feat: first commit with alpaca-lora Low-Rank LLaMA Instruct-Tuning example

Makefile

@@ -0,0 +1,24 @@
+SHELL=/bin/sh
+export PATH := ./venv/bin:$(PATH)
+.PHONY: help
+help: ## This help.
+	@awk 'BEGIN {FS = ":.*?## "} /^[a-zA-Z_-]+:.*?## / {printf " \033[36m%-20s\033[0m %s\n", $$1, $$2}' $(MAKEFILE_LIST)
+
+.DEFAULT_GOAL := help
+
+venv:
+	touch requirements.txt ;\
+	test -d venv || virtualenv --python=$$PYTHON3 venv
+
+pip-compile: venv
+	python -m pip install --upgrade pip;\
+	pip install pip-tools;\
+	touch requirements.in ;\
+	pip-compile --output-file requirements.txt requirements.in;\
+	pip install -r requirements.txt
+
+autopep8:
+	autopep8 -i *.py
+
+clean:
+	rm -fr venv

app.py

@@ -0,0 +1,79 @@
+import gradio as gr
+import random
+import time
+from peft import PeftModel
+from transformers import LLaMATokenizer, LLaMAForCausalLM, GenerationConfig
+
+tokenizer = LLaMATokenizer.from_pretrained("decapoda-research/llama-7b-hf")
+model = LLaMAForCausalLM.from_pretrained(
+    "decapoda-research/llama-7b-hf",
+    load_in_8bit=True,
+    device_map="auto",
+)
+model = PeftModel.from_pretrained(model, "tloen/alpaca-lora-7b")
+
+
+def generate_prompt(instruction, input=None):
+    if input:
+        return f"""Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.
+
+### Instruction:
+{instruction}
+
+### Input:
+{input}
+
+### Response:"""
+    else:
+        return f"""Below is an instruction that describes a task. Write a response that appropriately completes the request.
+
+### Instruction:
+{instruction}
+
+### Response:"""
+
+
+
+generation_config = GenerationConfig(
+    temperature=0.1,
+    top_p=0.75,
+    num_beams=4,
+)
+
+def evaluate(instruction, input=None):
+    prompt = generate_prompt(instruction, input)
+    inputs = tokenizer(prompt, return_tensors="pt")
+    input_ids = inputs["input_ids"].cuda()
+    generation_output = model.generate(
+        input_ids=input_ids,
+        generation_config=generation_config,
+        return_dict_in_generate=True,
+        output_scores=True,
+        max_new_tokens=256
+    )
+    for s in generation_output.sequences:
+        output = tokenizer.decode(s)
+        print("Response:", output.split("### Response:")[1].strip())
+
+
+
+with gr.Blocks() as demo:
+    chatbot = gr.Chatbot()
+    msg = gr.Textbox()
+    clear = gr.Button("Clear")
+
+    def user(user_message, history):
+        return "", history + [[user_message, None]]
+
+    def bot(history):
+        bot_message = evaluate(input("Instruction: {}".format(history)))
+        history[-1][1] = bot_message
+        time.sleep(1)
+        return history
+
+    msg.submit(user, [msg, chatbot], [msg, chatbot], queue=False).then(
+        bot, chatbot, chatbot
+    )
+    clear.click(lambda: None, None, chatbot, queue=False)
+
+demo.launch()

requirements.in

@@ -0,0 +1,7 @@
+gradio
+bitsandbytes
+datasets
+loralib
+sentencepiece
+git+https://github.com/zphang/transformers@c3dc391
+git+https://github.com/huggingface/peft.git

requirements.txt

@@ -0,0 +1,260 @@
+#
+# This file is autogenerated by pip-compile with Python 3.10
+# by the following command:
+#
+#    pip-compile --output-file=requirements.txt requirements.in
+#
+accelerate==0.17.1
+    # via peft
+aiofiles==23.1.0
+    # via gradio
+aiohttp==3.8.4
+    # via
+    #   datasets
+    #   fsspec
+    #   gradio
+aiosignal==1.3.1
+    # via aiohttp
+altair==4.2.2
+    # via gradio
+anyio==3.6.2
+    # via
+    #   httpcore
+    #   starlette
+async-timeout==4.0.2
+    # via aiohttp
+attrs==22.2.0
+    # via
+    #   aiohttp
+    #   jsonschema
+bitsandbytes==0.37.1
+    # via -r requirements.in
+certifi==2022.12.7
+    # via
+    #   httpcore
+    #   httpx
+    #   requests
+charset-normalizer==3.1.0
+    # via
+    #   aiohttp
+    #   requests
+click==8.1.3
+    # via uvicorn
+contourpy==1.0.7
+    # via matplotlib
+cycler==0.11.0
+    # via matplotlib
+datasets==2.10.1
+    # via -r requirements.in
+dill==0.3.6
+    # via
+    #   datasets
+    #   multiprocess
+entrypoints==0.4
+    # via altair
+fastapi==0.94.1
+    # via gradio
+ffmpy==0.3.0
+    # via gradio
+filelock==3.10.0
+    # via
+    #   huggingface-hub
+    #   torch
+    #   transformers
+fonttools==4.39.2
+    # via matplotlib
+frozenlist==1.3.3
+    # via
+    #   aiohttp
+    #   aiosignal
+fsspec[http]==2023.3.0
+    # via
+    #   datasets
+    #   gradio
+gradio==3.21.0
+    # via -r requirements.in
+h11==0.14.0
+    # via
+    #   httpcore
+    #   uvicorn
+httpcore==0.16.3
+    # via httpx
+httpx==0.23.3
+    # via gradio
+huggingface-hub==0.13.2
+    # via
+    #   datasets
+    #   gradio
+    #   transformers
+idna==3.4
+    # via
+    #   anyio
+    #   requests
+    #   rfc3986
+    #   yarl
+jinja2==3.1.2
+    # via
+    #   altair
+    #   gradio
+    #   torch
+jsonschema==4.17.3
+    # via altair
+kiwisolver==1.4.4
+    # via matplotlib
+linkify-it-py==2.0.0
+    # via markdown-it-py
+loralib==0.1.1
+    # via -r requirements.in
+markdown-it-py[linkify]==2.2.0
+    # via
+    #   gradio
+    #   mdit-py-plugins
+markupsafe==2.1.2
+    # via
+    #   gradio
+    #   jinja2
+matplotlib==3.7.1
+    # via gradio
+mdit-py-plugins==0.3.3
+    # via gradio
+mdurl==0.1.2
+    # via markdown-it-py
+mpmath==1.3.0
+    # via sympy
+multidict==6.0.4
+    # via
+    #   aiohttp
+    #   yarl
+multiprocess==0.70.14
+    # via datasets
+networkx==3.0
+    # via torch
+numpy==1.24.2
+    # via
+    #   accelerate
+    #   altair
+    #   contourpy
+    #   datasets
+    #   gradio
+    #   matplotlib
+    #   pandas
+    #   peft
+    #   pyarrow
+    #   transformers
+orjson==3.8.7
+    # via gradio
+packaging==23.0
+    # via
+    #   accelerate
+    #   datasets
+    #   huggingface-hub
+    #   matplotlib
+    #   peft
+    #   transformers
+pandas==1.5.3
+    # via
+    #   altair
+    #   datasets
+    #   gradio
+peft @ git+https://github.com/huggingface/peft.git
+    # via -r requirements.in
+pillow==9.4.0
+    # via
+    #   gradio
+    #   matplotlib
+psutil==5.9.4
+    # via
+    #   accelerate
+    #   peft
+pyarrow==11.0.0
+    # via datasets
+pydantic==1.10.6
+    # via
+    #   fastapi
+    #   gradio
+pydub==0.25.1
+    # via gradio
+pyparsing==3.0.9
+    # via matplotlib
+pyrsistent==0.19.3
+    # via jsonschema
+python-dateutil==2.8.2
+    # via
+    #   matplotlib
+    #   pandas
+python-multipart==0.0.6
+    # via gradio
+pytz==2022.7.1
+    # via pandas
+pyyaml==6.0
+    # via
+    #   accelerate
+    #   datasets
+    #   gradio
+    #   huggingface-hub
+    #   peft
+    #   transformers
+regex==2022.10.31
+    # via transformers
+requests==2.28.2
+    # via
+    #   datasets
+    #   fsspec
+    #   gradio
+    #   huggingface-hub
+    #   responses
+    #   transformers
+responses==0.18.0
+    # via datasets
+rfc3986[idna2008]==1.5.0
+    # via httpx
+sentencepiece==0.1.97
+    # via -r requirements.in
+six==1.16.0
+    # via python-dateutil
+sniffio==1.3.0
+    # via
+    #   anyio
+    #   httpcore
+    #   httpx
+starlette==0.26.1
+    # via fastapi
+sympy==1.11.1
+    # via torch
+tokenizers==0.13.2
+    # via transformers
+toolz==0.12.0
+    # via altair
+torch==2.0.0
+    # via
+    #   accelerate
+    #   peft
+tqdm==4.65.0
+    # via
+    #   datasets
+    #   huggingface-hub
+    #   transformers
+transformers @ git+https://github.com/zphang/transformers@c3dc391
+    # via
+    #   -r requirements.in
+    #   peft
+typing-extensions==4.5.0
+    # via
+    #   gradio
+    #   huggingface-hub
+    #   pydantic
+    #   torch
+uc-micro-py==1.0.1
+    # via linkify-it-py
+urllib3==1.26.15
+    # via
+    #   requests
+    #   responses
+uvicorn==0.21.1
+    # via gradio
+websockets==10.4
+    # via gradio
+xxhash==3.2.0
+    # via datasets
+yarl==1.8.2
+    # via aiohttp