update

data0805/compute_ce_fsdp_recom.py

@@ -112,21 +112,21 @@ class NetflixRatingDataset(Dataset):
 @torch.inference_mode()
 def run(underlying_model, lm_head, dataset, output_path):
     dataloader = DataLoader(dataset=dataset, batch_size=1, shuffle=False, drop_last=False, num_workers=0)
-    print(dataloader.dataset[0])
-    print("----hi---- start run")
+    # print(dataloader.dataset[0])
+    # print("----hi---- start run")
     results = []
     for idx, batch in enumerate(tqdm(dataloader, desc="Inference")):
-        print(f"----hi---- {idx}  id")
+        # print(f"----hi---- {idx}  id")
         if "Error" in batch:
             # Skip invalid items
-            print("----hi---- skip invalid items")
+            # print("----hi---- skip invalid items")
             continue
         # batch["input_ids"] = batch["input_ids"].to(model.device).to(torch.float16)
-        print("----hi-----id "+ str(idx))
+        # print("----hi-----id "+ str(idx))
         hidden_states = underlying_model(
             input_ids=batch["input_ids"].to(underlying_model.device),
         ).last_hidden_state
-        print("----hi----- calculate hiddenstate")
+        # print("----hi----- calculate hiddenstate")
 
         for cur_input_ids, cur_hidden_state, cur_prefix_length in zip(batch['input_ids'], hidden_states, batch["prefix_length"]):
             cur_input_ids = cur_input_ids.to(underlying_model.device)
@@ -159,9 +159,9 @@ def run(underlying_model, lm_head, dataset, output_path):
                 "losses": losses.cpu().float().numpy().tolist(),
                 "history_round_len": dataset.history_round_len,
             })
-            print("!!!!!!!!!!!!!!!!!!!-------------------------")
-            print(results[0])
-            break
+            # print("!!!!!!!!!!!!!!!!!!!-------------------------")
+            # print(results[0])
+            # break
     # Save results to JSON file
     os.makedirs(os.path.dirname(output_path), exist_ok=True)
     with open(output_path, 'w') as f:
@@ -184,7 +184,7 @@ def main():
         args.language = None
     # setup_distributed(rank, world_size)
         
-    model_path = f"/data-share/lijiaqi/model/{args.model_path}"
+    model_path = args.model_path
     tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
 
     with init_empty_weights():
@@ -250,13 +250,13 @@ def main():
     language = args.language
     round_threshold = args.round_threshold
     if args.data_name == "netflix":
-        dataset = NetflixRatingDataset("/data-share/lijiaqi/personlaw/data/netflix/ratings2000.csv","/data-share/lijiaqi/personlaw/data/netflix/movie_titles.csv",
+        dataset = NetflixRatingDataset("ratings2000.csv","movie_titles.csv",
                                         round_threshold=round_threshold, 
                                         tokenizer=tokenizer,
                                         max_seq_len=args.max_seq_len,
                                         max_item_per_user=args.max_item_per_user,
                                         max_history_len=args.max_history_len,)
-        output_filefolder = "/data-share/lijiaqi/personlaw/data/netflix/2000"
+        output_filefolder = "netflix/2000"
     else:
         raise NotImplementedError
     # for i in tqdm(range(round_threshold-1, 0, -1), desc="Main"):

data0805/compute_ce_fsdp_recom_test.py

@@ -72,8 +72,8 @@ class NetflixRatingDataset(Dataset):
             # exit()
             if len(history) < history_round_len:
                 continue
-            if len(history) > self.max_history_len:
-                history = history.tail(self.max_history_len)
+            if len(history) > self.history_round_len:
+                history = history.tail(self.history_round_len)
                 # 构建 Prompt
                 prefix = "历史评分记录：\n"
                 for _, row in history.iterrows():
@@ -112,21 +112,21 @@ class NetflixRatingDataset(Dataset):
 @torch.inference_mode()
 def run(underlying_model, lm_head, dataset, output_path):
     dataloader = DataLoader(dataset=dataset, batch_size=1, shuffle=False, drop_last=False, num_workers=0)
-    print(dataloader.dataset[0])
-    print("----hi---- start run")
+    # print(dataloader.dataset[0])
+    # print("----hi---- start run")
     results = []
     for idx, batch in enumerate(tqdm(dataloader, desc="Inference")):
-        print(f"----hi---- {idx}  id")
+        # print(f"----hi---- {idx}  id")
         if "Error" in batch:
             # Skip invalid items
-            print("----hi---- skip invalid items")
+            # print("----hi---- skip invalid items")
             continue
         # batch["input_ids"] = batch["input_ids"].to(model.device).to(torch.float16)
-        print("----hi-----id "+ str(idx))
+        # print("----hi-----id "+ str(idx))
         hidden_states = underlying_model(
             input_ids=batch["input_ids"].to(underlying_model.device),
         ).last_hidden_state
-        print("----hi----- calculate hiddenstate")
+        # print("----hi----- calculate hiddenstate")
 
         for cur_input_ids, cur_hidden_state, cur_prefix_length in zip(batch['input_ids'], hidden_states, batch["prefix_length"]):
             cur_input_ids = cur_input_ids.to(underlying_model.device)
@@ -159,9 +159,9 @@ def run(underlying_model, lm_head, dataset, output_path):
                 "losses": losses.cpu().float().numpy().tolist(),
                 "history_round_len": dataset.history_round_len,
             })
-            print("!!!!!!!!!!!!!!!!!!!-------------------------")
-            print(results[0])
-            break
+            # print("!!!!!!!!!!!!!!!!!!!-------------------------")
+            # print(results[0])
+            # break
     # Save results to JSON file
     os.makedirs(os.path.dirname(output_path), exist_ok=True)
     with open(output_path, 'w') as f:
@@ -184,7 +184,7 @@ def main():
         args.language = None
     # setup_distributed(rank, world_size)
         
-    model_path = f"/data-share/lijiaqi/model/{args.model_path}"
+    model_path = args.model_path
     tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
 
     with init_empty_weights():
@@ -250,13 +250,13 @@ def main():
     language = args.language
     round_threshold = args.round_threshold
     if args.data_name == "netflix":
-        dataset = NetflixRatingDataset("/data-share/lijiaqi/personlaw/data/netflix/ratings100.csv","/data-share/lijiaqi/personlaw/data/netflix/movie_titles.csv",
+        dataset = NetflixRatingDataset("ratings100.csv","movie_titles.csv",
                                         round_threshold=round_threshold, 
                                         tokenizer=tokenizer,
                                         max_seq_len=args.max_seq_len,
                                         max_item_per_user=args.max_item_per_user,
                                         max_history_len=args.max_history_len,)
-        output_filefolder = "/data-share/lijiaqi/personlaw/data/netflix/100"
+        output_filefolder = "data/netflix/100"
     else:
         raise NotImplementedError
     # for i in tqdm(range(round_threshold-1, 0, -1), desc="Main"):