Add group act order case to lm_eval test (#1080)

SUMMARY:
- Requires next transformers release/ when the following commit is
released: huggingface/transformers#35615
- Adds group act order case which previously could not be loaded through
the AutoModel pathway due to incorrect substring replacement of the
`weight_g_idx` parameter

Testing:
- Test passes locally with transformers 4.49

tests/e2e/vLLM/lm_eval_configs/w4a16_actorder_group.yaml

@@ -0,0 +1,11 @@
+cadence: "weekly"
+model: meta-llama/Meta-Llama-3-8B-Instruct
+recipe: tests/e2e/vLLM/recipes/actorder/recipe_w4a16_actorder_group.yaml
+num_fewshot: 5
+limit: 1000
+task: "gsm8k"
+dataset_id: HuggingFaceH4/ultrachat_200k
+dataset_split: train_sft
+exact_match,flexible-extract: 0.72
+exact_match,strict-match: 0.72
+scheme: W4A16_actorder_group

tests/e2e/vLLM/lm_eval_configs/w4a16_actorder_weight.yaml

@@ -8,4 +8,4 @@ dataset_id: HuggingFaceH4/ultrachat_200k
 dataset_split: train_sft
 exact_match,flexible-extract: 0.72
 exact_match,strict-match: 0.72
-scheme: W4A16_actorder_group
+scheme: W4A16_actorder_weight