feat: 接口对齐 + FP16 量化（第一版优化方案）

- CTRUserDataset → CTRTestSeqDataset，构造参数对齐评测接口
- load_model 签名修正：ckpt_path 作为第一参数
- FP16 量化：model.half() + Embedding 保留 FP32
- move_batch_to_device 自动 FP32→FP16 转换
- 缓存时预转 FP16，减少推理循环开销
- requirements.txt 精简（去除 nvidia-* 包）
- build_env.sh 标准化（set -e + pip install）
- CLAUDE.md 更新开发命令、代码架构、关键接口说明

代码/code/infer.py

@@ -118,15 +118,17 @@ def load_logids_from_file(file_path):
     return logids
 
 
-class CTRUserDataset(Dataset):
-    """按用户组织的 CTR 数据集"""
+class CTRTestSeqDataset(Dataset):
+    """按用户组织的 CTR 测试数据集（对齐评测接口）"""
 
-    def __init__(self, item_dict, user_seq=None, max_feasign_per_slot=None, pred_logids=None):
+    def __init__(self, test_logids_ordered, item_dict, user_seq=None,
+                 max_feasign_per_slot=None, max_ctx_len=None):
         super().__init__()
         self.item_dict = item_dict
         self.user_seq = user_seq if user_seq else {}
         self.max_feasign_per_slot = max_feasign_per_slot
-        self.pred_logids = pred_logids if pred_logids is not None else set()
+        self.max_ctx_len = max_ctx_len
+        self.pred_logids = set(test_logids_ordered) if test_logids_ordered else set()
 
         self.user_items = defaultdict(list)
         for logid, rec in item_dict.items():
@@ -236,7 +238,11 @@ def move_batch_to_device(batch, device):
     elif isinstance(batch, (list, tuple)):
         return [move_batch_to_device(x, device) for x in batch]
     elif torch.is_tensor(batch):
-        return batch.to(device)
+        x = batch.to(device)
+        # 浮点 tensor → FP16，整数 tensor 保持不变
+        if x.dtype == torch.float32:
+            x = x.half()
+        return x
     else:
         return batch
 
@@ -443,12 +449,12 @@ class CTRModel(nn.Module):
 # 模型加载入口
 # ============================================================
 
-def load_model(device='cuda:0', ckpt_path=None):
+def load_model(ckpt_path, device='cuda:0'):
     """加载模型并返回，供 evaluation.py 调用。
 
     Args:
+        ckpt_path: checkpoint 文件路径（评测系统传入 Path 对象）
         device: 推理设备（默认 'cuda:0'）
-        ckpt_path: checkpoint 文件路径，默认使用 infer.py 同目录下的 ckpt.pt
 
     Returns:
         (model, device) 元组
@@ -490,6 +496,11 @@ def load_model(device='cuda:0', ckpt_path=None):
         ckpt = torch.load(ckpt_path, map_location='cpu', weights_only=False)
         model.load_state_dict(ckpt['model_state_dict'])
         print(f"[INFO] Loaded checkpoint from {ckpt_path} (epoch={ckpt.get('epoch', '?')})")
+
+        # === FP16 量化：模型参数转半精度，Embedding 保留 FP32 ===
+        model = model.half()
+        model.rep_encoder.emb = model.rep_encoder.emb.to(torch.float32)
+        print("[INFO] Model converted to FP16 (embedding kept in FP32)")
     else:
         print(f"[WARNING] Checkpoint {ckpt_path} not found, using random weights")
 
@@ -616,10 +627,11 @@ def main():
         print(f'[INFO] Test pred logids count: {len(test_pred_logids)}')
 
         max_feasign_per_slot = {1: 2}
-        test_dataset = CTRUserDataset(
-            item_dict, user_seq,
+        test_dataset = CTRTestSeqDataset(
+            test_logids_ordered=list(test_pred_logids),
+            item_dict=item_dict,
+            user_seq=user_seq,
             max_feasign_per_slot=max_feasign_per_slot,
-            pred_logids=test_pred_logids,
         )
         print(f'[INFO] num_users={test_dataset.num_users}, '
               f'total_samples={test_dataset.total_samples}, '
@@ -634,9 +646,12 @@ def main():
             collate_fn=make_collate_fn(test_dataset.max_slot_id),
         )
 
-        # 收集 batches 并按分片缓存
-        print('[INFO] collecting batches and saving sharded cache...')
-        all_batches = [batch for batch in test_loader]
+        # 收集 batches，预转 FP16 后按分片缓存
+        print('[INFO] collecting batches (pre-converting to FP16) and saving sharded cache...')
+        all_batches = []
+        for batch in test_loader:
+            batch = move_batch_to_device(batch, torch.device('cpu'))
+            all_batches.append(batch)
 
         batches_cache_dir.mkdir(parents=True, exist_ok=True)
         shard_idx = 0