release v0.1.0

Former-commit-id: 63c8d3a

Author: hiyouga

README.md

@@ -10,7 +10,9 @@
 
 ## Changelog
 
-[23/07/11] Now we support training the **Baichuan-13B** model in this repo. Try `--model_name_or_path baichuan-inc/Baichuan-13B-Base`, `--padding_side right` and `--lora_target W_pack` arguments to train the Baichuan-13B model. Remember to use `--prompt_template baichuan` argument when you are using the Baichuan-13B-Chat model.
+[23/07/18] Now we develop an all-in-one Web UI for training, evaluation and inference. Try `train_web.py` to fine-tune models in your Web browser. Thank [@KanadeSiina](https://github.com/KanadeSiina) and [@codemayq](https://github.com/codemayq) for their efforts in the development.
+
+[23/07/11] Now we support training the **Baichuan-13B** model in this repo. Please replace the Baichuan-13B model file with `tests/modeling_baichuan.py` and try `--model_name_or_path path_to_baichuan_model` and `--lora_target W_pack` arguments to train the Baichuan-13B model. Remember to use `--prompt_template baichuan` argument when you are using the Baichuan-13B-Chat model.
 
 [23/07/09] Now we release [FastEdit](https://github.com/hiyouga/FastEdit)⚡🩹, an easy-to-use package for editing the factual knowledge of large language models efficiently. Please follow [FastEdit](https://github.com/hiyouga/FastEdit) if you are interested.
 
@@ -125,14 +127,10 @@ cd LLaMA-Efficient-Tuning
 pip install -r requirements.txt
 ```
 
-### LLaMA Weights Preparation (optional)
-
-1. Download the weights of the LLaMA models.
-2. Convert them to HF format using the following command.
+### All-in-one Web UI
 
 ```bash
-python -m transformers.models.llama.convert_llama_weights_to_hf \
-    --input_dir path_to_llama_weights --model_size 7B --output_dir path_to_llama_model
+python src/train_web.py
 ```
 
 ### (Continually) Pre-Training
@@ -275,10 +273,20 @@ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
 
 We recommend using `--per_device_eval_batch_size=1` and `--max_target_length 128` at 4/8-bit evaluation.
 
-### API / CLI / Web Demo
+### API Demo
+
+```bash
+python src/api_demo.py \
+    --model_name_or_path path_to_your_model \
+    --checkpoint_dir path_to_checkpoint
+```
+
+See `http://localhost:8000/docs` for API documentation.
+
+### CLI Demo
 
 ```bash
-python src/xxx_demo.py \
+python src/cli_demo.py \
     --model_name_or_path path_to_your_model \
     --checkpoint_dir path_to_checkpoint
 ```

requirements.txt

@@ -3,14 +3,14 @@ transformers>=4.29.1
 datasets>=2.12.0
 accelerate>=0.19.0
 peft>=0.3.0
-trl==0.4.4
+trl>=0.4.7
 sentencepiece
 jieba
 rouge-chinese
 nltk
 gradio>=3.36.0
 uvicorn
-pydantic==1.10.7
+pydantic
 fastapi
 sse-starlette
 matplotlib

src/llmtuner/__init__.py

@@ -1,6 +1,7 @@
 from llmtuner.api import create_app
 from llmtuner.chat import ChatModel
 from llmtuner.tuner import get_train_args, get_infer_args, load_model_and_tokenizer, run_pt, run_sft, run_rm, run_ppo
+from llmtuner.webui import create_ui
 
 
-__version__ = "0.0.9"
+__version__ = "0.1.0"

src/llmtuner/api/app.py

@@ -1,3 +1,4 @@
+import json
 import uvicorn
 from fastapi import FastAPI, HTTPException
 from fastapi.middleware.cors import CORSMiddleware
@@ -93,7 +94,7 @@ async def predict(query: str, history: List[Tuple[str, str]], prefix: str, reque
             finish_reason=None
         )
         chunk = ChatCompletionStreamResponse(model=request.model, choices=[choice_data], object="chat.completion.chunk")
-        yield chunk.json(exclude_unset=True, ensure_ascii=False)
+        yield json.dumps(chunk, ensure_ascii=False)
 
         for new_text in chat_model.stream_chat(
             query, history, prefix, temperature=request.temperature, top_p=request.top_p, max_new_tokens=request.max_tokens
@@ -107,15 +108,15 @@ async def predict(query: str, history: List[Tuple[str, str]], prefix: str, reque
                 finish_reason=None
             )
             chunk = ChatCompletionStreamResponse(model=request.model, choices=[choice_data], object="chat.completion.chunk")
-            yield chunk.json(exclude_unset=True, ensure_ascii=False)
+            yield json.dumps(chunk, ensure_ascii=False)
 
         choice_data = ChatCompletionResponseStreamChoice(
             index=0,
             delta=DeltaMessage(),
             finish_reason="stop"
         )
         chunk = ChatCompletionStreamResponse(model=request.model, choices=[choice_data], object="chat.completion.chunk")
-        yield chunk.json(exclude_unset=True, ensure_ascii=False)
+        yield json.dumps(chunk, ensure_ascii=False)
         yield "[DONE]"
 
     return app

src/llmtuner/extras/constants.py

@@ -5,3 +5,27 @@
 FINETUNING_ARGS_NAME = "finetuning_args.json"
 
 LAYERNORM_NAMES = ["norm", "ln_f", "ln_attn", "ln_mlp"] # for LLaMA, BLOOM and Falcon settings
+
+METHODS = ["full", "freeze", "lora"]
+
+SUPPORTED_MODELS = {
+    "LLaMA-7B": "huggyllama/llama-7b",
+    "LLaMA-13B": "huggyllama/llama-13b",
+    "LLaMA-30B": "huggyllama/llama-30b",
+    "LLaMA-65B": "huggyllama/llama-65b",
+    "BLOOM-560M": "bigscience/bloom-560m",
+    "BLOOM-3B": "bigscience/bloom-3b",
+    "BLOOM-7B1": "bigscience/bloom-7b1",
+    "BLOOMZ-560M": "bigscience/bloomz-560m",
+    "BLOOMZ-3B": "bigscience/bloomz-3b",
+    "BLOOMZ-7B1-mt": "bigscience/bloomz-7b1-mt",
+    "Falcon-7B-Base": "tiiuae/falcon-7b",
+    "Falcon-7B-Chat": "tiiuae/falcon-7b-instruct",
+    "Falcon-40B-Base": "tiiuae/falcon-40b",
+    "Falcon-40B-Chat": "tiiuae/falcon-40b-instruct",
+    "Baichuan-7B": "baichuan-inc/Baichuan-7B",
+    "Baichuan-13B-Base": "baichuan-inc/Baichuan-13B-Base",
+    "Baichuan-13B-Chat": "baichuan-inc/Baichuan-13B-Chat",
+    "InternLM-7B-Base": "internlm/internlm-7b",
+    "InternLM-7B-Chat": "internlm/internlm-chat-7b"
+}

src/llmtuner/extras/logging.py

@@ -2,6 +2,20 @@
 import logging
 
 
+class LoggerHandler(logging.Handler):
+
+    def __init__(self):
+        super().__init__()
+        self.log = ""
+
+    def emit(self, record):
+        if record.name == "httpx":
+            return
+        log_entry = self.format(record)
+        self.log += log_entry
+        self.log += "\n\n"
+
+
 def get_logger(name: str) -> logging.Logger:
 
     formatter = logging.Formatter(

src/llmtuner/extras/ploting.py

@@ -1,4 +1,5 @@
 import os
+import math
 import json
 import matplotlib.pyplot as plt
 from typing import List, Optional
@@ -10,12 +11,13 @@
 logger = get_logger(__name__)
 
 
-def smooth(scalars: List[float], weight: Optional[float] = 0.9) -> List[float]:
+def smooth(scalars: List[float]) -> List[float]:
     r"""
     EMA implementation according to TensorBoard.
     """
     last = scalars[0]
     smoothed = list()
+    weight = 1.8 * (1 / (1 + math.exp(-0.05 * len(scalars))) - 0.5) # a sigmoid function
     for next_val in scalars:
         smoothed_val = last * weight + (1 - weight) * next_val
         smoothed.append(smoothed_val)