[AMD] Support AMD MXFP4 Qwen3.5-397B-A17B model (#21234)

python/sglang/srt/models/qwen3_5.py

@@ -88,6 +88,7 @@ from sglang.srt.utils import (
     cpu_has_amx_support,
     is_cpu,
     is_cuda,
+    is_gfx95_supported,
     is_npu,
     make_layers,
     set_weight_attrs,
@@ -98,6 +99,7 @@ logger = logging.getLogger(__name__)
 _is_cuda = is_cuda()
 _is_npu = is_npu()
 _is_cpu = is_cpu()
+_is_gfx95 = is_gfx95_supported()
 _is_amx_available = cpu_has_amx_support()
 
 
@@ -879,6 +881,14 @@ ALL_DECODER_LAYER_TYPES = {
 class Qwen3_5ForCausalLM(nn.Module):
     """Qwen3.5 Model with support for dense variant."""
 
+    if _is_gfx95:
+        packed_modules_mapping = {
+            "qkv_proj": ["q_proj", "k_proj", "v_proj"],
+            "gate_up_proj": ["gate_proj", "up_proj"],
+            "in_proj_qkvz": ["in_proj_qkv", "in_proj_z"],
+            "in_proj_ba": ["in_proj_b", "in_proj_a"],
+        }
+
     def __init__(
         self,
         config: Qwen3_5TextConfig,
@@ -1300,6 +1310,10 @@ class Qwen3_5MoeForCausalLM(Qwen3_5ForCausalLM):
 
 
 class Qwen3_5ForConditionalGeneration(Qwen3VLForConditionalGeneration):
+    if _is_gfx95:
+        packed_modules_mapping = Qwen3_5ForCausalLM.packed_modules_mapping
+        hf_to_sglang_mapper = None
+
     def __init__(
         self,
         config: Qwen3_5Config,
@@ -1422,6 +1436,10 @@ class Qwen3_5ForConditionalGeneration(Qwen3VLForConditionalGeneration):
 class Qwen3_5MoeForConditionalGeneration(Qwen3VLForConditionalGeneration):
     """Qwen3.5 MoE Vision-Language Model."""
 
+    if _is_gfx95:
+        packed_modules_mapping = Qwen3_5ForCausalLM.packed_modules_mapping
+        hf_to_sglang_mapper = None
+
     def __init__(
         self,
         config: Qwen3_5MoeConfig,