Merge pull request #15 from probcomp/alexlew-hf-updates

Use BitsAndBytesConfig instead of load_in_8bit and token instead of u…

Author: alex-lew

hfppl/llms.py

@@ -1,7 +1,7 @@
 """Utilities for working with HuggingFace language models, including caching and auto-batching."""
 
 import torch
-from transformers import AutoTokenizer, AutoModelForCausalLM
+from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig
 import asyncio
 import string
 
@@ -266,18 +266,22 @@ def from_pretrained(cls, model_id, auth_token=False, load_in_8bit=True):
         Returns:
             model (hfppl.llms.CachedCausalLM): the LLaMPPL-compatible interface to the HuggingFace model.
         """
+        bnb_config = BitsAndBytesConfig(load_in_8bit=load_in_8bit)
+
         if not auth_token:
             tok = AutoTokenizer.from_pretrained(model_id)
             mod = AutoModelForCausalLM.from_pretrained(
-                model_id, device_map="auto", load_in_8bit=load_in_8bit
+                model_id,
+                device_map="auto",
+                quantization_config=bnb_config,
             )
         else:
-            tok = AutoTokenizer.from_pretrained(model_id, use_auth_token=auth_token)
+            tok = AutoTokenizer.from_pretrained(model_id, token=auth_token)
             mod = AutoModelForCausalLM.from_pretrained(
                 model_id,
-                use_auth_token=auth_token,
+                token=auth_token,
                 device_map="auto",
-                load_in_8bit=load_in_8bit,
+                quantization_config=bnb_config,
             )
 
         return CachedCausalLM(mod, tok)