Revert PR #1473 (#1582)

examples/text-generation/requirements_lm_eval.txt

@@ -1 +1,2 @@
-https://github.com/EleutherAI/lm-evaluation-harness/archive/c1d8795da7610d507cb191c2769c5e7bf1060a35.zip
+https://github.com/EleutherAI/lm-evaluation-harness/archive/0bf683b4e6a9df359b3156ba9ba8d62bdd47e0c0.zip
+datasets==2.21.0

examples/text-generation/run_lm_eval.py

@@ -29,7 +29,6 @@
 import psutil
 import torch
 import torch.nn.functional as F
-from lm_eval.models.huggingface import HFLM
 
 # Local imports
 from run_generation import setup_parser
@@ -92,15 +91,17 @@ def setup_lm_eval_parser():
     return args
 
 
-class HabanaModelAdapter(HFLM):
+class HabanaModelAdapter(lm_eval.base.BaseLM):
     def __init__(self, tokenizer, model, args, options):
-        super().__init__(pretrained=model, tokenizer=tokenizer, batch_size=args.batch_size)
+        super().__init__()
         self.tokenizer = tokenizer
+        self.model = model
+        self._batch_size = args.batch_size
         self.buckets = sorted(args.buckets)
         self.options = options
         self._device = args.device
         self.model_inputs = {"use_cache": self.options.use_cache}
-        if self._model.config.model_type in [
+        if self.model.config.model_type in [
             "llama",
             "mistral",
             "falcon",
@@ -136,7 +137,7 @@ def __init__(self, tokenizer, model, args, options):
 
     def warm_up(self):
         for bucket_size in reversed(self.buckets):
-            inps = torch.ones((self.batch_size, bucket_size), dtype=torch.int64)
+            inps = torch.ones((self._batch_size, bucket_size), dtype=torch.int64)
             self._model_call(inps)
             pass
 
@@ -148,6 +149,14 @@ def eot_token_id(self):
     def max_length(self):
         return self.buckets[-1]
 
+    @property
+    def max_gen_toks(self):
+        raise NotImplementedError()
+
+    @property
+    def batch_size(self):
+        return self._batch_size
+
     @property
     def device(self):
         # We need to do padding ourselves, otherwise we'll end up with recompilations