Tim Salimans, Ben Poole, Diederik Kingma, Jonathan Ho
29 of 38 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (18 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 1 violated it and 19 have no contract verdict; 9 failures of our instrument; 9 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/2107.00630.json · badge: /api/ran/2107.00630.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.
| Repository | Role | Ran |
|---|---|---|
| google-research/vdm | canonical (classifier label) classifier basis: classifier_2025 |
11 ran · 5 no recorded run |
| hojonathanho/diffusion | canonical (classifier label) classifier basis: classifier_2025 |
2 ran · 0 no recorded run |
| revsic/jax-variational-diffwave | community Papers with Code archive |
7 ran · 2 no recorded run |
| addtt/variational-diffusion-models | reimplementation (classifier label) classifier basis: classifier_2025 |
5 ran · 1 no recorded run |
| martenlienen/bsi | community Papers with Code archive |
2 ran · 1 no recorded run |
| yoyololicon/variational-diffwave | community Papers with Code archive |
2 ran · 0 no recorded run |
| Function | Status | Where it lives |
|---|---|---|
| AttnBlock | Ran ran on a synthesized input · constructed an object; no result was computed |
google-research/vdm/model_vdm.py code served (permissive licence) · get_code("561db3edf2a73d49") How it ranRan 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, AttnBlock: 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.
|
| Base2FourierFeatures | Ran ran on a synthesized input · constructed an object; no result was computed |
google-research/vdm/model_vdm.py code served (permissive licence) · get_code("1fd721cbc30c55b2") How it ranRan 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, Base2FourierFeatures: 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.
|
| Config | Ran ran on a synthesized input · constructed an object; no result was computed |
revsic/jax-variational-diffwave/vlbdiffwave/impl.py code served (permissive licence) · get_code("59e0526c50f31501") How it ranRan 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, Config: 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.
|
| DiffWave | Ran ran on a synthesized input · constructed an object; no result was computed |
yoyololicon/variational-diffwave/src/diffwave/model.py code served (permissive licence) · get_code("ae2c486b1f90a559") How it ranRan 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, DiffWave: 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.
|
| Discretization | Ran ran on a synthesized input |
martenlienen/bsi/bsi/vdm.py code served (permissive licence) · get_code("bb89d1300208329a") How it ranRan in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 1, dtype family int, shape (8,) 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.
|
| EncDec | Ran ran on a synthesized input · constructed an object; no result was computed |
google-research/vdm/model_vdm.py code served (permissive licence) · get_code("491caa588adc4c8f") How it ranRan 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, EncDec: 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.
|
| FixedLinearSchedule | Ran ran on a synthesized input · constructed an object; no result was computed |
addtt/variational-diffusion-models/vdm.py code served (permissive licence) · get_code("d212b472730598ed") How it ranRan 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, FixedLinearSchedule: 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.
|
| GaussianDiffusion2 | Ran ran on a synthesized input · constructed an object; no result was computed |
hojonathanho/diffusion/diffusion_tf/diffusion_utils_2.py pointer only (licence: NONE) · get_code("16c2edc6a8669837") How it ranThe body of this sample is withheld: GitHub detected no licence in this paper's repository, hojonathanho/diffusion when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded. |
| LearnedLinearSchedule | Ran ran on a synthesized input · constructed an object; no result was computed |
addtt/variational-diffusion-models/vdm.py code served (permissive licence) · get_code("ea5520290d73f53d") How it ranRan 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, LearnedLinearSchedule: 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.
|
| LogSNR | Ran ran on a synthesized input · constructed an object; no result was computed |
revsic/jax-variational-diffwave/vlbdiffwave/impl.py code served (permissive licence) · get_code("212e87de38149204") How it ranRan 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, LogSNR: 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.
|
| NoiseSchedule_FixedLinear | Ran ran on a synthesized input · constructed an object; no result was computed |
google-research/vdm/model_vdm.py code served (permissive licence) · get_code("069d8acfc9e74bee") How it ranRan 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, NoiseSchedule_FixedLinear: 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.
|
| NoiseSchedule_Scalar | Ran ran on a synthesized input · constructed an object; no result was computed |
google-research/vdm/model_vdm.py code served (permissive licence) · get_code("529196025d205b76") How it ranRan 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, NoiseSchedule_Scalar: 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.
|
| PosDense | Ran ran on a synthesized input · constructed an object; no result was computed |
revsic/jax-variational-diffwave/vlbdiffwave/impl.py code served (permissive licence) · get_code("cb2f69aa1655d268") How it ranRan 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, PosDense: 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.
|
| ResidualBlock | Ran ran on a synthesized input · constructed an object; no result was computed |
yoyololicon/variational-diffwave/src/diffwave/model.py code served (permissive licence) · get_code("7fcce4b4613629ad") How it ranRan 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, ResidualBlock: 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.
|
| VDMConfig | Ran ran on a synthesized input · constructed an object; no result was computed |
google-research/vdm/model_vdm.py code served (permissive licence) · get_code("f447b1a658e14845") How it ranRan 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, VDMConfig: 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.
|
| VDMOutput | Ran ran on a synthesized input · constructed an object; no result was computed |
google-research/vdm/model_vdm.py code served (permissive licence) · get_code("a08f56d5383b4deb") How it ranRan 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, VDMOutput: 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.
|
| WNDense | Ran ran on a synthesized input · constructed an object; no result was computed |
revsic/jax-variational-diffwave/vlbdiffwave/impl.py code served (permissive licence) · get_code("3421a6b0be94af74") How it ranRan 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, WNDense: 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.
|
| WNDilatedConv | Ran ran on a synthesized input · constructed an object; no result was computed |
revsic/jax-variational-diffwave/vlbdiffwave/impl.py code served (permissive licence) · get_code("a4d6fbb9fe89feb5") How it ranRan 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, WNDilatedConv: 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.
|
| WaveNetBlock | Ran ran on a synthesized input · constructed an object; no result was computed |
revsic/jax-variational-diffwave/vlbdiffwave/impl.py code served (permissive licence) · get_code("f55545fd60b5575f") How it ranRan 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, WaveNetBlock: 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.
|
| _invert_perm | Ran ran; our contract draft was wrong, not the code |
google-research/vdm/model_vdm.py code served (permissive licence) · get_code("6f2b76f02949e8a2") How it ranRan 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 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.
|
| broadcast_right | Ran ran, violated the contract we drafted |
martenlienen/bsi/bsi/vdm.py code served (permissive licence) · get_code("5c967875cb5a67f1") How it ranRan in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned Tensor, ndim 4, dtype family float, shape (2, 4, 1, 1) 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.
|
| constant | Ran ran; our contract draft was wrong, not the code |
revsic/jax-variational-diffwave/vlbdiffwave/impl.py code served (permissive licence) · get_code("4f4db12a92cab549") How it ranRan 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 function. 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.
|
| constant_init | Ran ran; our contract draft was wrong, not the code |
google-research/vdm/model_vdm.py code served (permissive licence) · get_code("efd5fbe71f5d4417") How it ranRan 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 function. 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.
|
| dot_product_attention | Ran ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT) |
google-research/vdm/model_vdm.py code served (permissive licence) · get_code("3650a2dfe2e97298") How it ranRan 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 ArrayImpl, ndim 4, dtype family float, shape (2, 6, 3, 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.
|
| get_timestep_embedding | Ran ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT) |
google-research/vdm/model_vdm.py code served (permissive licence) · get_code("0a5352fc3e23401b") How it ranRan 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 ArrayImpl, 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.
|
| kl_std_normal | Ran ran; our contract draft was wrong, not the code |
addtt/variational-diffusion-models/vdm.py code served (permissive licence) · get_code("abd7e936a4a2330d") How it ranRan 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.
|
| maybe_unpack_batch | Ran ran; our contract draft was wrong, not the code |
addtt/variational-diffusion-models/vdm.py code served (permissive licence) · get_code("656e283a7a532e0d") How it ranRan 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. 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.
|
| normal_kl | Ran ran; our contract draft was wrong, not the code |
hojonathanho/diffusion/diffusion_tf/diffusion_utils_2.py pointer only (licence: NONE) · get_code("5f5e07eb94d1c5d2") How it ranThe body of this sample is withheld: GitHub detected no licence in this paper's repository, hojonathanho/diffusion when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded. |
| unsqueeze_right | Ran ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT) |
addtt/variational-diffusion-models/vdm.py code served (permissive licence) · get_code("9081710b06dc58b5") How it ranRan 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, 1) 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.
|
| DenseMonotone | Not yet run unverified (harvested, no recorded run) |
google-research/vdm/model_vdm.py code served (permissive licence) · get_code("2bfdfec42cfdcdcf") How it ranno invocation recorded. |
| DiffWave | Not yet run unverified (harvested, no recorded run) |
revsic/jax-variational-diffwave/vlbdiffwave/impl.py code served (permissive licence) · get_code("dacd58542a91f9cc") How it ranno invocation recorded. |
| NoiseSchedule_NNet | Not yet run unverified (harvested, no recorded run) |
google-research/vdm/model_vdm.py code served (permissive licence) · get_code("6f9deb29cfceeb7f") How it ranno invocation recorded. |
| ResnetBlock | Not yet run unverified (harvested, no recorded run) |
google-research/vdm/model_vdm.py code served (permissive licence) · get_code("32c40355362a0b32") How it ranno invocation recorded. |
| ScoreUNet | Not yet run unverified (harvested, no recorded run) |
google-research/vdm/model_vdm.py code served (permissive licence) · get_code("3c07ccd2fae8c079") How it ranno invocation recorded. |
| VDM | Not yet run unverified (harvested, no recorded run) |
google-research/vdm/model_vdm.py code served (permissive licence) · get_code("29cb7af455c82dc4") How it ranno invocation recorded. |
| VDM | Not yet run unverified (harvested, no recorded run) |
addtt/variational-diffusion-models/vdm.py code served (permissive licence) · get_code("d69b5df2d07b253c") How it ranno invocation recorded. |
| VDM | Not yet run unverified (harvested, no recorded run) |
martenlienen/bsi/bsi/vdm.py code served (permissive licence) · get_code("b780d07f50dbb622") How it ranno invocation recorded. |
| VLBDiffWave | Not yet run unverified (harvested, no recorded run) |
revsic/jax-variational-diffwave/vlbdiffwave/impl.py code served (permissive licence) · get_code("9f2532a370c4d0c4") How it ranno invocation recorded. |
Of 38 samples lifted from this paper's linked repositories, 38 carry a function identity and 0 do not. Of the 38 distinct functions with an identity, 10 are also present in other papers whose code we hold (6 with at least one paper dated earlier, 0 the same month, 4 later) and 28 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.
| Function | Also present in | Earliest-dated | Others named |
|---|---|---|---|
| constant_init brick of this paper's assembly: helper |
3 other papers 3 later |
2310.19789 (2023-10) DiffEnc: Variational Diffusion with a Learned Encoder |
2403.16382 (2024-03) An image-computable model of speeded decision-making 2501.06999 (2025-01) Likelihood Training of Cascaded Diffusion Models via Hierarchical Volume-preserving Maps |
| get_timestep_embedding brick of this paper's assembly: helper |
3 other papers 3 later |
2305.03935 (2023-05) Improved Techniques for Maximum Likelihood Estimation for Diffusion ODEs |
2312.13236 (2023-12) Diffusion Models With Learned Adaptive Noise 2501.06999 (2025-01) Likelihood Training of Cascaded Diffusion Models via Hierarchical Volume-preserving Maps |
| dot_product_attention brick of this paper's assembly: helper |
2 other papers 2 later |
2305.03935 (2023-05) Improved Techniques for Maximum Likelihood Estimation for Diffusion ODEs |
2501.06999 (2025-01) Likelihood Training of Cascaded Diffusion Models via Hierarchical Volume-preserving Maps |
| kl_std_normal brick of this paper's assembly: loss |
2 other papers 2 later |
2403.10648 (2024-03) Debiasing with Diffusion: Probabilistic reconstruction of Dark Matter fields from galaxies with CAMELS |
2412.01003 (2024-12) Competition Dynamics Shape Algorithmic Phases of In-Context Learning |
| Config brick of this paper's assembly: helper |
1 other paper 1 dated earlier · 1 only as an assembly brick |
2009.09761 (2020-09) DiffWave: A Versatile Diffusion Model for Audio Synthesis |
|
| DiffWave brick of this paper's assembly: model |
1 other paper 1 dated earlier |
2009.09761 (2020-09) DiffWave: A Versatile Diffusion Model for Audio Synthesis |
|
| WNDense brick of this paper's assembly: helper |
1 other paper 1 dated earlier · 1 only as an assembly brick |
2009.09761 (2020-09) DiffWave: A Versatile Diffusion Model for Audio Synthesis |
|
| WNDilatedConv brick of this paper's assembly: helper |
1 other paper 1 dated earlier · 1 only as an assembly brick |
2009.09761 (2020-09) DiffWave: A Versatile Diffusion Model for Audio Synthesis |
|
| WaveNetBlock brick of this paper's assembly: model |
1 other paper 1 dated earlier |
2009.09761 (2020-09) DiffWave: A Versatile Diffusion Model for Audio Synthesis |
|
| constant brick of this paper's assembly: helper |
1 other paper 1 dated earlier · 1 only as an assembly brick |
2009.09761 (2020-09) DiffWave: A Versatile Diffusion Model for Audio Synthesis |
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.
Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.
Diffusion-based generative models have demonstrated a capacity for perceptually impressive synthesis, but can they also be great likelihood-based models? We answer this in the affirmative, and introduce a family of diffusion-based generative models that obtain state-of-the-art likelihoods on standard image density estimation benchmarks. Unlike other diffusion-based models, our method allows for efficient optimization of the noise schedule jointly with the rest of the model. We show that the variational lower bound (VLB) simplifies to a remarkably short expression in terms of the signal-to-noise ratio of the diffused data, thereby improving our theoretical understanding of this model class. Using this insight, we prove an equivalence between several models proposed in the literature. In addition, we show that the continuous-time VLB is invariant to the noise schedule, except for the signal-to-noise ratio at its endpoints. This enables us to learn a noise schedule that minimizes the variance of the resulting VLB estimator, leading to faster optimization. Combining these advances with architectural improvements, we obtain state-of-the-art likelihoods on image density estimation benchmarks, outperforming autoregressive models that have dominated these benchmarks for many years, with often significantly faster optimization. In addition, we show how to use the model as part of a bits-back compression scheme, and demonstrate lossless compression rates close to the theoretical optimum. Code is available at https://github.com/google-research/vdm.
The same record, over MCP at https://syntology.ai/mcp:
get_harvested_code_for_paper("2107.00630")
get_code_for_paper("2107.00630")
have("2107.00630")
The run record, dated, one paper per request, free:
curl https://syntology.ai/api/ran/2107.00630.json
A badge for a README (the split and the date, never a ratio):
[](https://syntology.ai/paper/2107.00630)
Connect an agent — have() is free.