correct file

epfl-dlab · Jul 5, 2024 · aaec04b · aaec04b
1 parent 4b62393
commit aaec04b
Show file tree

Hide file tree

Showing 29 changed files with 288 additions and 60,495 deletions.
diff --git a/outputs/result/mistral_v1_outputs_test.json b/outputs/result/mistral_v1_outputs_test.json
diff --git a/outputs/result/mistral_v1_outputs_test_pause_token.json b/outputs/result/mistral_v1_outputs_test_pause_token.json
diff --git a/outputs/result/mistral_v1_outputs_test_pause_token_random.json b/outputs/result/mistral_v1_outputs_test_pause_token_random.json
diff --git a/outputs/result/mistral_v1_outputs_train.json b/outputs/result/mistral_v1_outputs_train.json
diff --git a/outputs/result/output_mistral_instruct_v2_pause_token_sft_gsm8k.json b/outputs/result/output_mistral_instruct_v2_pause_token_sft_gsm8k.json
diff --git a/outputs/result/output_mistral_instruct_v2_sft_gsm8k.json b/outputs/result/output_mistral_instruct_v2_sft_gsm8k.json
diff --git a/outputs/result/output_mistral_instruct_v2_sft_gsm8k_updated_embedding_layer.json b/outputs/result/output_mistral_instruct_v2_sft_gsm8k_updated_embedding_layer.json
diff --git a/outputs/result/rc_mistral_correctness_reward_outer_loop_1.json b/outputs/result/rc_mistral_correctness_reward_outer_loop_1.json
diff --git a/outputs/result/rc_mistral_correctness_reward_outer_loop_2.json b/outputs/result/rc_mistral_correctness_reward_outer_loop_2.json
diff --git a/outputs/result/rc_mistral_delta_reward_outer_loop_1.json b/outputs/result/rc_mistral_delta_reward_outer_loop_1.json
diff --git a/outputs/result/rc_mistral_delta_reward_outer_loop_2.json b/outputs/result/rc_mistral_delta_reward_outer_loop_2.json
diff --git a/outputs/result/rc_mistral_outer_loop_0.json b/outputs/result/rc_mistral_outer_loop_0.json
diff --git a/pip_requirements.txt b/pip_requirements.txt
@@ -4,8 +4,8 @@ datasets==2.18.0
 transformers==4.39.3
 trl==0.8.1
 peft==0.10.0
-sentencepiece==0.2.0
 scipy==1.13.0
+sentencepiece==0.2.0
 protobuf==5.26.1
 matplotlib==3.8.4
 

diff --git a/src/tariners/Readme.md b/src/tariners/Readme.md
diff --git a/src/tariners/behavior_cloning.py b/src/tariners/behavior_cloning.py
diff --git a/src/tariners/collators.py b/src/tariners/collators.py
@@ -36,6 +36,8 @@ def torch_call(self, examples: List[Union[List[int], Any, Dict[str, Any]]]) -> D
         train_method = examples[0][self.name_col]
         collator = self.name_to_collator[train_method]
         data = [example["data"] for example in examples]
+        #remove any field of data that is None
+        data = [{key: value for key, value in example.items() if value is not None} for example in data]
         tensor = collator(data)
         tensor[self.name_col] = torch.tensor([example[self.name_col] for example in examples]).long()
         return tensor
diff --git a/src/tariners/constraints.py b/src/tariners/constraints.py
@@ -39,7 +39,6 @@ def process_logits(self, input_ids, scores):
         log_probs = self.softmax_temperature * torch.log(probs)
         #check if `inf`, `nan` are in probs
         if torch.isnan(probs).any() or torch.isinf(probs).any():
-            breakpoint()
             raise ValueError("LogitsProcessor: the model generated `inf` or `nan` values")
         return log_probs
 

diff --git a/src/tariners/eval_script.py b/src/tariners/eval_script.py