diff --git a/.gitmodules b/.gitmodules
index c6f4b308..4a1fb7c9 100644
--- a/.gitmodules
+++ b/.gitmodules
@@ -8,3 +8,7 @@
 	path = repos/gptq
 	url = https://github.com/0cc4m/GPTQ-for-LLaMa
 	branch = a8303654c200c25577130466e5f9bc1e70fc8a50
+[submodule "repos/hf_bleeding_edge"]
+	path = repos/hf_bleeding_edge
+	url = https://github.com/0cc4m/hf_bleeding_edge
+	branch = b5d0b80c6947605b9ccf080fc17b68a516ea5857
diff --git a/aiserver.py b/aiserver.py
index 80518450..bb6cc171 100644
--- a/aiserver.py
+++ b/aiserver.py
@@ -1621,7 +1621,7 @@ def get_layer_count(model, directory=""):
         else:
             if(directory):
                 model = directory
-            from transformers import AutoConfig
+            from repos.hf_bleeding_edge import AutoConfig
             if(os.path.isdir(model.replace('/', '_'))):
                 model_config = AutoConfig.from_pretrained(model.replace('/', '_'), revision=koboldai_vars.revision, cache_dir="cache")
             elif(is_model_downloaded(model)):
diff --git a/environments/huggingface.yml b/environments/huggingface.yml
index e5fb939c..a179c468 100644
--- a/environments/huggingface.yml
+++ b/environments/huggingface.yml
@@ -49,3 +49,4 @@ dependencies:
     - diffusers
     - --find-links=https://0cc4m.github.io/KoboldAI/gptq-whl-links.html
     - quant_cuda
+    - einops
diff --git a/environments/rocm.yml b/environments/rocm.yml
index 9358575d..d0daf4f2 100644
--- a/environments/rocm.yml
+++ b/environments/rocm.yml
@@ -44,3 +44,4 @@ dependencies:
     - diffusers
     - --find-links=https://0cc4m.github.io/KoboldAI/gptq-whl-links.html
     - quant_rocm
+    - einops
diff --git a/modeling/inference_models/generic_hf_torch.py b/modeling/inference_models/generic_hf_torch.py
index 9e30a7fd..61004db5 100644
--- a/modeling/inference_models/generic_hf_torch.py
+++ b/modeling/inference_models/generic_hf_torch.py
@@ -6,7 +6,8 @@ import torch
 import shutil
 from typing import Union
 
-from transformers import AutoModelForCausalLM, GPTNeoForCausalLM, GPT2LMHeadModel
+from transformers import GPTNeoForCausalLM, GPT2LMHeadModel
+from repos.hf_bleeding_edge import AutoModelForCausalLM
 
 import utils
 import modeling.lazy_loader as lazy_loader
diff --git a/modeling/inference_models/hf.py b/modeling/inference_models/hf.py
index 99e55be4..8c797940 100644
--- a/modeling/inference_models/hf.py
+++ b/modeling/inference_models/hf.py
@@ -1,6 +1,6 @@
 import os
 from typing import Optional
-from transformers import AutoConfig
+from repos.hf_bleeding_edge import AutoConfig
 
 import utils
 import koboldai_settings
diff --git a/modeling/inference_models/hf_torch.py b/modeling/inference_models/hf_torch.py
index 3cc28291..e0081c90 100644
--- a/modeling/inference_models/hf_torch.py
+++ b/modeling/inference_models/hf_torch.py
@@ -19,9 +19,9 @@ from transformers import (
     StoppingCriteria,
     GPTNeoForCausalLM,
     GPT2LMHeadModel,
-    AutoModelForCausalLM,
     LogitsProcessorList,
 )
+from repos.hf_bleeding_edge import AutoModelForCausalLM
 
 import utils
 import modeling.lazy_loader as lazy_loader
diff --git a/modeling/inference_models/hf_torch_4bit.py b/modeling/inference_models/hf_torch_4bit.py
index 5eb8d60c..75fb9ddf 100644
--- a/modeling/inference_models/hf_torch_4bit.py
+++ b/modeling/inference_models/hf_torch_4bit.py
@@ -9,7 +9,8 @@ import shutil
 import sys
 from typing import Union
 
-from transformers import AutoModelForCausalLM, GPTNeoForCausalLM, AutoTokenizer, LlamaTokenizer
+from transformers import GPTNeoForCausalLM, AutoTokenizer, LlamaTokenizer
+from repos.hf_bleeding_edge import AutoModelForCausalLM
 
 import utils
 import modeling.lazy_loader as lazy_loader
diff --git a/repos/__init__.py b/repos/__init__.py
new file mode 100644
index 00000000..af438273
--- /dev/null
+++ b/repos/__init__.py
@@ -0,0 +1 @@
+from . import hf_bleeding_edge
diff --git a/repos/hf_bleeding_edge b/repos/hf_bleeding_edge
new file mode 160000
index 00000000..b5d0b80c
--- /dev/null
+++ b/repos/hf_bleeding_edge
@@ -0,0 +1 @@
+Subproject commit b5d0b80c6947605b9ccf080fc17b68a516ea5857
diff --git a/utils.py b/utils.py
index 13ebb6a3..89b9fb4f 100644
--- a/utils.py
+++ b/utils.py
@@ -184,7 +184,7 @@ def decodenewlines(txt):
 #  Returns number of layers given an HF model config
 #==================================================================#
 def num_layers(config):
-    return config["n_layer"] if isinstance(config, dict) else config.num_layers if hasattr(config, "num_layers") else config.n_layer if hasattr(config, "n_layer") else config.num_hidden_layers if hasattr(config, 'num_hidden_layers') else None
+    return config["n_layer"] if isinstance(config, dict) else config.num_layers if hasattr(config, "num_layers") else config.n_layer if hasattr(config, "n_layer") else config.num_hidden_layers if hasattr(config, 'num_hidden_layers') else config.n_layers if hasattr(config, "n_layers") else None
 
 #==================================================================#
 #  Downloads huggingface checkpoints using aria2c if possible