add alpaca

jianzhnie · May 25, 2023 · b754fa5 · b754fa5
1 parent a256bdf
commit b754fa5
Show file tree

Hide file tree

Showing 5 changed files with 62 additions and 63 deletions.
diff --git a/examples/alpaca/generate_server.py b/examples/alpaca/generate_server.py
@@ -80,6 +80,8 @@ def main(args):
                                   args.lora_model_name_or_path,
                                   load_8bit=args.load_8bit)
 
+    if not args.load_8bit:
+        model.half()  # seems to fix bugs for some users.
     # unwind broken decapoda-research config
     model.config.pad_token_id = tokenizer.pad_token_id = 0  # unk
     model.config.bos_token_id = 1
@@ -160,54 +162,54 @@ def generate_with_streaming(**kwargs):
         output = tokenizer.decode(s)
         yield prompter.get_response(output)
 
-        description = 'Alpaca-LoRA is a 7B-parameter LLaMA model finetuned to follow instructions. '
-        'It is trained on the [Stanford Alpaca](https://github.com/tatsu-lab/stanford_alpaca) '
-        'dataset and makes use of the Huggingface LLaMA implementation. For more information, '
-        "please visit [the project's website](https://github.com/tloen/alpaca-lora).",
-
-        server = gr.Interface(
-            fn=evaluate,
-            inputs=[
-                gr.components.Textbox(lines=2,
-                                      label='Instruction',
-                                      placeholder='Tell me about alpacas.'),
-                gr.components.Textbox(lines=2,
-                                      label='Input',
-                                      placeholder='none'),
-                gr.components.Slider(minimum=0,
-                                     maximum=1,
-                                     value=0.1,
-                                     label='Temperature'),
-                gr.components.Slider(minimum=0,
-                                     maximum=1,
-                                     value=0.75,
-                                     label='Top p'),
-                gr.components.Slider(minimum=0,
-                                     maximum=100,
-                                     step=1,
-                                     value=40,
-                                     label='Top k'),
-                gr.components.Slider(minimum=1,
-                                     maximum=4,
-                                     step=1,
-                                     value=4,
-                                     label='Beams'),
-                gr.components.Slider(minimum=1,
-                                     maximum=2000,
-                                     step=1,
-                                     value=128,
-                                     label='Max tokens'),
-                gr.components.Checkbox(label='Stream output'),
-            ],
-            outputs=[gr.inputs.Textbox(
-                lines=5,
-                label='Output',
-            )],
-            title='🦙🌲 Alpaca-LoRA',
-            description=description,
-        )
-
-        server.queue().launch(server_name='0.0.0.0', share=False)
+    description = 'Alpaca-LoRA is a 7B-parameter LLaMA model finetuned to follow instructions. '
+    'It is trained on the [Stanford Alpaca](https://github.com/tatsu-lab/stanford_alpaca) '
+    'dataset and makes use of the Huggingface LLaMA implementation. For more information, '
+    "please visit [the project's website](https://github.com/tloen/alpaca-lora).",
+
+    server = gr.Interface(
+        fn=evaluate,
+        inputs=[
+            gr.components.Textbox(lines=2,
+                                    label='Instruction',
+                                    placeholder='Tell me about alpacas.'),
+            gr.components.Textbox(lines=2,
+                                    label='Input',
+                                    placeholder='none'),
+            gr.components.Slider(minimum=0,
+                                    maximum=1,
+                                    value=0.1,
+                                    label='Temperature'),
+            gr.components.Slider(minimum=0,
+                                    maximum=1,
+                                    value=0.75,
+                                    label='Top p'),
+            gr.components.Slider(minimum=0,
+                                    maximum=100,
+                                    step=1,
+                                    value=40,
+                                    label='Top k'),
+            gr.components.Slider(minimum=1,
+                                    maximum=4,
+                                    step=1,
+                                    value=4,
+                                    label='Beams'),
+            gr.components.Slider(minimum=1,
+                                    maximum=2000,
+                                    step=1,
+                                    value=128,
+                                    label='Max tokens'),
+            gr.components.Checkbox(label='Stream output'),
+        ],
+        outputs=[gr.inputs.Textbox(
+            lines=5,
+            label='Output',
+        )],
+        title='🦙🌲 Alpaca-LoRA',
+        description=description,
+    )
+
+    server.queue().launch(server_name='0.0.0.0', share=False)
 
 
 if __name__ == '__main__':

diff --git a/examples/alpaca/infer.sh b/examples/alpaca/infer.sh
@@ -14,3 +14,8 @@ python generate_server.py \
     --model_name_or_path decapoda-research/llama-7b-hf \
     --lora_model_name_or_path  tloen/alpaca-lora-7b \
     --load_8bit
+
+python generate_server.py \
+    --model_name_or_path decapoda-research/llama-7b-hf \
+    --lora_model_name_or_path  ./work_dir_lora \
+    --load_8bit
diff --git a/examples/alpaca/run.sh b/examples/alpaca/run.sh
@@ -1,12 +1,2 @@
 nohup sh scripts/run_alpaca_ds.sh > run2.log 2>&1 &
-nohup sh scripts/run_alpaca_lora.sh > run2.log 2>&1 &
-
-
-python generate.py \
-    --model_name_or_path decapoda-research/llama-7b-hf \
-    --lora_model_name_or_path  \
-    --temperature 1.0 \
-    --repetition_penalty 1.0 \
-    --top_k  40 \
-    --top_p  0.9  \
-    --num_beams 4 \
+nohup sh scripts/run_alpaca_lora.sh > run2.log 2>&1 &
diff --git a/examples/alpaca/scripts/run_alpaca_lora.sh b/examples/alpaca/scripts/run_alpaca_lora.sh
@@ -8,10 +8,12 @@ python train_alpaca_lora.py \
     --gradient_accumulation_steps 8 \
     --evaluation_strategy "no" \
     --save_strategy "steps" \
-    --save_steps 2000 \
+    --save_steps 500 \
     --save_total_limit 5 \
-    --learning_rate 2e-5 \
+    --learning_rate 1e-4 \
     --weight_decay 0. \
     --warmup_ratio 0.03 \
     --lr_scheduler_type "cosine" \
-    --logging_steps 1
+    --model_max_length 2048 \
+    --logging_steps 1 \
+    --fp16 True
diff --git a/examples/alpaca/train_alpaca_lora.py b/examples/alpaca/train_alpaca_lora.py
@@ -335,4 +335,4 @@ def train(load_in_8bit=False) -> None:
 
 
 if __name__ == '__main__':
-    train()
+    train(load_in_8bit=True)
Original file line number	Diff line number	Diff line change
Expand Up		@@ -335,4 +335,4 @@ def train(load_in_8bit=False) -> None:


		if __name__ == '__main__':
		train()
		train(load_in_8bit=True)