[Bugfix]Fix Phi-3 BNB online quantization (#10417)

Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>

vllm/model_executor/layers/linear.py

@@ -470,7 +470,8 @@ class MergedColumnParallelLinear(ColumnParallelLinear):
         needs_scalar_to_array = getattr(param, "needs_scalar_to_array", False)
 
         if loaded_shard_id is None:
-            # Loaded weight is already fused on disk (qkv/mlp).
+            # Loaded weight is already fused on disk (mlp).
+            # (e.g., Phi-3's gate_up_proj).
             if output_dim is None:
                 if needs_scalar_to_array:
                     param_data, loaded_weight = adjust_scalar_to_fused_array(
@@ -480,6 +481,8 @@ class MergedColumnParallelLinear(ColumnParallelLinear):
                 param_data.copy_(loaded_weight)
                 return
             current_shard_offset = 0
+            use_bitsandbytes_4bit = getattr(param, "use_bitsandbytes_4bit",
+                                            False)
             shard_offsets: List[Tuple[int, int, int]] = []
             for i, output_size in enumerate(self.output_sizes):
                 shard_offsets.append((i, current_shard_offset, output_size))
@@ -495,7 +498,9 @@ class MergedColumnParallelLinear(ColumnParallelLinear):
                     # Special case for Marlin.
                     shard_size, shard_offset = adjust_marlin_shard(
                         param, shard_size, shard_offset)
-
+                if use_bitsandbytes_4bit:
+                    shard_size = loaded_weight.shape[output_dim] // 2
+                    shard_offset = shard_size * shard_id
                 loaded_weight_shard = loaded_weight.narrow(
                     output_dim, shard_offset, shard_size)
                 self.weight_loader(param, loaded_weight_shard, shard_id)
@@ -808,7 +813,8 @@ class QKVParallelLinear(ColumnParallelLinear):
         needs_scalar_to_array = getattr(param, "needs_scalar_to_array", False)
 
         if loaded_shard_id is None:
-            # Loaded weight is already fused on disk (qkv/mlp).
+            # Loaded weight is already fused on disk (qkv).
+            # (e.g., Phi-3's qkv_proj).
             if output_dim is None:
                 if needs_scalar_to_array:
                     param_data, loaded_weight = adjust_scalar_to_fused_array(

vllm/model_executor/models/phi3.py

@@ -14,3 +14,13 @@ class Phi3ForCausalLM(LlamaForCausalLM):
             "gate_up_proj",
         ],
     }
+
+    # BitandBytes specific attributes
+    default_bitsandbytes_target_modules = [
+        ".gate_up_proj.",
+        ".down_proj.",
+        ".qkv_proj.",
+        ".o_proj.",
+    ]
+    # Initialize an empty dict when there is no stacked parameter mapping.
+    bitsandbytes_stacked_params_mapping = {}