vrr/kvcache-ai-ktransformers
2
0
Fork 0
mirror of https://github.com/kvcache-ai/ktransformers.git synced 2025-09-14 17:19:42 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

fix rope; update moegate

Browse source
This commit is contained in:
Azure 2025-02-01 18:05:45 +00:00
parent f873558a89
commit f748cd29f0
5 changed files with 54 additions and 21 deletions
Download patch file Download diff file Expand all files Collapse all files

8
ktransformers/optimize/optimize_rules/DeepSeek-V3-Chat-multi-gpu.yaml
View file

@ -8,17 +8,17 @@
- match:
name: "^model\\.layers\\.(0|[1-9]|[12][0-9])\\."
class: ktransformers.models.modeling_deepseek.DeepseekV2YarnRotaryEmbedding
class: ktransformers.models.modeling_deepseek_v3.DeepseekV3RotaryEmbedding
replace:
class: ktransformers.operators.RoPE.YarnRotaryEmbedding
class: ktransformers.operators.RoPE.DeepSeekV3YarnRotaryEmbedding
kwargs:
generate_device: "cuda:0"
prefill_device: "cuda:0"
- match:
name: "^model\\.layers\\.([3456][0-9])\\."
class: ktransformers.models.modeling_deepseek.DeepseekV2YarnRotaryEmbedding
class: ktransformers.models.modeling_deepseek_v3.DeepseekV3RotaryEmbedding
replace:
class: ktransformers.operators.RoPE.YarnRotaryEmbedding
class: ktransformers.operators.RoPE.DeepSeekV3YarnRotaryEmbedding
kwargs:
generate_device: "cuda:1"
prefill_device: "cuda:1"