improve the eval perf with kv cache

Summary:
The original implementation was too slow, and because of the frequent travel: cpu->gpu->cpu->gpu-> , it's inefficient. Change it to batch process the sequence so the compute remains in gpu

When evaluate stories model, before the change:
```
2024-05-23:23:42:25,115 INFO     [evaluator.py:362] Running loglikelihood_rolling requests
100%|██████████| 5/5 [02:37<00:00, 31.50s/it]
wikitext: {'word_perplexity,none': 10589.525426446424, 'word_perplexity_stderr,none': 'N/A', 'byte_perplexity,none': 6.111053701258041, 'byte_perplexity_stderr,none': 'N/A', 'bits_per_byte,none': 2.6114211588515417, 'bits_per_byte_stderr,none': 'N/A', 'alias': 'wikitext'}
```

After the change:
```
2024-05-23:23:36:50,339 INFO     [evaluator.py:362] Running loglikelihood_rolling requests
100%|██████████| 5/5 [00:03<00:00,  1.55it/s]
wikitext: {'word_perplexity,none': 10589.52618994558, 'word_perplexity_stderr,none': 'N/A', 'byte_perplexity,none': 6.111053787314264, 'byte_perplexity_stderr,none': 'N/A', 'bits_per_byte,none': 2.611421179167659, 'bits_per_byte_stderr,none': 'N/A', 'alias': 'wikitext'}
```

Differential Revision: D57764318

Author: cccclai

examples/models/llama2/eval_llama_lib.py

@@ -88,12 +88,13 @@ def tok_decode(self, tokens):
 
     def _model_call(self, inps):
         if self._use_kv_cache:
-            result_logits = []
-            for pos in range(self._max_seq_length):
-                pos_tensor = torch.tensor([pos], dtype=torch.int64)
-                logits = self._model(inps[:, pos : pos + 1], pos_tensor)
-                result_logits.append(logits)
-            return torch.cat(result_logits, dim=1)
+            pos_tensor = torch.arange(
+                self._max_seq_length, dtype=torch.int64, device=self.device
+            )
+
+            # Batch process the whole sequence.
+            logits = self._model(inps[:, : self._max_seq_length], pos_tensor)
+            return logits
         else:
             return self._model(inps)