SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2407.17353 · ICML · 2024

Scalify: scale propagation for efficient low-precision LLM training

Carlo Luschi, Andrew Fitzgibbon, Paul Balanc ¸a, Sam Hosegood

arXiv · PDF · Open in the Atlas

Code that ran

4 of 8 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; 1 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 3 failures of our instrument; 4 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/2407.17353.json · badge: /api/ran/2407.17353.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
graphcore-research/jax-scalify canonical (classifier label)
classifier basis: census
4 ran · 4 no recorded run
FunctionStatusWhere it lives
get_debug_callback_effect Ran
ran; our contract draft was wrong, not the code
graphcore-research/jax-scalify/jax_scalify/core/debug.py
code served (permissive licence) · get_code("38d4748ec4e5294b")
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 OrderedDebugEffect. 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

# The function get_debug_callback_effect takes a single boolean parameter 'ordered'
# It attempts to import JAX debugging functions and returns either:
# - ordered_debug_effect or debug_effect (if import succeeds)
# - DebugEffect.ORDERED_PRINT or DebugEffect.PRINT (if import fails)
# The parameter just needs to be a boolean value

ARGS = {
    "ordered": True
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
jax, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type OrderedDebugEffect — a OrderedDebugEffect output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a OrderedDebugEffect 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
get_numpy_api Ran
ran; our contract draft was wrong, not the code
graphcore-research/jax-scalify/jax_scalify/core/typing.py
code served (permissive licence) · get_code("c3405e97962dbc40")
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 module. 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 numpy as np
import jax.numpy as jnp

# Define ArrayTypes - this is what the function checks for
# In JAX, this would typically be jax.Array or similar
ArrayTypes = (jnp.ndarray, type(jnp.array([1])))

# Test case 1: numpy array
numpy_array = np.array([1, 2, 3, 4], dtype=np.float32)

# Test case 2: numpy scalar
numpy_scalar = np.float32(3.14)

# Test case 3: JAX array
jax_array = jnp.array([1.0, 2.0, 3.0, 4.0])

# Create ARGS with different test cases
# The function takes one parameter: val
ARGS = {
    "val": numpy_array
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
jax, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type module — a module output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a module 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
python_scalar_as_numpy Ran
ran, honoured the contract we drafted
graphcore-research/jax-scalify/jax_scalify/core/utils.py
code served (permissive licence) · get_code("6de40b822f908408")
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 float32, 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 numpy as np
from typing import Any

# Test various Python scalar types
ARGS = {
    "val": 42  # Python int - will be converted to np.int32
}

# Additional test cases that could be used individually:
test_cases = [
    {"val": True},           # Python bool - will be converted to np.bool_
    {"val": 3.14},          # Python float - will be converted to np.float32
    {"val": "string"},      # String - will be returned unchanged
    {"val": [1, 2, 3]},     # List - will be returned unchanged
    {"val": np.array([1, 2])},  # Numpy array - will be returned unchanged
]

# Using the first test case as the main ARGS
ARGS = test_cases[1]  # bool example
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type float32, 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
set_scaling_abstract_eval Ran
ran; our contract draft was wrong, not the code
graphcore-research/jax-scalify/jax_scalify/lax/base_scaling_primitives.py
code served (permissive licence) · get_code("10666eae469c2eed")
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 ShapedArray. 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 jax
from jax import core
import numpy as np

# Create ShapedArray objects for the function parameters
# ShapedArray is a simple container for shape and dtype information

# values parameter: a shaped array with shape (2, 4) and dtype float32
values = core.ShapedArray(shape=(2, 4), dtype=np.float32)

# scale parameter: a shaped array with shape (4,) and dtype float32
scale = core.ShapedArray(shape=(4,), dtype=np.float32)

ARGS = {
    "values": values,
    "scale": scale,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
jax, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ShapedArray — a ShapedArray output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a ShapedArray 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
debug_callback Not yet run
unverified (harvested, no recorded run)
graphcore-research/jax-scalify/jax_scalify/core/debug.py
code served (permissive licence) · get_code("ccdb8081b32f9f85")
How it ran

no invocation recorded.

debug_callback_grad Not yet run
unverified (harvested, no recorded run)
graphcore-research/jax-scalify/jax_scalify/ops/debug.py
code served (permissive licence) · get_code("ce0d08e42d91f35b")
How it ran

no invocation recorded.

debug_callback_grad_fwd Not yet run
unverified (harvested, no recorded run)
graphcore-research/jax-scalify/jax_scalify/ops/debug.py
code served (permissive licence) · get_code("7c5e52fa74a762c4")
How it ran

no invocation recorded.

dtype_exponent_mask Not yet run
unverified (harvested, no recorded run)
graphcore-research/jax-scalify/jax_scalify/core/pow2.py
code served (permissive licence) · get_code("4fbd7e698d2b18f7")
How it ran

no invocation recorded.

Also present in other papers' code

Of 8 samples lifted from this paper's linked repositories, 8 carry a function identity and 0 do not. Of the 8 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 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.

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

Low-precision formats such as float8 have been introduced in machine learning accelerated hardware to improve computational efficiency for large language models training and inference. Nevertheless, adoption by the ML community has been slowed down by the complex, and sometimes brittle, techniques required to match higher precision training accuracy. In this work, we present SCALIFY, a end-to-end scale propagation paradigm for computational graphs, generalizing and formalizing existing tensor scaling methods. Experiment results show that SCAL-IFY supports out-of-the-box float8 matrix multiplication and gradients representation, as well as float16 optimizer state storage. Our JAX implementation of SCALIFY is open-sourced at github.com/graphcore-research/jax-scalify.

For agents

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

get_harvested_code_for_paper("2407.17353")
get_code_for_paper("2407.17353")
have("2407.17353")

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

curl https://syntology.ai/api/ran/2407.17353.json

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

[![Syntology run record](https://syntology.ai/api/ran/2407.17353.svg)](https://syntology.ai/paper/2407.17353)

Connect an agent — have() is free.