loss Ran
ran; our contract draft was wrong, not the code
VectorInstitute/odyssey/odyssey/models/value_head.py
code served (permissive licence) · get_code("ab16e79712613f58")
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 1, dtype family float, shape (2,) 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

torch.manual_seed(0)

# Create small tensors with distinct dimensions
# quantiles: shape (batch=2, num_quantiles=4)
quantiles = torch.randn(2, 4)

# target: shape (batch=2,) - will be unsqueezed to (2, 1) to broadcast against quantiles
target = torch.randn(2)

# levels: sequence of quantile levels (tau values)
levels = [0.25, 0.5, 0.75, 0.9]

ARGS = {
    "quantiles": quantiles,
    "target": target,
    "levels": levels,
}
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 1, dtype family float, shape (2,) — shape (2,) 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
select_diverse_cases Ran
ran, honoured the contract we drafted
VectorInstitute/odyssey/odyssey/inference/case_study.py
code served (permissive licence) · get_code("20f21b3aa186f2d8")
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 int; 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 torch
import polars as pl
import random

torch.manual_seed(0)
random.seed(0)

# Create a sample events DataFrame with subject_id and other columns
events = pl.DataFrame({
    'subject_id': [1, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 3, 4, 4, 5, 5, 5, 5, 5, 5],
    'event_type': ['A', 'B', 'C', 'A', 'B', 'C', 'D', 'A', 'B', 'C', 'D', 'E', 'A', 'B', 'A', 'B', 'C', 'D', 'E', 'F'],
})

# Create concept_labels dict with tensors
# Each subject has a tensor of concept labels (binary or count)
concept_labels = {
    1: torch.tensor([1, 0, 1, 0], dtype=torch.float32),  # 2 concepts triggered
    2: torch.tensor([1, 1, 1, 0], dtype=torch.float32),  # 3 concepts triggered
    3: torch.tensor([0, 0, 0, 0], dtype=torch.float32),  # 0 concepts triggered
    4: torch.tensor([1, 0, 0, 0], dtype=torch.float32),  # 1 concept triggered
    5: torch.tensor([1, 1, 0, 0], dtype=torch.float32),  # 2 concepts triggered
}

ARGS = {
    'events': events,
    'concept_labels': concept_labels,
    'n_cases': 3,
    'min_events': 2,
    'seed': 42,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
polars, torch — 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 int — 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
stream_injection Ran
ran; our contract draft was wrong, not the code
VectorInstitute/odyssey/odyssey/models/injection.py
code served (permissive licence) · get_code("04ebfb65c3483398")
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 _GeneratorContextManager. 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 simple block-structured backbone with a layers attribute
class SimpleBlock(nn.Module):
    def __init__(self, input_dim, output_dim):
        super().__init__()
        self.linear = nn.Linear(input_dim, output_dim)
    
    def forward(self, x):
        return self.linear(x)

class BlockBackbone(nn.Module):
    def __init__(self, num_layers=3, hidden_dim=4):
        super().__init__()
        self.layers = nn.ModuleList([
            SimpleBlock(hidden_dim, hidden_dim) for _ in range(num_layers)
        ])
    
    def forward(self, x):
        for layer in self.layers:
            x = layer(x)
        return x

# Construct the arguments
backbone = BlockBackbone(num_layers=3, hidden_dim=4)
layer_index = 1
vector = torch.randn(4)  # Shape (d,) where d=4

ARGS = {
    "backbone": backbone,
    "layer_index": layer_index,
    "vector": vector,
}
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 _GeneratorContextManager — a _GeneratorContextManager output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a _GeneratorContextManager 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
value_features Ran
ran; our contract draft was wrong, not the code
VectorInstitute/odyssey/odyssey/models/embeddings.py
code served (permissive licence) · get_code("816376e6520ac858")
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 4, dtype family float, shape (2, 4, 3, 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

torch.manual_seed(0)

# Construct test values tensor with shape (..., 3)
# Using a batch of 2 samples with 4 features each, resulting in (..., 3) compatibility
# The function expects values to potentially contain NaN and work with any shape ending in a dimension
values = torch.tensor([
    [[1.0, 2.0, 3.0],
     [4.0, 5.0, 6.0],
     [7.0, 8.0, 9.0],
     [10.0, 11.0, 12.0]],
    [[float('nan'), 2.5, 3.5],
     [4.5, float('nan'), 6.5],
     [7.5, 8.5, float('nan')],
     [10.5, 11.5, 12.5]]
], dtype=torch.float32)

ARGS = {
    "values": values
}
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 4, dtype family float, shape (2, 4, 3, 3) — shape (2, 4, 3, 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
value_quantile_loss Ran
ran; our contract draft was wrong, not the code
VectorInstitute/odyssey/odyssey/models/value_head.py
code served (permissive licence) · get_code("0a2999efd7fc3d1b")
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 0, dtype family float, 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
from collections.abc import Sequence

torch.manual_seed(0)

# Construct tensors for the function
# quantiles: shape (..., K) where K is the number of quantile levels
# Let's use shape (2, 3, 4) where batch=2, some intermediate dim=3, quantiles=4
quantiles = torch.randn(2, 3, 4)

# target: shape (...) that broadcasts against quantiles' last dimension
# Should have shape (2, 3) to broadcast with (2, 3, 4)
target = torch.randn(2, 3)

# valid: boolean mask with same shape as target
# Shape (2, 3) to match target
valid = torch.randint(0, 2, (2, 3), dtype=torch.bool)

# levels: sequence of floats representing quantile levels
# Must have length matching the last dimension of quantiles (4 in this case)
levels = [0.25, 0.5, 0.75, 0.95]

ARGS = {
    "quantiles": quantiles,
    "target": target,
    "valid": valid,
    "levels": levels,
}
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 0, dtype family float, 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
value_target_valid_mask Ran
ran; our contract draft was wrong, not the code
VectorInstitute/odyssey/odyssey/models/value_head.py
code served (permissive licence) · get_code("ab2ebe14caf34af0")
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, ndim 3, 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 torch

torch.manual_seed(0)

# Create small test tensors
# values: (lanes, T) - standardized values, already shifted
# real_mask: (lanes, T) - boolean mask indicating real tokens
lanes = 2
T = 4

values = torch.randn(lanes, T)
# Set some values to NaN to test the isnan logic
values[0, 2] = float('nan')

real_mask = torch.tensor([
    [True, True, False, True],
    [True, True, True, False]
], dtype=torch.bool)

ARGS = {
    "values": values,
    "real_mask": real_mask,
}
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, ndim 3, dtype family float — a tuple 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

Also present in other papers' code

Of 14 samples lifted from this paper's linked repositories, 14 carry a function identity and 0 do not. Of the 14 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 14 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

Transformers have significantly advanced the modeling of Electronic Health Records (EHR), yet their deployment in real-world healthcare is limited by several key challenges. Firstly, the quadratic computational cost and insufficient context length of these models hinder hospitals' ability in processing the extensive medical histories typical in EHR data. Additionally, existing models employ separate finetuning for each clinical task, complicating maintenance in healthcare environments. Moreover, these models focus exclusively on either clinical prediction or EHR forecasting, lacking proficiency in both tasks. To overcome these limitations, we introduce EHRMamba, a robust foundation model built on the Mamba architecture. EHRMamba can process sequences up to 300% longer than previous models due to its linear computational cost. We also introduce a novel approach to Multitask Prompted Finetuning (MPF) for EHR data, which enables EHRMamba to simultaneously learn multiple clinical tasks in a single finetuning phase, significantly enhancing deployment and cross-task generalization. Furthermore, our model leverages the HL7 FHIR data standard to simplify integration into existing hospital systems. Alongside EHRMamba, we open-source Odyssey, a toolkit designed to support the development and deployment of EHR foundation models, with an emphasis on data standardization and interpretability. Our evaluations on the MIMIC-IV dataset demonstrate that EHRMamba advances state-of-the-art performance across 6 major clinical tasks and excels in EHR forecasting, marking a significant leap forward in the field.

For agents

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

get_harvested_code_for_paper("2405.14567")
get_code_for_paper("2405.14567")
have("2405.14567")

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

curl https://syntology.ai/api/ran/2405.14567.json

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

[![Syntology run record](https://syntology.ai/api/ran/2405.14567.svg)](https://syntology.ai/paper/2405.14567)

Connect an agent — have() is free.