Add model configs and bump transformers version

scripts/configs/eval_configs.yaml

@@ -852,4 +852,43 @@ Ray2333/GRM-llama3.2-3B-rewardmodel-ft:
   batch_size: 16
   trust_remote_code: False
   dpo: False
-  quantized: False
+  quantized: False
+infly/INF-ORM-Llama3.1-70B:
+  model: infly/INF-ORM-Llama3.1-70B
+  tokenizer: infly/INF-ORM-Llama3.1-70B
+  chat_template: # none for tokenizer
+  batch_size: 16
+  torch_dtype: bfloat16
+  attention_implementation: flash_attention_2
+  trust_remote_code: False
+  dpo: False
+  quantized: False
+SultanR/SmolTulu-1.7b-RM:
+  model: SultanR/SmolTulu-1.7b-RM
+  tokenizer: SultanR/SmolTulu-1.7b-RM
+  chat_template: # none for tokenizer
+  batch_size: 16
+  trust_remote_code: False
+  dpo: False
+  quantized: False
+nicolinho/QRM-Gemma-2-27B:
+  model: nicolinho/QRM-Gemma-2-27B
+  tokenizer: nicolinho/QRM-Gemma-2-27B
+  chat_template: # none for tokenizer
+  batch_size: 1
+  torch_dtype: bfloat16
+  attention_implementation: flash_attention_2
+  max_length: 4096
+  quantized: False
+  dpo: False
+  trust_remote_code: True
+nicolinho/QRM-Llama3.1-8B-v2:
+  model: nicolinho/QRM-Llama3.1-8B-v2
+  tokenizer: nicolinho/QRM-Llama3.1-8B-v2
+  chat_template: # none for tokenizer
+  batch_size: 1
+  torch_dtype: bfloat16
+  attention_implementation: flash_attention_2
+  quantized: False
+  dpo: False
+  trust_remote_code: True

setup.py

@@ -60,7 +60,7 @@
         "tokenizers",
         "torch",
         "tiktoken==0.6.0",  # added for llama 3
-        "transformers==4.43.4",  # pinned at llama 3
+        "transformers==4.47.1",  # pinned at most recent version
         "trl>=0.8.2",  # fixed transformers import error, for DPO
         "wandb",  # for loading model path / reivisions from wandb
     ],