Skip to content

Add DramSsdKVEmbeddingCache composite backend (DRAM forwarding) (#5926)#5926

Open
lizhe-ji wants to merge 1 commit into
pytorch:mainfrom
lizhe-ji:export-D108964307
Open

Add DramSsdKVEmbeddingCache composite backend (DRAM forwarding) (#5926)#5926
lizhe-ji wants to merge 1 commit into
pytorch:mainfrom
lizhe-ji:export-D108964307

Conversation

@lizhe-ji

@lizhe-ji lizhe-ji commented Jun 18, 2026

Copy link
Copy Markdown

Summary:
X-link: https://github.com/facebookresearch/FBGEMM/pull/2844

Introduce the DramSsdKVEmbeddingCache composite backend and its
TorchScript wrapper as a forwarding skeleton: every EmbeddingKVDB method
delegates to the wrapped DRAM (L2) cache. This establishes the class and
build wiring; the SSD (L3) tier and its orchestration are added in the
follow-up.

  • DramSsdKVEmbeddingCache: wraps DramKVEmbeddingCache and forwards all calls
  • DramSsdKVEmbeddingCacheWrapper: TorchScript wrapper that builds the DRAM
    cache, wraps it in the composite, and forwards calls
  • BUCK: register the two new headers

Differential Revision: D108964307

@meta-cla meta-cla Bot added the cla signed label Jun 18, 2026
@meta-codesync meta-codesync Bot changed the title Add DramSsdKVEmbeddingCache composite backend (DRAM forwarding) Add DramSsdKVEmbeddingCache composite backend (DRAM forwarding) (#5926) Jun 19, 2026
lizhe-ji pushed a commit to lizhe-ji/FBGEMM-1 that referenced this pull request Jun 19, 2026
…rch#5926)

Summary:
X-link: facebookresearch/FBGEMM#2844


Introduce the DramSsdKVEmbeddingCache composite backend and its
TorchScript wrapper as a forwarding skeleton: every EmbeddingKVDB method
delegates to the wrapped DRAM (L2) cache. This establishes the class and
build wiring; the SSD (L3) tier and its orchestration are added in the
follow-up.

- DramSsdKVEmbeddingCache: wraps DramKVEmbeddingCache and forwards all calls
- DramSsdKVEmbeddingCacheWrapper: TorchScript wrapper that builds the DRAM
  cache, wraps it in the composite, and forwards calls
- BUCK: register the two new headers

Differential Revision: D108964307
@lizhe-ji lizhe-ji force-pushed the export-D108964307 branch 2 times, most recently from e11a4b8 to 1831fa1 Compare June 23, 2026 19:00
lizhe-ji pushed a commit to lizhe-ji/FBGEMM-1 that referenced this pull request Jun 23, 2026
…rch#5926)

Summary:
X-link: facebookresearch/FBGEMM#2844


Introduce the DramSsdKVEmbeddingCache composite backend and its
TorchScript wrapper as a forwarding skeleton: every EmbeddingKVDB method
delegates to the wrapped DRAM (L2) cache. This establishes the class and
build wiring; the SSD (L3) tier and its orchestration are added in the
follow-up.

- DramSsdKVEmbeddingCache: wraps DramKVEmbeddingCache and forwards all calls
- DramSsdKVEmbeddingCacheWrapper: TorchScript wrapper that builds the DRAM
  cache, wraps it in the composite, and forwards calls
- BUCK: register the two new headers

Differential Revision: D108964307
…rch#5926)

Summary:
X-link: facebookresearch/FBGEMM#2844


Introduce the DramSsdKVEmbeddingCache composite backend and its
TorchScript wrapper as a forwarding skeleton: every EmbeddingKVDB method
delegates to the wrapped DRAM (L2) cache. This establishes the class and
build wiring; the SSD (L3) tier and its orchestration are added in the
follow-up.

- DramSsdKVEmbeddingCache: wraps DramKVEmbeddingCache and forwards all calls
- DramSsdKVEmbeddingCacheWrapper: TorchScript wrapper that builds the DRAM
  cache, wraps it in the composite, and forwards calls
- BUCK: register the two new headers

Differential Revision: D108964307
@lizhe-ji lizhe-ji force-pushed the export-D108964307 branch from 1831fa1 to 66903bc Compare June 24, 2026 17:58
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant