Merge branch 'main' into main

Former-commit-id: e8e6af2651

src/llamafactory/webui/components/__init__.py

@@ -1,3 +1,17 @@
+# Copyright 2024 the LlamaFactory team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
 from .chatbot import create_chat_box
 from .eval import create_eval_tab
 from .export import create_export_tab

src/llamafactory/webui/components/chatbot.py

@@ -1,3 +1,17 @@
+# Copyright 2024 the LlamaFactory team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
 from typing import TYPE_CHECKING, Dict, Tuple
 
 from ...data import Role

src/llamafactory/webui/components/data.py

@@ -1,3 +1,17 @@
+# Copyright 2024 the LlamaFactory team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
 import json
 import os
 from typing import TYPE_CHECKING, Any, Dict, List, Tuple

src/llamafactory/webui/components/eval.py

@@ -1,3 +1,17 @@
+# Copyright 2024 the LlamaFactory team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
 from typing import TYPE_CHECKING, Dict
 
 from ...extras.packages import is_gradio_available

src/llamafactory/webui/components/export.py

@@ -1,10 +1,24 @@
+# Copyright 2024 the LlamaFactory team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
 from typing import TYPE_CHECKING, Dict, Generator, List, Union
 
 from ...extras.constants import PEFT_METHODS
 from ...extras.misc import torch_gc
 from ...extras.packages import is_gradio_available
 from ...train.tuner import export_model
-from ..common import get_save_dir
+from ..common import GPTQ_BITS, get_save_dir
 from ..locales import ALERTS
 
 
@@ -18,7 +32,11 @@ if TYPE_CHECKING:
     from ..engine import Engine
 
 
-GPTQ_BITS = ["8", "4", "3", "2"]
+def can_quantize(checkpoint_path: Union[str, List[str]]) -> "gr.Dropdown":
+    if isinstance(checkpoint_path, list) and len(checkpoint_path) != 0:
+        return gr.Dropdown(value="none", interactive=False)
+    else:
+        return gr.Dropdown(interactive=True)
 
 
 def save_model(
@@ -96,6 +114,9 @@ def create_export_tab(engine: "Engine") -> Dict[str, "Component"]:
         export_dir = gr.Textbox()
         export_hub_model_id = gr.Textbox()
 
+    checkpoint_path: gr.Dropdown = engine.manager.get_elem_by_id("top.checkpoint_path")
+    checkpoint_path.change(can_quantize, [checkpoint_path], [export_quantization_bit], queue=False)
+
     export_btn = gr.Button()
     info_box = gr.Textbox(show_label=False, interactive=False)
 

src/llamafactory/webui/components/infer.py

@@ -1,3 +1,17 @@
+# Copyright 2024 the LlamaFactory team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
 from typing import TYPE_CHECKING, Dict
 
 from ...extras.packages import is_gradio_available
@@ -18,15 +32,26 @@ def create_infer_tab(engine: "Engine") -> Dict[str, "Component"]:
     input_elems = engine.manager.get_base_elems()
     elem_dict = dict()
 
-    infer_backend = gr.Dropdown(choices=["huggingface", "vllm"], value="huggingface")
+    with gr.Row():
+        infer_backend = gr.Dropdown(choices=["huggingface", "vllm"], value="huggingface")
+        infer_dtype = gr.Dropdown(choices=["auto", "float16", "bfloat16", "float32"], value="auto")
+
     with gr.Row():
         load_btn = gr.Button()
         unload_btn = gr.Button()
 
     info_box = gr.Textbox(show_label=False, interactive=False)
 
-    input_elems.update({infer_backend})
-    elem_dict.update(dict(infer_backend=infer_backend, load_btn=load_btn, unload_btn=unload_btn, info_box=info_box))
+    input_elems.update({infer_backend, infer_dtype})
+    elem_dict.update(
+        dict(
+            infer_backend=infer_backend,
+            infer_dtype=infer_dtype,
+            load_btn=load_btn,
+            unload_btn=unload_btn,
+            info_box=info_box,
+        )
+    )
 
     chatbot, messages, chat_elems = create_chat_box(engine, visible=False)
     elem_dict.update(chat_elems)

src/llamafactory/webui/components/top.py

@@ -1,10 +1,24 @@
+# Copyright 2024 the LlamaFactory team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
 from typing import TYPE_CHECKING, Dict
 
 from ...data import TEMPLATES
 from ...extras.constants import METHODS, SUPPORTED_MODELS
 from ...extras.packages import is_gradio_available
 from ..common import get_model_info, list_checkpoints, save_config
-from ..utils import can_quantize
+from ..utils import can_quantize, can_quantize_to
 
 
 if is_gradio_available():
@@ -29,17 +43,23 @@ def create_top() -> Dict[str, "Component"]:
 
     with gr.Accordion(open=False) as advanced_tab:
         with gr.Row():
-            quantization_bit = gr.Dropdown(choices=["none", "8", "4"], value="none", scale=2)
-            template = gr.Dropdown(choices=list(TEMPLATES.keys()), value="default", scale=2)
-            rope_scaling = gr.Radio(choices=["none", "linear", "dynamic"], value="none", scale=3)
-            booster = gr.Radio(choices=["none", "flashattn2", "unsloth"], value="none", scale=3)
+            quantization_bit = gr.Dropdown(choices=["none", "8", "4"], value="none", allow_custom_value=True, scale=1)
+            quantization_method = gr.Dropdown(choices=["bitsandbytes", "hqq", "eetq"], value="bitsandbytes", scale=1)
+            template = gr.Dropdown(choices=list(TEMPLATES.keys()), value="default", scale=1)
+            rope_scaling = gr.Radio(choices=["none", "linear", "dynamic"], value="none", scale=2)
+            booster = gr.Radio(choices=["auto", "flashattn2", "unsloth"], value="auto", scale=2)
             visual_inputs = gr.Checkbox(scale=1)
 
-    model_name.change(get_model_info, [model_name], [model_path, template, visual_inputs], queue=False)
+    model_name.change(get_model_info, [model_name], [model_path, template, visual_inputs], queue=False).then(
+        list_checkpoints, [model_name, finetuning_type], [checkpoint_path], queue=False
+    )
     model_name.input(save_config, inputs=[lang, model_name], queue=False)
     model_path.input(save_config, inputs=[lang, model_name, model_path], queue=False)
-    finetuning_type.change(can_quantize, [finetuning_type], [quantization_bit], queue=False)
+    finetuning_type.change(can_quantize, [finetuning_type], [quantization_bit], queue=False).then(
+        list_checkpoints, [model_name, finetuning_type], [checkpoint_path], queue=False
+    )
     checkpoint_path.focus(list_checkpoints, [model_name, finetuning_type], [checkpoint_path], queue=False)
+    quantization_method.change(can_quantize_to, [quantization_method], [quantization_bit], queue=False)
 
     return dict(
         lang=lang,
@@ -49,6 +69,7 @@ def create_top() -> Dict[str, "Component"]:
         checkpoint_path=checkpoint_path,
         advanced_tab=advanced_tab,
         quantization_bit=quantization_bit,
+        quantization_method=quantization_method,
         template=template,
         rope_scaling=rope_scaling,
         booster=booster,

src/llamafactory/webui/components/train.py

@@ -1,3 +1,17 @@
+# Copyright 2024 the LlamaFactory team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
 from typing import TYPE_CHECKING, Dict
 
 from transformers.trainer_utils import SchedulerType
@@ -40,7 +54,7 @@ def create_train_tab(engine: "Engine") -> Dict[str, "Component"]:
         num_train_epochs = gr.Textbox(value="3.0")
         max_grad_norm = gr.Textbox(value="1.0")
         max_samples = gr.Textbox(value="100000")
-        compute_type = gr.Dropdown(choices=["fp16", "bf16", "fp32", "pure_bf16"], value="fp16")
+        compute_type = gr.Dropdown(choices=["bf16", "fp16", "fp32", "pure_bf16"], value="bf16")
 
     input_elems.update({learning_rate, num_train_epochs, max_grad_norm, max_samples, compute_type})
     elem_dict.update(
@@ -152,10 +166,9 @@ def create_train_tab(engine: "Engine") -> Dict[str, "Component"]:
             create_new_adapter = gr.Checkbox()
 
         with gr.Row():
-            with gr.Column(scale=1):
-                use_rslora = gr.Checkbox()
-                use_dora = gr.Checkbox()
-
+            use_rslora = gr.Checkbox()
+            use_dora = gr.Checkbox()
+            use_pissa = gr.Checkbox()
             lora_target = gr.Textbox(scale=2)
             additional_target = gr.Textbox(scale=2)
 
@@ -168,6 +181,7 @@ def create_train_tab(engine: "Engine") -> Dict[str, "Component"]:
             create_new_adapter,
             use_rslora,
             use_dora,
+            use_pissa,
             lora_target,
             additional_target,
         }
@@ -182,6 +196,7 @@ def create_train_tab(engine: "Engine") -> Dict[str, "Component"]:
             create_new_adapter=create_new_adapter,
             use_rslora=use_rslora,
             use_dora=use_dora,
+            use_pissa=use_pissa,
             lora_target=lora_target,
             additional_target=additional_target,
         )
@@ -279,7 +294,7 @@ def create_train_tab(engine: "Engine") -> Dict[str, "Component"]:
         with gr.Column(scale=1):
             loss_viewer = gr.Plot()
 
-    input_elems.update({output_dir, config_path, device_count, ds_stage, ds_offload})
+    input_elems.update({output_dir, config_path, ds_stage, ds_offload})
     elem_dict.update(
         dict(
             cmd_preview_btn=cmd_preview_btn,