Adding CPU (float32) support

scripts/create_request_file.py

@@ -12,7 +12,7 @@ from src.envs import TOKEN
 EVAL_REQUESTS_PATH = "eval-queue"
 QUEUE_REPO = "open-llm-leaderboard/requests"
 
-precisions = ("float16", "bfloat16", "8bit (LLM.int8)", "4bit (QLoRA / FP4)", "GPTQ")
+precisions = ("float16", "bfloat16", "8bit (LLM.int8)", "4bit (QLoRA / FP4)", "GPTQ", "float32")
 model_types = ("pretrained", "fine-tuned", "RL-tuned", "instruction-tuned")
 weight_types = ("Original", "Delta", "Adapter")
 

src/backend/manage_requests.py

@@ -26,7 +26,7 @@ class EvalRequest:
     def get_model_args(self):
         model_args = f"pretrained={self.model},revision={self.revision}"
 
-        if self.precision in ["float16", "bfloat16"]:
+        if self.precision in ["float16", "bfloat16", "float32"]:
             model_args += f",dtype={self.precision}"
         # Quantized models need some added config, the install of bits and bytes, etc
         #elif self.precision == "8bit":