add evaluations

evals/arc/arc_nl_Llama-2-13b-chat-dutch.json

@@ -0,0 +1,23 @@
+{
+  "results": {
+    "arc_nl": {
+      "acc": 0.40975192472198463,
+      "acc_stderr": 0.014389853401438972,
+      "acc_norm": 0.3763900769888794,
+      "acc_norm_stderr": 0.014176015188369369
+    }
+  },
+  "versions": {
+    "arc_nl": 0
+  },
+  "config": {
+    "model": "hf-auto",
+    "model_args": "pretrained=BramVanroy/Llama-2-13b-chat-dutch,use_accelerate=True,device_map_option=auto,dtype=bfloat16,load_in_8bit=True",
+    "batch_size": 2,
+    "device": "cuda",
+    "no_cache": false,
+    "limit": null,
+    "bootstrap_iters": 100000,
+    "description_dict": {}
+  }
+}

evals/arc/arc_nl_Llama-2-13b-chat-hf.json

@@ -0,0 +1,23 @@
+{
+  "results": {
+    "arc_nl": {
+      "acc": 0.40290846877673225,
+      "acc_stderr": 0.014351663146567216,
+      "acc_norm": 0.40889649272882805,
+      "acc_norm_stderr": 0.014385237598173261
+    }
+  },
+  "versions": {
+    "arc_nl": 0
+  },
+  "config": {
+    "model": "hf-auto",
+    "model_args": "pretrained=meta-llama/Llama-2-13b-chat-hf,use_accelerate=True,device_map_option=auto,dtype=bfloat16,load_in_8bit=True",
+    "batch_size": 2,
+    "device": "cuda",
+    "no_cache": false,
+    "limit": null,
+    "bootstrap_iters": 100000,
+    "description_dict": {}
+  }
+}

evals/arc/arc_nl_Llama-2-13b-hf.json

@@ -0,0 +1,23 @@
+{
+  "results": {
+    "arc_nl": {
+      "acc": 0.40889649272882805,
+      "acc_stderr": 0.01438523759817326,
+      "acc_norm": 0.3798118049615056,
+      "acc_norm_stderr": 0.014201184043263487
+    }
+  },
+  "versions": {
+    "arc_nl": 0
+  },
+  "config": {
+    "model": "hf-auto",
+    "model_args": "pretrained=meta-llama/Llama-2-13b-hf,use_accelerate=True,device_map_option=auto,dtype=bfloat16,load_in_8bit=True",
+    "batch_size": 2,
+    "device": "cuda",
+    "no_cache": false,
+    "limit": null,
+    "bootstrap_iters": 100000,
+    "description_dict": {}
+  }
+}

evals/arc/arc_nl_Orca-2-13b.json

@@ -0,0 +1,23 @@
+{
+  "results": {
+    "arc_nl": {
+      "acc": 0.40290846877673225,
+      "acc_stderr": 0.014351663146567216,
+      "acc_norm": 0.4174508126603935,
+      "acc_norm_stderr": 0.014429375469140367
+    }
+  },
+  "versions": {
+    "arc_nl": 0
+  },
+  "config": {
+    "model": "hf-auto",
+    "model_args": "pretrained=microsoft/Orca-2-13b,use_accelerate=True,device_map_option=auto,dtype=bfloat16,load_in_8bit=True",
+    "batch_size": 2,
+    "device": "cuda",
+    "no_cache": false,
+    "limit": null,
+    "bootstrap_iters": 100000,
+    "description_dict": {}
+  }
+}

evals/arc/arc_nl_llama2-13b-ft-mc4_nl_cleaned_tiny.json

@@ -0,0 +1,23 @@
+{
+  "results": {
+    "arc_nl": {
+      "acc": 0.40804106073567153,
+      "acc_stderr": 0.014380576747005828,
+      "acc_norm": 0.39863130881094955,
+      "acc_norm_stderr": 0.014326321344263279
+    }
+  },
+  "versions": {
+    "arc_nl": 0
+  },
+  "config": {
+    "model": "hf-auto",
+    "model_args": "pretrained=BramVanroy/llama2-13b-ft-mc4_nl_cleaned_tiny,use_accelerate=True,device_map_option=auto,dtype=bfloat16,load_in_8bit=True",
+    "batch_size": 2,
+    "device": "cuda",
+    "no_cache": false,
+    "limit": null,
+    "bootstrap_iters": 100000,
+    "description_dict": {}
+  }
+}

evals/mmlu/mmlu_nl_Orca-2-13b.json

@@ -0,0 +1,23 @@
+{
+  "results": {
+    "mmlu_nl": {
+      "acc": 0.42285801016923424,
+      "acc_stderr": 0.004303747902499352,
+      "acc_norm": 0.366623662442134,
+      "acc_norm_stderr": 0.004198067042438322
+    }
+  },
+  "versions": {
+    "mmlu_nl": 0
+  },
+  "config": {
+    "model": "hf-auto",
+    "model_args": "pretrained=microsoft/Orca-2-13b,use_accelerate=True,device_map_option=auto,dtype=bfloat16,load_in_8bit=True",
+    "batch_size": 2,
+    "device": "cuda",
+    "no_cache": false,
+    "limit": null,
+    "bootstrap_iters": 100000,
+    "description_dict": {}
+  }
+}

evals/truthfulqa/truthfulqa_nl_Llama-2-13b-chat-dutch.json

@@ -0,0 +1,23 @@
+{
+  "results": {
+    "truthfulqa_nl": {
+      "mc1": 0.31337579617834393,
+      "mc1_stderr": 0.016566632870878895,
+      "mc2": 0.43607766122269953,
+      "mc2_stderr": 0.015030932806527442
+    }
+  },
+  "versions": {
+    "truthfulqa_nl": 1
+  },
+  "config": {
+    "model": "hf-auto",
+    "model_args": "pretrained=BramVanroy/Llama-2-13b-chat-dutch,use_accelerate=True,device_map_option=auto,dtype=bfloat16,load_in_8bit=True",
+    "batch_size": 2,
+    "device": "cuda",
+    "no_cache": false,
+    "limit": null,
+    "bootstrap_iters": 100000,
+    "description_dict": {}
+  }
+}

evals/truthfulqa/truthfulqa_nl_Llama-2-13b-chat-hf.json

@@ -0,0 +1,23 @@
+{
+  "results": {
+    "truthfulqa_nl": {
+      "mc1": 0.2929936305732484,
+      "mc1_stderr": 0.0162548408416286,
+      "mc2": 0.42952607696898126,
+      "mc2_stderr": 0.016098036045447243
+    }
+  },
+  "versions": {
+    "truthfulqa_nl": 1
+  },
+  "config": {
+    "model": "hf-auto",
+    "model_args": "pretrained=meta-llama/Llama-2-13b-chat-hf,use_accelerate=True,device_map_option=auto,dtype=bfloat16,load_in_8bit=True",
+    "batch_size": 2,
+    "device": "cuda",
+    "no_cache": false,
+    "limit": null,
+    "bootstrap_iters": 100000,
+    "description_dict": {}
+  }
+}

evals/truthfulqa/truthfulqa_nl_Llama-2-13b-hf.json

@@ -0,0 +1,23 @@
+{
+  "results": {
+    "truthfulqa_nl": {
+      "mc1": 0.28662420382165604,
+      "mc1_stderr": 0.016149444357869684,
+      "mc2": 0.4106464775956211,
+      "mc2_stderr": 0.014882253230962721
+    }
+  },
+  "versions": {
+    "truthfulqa_nl": 1
+  },
+  "config": {
+    "model": "hf-auto",
+    "model_args": "pretrained=meta-llama/Llama-2-13b-hf,use_accelerate=True,device_map_option=auto,dtype=bfloat16,load_in_8bit=True",
+    "batch_size": 2,
+    "device": "cuda",
+    "no_cache": false,
+    "limit": null,
+    "bootstrap_iters": 100000,
+    "description_dict": {}
+  }
+}

evals/truthfulqa/truthfulqa_nl_Orca-2-13b.json

@@ -0,0 +1,23 @@
+{
+  "results": {
+    "truthfulqa_nl": {
+      "mc1": 0.34394904458598724,
+      "mc1_stderr": 0.016965154087337654,
+      "mc2": 0.4961166743388786,
+      "mc2_stderr": 0.016446183810227374
+    }
+  },
+  "versions": {
+    "truthfulqa_nl": 1
+  },
+  "config": {
+    "model": "hf-auto",
+    "model_args": "pretrained=microsoft/Orca-2-13b,use_accelerate=True,device_map_option=auto,dtype=bfloat16,load_in_8bit=True",
+    "batch_size": 2,
+    "device": "cuda",
+    "no_cache": false,
+    "limit": null,
+    "bootstrap_iters": 100000,
+    "description_dict": {}
+  }
+}

evals/truthfulqa/truthfulqa_nl_llama2-13b-ft-mc4_nl_cleaned_tiny.json

@@ -0,0 +1,23 @@
+{
+  "results": {
+    "truthfulqa_nl": {
+      "mc1": 0.2713375796178344,
+      "mc1_stderr": 0.01588035212305625,
+      "mc2": 0.4181285992539256,
+      "mc2_stderr": 0.01475460676399081
+    }
+  },
+  "versions": {
+    "truthfulqa_nl": 1
+  },
+  "config": {
+    "model": "hf-auto",
+    "model_args": "pretrained=BramVanroy/llama2-13b-ft-mc4_nl_cleaned_tiny,use_accelerate=True,device_map_option=auto,dtype=bfloat16,load_in_8bit=True",
+    "batch_size": 2,
+    "device": "cuda",
+    "no_cache": false,
+    "limit": null,
+    "bootstrap_iters": 100000,
+    "description_dict": {}
+  }
+}