[https://nvbugs/5449155][fix] Fix DeepSeek R1 weight loading for TP16 (#6913)

Signed-off-by: Aurelien Chartier <[email protected]>
Signed-off-by: Wangshanshan <[email protected]>

Author: achartier

tensorrt_llm/_torch/modules/linear.py

@@ -650,9 +650,12 @@ def load_weights_vanilla(self, module: Linear, weights: List[Dict]) -> None:
         load_weights_vanilla_helper(module, weights)
 
         scale_name = self._get_scale_name(weights)
-        weight_scale = load_weight_shard(weights[0][scale_name], module.tp_size,
-                                         module.tp_rank,
-                                         module.tp_mode).squeeze()
+        full_weight_scale = weights[0][scale_name]
+        # modelopt fp8_pb_wo can have 2 extra singleton dimensions
+        if full_weight_scale.dim() == 4:
+            full_weight_scale = full_weight_scale.squeeze(1).squeeze(-1)
+        weight_scale = load_weight_shard(full_weight_scale, module.tp_size,
+                                         module.tp_rank, module.tp_mode)
         copy_weight(module.weight_scale, weight_scale)
         if "input_scale" in weights[0]:
             copy_weight(module.input_scale, weights[0]["input_scale"])
@@ -665,13 +668,23 @@ def load_weights_fused_qkv_linear(self, module: Linear,
         fused_weight = torch.cat((q_weight, k_weight, v_weight))
 
         scale_name = self._get_scale_name(weights)
-        q_scale = load_weight_shard(weights[0][scale_name], module.tp_size,
+        full_q_scale = weights[0][scale_name]
+        full_k_scale = weights[1][scale_name]
+        full_v_scale = weights[2][scale_name]
+        # modelopt fp8_pb_wo can have 2 extra singleton dimensions
+        if full_q_scale.dim() == 4:
+            full_q_scale = full_q_scale.squeeze(1).squeeze(-1)
+        if full_k_scale.dim() == 4:
+            full_k_scale = full_k_scale.squeeze(1).squeeze(-1)
+        if full_v_scale.dim() == 4:
+            full_v_scale = full_v_scale.squeeze(1).squeeze(-1)
+        q_scale = load_weight_shard(full_q_scale, module.tp_size,
                                     module.tp_rank, module.tp_mode)
-        k_scale = load_weight_shard(weights[1][scale_name], module.tp_size,
+        k_scale = load_weight_shard(full_k_scale, module.tp_size,
                                     module.tp_rank, module.tp_mode)
-        v_scale = load_weight_shard(weights[2][scale_name], module.tp_size,
+        v_scale = load_weight_shard(full_v_scale, module.tp_size,
                                     module.tp_rank, module.tp_mode)
-        fused_fp8_block_scale = torch.cat((q_scale, k_scale, v_scale)).squeeze()
+        fused_fp8_block_scale = torch.cat((q_scale, k_scale, v_scale))
 
         copy_weight(module.weight, fused_weight)
         copy_weight(module.weight_scale, fused_fp8_block_scale)
@@ -683,11 +696,18 @@ def load_weights_fused_gate_up_linear(self, module: Linear,
         fused_weight = torch.cat((gate_weight, up_weight))
 
         scale_name = self._get_scale_name(weights)
-        left_scale = load_weight_shard(weights[0][scale_name], module.tp_size,
+        full_left_scale = weights[0][scale_name]
+        full_right_scale = weights[1][scale_name]
+        # modelopt fp8_pb_wo can have 2 extra singleton dimensions
+        if full_left_scale.dim() == 4:
+            full_left_scale = full_left_scale.squeeze(1).squeeze(-1)
+        if full_right_scale.dim() == 4:
+            full_right_scale = full_right_scale.squeeze(1).squeeze(-1)
+        left_scale = load_weight_shard(full_left_scale, module.tp_size,
                                        module.tp_rank, module.tp_mode)
-        right_scale = load_weight_shard(weights[1][scale_name], module.tp_size,
+        right_scale = load_weight_shard(full_right_scale, module.tp_size,
                                         module.tp_rank, module.tp_mode)
-        fused_scale = torch.cat([left_scale, right_scale], dim=0).squeeze()
+        fused_scale = torch.cat([left_scale, right_scale], dim=0)
         copy_weight(module.weight, fused_weight)
         copy_weight(module.weight_scale, fused_scale)