LLama tokenizer settings

aiserver.py

@@ -3218,6 +3218,11 @@ def load_model(use_gpu=True, gpu_layers=None, disk_layers=None, initial_load=Fal
             #for key in koboldai_vars.badwords:
             #    koboldai_vars.badwordsids.append([vocab[key]])
 
+            # These are model specific overrides if a model has bad defaults
+            if koboldai_vars.model_type == "llama":
+                tokenizer.decode_with_prefix_space = True
+                tokenizer.add_bos_token = False
+                        
             logger.info(f"Pipeline created: {koboldai_vars.model}")
         
         else:

install_git_transformers.bat

@@ -0,0 +1,41 @@
+@echo off
+cd /D %~dp0
+SET CONDA_SHLVL=
+
+TITLE KoboldAI - Git Transformers Installer
+ECHO This script will replace the Transformers version with the latest Git Transformers which may contain breaking changes.
+ECHO If you wish to return to the approved version of transformers you can run the install_requirements.bat script or KoboldAI Updater.
+pause
+
+SET /P M=<loader.settings
+IF %M%==1 GOTO drivemap
+IF %M%==2 GOTO subfolder
+IF %M%==3 GOTO drivemap_B
+
+:subfolder
+ECHO Runtime launching in subfolder mode
+SET TEMP=%~DP0MINICONDA3
+SET TMP=%~DP0MINICONDA3
+call miniconda3\condabin\activate
+pip install git+https://github.com/huggingface/transformers
+cmd /k
+
+:drivemap
+ECHO Runtime launching in K: drive mode
+subst /D K: >nul
+subst K: miniconda3 >nul
+SET TEMP=K:\
+SET TMP=K:\
+call K:\python\condabin\activate
+pip install git+https://github.com/huggingface/transformers
+cmd /k
+
+:drivemap_B
+ECHO Runtime launching in B: drive mode
+subst /D B: >nul
+subst B: miniconda3 >nul
+SET TEMP=B:\
+SET TMP=B:\
+call B:\python\condabin\activate
+pip install git+https://github.com/huggingface/transformers
+cmd /k