Use pre-compiled python module for ExLlama (#2770)

modules/exllama_hf.py

@@ -9,10 +9,19 @@ from transformers.modeling_outputs import CausalLMOutputWithPast
 
 from modules import shared
 from modules.logging_colors import logger
-from modules.relative_imports import RelativeImport
 
-with RelativeImport("repositories/exllama"):
-    from model import ExLlama, ExLlamaCache, ExLlamaConfig
+try:
+    from exllama.model import ExLlama, ExLlamaCache, ExLlamaConfig
+except:
+    logger.warning('Exllama module failed to load. Will attempt to load from repositories.')
+    try:
+        from modules.relative_imports import RelativeImport
+
+        with RelativeImport("repositories/exllama"):
+            from model import ExLlama, ExLlamaCache, ExLlamaConfig
+    except:
+        logger.error("Could not find repositories/exllama/. Make sure that exllama is cloned inside repositories/ and is up to date.")
+        raise
 
 
 class ExllamaHF(PreTrainedModel):
@@ -68,7 +77,7 @@ class ExllamaHF(PreTrainedModel):
         assert len(model_args) == 0 and len(kwargs) == 0, "extra args is currently not supported"
         if isinstance(pretrained_model_name_or_path, str):
             pretrained_model_name_or_path = Path(pretrained_model_name_or_path)
-            
+
         pretrained_model_name_or_path = Path(f'{shared.args.model_dir}') / Path(pretrained_model_name_or_path)
         config = ExLlamaConfig(pretrained_model_name_or_path / 'config.json')
 
@@ -86,7 +95,7 @@ class ExllamaHF(PreTrainedModel):
         if shared.args.gpu_split:
             config.set_auto_map(shared.args.gpu_split)
             config.gpu_peer_fix = True
-        
+
         # This slowes down a bit but align better with autogptq generation.
         # TODO: Should give user choice to tune the exllama config
         # config.fused_attn = False