Skip to content

feat(bailingmoe3): add bailingmoe3 arch support - #32

Open
oscar1229 wants to merge 1 commit into
spacemit-com:mtmd-backendfrom
oscar1229:feat/bailingmoe3-support
Open

feat(bailingmoe3): add bailingmoe3 arch support#32
oscar1229 wants to merge 1 commit into
spacemit-com:mtmd-backendfrom
oscar1229:feat/bailingmoe3-support

Conversation

@oscar1229

Copy link
Copy Markdown

Overview

Add bailingmoe3 arch support

Additional information

  • Add LLM_ARCH_BAILINGMOE3 enum and arch name mapping
  • Add LLM_KV_KDA_SAFE_GATE / LLM_KV_KDA_GATE_LOWER_BOUND KV keys
  • Register llama_model_bailingmoe3 in model factory and rope_type
  • Add kda_safe_gate / kda_gate_lower_bound to llama_hparams
  • Add MXFP4 quant guard for 3D MLA tensors
  • Implement llama_model_bailingmoe3 (hybrid KDA+MLA+MoE graph)
  • Support Q-LoRA bottleneck for Ling-3.0-tiny (q_lora_rank>0)
  • Make kda_safe_gate optional; load SwiGLU clamps from GGUF

Requirements

@github-actions github-actions Bot added the model label Aug 19, 2026
@oscar1229
oscar1229 marked this pull request as draft August 20, 2026 01:09
- Add LLM_ARCH_BAILINGMOE3 enum and arch name mapping
- Add LLM_KV_KDA_SAFE_GATE / LLM_KV_KDA_GATE_LOWER_BOUND KV keys
- Register llama_model_bailingmoe3 in model factory and rope_type
- Add kda_safe_gate / kda_gate_lower_bound to llama_hparams
- Add MXFP4 quant guard for 3D MLA tensors
- Implement llama_model_bailingmoe3 (hybrid KDA+MLA+MoE graph)
- Support Q-LoRA bottleneck for Ling-3.0-tiny (q_lora_rank>0)
- Make kda_safe_gate optional; load SwiGLU clamps from GGUF
@oscar1229
oscar1229 force-pushed the feat/bailingmoe3-support branch from 5102433 to 2ea09ec Compare August 20, 2026 02:40
@oscar1229
oscar1229 marked this pull request as ready for review August 20, 2026 03:02
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