add mmlu evals

evals/mmlu/mmlu_nl_Llama-2-7b-chat-hf.json

@@ -0,0 +1,23 @@
+{
+  "results": {
+    "mmlu_nl": {
+      "acc": 0.35812400394627003,
+      "acc_stderr": 0.0041768654938793455,
+      "acc_norm": 0.33399104500265614,
+      "acc_norm_stderr": 0.0041088061542381964
+    }
+  },
+  "versions": {
+    "mmlu_nl": 0
+  },
+  "config": {
+    "model": "hf-auto",
+    "model_args": "pretrained=meta-llama/Llama-2-7b-chat-hf,use_accelerate=True,device_map_option=auto,dtype=bfloat16,load_in_8bit=True",
+    "batch_size": "auto",
+    "device": "cuda",
+    "no_cache": false,
+    "limit": null,
+    "bootstrap_iters": 100000,
+    "description_dict": {}
+  }
+}

evals/mmlu/mmlu_nl_Llama-2-7b-hf.json

@@ -0,0 +1,23 @@
+{
+  "results": {
+    "mmlu_nl": {
+      "acc": 0.346436973514457,
+      "acc_stderr": 0.004145377277152223,
+      "acc_norm": 0.3213933368748577,
+      "acc_norm_stderr": 0.004068512934976491
+    }
+  },
+  "versions": {
+    "mmlu_nl": 0
+  },
+  "config": {
+    "model": "hf-auto",
+    "model_args": "pretrained=meta-llama/Llama-2-7b-hf,use_accelerate=True,device_map_option=auto,dtype=bfloat16,load_in_8bit=True",
+    "batch_size": "auto",
+    "device": "cuda",
+    "no_cache": false,
+    "limit": null,
+    "bootstrap_iters": 100000,
+    "description_dict": {}
+  }
+}