Merge pull request #224 from ItsLogic/llama-bits

Allow users to load 2, 3 and 4 bit llama models

README.md

@@ -138,7 +138,8 @@ Optionally, you can use the following command-line flags:
 | `--cai-chat`  | Launch the web UI in chat mode with a style similar to Character.AI's. If the file `img_bot.png` or `img_bot.jpg` exists in the same folder as server.py, this image will be used as the bot's profile picture. Similarly, `img_me.png` or `img_me.jpg` will be used as your profile picture. |
 | `--cpu`       | Use the CPU to generate text.|
 | `--load-in-8bit`  | Load the model with 8-bit precision.|
-| `--load-in-4bit`  |  Load the model with 4-bit precision. Currently only works with LLaMA. |
+| `--load-in-4bit`  | Load the model with 4-bit precision. Currently only works with LLaMA.|
+| `--gptq-bits`  |  Load a pre-quantized model with specified precision. 2, 3, 4 and 8bit are supported. Currently only works with LLaMA. |
 | `--bf16`  | Load the model with bfloat16 precision. Requires NVIDIA Ampere GPU. |
 | `--auto-devices` | Automatically split the model across the available GPU(s) and CPU.|
 | `--disk` | If the model is too large for your GPU(s) and CPU combined, send the remaining layers to the disk. |

modules/models.py

@@ -42,7 +42,7 @@ def load_model(model_name):
     shared.is_RWKV = model_name.lower().startswith('rwkv-')
 
     # Default settings
-    if not (shared.args.cpu or shared.args.load_in_8bit or shared.args.load_in_4bit or shared.args.auto_devices or shared.args.disk or shared.args.gpu_memory is not None or shared.args.cpu_memory is not None or shared.args.deepspeed or shared.args.flexgen or shared.is_RWKV):
+    if not any([shared.args.cpu, shared.args.load_in_8bit, shared.args.load_in_4bit, shared.args.gptq_bits > 0, shared.args.auto_devices, shared.args.disk, shared.args.gpu_memory is not None, shared.args.cpu_memory is not None, shared.args.deepspeed, shared.args.flexgen, shared.is_RWKV]):
         if any(size in shared.model_name.lower() for size in ('13b', '20b', '30b')):
             model = AutoModelForCausalLM.from_pretrained(Path(f"models/{shared.model_name}"), device_map='auto', load_in_8bit=True)
         else:
@@ -88,51 +88,10 @@ def load_model(model_name):
         return model, tokenizer
 
     # 4-bit LLaMA
-    elif shared.args.load_in_4bit:
-        sys.path.insert(0, os.path.abspath(Path("repositories/GPTQ-for-LLaMa")))
-
-        from llama import load_quant
-
-        path_to_model = Path(f'models/{model_name}')
-        pt_model = ''
-        if path_to_model.name.lower().startswith('llama-7b'):
-            pt_model = 'llama-7b-4bit.pt'
-        elif path_to_model.name.lower().startswith('llama-13b'):
-            pt_model = 'llama-13b-4bit.pt'
-        elif path_to_model.name.lower().startswith('llama-30b'):
-            pt_model = 'llama-30b-4bit.pt'
-        elif path_to_model.name.lower().startswith('llama-65b'):
-            pt_model = 'llama-65b-4bit.pt'
-        else:
-            pt_model = f'{model_name}-4bit.pt'
-
-        # Try to find the .pt both in models/ and in the subfolder
-        pt_path = None
-        for path in [Path(p) for p in [f"models/{pt_model}", f"{path_to_model}/{pt_model}"]]:
-            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, Path(f"models/{pt_model}"), 4)
+    elif shared.args.gptq_bits > 0 or shared.args.load_in_4bit:
+        from modules.quantized_LLaMA import load_quantized_LLaMA
 
-        # Multi-GPU setup
-        if shared.args.gpu_memory:
-            import accelerate
-
-            max_memory = {}
-            for i in range(len(shared.args.gpu_memory)):
-                max_memory[i] = f"{shared.args.gpu_memory[i]}GiB"
-            max_memory['cpu'] = f"{shared.args.cpu_memory or '99'}GiB"
-
-            device_map = accelerate.infer_auto_device_map(model, max_memory=max_memory, no_split_module_classes=["LLaMADecoderLayer"])
-            model = accelerate.dispatch_model(model, device_map=device_map)
-
-        # Single GPU
-        else:
-            model = model.to(torch.device('cuda:0'))
+        model = load_quantized_LLaMA(model_name)
 
     # Custom
     else:

modules/quantized_LLaMA.py

@@ -0,0 +1,60 @@
+import os
+import sys
+from pathlib import Path
+
+import accelerate
+import torch
+
+import modules.shared as shared
+
+sys.path.insert(0, os.path.abspath(Path("repositories/GPTQ-for-LLaMa")))
+from llama import load_quant
+
+
+# 4-bit LLaMA
+def load_quantized_LLaMA(model_name):
+    if shared.args.load_in_4bit:
+        bits = 4
+    else:
+        bits = shared.args.gptq_bits
+
+    path_to_model = Path(f'models/{model_name}')
+    pt_model = ''
+    if path_to_model.name.lower().startswith('llama-7b'):
+        pt_model = f'llama-7b-{bits}bit.pt'
+    elif path_to_model.name.lower().startswith('llama-13b'):
+        pt_model = f'llama-13b-{bits}bit.pt'
+    elif path_to_model.name.lower().startswith('llama-30b'):
+        pt_model = f'llama-30b-{bits}bit.pt'
+    elif path_to_model.name.lower().startswith('llama-65b'):
+        pt_model = f'llama-65b-{bits}bit.pt'
+    else:
+        pt_model = f'{model_name}-{bits}bit.pt'
+
+    # Try to find the .pt both in models/ and in the subfolder
+    pt_path = None
+    for path in [Path(p) for p in [f"models/{pt_model}", f"{path_to_model}/{pt_model}"]]:
+        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, bits)
+
+    # Multi-GPU setup
+    if shared.args.gpu_memory:
+        max_memory = {}
+        for i in range(len(shared.args.gpu_memory)):
+            max_memory[i] = f"{shared.args.gpu_memory[i]}GiB"
+        max_memory['cpu'] = f"{shared.args.cpu_memory or '99'}GiB"
+
+        device_map = accelerate.infer_auto_device_map(model, max_memory=max_memory, no_split_module_classes=["LLaMADecoderLayer"])
+        model = accelerate.dispatch_model(model, device_map=device_map)
+
+    # Single GPU
+    else:
+        model = model.to(torch.device('cuda:0'))
+
+    return model

modules/shared.py

@@ -69,6 +69,7 @@ parser.add_argument('--cai-chat', action='store_true', help='Launch the web UI i
 parser.add_argument('--cpu', action='store_true', help='Use the CPU to generate text.')
 parser.add_argument('--load-in-8bit', action='store_true', help='Load the model with 8-bit precision.')
 parser.add_argument('--load-in-4bit', action='store_true', help='Load the model with 4-bit precision. Currently only works with LLaMA.')
+parser.add_argument('--gptq-bits', type=int, default=0, help='Load a pre-quantized model with specified precision. 2, 3, 4 and 8bit are supported. Currently only works with LLaMA.')
 parser.add_argument('--bf16', action='store_true', help='Load the model with bfloat16 precision. Requires NVIDIA Ampere GPU.')
 parser.add_argument('--auto-devices', action='store_true', help='Automatically split the model across the available GPU(s) and CPU.')
 parser.add_argument('--disk', action='store_true', help='If the model is too large for your GPU(s) and CPU combined, send the remaining layers to the disk.')