[Bugfix] Enable loading FP8 checkpoints for gpt_bigcode models (vllm-project#5460)

Signed-off-by: Thomas Parnell <tpa@zurich.ibm.com>

Author: tdoublep

vllm/model_executor/models/gpt_bigcode.py

@@ -299,4 +299,10 @@ def load_weights(self, weights: Iterable[Tuple[str, torch.Tensor]]):
             param = params_dict[name]
             weight_loader = getattr(param, "weight_loader",
                                     default_weight_loader)
-            weight_loader(param, loaded_weight)
+            # TODO (@robertgshaw2-neuralmagic): move to fp8 linear method
+            if "c_attn.input_scale" in name or "c_attn.weight_scale" in name:
+                weight_loader(param, loaded_weight, 'q')
+                weight_loader(param, loaded_weight, 'k')
+                weight_loader(param, loaded_weight, 'v')
+            else:
+                weight_loader(param, loaded_weight)