transformers
cb0adddc - fix(DSV3): parity between native `DeepseekV3MoE` and remote official implementation (#45441)

Commit

193 days ago

fix(DSV3): parity between native `DeepseekV3MoE` and remote official implementation (#45441) fix(`DeepseekV3MoE`): correct expert masking with negative bias

References

#45441 - fix(DSV3): parity between native `DeepseekV3MoE` and remote official implementation

Author

casinca

Parents

08244b96

transformers cb0adddc - fix(DSV3): parity between native `DeepseekV3MoE` and remote official implementation (#45441)

transformers
cb0adddc - fix(DSV3): parity between native `DeepseekV3MoE` and remote official implementation (#45441)