Skip to content

Add support for deepseek_v32 and glm_moe_dsa architectures - #18

Merged
sharma-open-source merged 1 commit into
mainfrom
version2.0
Jul 20, 2026
Merged

sharma-open-source merged 1 commit into
mainfrom
version2.0

Conversation

@sharma-open-source

Copy link
Copy Markdown
Owner
  • Introduced new architecture specifications for deepseek_v32 and glm_moe_dsa in architectures.py.
  • Implemented layer routing and cache management specific to deepseek_v32 and glm_moe_dsa in archspec.py.
  • Enhanced StreamingEngine to handle paired cache structures and modified forward methods for compatibility with new architectures.
  • Added tests for glm_moe_dsa to ensure byte-identical reassembly and streamed forward consistency with reference models.
  • Updated KVStore and prefix cache handling to support new cache types and ensure correct serialization/deserialization.
  • Adjusted generate.py to prevent batched generation with incompatible cache types.

- Introduced new architecture specifications for deepseek_v32 and glm_moe_dsa in architectures.py.
- Implemented layer routing and cache management specific to deepseek_v32 and glm_moe_dsa in archspec.py.
- Enhanced StreamingEngine to handle paired cache structures and modified forward methods for compatibility with new architectures.
- Added tests for glm_moe_dsa to ensure byte-identical reassembly and streamed forward consistency with reference models.
- Updated KVStore and prefix cache handling to support new cache types and ensure correct serialization/deserialization.
- Adjusted generate.py to prevent batched generation with incompatible cache types.
@sharma-open-source
sharma-open-source merged commit 9a6e1fd into main Jul 20, 2026
2 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant