From 826e297b0ec40299318f1002f9165e7ac9c9c257 Mon Sep 17 00:00:00 2001
From: rohvani <3782201+rohvani@users.noreply.github.com>
Date: Thu, 9 Mar 2023 18:31:32 -0800
Subject: [PATCH 1/4] add llama-65b-4bit support & multiple pt paths

---
 modules/models.py | 24 ++++++++++++++++--------
 1 file changed, 16 insertions(+), 8 deletions(-)

diff --git a/modules/models.py b/modules/models.py
index 3e6cea18..062ccb1f 100644
--- a/modules/models.py
+++ b/modules/models.py
@@ -97,19 +97,27 @@ def load_model(model_name):
         pt_model = ''
         if path_to_model.name.lower().startswith('llama-7b'):
             pt_model = 'llama-7b-4bit.pt'
-        if path_to_model.name.lower().startswith('llama-13b'):
+        elif path_to_model.name.lower().startswith('llama-13b'):
             pt_model = 'llama-13b-4bit.pt'
-        if path_to_model.name.lower().startswith('llama-30b'):
+        elif path_to_model.name.lower().startswith('llama-30b'):
             pt_model = 'llama-30b-4bit.pt'
-
-        if not Path(f"models/{pt_model}").exists():
-            print(f"Could not find models/{pt_model}, exiting...")
-            exit()
-        elif pt_model == '':
+        elif path_to_model.name.lower().startswith('llama-65b'):
+            pt_model = 'llama-65b-4bit.pt'
+        else:
             print(f"Could not find the .pt model for {model_name}, exiting...")
             exit()
 
-        model = load_quant(path_to_model, Path(f"models/{pt_model}"), 4)
+        # check root of models folder, and model path root
+        paths = [  f"{path_to_model}/{pt_model}", f"models/{pt_model}" ]
+        for path in [ Path(p) for p in paths ]:
+            if path.exists():
+                pt_path = path
+
+        if not pt_path:
+            print(f"Could not find {pt_model}, exiting...")
+            exit()
+
+        model = load_quant(path_to_model, pt_path, 4)
         model = model.to(torch.device('cuda:0'))
 
     # Custom

From 5ee376c580e8c2cf2e3b34e1822c43e6754b2649 Mon Sep 17 00:00:00 2001
From: rohvani <3782201+rohvani@users.noreply.github.com>
Date: Thu, 9 Mar 2023 18:31:41 -0800
Subject: [PATCH 2/4] add LLaMA preset

---
 presets/LLaMA-Default.txt | 12 ++++++++++++
 1 file changed, 12 insertions(+)
 create mode 100644 presets/LLaMA-Default.txt

diff --git a/presets/LLaMA-Default.txt b/presets/LLaMA-Default.txt
new file mode 100644
index 00000000..3df8209a
--- /dev/null
+++ b/presets/LLaMA-Default.txt
@@ -0,0 +1,12 @@
+do_sample=False
+temperature=0.7
+top_p=0
+typical_p=1
+repetition_penalty=1.15
+top_k=40
+num_beams=1
+penalty_alpha=0
+min_length=0
+length_penalty=1
+no_repeat_ngram_size=0
+early_stopping=True

From 2ac29137470396733e95e7efa77e091d5e8a5ef5 Mon Sep 17 00:00:00 2001
From: rohvani <3782201+rohvani@users.noreply.github.com>
Date: Thu, 9 Mar 2023 20:13:23 -0800
Subject: [PATCH 3/4] fix reference issue

---
 modules/models.py | 1 +
 1 file changed, 1 insertion(+)

diff --git a/modules/models.py b/modules/models.py
index 062ccb1f..a2256b98 100644
--- a/modules/models.py
+++ b/modules/models.py
@@ -109,6 +109,7 @@ def load_model(model_name):
 
         # check root of models folder, and model path root
         paths = [  f"{path_to_model}/{pt_model}", f"models/{pt_model}" ]
+        pt_path = None
         for path in [ Path(p) for p in paths ]:
             if path.exists():
                 pt_path = path

From 706a03b2cb5bf3c0667d8c13b3a47f1a6e33cc81 Mon Sep 17 00:00:00 2001
From: oobabooga <112222186+oobabooga@users.noreply.github.com>
Date: Fri, 10 Mar 2023 11:02:25 -0300
Subject: [PATCH 4/4] Minor changes

---
 modules/models.py         |  8 +++-----
 presets/LLaMA-Default.txt | 12 ------------
 2 files changed, 3 insertions(+), 17 deletions(-)
 delete mode 100644 presets/LLaMA-Default.txt

diff --git a/modules/models.py b/modules/models.py
index a2256b98..a23f1fa9 100644
--- a/modules/models.py
+++ b/modules/models.py
@@ -104,13 +104,11 @@ def load_model(model_name):
         elif path_to_model.name.lower().startswith('llama-65b'):
             pt_model = 'llama-65b-4bit.pt'
         else:
-            print(f"Could not find the .pt model for {model_name}, exiting...")
-            exit()
+            pt_model = f'{model_name}-4bit.pt'
 
-        # check root of models folder, and model path root
-        paths = [  f"{path_to_model}/{pt_model}", f"models/{pt_model}" ]
+        # Try to find the .pt both in models/ and in the subfolder
         pt_path = None
-        for path in [ Path(p) for p in paths ]:
+        for path in [Path(p) for p in [f"models/{pt_model}", f"{path_to_model}/{pt_model}"]]:
             if path.exists():
                 pt_path = path
 
diff --git a/presets/LLaMA-Default.txt b/presets/LLaMA-Default.txt
deleted file mode 100644
index 3df8209a..00000000
--- a/presets/LLaMA-Default.txt
+++ /dev/null
@@ -1,12 +0,0 @@
-do_sample=False
-temperature=0.7
-top_p=0
-typical_p=1
-repetition_penalty=1.15
-top_k=40
-num_beams=1
-penalty_alpha=0
-min_length=0
-length_penalty=1
-no_repeat_ngram_size=0
-early_stopping=True