add balance-serve, support concurrence

ktransformers/optimize/optimize_rules/DeepSeek-V2-Lite-Chat.yaml

@@ -22,7 +22,7 @@
   replace:
     class: ktransformers.operators.linear.KTransformersLinear
     kwargs:
-      generate_device: "cpu"
+      generate_device: "cuda"
       prefill_device: "cuda"
       generate_op: "KLinearMarlin"
       prefill_op: "KLinearTorch"

ktransformers/optimize/optimize_rules/DeepSeek-V3-Chat-fp8-linear-ggml-experts-serve.yaml

@@ -0,0 +1,90 @@
+- match:
+    class: ktransformers.models.modeling_deepseek_v3.DeepseekV3RotaryEmbedding
+  replace:
+    class: ktransformers.operators.RoPE.YarnRotaryEmbeddingV3
+    kwargs:
+      generate_device: "cuda"
+      prefill_device: "cuda"
+- match:
+    name: "^model\\.layers\\.(?!.*self_attn\\.kv_b_proj).*$"  # regular expression 
+    class: torch.nn.Linear  # only match modules matching name and class simultaneously
+  replace:
+    class: ktransformers.operators.linear.KTransformersLinear  # optimized Kernel on quantized data types
+    kwargs:
+      generate_device: "cuda"
+      prefill_device: "cuda"
+      generate_op: "KLinearFP8"
+      prefill_op: "KLinearTorch"
+- match:
+    name: "^model\\.layers\\..*\\.mlp$"
+    class: ktransformers.models.modeling_deepseek_v3.DeepseekV3MoE
+  replace:
+    class: ktransformers.operators.experts.KDeepseekV3MoEV2     # mlp module with custom forward function
+    kwargs:
+      generate_device: "cuda"
+      prefill_device: "cuda"
+- match:
+    class: ktransformers.models.modeling_deepseek_v3.MoEGate
+  replace:
+    class: ktransformers.operators.gate.KMoEGate
+    kwargs:
+      generate_device: "cuda:0"
+      prefill_device: "cuda:0"
+- match:
+    name: "^model\\.layers\\..*\\.mlp\\.experts$"
+  replace:
+    class: ktransformers.operators.experts.KTransformersExpertsV2     # custom MoE Kernel with expert paralleism
+    kwargs:
+      prefill_device: "cuda"
+      prefill_op: "KExpertsTorch"
+      generate_device: "cpu"
+      generate_op: "KExpertsCPU"
+      out_device: "cuda"
+  recursive: False # don't recursively inject submodules of this module
+- match:
+    name: "^model\\.layers\\..*\\.self_attn$"
+  replace:
+    class: ktransformers.operators.attention.flashinfer_attn # optimized MLA implementation
+    kwargs:
+      generate_device: "cuda"
+      prefill_device: "cuda"
+- match:
+    name: "^model$"
+  replace:
+    class: "ktransformers.operators.models.KDeepseekV2Model"
+    kwargs:
+      per_layer_prefill_intput_threshold: 0 # 0 is close layer wise prefill
+- match:
+    name: "^model.embed_tokens"
+  replace:
+    class: "default"
+    kwargs:
+      generate_device: "cpu"
+      prefill_device: "cpu"
+
+- match:
+    class: ktransformers.models.modeling_deepseek_v3.DeepseekV3RMSNorm
+  replace:
+    class: ktransformers.operators.layernorm.RMSNorm
+    kwargs:
+      generate_device: "cuda"
+      prefill_device: "cuda"
+
+- match:
+    class: ktransformers.models.modeling_deepseek_v3.DeepseekV3MLP
+  replace:
+    class:  ktransformers.operators.mlp.kDeepseekV3MLP
+    kwargs:
+      generate_device: "cuda"
+      prefill_device: "cuda"
+
+- match:
+    name: "^lm_head$"  # regular expression 
+    class: torch.nn.Linear  # only match modules matching name and class simultaneously
+  replace:
+    class: ktransformers.operators.linear.KTransformersLinear  # optimized Kernel on quantized data types
+    kwargs:
+      generate_device: "cuda"
+      prefill_device: "cuda"
+      generate_op: "VLinearMarlin"
+      prefill_op: "KLinearTorch"

ktransformers/optimize/optimize_rules/DeepSeek-V3-Chat-multi-gpu-fp8-linear-ggml-experts.yaml

@@ -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"
@@ -18,7 +18,7 @@
     name: "^model\\.layers\\.([3456][0-9])\\."
     class: ktransformers.models.modeling_deepseek_v3.DeepseekV3RotaryEmbedding
   replace:
-    class: ktransformers.operators.RoPE.KMoEGateDeepSeekV3
+    class: ktransformers.operators.RoPE.YarnRotaryEmbeddingV3
     kwargs:
       generate_device: "cuda:1"
       prefill_device: "cuda:1"
@@ -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"

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

@@ -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"

ktransformers/optimize/optimize_rules/DeepSeek-V3-Chat-serve.yaml

@@ -0,0 +1,92 @@
+- match:
+    class: ktransformers.models.modeling_deepseek_v3.DeepseekV3RotaryEmbedding
+  replace:
+    class: ktransformers.operators.RoPE.YarnRotaryEmbeddingV3
+    kwargs:
+      generate_device: "cuda"
+      prefill_device: "cuda"
+
+- match:
+    name: "^lm_head$"  # regular expression 
+    class: torch.nn.Linear  # only match modules matching name and class simultaneously
+  replace:
+    class: ktransformers.operators.linear.KTransformersLinear  # optimized Kernel on quantized data types
+    kwargs:
+      generate_device: "cuda"
+      prefill_device: "cuda"
+      generate_op: "VLinearMarlin"
+      prefill_op: "KLinearTorch"
+
+- match:
+    name: "^model\\.layers\\.(?!.*self_attn\\.kv_b_proj).*$"  # regular expression 
+    class: torch.nn.Linear  # only match modules matching name and class simultaneously
+  replace:
+    class: ktransformers.operators.linear.KTransformersLinear  # optimized Kernel on quantized data types
+    kwargs:
+      generate_device: "cuda"
+      prefill_device: "cuda"
+      generate_op: "VLinearMarlin"
+      prefill_op: "KLinearTorch"
+- match:
+    name: "^model\\.layers\\..*\\.mlp$"
+    class: ktransformers.models.modeling_deepseek_v3.DeepseekV3MoE
+  replace:
+    class: ktransformers.operators.experts.KDeepseekV3MoEV2     # mlp module with custom forward function
+    kwargs:
+      generate_device: "cuda"
+      prefill_device: "cuda"
+- match:
+    class: ktransformers.models.modeling_deepseek_v3.MoEGate
+  replace:
+    class: ktransformers.operators.gate.KMoEGate
+    kwargs:
+      generate_device: "cuda:0"
+      prefill_device: "cuda:0"
+- match:
+    name: "^model\\.layers\\..*\\.mlp\\.experts$"
+  replace:
+    class: ktransformers.operators.experts.KTransformersExpertsV2     # custom MoE Kernel with expert paralleism
+    kwargs:
+      prefill_device: "cuda"
+      prefill_op: "KExpertsTorch"
+      generate_device: "cpu"
+      generate_op: "KExpertsCPU"
+      out_device: "cuda"
+  recursive: False # don't recursively inject submodules of this module
+- match:
+    name: "^model\\.layers\\..*\\.self_attn$"
+  replace:
+    class: ktransformers.operators.attention.flashinfer_attn # optimized MLA implementation
+    kwargs:
+      generate_device: "cuda"
+      prefill_device: "cuda"
+      absorb_for_prefill: False # change this to True to enable long context(prefill may slower).
+- match:
+    name: "^model$"
+  replace:
+    class: "ktransformers.operators.models.KDeepseekV2Model"
+    kwargs:
+      per_layer_prefill_intput_threshold: 0 # 0 is close layer wise prefill
+- match:
+    name: "^model.embed_tokens"
+  replace:
+    class: "default"
+    kwargs:
+      generate_device: "cpu"
+      prefill_device: "cpu"
+
+- match:
+    class: ktransformers.models.modeling_deepseek_v3.DeepseekV3RMSNorm
+  replace:
+    class: ktransformers.operators.layernorm.RMSNorm
+    kwargs:
+      generate_device: "cuda"
+      prefill_device: "cuda"
+
+- match:
+    class: ktransformers.models.modeling_deepseek_v3.DeepseekV3MLP
+  replace:
+    class:  ktransformers.operators.mlp.kDeepseekV3MLP
+    kwargs:
+      generate_device: "cuda"
+      prefill_device: "cuda"

ktransformers/optimize/optimize_rules/DeepSeek-V3-Chat.yaml

@@ -38,7 +38,7 @@
 - match:
     class: ktransformers.models.modeling_deepseek_v3.MoEGate
   replace:
-    class: ktransformers.operators.gate.KMoEGateDeepSeekV3
+    class: ktransformers.operators.gate.KMoEGate
     kwargs:
       generate_device: "cuda:0"
       prefill_device: "cuda:0"

ktransformers/optimize/optimize_rules/Moonlight-16B-A3B-serve.yaml

@@ -0,0 +1,94 @@
+
+
+- match:
+    name: "^lm_head$"  # regular expression 
+    class: torch.nn.Linear  # only match modules matching name and class simultaneously
+  replace:
+    class: ktransformers.operators.linear.KTransformersLinear  # optimized Kernel on quantized data types
+    kwargs:
+      generate_device: "cuda"
+      prefill_device: "cuda"
+      generate_op: "VLinearMarlin"
+      prefill_op: "KLinearTorch"
+
+- match:
+    name: "^model\\.layers\\.(?!.*self_attn\\.kv_b_proj).*$"  # regular expression 
+    class: torch.nn.Linear  # only match modules matching name and class simultaneously
+  replace:
+    class: ktransformers.operators.linear.KTransformersLinear  # optimized Kernel on quantized data types
+    kwargs:
+      generate_device: "cuda"
+      prefill_device: "cuda"
+      generate_op: "VLinearMarlin"
+      prefill_op: "KLinearTorch"
+- match:
+    name: "^model\\.layers\\..*\\.mlp$"
+    class: ktransformers.models.modeling_deepseek_v3.DeepseekV3MoE
+  replace:
+    class: ktransformers.operators.experts.KDeepseekV3MoEV2     # mlp module with custom forward function
+    kwargs:
+      generate_device: "cuda"
+      prefill_device: "cuda"
+- match:
+    class: ktransformers.models.modeling_deepseek_v3.MoEGate
+  replace:
+    class: ktransformers.operators.gate.KMoEGateDeepSeekV3
+    kwargs:
+      generate_device: "cuda:0"
+      prefill_device: "cuda:0"
+- match:
+    name: "^model\\.layers\\..*\\.mlp\\.experts$"
+  replace:
+    class: ktransformers.operators.experts.KTransformersExpertsV2     # custom MoE Kernel with expert paralleism
+    kwargs:
+      prefill_device: "cuda"
+      prefill_op: "KExpertsTorch"
+      generate_device: "cpu"
+      generate_op: "KExpertsCPU"
+      out_device: "cuda"
+  recursive: False # don't recursively inject submodules of this module
+- match:
+    name: "^model\\.layers\\..*\\.self_attn$"
+  replace:
+    class: ktransformers.operators.attention.flashinfer_attn # optimized MLA implementation
+    kwargs:
+      generate_device: "cuda"
+      prefill_device: "cuda"
+      absorb_for_prefill: False # change this to True to enable long context(prefill may slower).
+- match:
+    name: "^model$"
+  replace:
+    class: "ktransformers.operators.models.KDeepseekV2Model"
+    kwargs:
+      per_layer_prefill_intput_threshold: 0 # 0 is close layer wise prefill
+- match:
+    name: "^model.embed_tokens"
+  replace:
+    class: "default"
+    kwargs:
+      generate_device: "cpu"
+      prefill_device: "cpu"
+
+- match:
+    class: ktransformers.models.modeling_deepseek_v3.DeepseekV3RMSNorm
+  replace:
+    class: ktransformers.operators.layernorm.RMSNorm
+    kwargs:
+      generate_device: "cuda"
+      prefill_device: "cuda"
+
+- match:
+    class: ktransformers.models.modeling_deepseek_v3.DeepseekV3MLP
+  replace:
+    class:  ktransformers.operators.mlp.kDeepseekV3MLP
+    kwargs:
+      generate_device: "cuda"
+      prefill_device: "cuda"
+
+- match:
+    class: ktransformers.models.modeling_deepseek_v3.DeepseekV3RotaryEmbedding
+  replace:
+    class: ktransformers.operators.RoPE.RotaryEmbeddingV4
+    kwargs:
+      generate_device: "cuda"
+      prefill_device: "cuda"

ktransformers/optimize/optimize_rules/Moonlight-16B-A3B.yaml

@@ -38,7 +38,7 @@
 - match:
     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"