SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2209.05433 · 2022

FP8 Formats for Deep Learning

arXiv · PDF · Open in the Atlas

Code that ran

6 of 7 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 6 have no contract verdict; 0 failures of our instrument; 1 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/2209.05433.json · badge: /api/ran/2209.05433.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
nvidia/transformerengine community (classifier label)
classifier basis not recorded on the link
6 ran · 1 no recorded run
FunctionStatusWhere it lives
all_files_in_dir Ran
ran on a synthesized input
nvidia/transformerengine/build_tools/utils.py
code served (permissive licence) · get_code("7fd2e1e68bc3372f")
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 os
import tempfile
from pathlib import Path

# Create a temporary directory structure for testing
tmpdir = tempfile.mkdtemp()
test_dir = Path(tmpdir)

# Create some test files with various extensions
(test_dir / "file1.txt").touch()
(test_dir / "file2.txt").touch()
(test_dir / "file3.py").touch()
(test_dir / "file4.json").touch()

# Create a subdirectory with more files
subdir = test_dir / "subdir"
subdir.mkdir()
(subdir / "file5.txt").touch()
(subdir / "file6.md").touch()

# Define arguments for the function
ARGS = {
    "path": str(test_dir),
    "name_extension": "txt"
}
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
dump_kernel_counts Ran
ran on a synthesized input
nvidia/transformerengine/benchmarks/profile_rht_cast_swizzle_fusion.py
code served (permissive licence) · get_code("2491511b79d5aa62")
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.

import torch

torch.manual_seed(0)

class MockProfilerEvent:
    def __init__(self, name, device_type):
        self.name = name
        self.device_type = device_type

class MockProfiler:
    def __init__(self):
        self._events = [
            MockProfilerEvent("aten::matmul", torch.autograd.DeviceType.CUDA),
            MockProfilerEvent("aten::add", torch.autograd.DeviceType.CUDA),
            MockProfilerEvent("multi_tensor_swizzle_row_scaling_kernel", torch.autograd.DeviceType.CUDA),
            MockProfilerEvent("aten::matmul", torch.autograd.DeviceType.CUDA),
            MockProfilerEvent("swizzle_col_scaling_kernel", torch.autograd.DeviceType.CUDA),
            MockProfilerEvent("aten::mul", torch.autograd.DeviceType.CPU),
            MockProfilerEvent("multi_tensor_unswizzle_row_scaling_kernel", torch.autograd.DeviceType.CUDA),
            MockProfilerEvent("aten::add", torch.autograd.DeviceType.CUDA),
        ]
    
    def events(self):
        return self._events

prof = MockProfiler()
label = "test_profiler_output"

ARGS = {
    "prof": prof,
    "label": label,
}
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 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
get_build_ext Ran
ran on a synthesized input
nvidia/transformerengine/build_tools/build_ext.py
code served (permissive licence) · get_code("ec12d51e3d0d52ba")
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 type. 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 setuptools
from pathlib import Path
from typing import Type, Optional, List


class MockExtension(setuptools.Extension):
    """Mock extension class for testing"""
    def __init__(self, name: str, sources=None):
        super().__init__(name, sources=sources or [])


ARGS = {
    "extension_cls": MockExtension,
    "framework_extension_only": False,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
pybind11, setuptools — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type type — a type output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a type 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
nccl_ep_enabled Ran
ran on a synthesized input
nvidia/transformerengine/build_tools/utils.py
code served (permissive licence) · get_code("eb194c4d49f9493a")
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 bool, 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 os

ARGS = {
    'archs': '75;80;89;90',
}
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 bool, 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
remove_dups Ran
ran on a synthesized input
nvidia/transformerengine/build_tools/utils.py
code served (permissive licence) · get_code("d726e43ffe51bcf4")
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

torch.manual_seed(0)

# The function remove_dups takes a list and returns unique elements
# It uses set() to remove duplicates, so we need a list with some duplicate elements

ARGS = {
    "_list": [1, 2, 3, 2, 1, 4, 5, 4]
}
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 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
setup_pytorch_extension Ran
ran on a synthesized input
nvidia/transformerengine/build_tools/pytorch.py
code served (permissive licence) · get_code("b6b66114e7f5ad56")
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 Extension. 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 os
from pathlib import Path
import tempfile
import setuptools

# Create temporary directories and files for the setup
temp_dir = tempfile.mkdtemp()
csrc_dir = Path(temp_dir) / "csrc"
header_dir = Path(temp_dir) / "csrc_headers"
common_dir = Path(temp_dir) / "common_headers"

csrc_dir.mkdir(exist_ok=True)
header_dir.mkdir(exist_ok=True)
common_dir.mkdir(exist_ok=True)
(common_dir / "common").mkdir(exist_ok=True)
(common_dir / "common" / "include").mkdir(exist_ok=True)

# Create dummy C++ source files
cpp_file1 = csrc_dir / "test1.cpp"
cpp_file1.write_text("// dummy cpp file 1\n")

cpp_file2 = csrc_dir / "test2.cpp"
cpp_file2.write_text("// dummy cpp file 2\n")

# Create dummy header files
header_file = header_dir / "test.h"
header_file.write_text("// dummy header file\n")

# Mock the helper functions that are called by setup_pytorch_extension
def all_files_in_dir(directory, name_extension='cpp'):
    """Return all files with given extension in directory"""
    if not directory.exists():
        return []
    return list(directory.glob(f'*.{name_extension}'))

def get_cuda_include_dirs():
    """Return mock CUDA include directories"""
    return [Path(temp_dir) / "cuda_includes"]

def debug_build_enabled():
    """Check if debug build is enabled"""
    return False

def cuda_version():
    """Return mock CUDA version"""
    return (12, 0)

def setup_mpi_flags(include_dirs, cxx_flags):
    """Setup MPI flags (mock)"""
    pass

def nccl_ep_enabled():
    """Check if NCCL EP is enabled"""
    return False

# Inject mocks into the module namespace
import sys
current_module = sys.modules[__name__]
current_module.all_files_in_dir = all_files_in_dir
current_module.get_cuda_include_dirs = get_cuda_include_dirs
current_module.debug_build_enabled = debug_build_enabled
current_module.cuda_version = cuda_version
current_module.setup_mpi_flags = setup_mpi_flags
current_module.nccl_ep_enabled = nccl_ep_enabled

# Clean environment variables
os.environ.pop('NVTE_ENABLE_NVSHMEM', None)
os.environ.pop('NVSHMEM_HOME', None)
os.environ.pop('NVTE_WITH_CUBLASMP', None)

ARGS = {
    'csrc_source_files': csrc_dir,
    'csrc_header_files': header_dir,
    'common_header_files': common_dir,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
setuptools, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Extension — a Extension output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Extension 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
setup_jax_extension Not yet run
unverified (harvested, no recorded run)
nvidia/transformerengine/build_tools/jax.py
code served (permissive licence) · get_code("ec96f6c9e40998bf")
How it ran

no invocation recorded.

Also present in other papers' code

Of 7 samples lifted from this paper's linked repositories, 7 carry a function identity and 0 do not. Of the 7 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 7 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

FP8 is a natural progression for accelerating deep learning training inference beyond the 16-bit formats common in modern processors. In this paper we propose an 8-bit floating point (FP8) binary interchange format consisting of two encodings - E4M3 (4-bit exponent and 3-bit mantissa) and E5M2 (5-bit exponent and 2-bit mantissa). While E5M2 follows IEEE 754 conventions for representatio of special values, E4M3's dynamic range is extended by not representing infinities and having only one mantissa bit-pattern for NaNs. We demonstrate the efficacy of the FP8 format on a variety of image and language tasks, effectively matching the result quality achieved by 16-bit training sessions. Our study covers the main modern neural network architectures - CNNs, RNNs, and Transformer-based models, leaving all the hyperparameters unchanged from the 16-bit baseline training sessions. Our training experiments include large, up to 175B parameter, language models. We also examine FP8 post-training-quantization of language models trained using 16-bit formats that resisted fixed point int8 quantization.

For agents

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

get_harvested_code_for_paper("2209.05433")
get_code_for_paper("2209.05433")
have("2209.05433")

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

curl https://syntology.ai/api/ran/2209.05433.json

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

[![Syntology run record](https://syntology.ai/api/ran/2209.05433.svg)](https://syntology.ai/paper/2209.05433)

Connect an agent — have() is free.