vrr/kvcache-ai-ktransformers
2
0
Fork 0
mirror of https://github.com/kvcache-ai/ktransformers.git synced 2026-04-28 11:49:51 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 4

add balance-serve, support concurrence

Browse source
This commit is contained in:
Atream 2025-03-31 22:55:32 +08:00
parent 8d0292aa44
commit 25cee5810e
196 changed files with 22077 additions and 565 deletions
Download patch file Download diff file Expand all files Collapse all files

6
ktransformers/optimize/optimize_rules/DeepSeek-V3-Chat-multi-gpu-marlin.yaml
View file

@ -10,7 +10,7 @@
name: "^model\\.layers\\.(0|[1-9]|[12][0-9])\\."
class: ktransformers.models.modeling_deepseek_v3.DeepseekV3RotaryEmbedding
replace:
class: ktransformers.operators.RoPE.KMoEGateDeepSeekV3
class: ktransformers.operators.RoPE.YarnRotaryEmbeddingV3
kwargs:
generate_device: "cuda:0"
prefill_device: "cuda:0"
@ -66,7 +66,7 @@
name: "^model\\.layers\\.(0|[1-9]|[12][0-9])\\.mlp\\.gate$"
class: ktransformers.models.modeling_deepseek_v3.MoEGate
replace:
class: ktransformers.operators.gate.KMoEGate
class: ktransformers.operators.gate.KMoEGateDeepSeekV3
kwargs:
generate_device: "cuda:0"
prefill_device: "cuda:0"
@ -74,7 +74,7 @@
name: "^model\\.layers\\.([3456][0-9])\\.mlp\\.gate$"
class: ktransformers.models.modeling_deepseek_v3.MoEGate
replace:
class: ktransformers.operators.gate.KMoEGate # mlp module with custom forward function
class: ktransformers.operators.gate.KMoEGateDeepSeekV3 # mlp module with custom forward function
kwargs:
generate_device: "cuda:1"
prefill_device: "cuda:1"