for gpu

.gitignore

@@ -0,0 +1,2 @@
+.venv
+.env

Dockerfile

@@ -0,0 +1,25 @@
+# Use an official Python runtime as a base image
+FROM python:3.9-slim
+
+# Set the working directory in the container to /app
+WORKDIR /app
+
+# Copy the current directory contents into the container at /app
+ADD . /app
+
+# Install any needed packages specified in requirements.txt
+RUN apt-get update && \
+    apt-get install -y aria2 curl gcc python3-dev && \
+    curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y && \
+    . $HOME/.cargo/env && \
+    pip install --upgrade pip && \
+    pip install --no-cache-dir -r requirements.txt
+
+# Run download_model.sh
+RUN bash download_model.sh
+
+# Make port 80 available to the world outside this container
+EXPOSE 80
+
+# Run app.py when the container launches
+CMD ["python", "app.py"]

Makefile.venv

@@ -0,0 +1,272 @@
+#
+# SEAMLESSLY MANAGE PYTHON VIRTUAL ENVIRONMENT WITH A MAKEFILE
+#
+# https://github.com/sio/Makefile.venv             v2023.04.17
+#
+#
+# Insert `include Makefile.venv` at the bottom of your Makefile to enable these
+# rules.
+#
+# When writing your Makefile use '$(VENV)/python' to refer to the Python
+# interpreter within virtual environment and '$(VENV)/executablename' for any
+# other executable in venv.
+#
+# This Makefile provides the following targets:
+#   venv
+#       Use this as a dependency for any target that requires virtual
+#       environment to be created and configured
+#   python, ipython
+#       Use these to launch interactive Python shell within virtual environment
+#   shell, bash, zsh
+#       Launch interactive command line shell. "shell" target launches the
+#       default shell Makefile executes its rules in (usually /bin/sh).
+#       "bash" and "zsh" can be used to refer to the specific desired shell.
+#   show-venv
+#       Show versions of Python and pip, and the path to the virtual environment
+#   clean-venv
+#       Remove virtual environment
+#   $(VENV)/executable_name
+#       Install `executable_name` with pip. Only packages with names matching
+#       the name of the corresponding executable are supported.
+#       Use this as a lightweight mechanism for development dependencies
+#       tracking. E.g. for one-off tools that are not required in every
+#       developer's environment, therefore are not included into
+#       requirements.txt or setup.py.
+#       Note:
+#           Rules using such target or dependency MUST be defined below
+#           `include` directive to make use of correct $(VENV) value.
+#       Example:
+#           codestyle: $(VENV)/pyflakes
+#               $(VENV)/pyflakes .
+#       See `ipython` target below for another example.
+#
+# This Makefile can be configured via following variables:
+#   PY
+#       Command name for system Python interpreter. It is used only initially to
+#       create the virtual environment
+#       Default: python3
+#   REQUIREMENTS_TXT
+#       Space separated list of paths to requirements.txt files.
+#       Paths are resolved relative to current working directory.
+#       Default: requirements.txt
+#
+#       Non-existent files are treated as hard dependencies,
+#       recipes for creating such files must be provided by the main Makefile.
+#       Providing empty value (REQUIREMENTS_TXT=) turns off processing of
+#       requirements.txt even when the file exists.
+#   SETUP_PY, SETUP_CFG, PYPROJECT_TOML, VENV_LOCAL_PACKAGE
+#       Space separated list of paths to files that contain build instructions
+#       for local Python packages. Corresponding packages will be installed
+#       into venv in editable mode along with all their dependencies.
+#       Default: setup.py setup.cfg pyproject.toml (whichever present)
+#
+#       Non-existent and empty values are treated in the same way as for REQUIREMENTS_TXT.
+#   WORKDIR
+#       Parent directory for the virtual environment.
+#       Default: current working directory.
+#   VENVDIR
+#       Python virtual environment directory.
+#       Default: $(WORKDIR)/.venv
+#
+# This Makefile was written for GNU Make and may not work with other make
+# implementations.
+#
+#
+# Copyright (c) 2019-2023 Vitaly Potyarkin
+#
+# Licensed under the Apache License, Version 2.0
+#    <http://www.apache.org/licenses/LICENSE-2.0>
+#
+
+
+#
+# Configuration variables
+#
+
+WORKDIR?=.
+VENVDIR?=$(WORKDIR)/.venv
+REQUIREMENTS_TXT?=$(wildcard requirements.txt)  # Multiple paths are supported (space separated)
+SETUP_PY?=$(wildcard setup.py)                  # Multiple paths are supported (space separated)
+SETUP_CFG?=$(foreach s,$(SETUP_PY),$(wildcard $(patsubst %setup.py,%setup.cfg,$(s))))
+PYPROJECT_TOML?=$(wildcard pyproject.toml)
+VENV_LOCAL_PACKAGE?=$(SETUP_PY) $(SETUP_CFG) $(PYPROJECT_TOML)
+MARKER=.initialized-with-Makefile.venv
+
+
+#
+# Python interpreter detection
+#
+
+_PY_AUTODETECT_MSG=Detected Python interpreter: $(PY). Use PY environment variable to override
+
+ifeq (ok,$(shell test -e /dev/null 2>&1 && echo ok))
+NULL_STDERR=2>/dev/null
+else
+NULL_STDERR=2>NUL
+endif
+
+ifndef PY
+_PY_OPTION:=python3
+ifeq (ok,$(shell $(_PY_OPTION) -c "print('ok')" $(NULL_STDERR)))
+PY=$(_PY_OPTION)
+endif
+endif
+
+ifndef PY
+_PY_OPTION:=$(VENVDIR)/bin/python
+ifeq (ok,$(shell $(_PY_OPTION) -c "print('ok')" $(NULL_STDERR)))
+PY=$(_PY_OPTION)
+$(info $(_PY_AUTODETECT_MSG))
+endif
+endif
+
+ifndef PY
+_PY_OPTION:=$(subst /,\,$(VENVDIR)/Scripts/python)
+ifeq (ok,$(shell $(_PY_OPTION) -c "print('ok')" $(NULL_STDERR)))
+PY=$(_PY_OPTION)
+$(info $(_PY_AUTODETECT_MSG))
+endif
+endif
+
+ifndef PY
+_PY_OPTION:=py -3
+ifeq (ok,$(shell $(_PY_OPTION) -c "print('ok')" $(NULL_STDERR)))
+PY=$(_PY_OPTION)
+$(info $(_PY_AUTODETECT_MSG))
+endif
+endif
+
+ifndef PY
+_PY_OPTION:=python
+ifeq (ok,$(shell $(_PY_OPTION) -c "print('ok')" $(NULL_STDERR)))
+PY=$(_PY_OPTION)
+$(info $(_PY_AUTODETECT_MSG))
+endif
+endif
+
+ifndef PY
+define _PY_AUTODETECT_ERR
+Could not detect Python interpreter automatically.
+Please specify path to interpreter via PY environment variable.
+endef
+$(error $(_PY_AUTODETECT_ERR))
+endif
+
+
+#
+# Internal variable resolution
+#
+
+VENV=$(VENVDIR)/bin
+EXE=
+# Detect windows
+ifeq (win32,$(shell $(PY) -c "import __future__, sys; print(sys.platform)"))
+VENV=$(VENVDIR)/Scripts
+EXE=.exe
+endif
+
+touch=touch $(1)
+ifeq (,$(shell command -v touch $(NULL_STDERR)))
+# https://ss64.com/nt/touch.html
+touch=type nul >> $(subst /,\,$(1)) && copy /y /b $(subst /,\,$(1))+,, $(subst /,\,$(1))
+endif
+
+RM?=rm -f
+ifeq (,$(shell command -v $(firstword $(RM)) $(NULL_STDERR)))
+RMDIR:=rd /s /q
+else
+RMDIR:=$(RM) -r
+endif
+
+
+#
+# Virtual environment
+#
+
+.PHONY: venv
+venv: $(VENV)/$(MARKER)
+
+.PHONY: clean-venv
+clean-venv:
+	-$(RMDIR) "$(VENVDIR)"
+
+.PHONY: show-venv
+show-venv: venv
+	@$(VENV)/python -c "import sys; print('Python ' + sys.version.replace('\n',''))"
+	@$(VENV)/pip --version
+	@echo venv: $(VENVDIR)
+
+.PHONY: debug-venv
+debug-venv:
+	@echo "PATH (Shell)=$$PATH"
+	@$(MAKE) --version
+	$(info PATH (GNU Make)="$(PATH)")
+	$(info SHELL="$(SHELL)")
+	$(info PY="$(PY)")
+	$(info REQUIREMENTS_TXT="$(REQUIREMENTS_TXT)")
+	$(info VENV_LOCAL_PACKAGE="$(VENV_LOCAL_PACKAGE)")
+	$(info VENVDIR="$(VENVDIR)")
+	$(info VENVDEPENDS="$(VENVDEPENDS)")
+	$(info WORKDIR="$(WORKDIR)")
+
+
+#
+# Dependencies
+#
+
+ifneq ($(strip $(REQUIREMENTS_TXT)),)
+VENVDEPENDS+=$(REQUIREMENTS_TXT)
+endif
+
+ifneq ($(strip $(VENV_LOCAL_PACKAGE)),)
+VENVDEPENDS+=$(VENV_LOCAL_PACKAGE)
+endif
+
+$(VENV):
+	$(PY) -m venv $(VENVDIR)
+	$(VENV)/python -m pip install --upgrade pip setuptools wheel
+
+$(VENV)/$(MARKER): $(VENVDEPENDS) | $(VENV)
+ifneq ($(strip $(REQUIREMENTS_TXT)),)
+	$(VENV)/pip install $(foreach path,$(REQUIREMENTS_TXT),-r $(path))
+endif
+ifneq ($(strip $(VENV_LOCAL_PACKAGE)),)
+	$(VENV)/pip install $(foreach path,$(sort $(VENV_LOCAL_PACKAGE)),-e $(dir $(path)))
+endif
+	$(call touch,$(VENV)/$(MARKER))
+
+
+#
+# Interactive shells
+#
+
+.PHONY: python
+python: venv
+	exec $(VENV)/python
+
+.PHONY: ipython
+ipython: $(VENV)/ipython
+	exec $(VENV)/ipython
+
+.PHONY: shell
+shell: venv
+	. $(VENV)/activate && exec $(notdir $(SHELL))
+
+.PHONY: bash zsh
+bash zsh: venv
+	. $(VENV)/activate && exec $@
+
+
+#
+# Commandline tools (wildcard rule, executable name must match package name)
+#
+
+ifneq ($(EXE),)
+$(VENV)/%: $(VENV)/%$(EXE) ;
+.PHONY:    $(VENV)/%
+.PRECIOUS: $(VENV)/%$(EXE)
+endif
+
+$(VENV)/%$(EXE): $(VENV)/$(MARKER)
+	$(VENV)/pip install --upgrade $*
+	$(call touch,$@)

app.py

@@ -0,0 +1,70 @@
+import os
+import gradio as gr
+import transformers
+from torch import bfloat16
+from threading import Thread
+from gradio.themes.utils.colors import Color
+
+# Download model and tokenizer files
+os.system('bash download_model.sh')
+
+model_id = "/app/medllama2_7b"
+tokenizer = transformers.AutoTokenizer.from_pretrained(model_id)
+model_config = transformers.AutoConfig.from_pretrained(model_id)
+
+bnb_config = transformers.BitsAndBytesConfig(
+    load_in_4bit=True,
+    bnb_4bit_quant_type='nf4',
+    bnb_4bit_use_double_quant=True,
+    bnb_4bit_compute_dtype=bfloat16
+)
+model = transformers.AutoModelForCausalLM.from_pretrained(
+    model_id,
+    trust_remote_code=True,
+    config=model_config,
+    quantization_config=bnb_config,
+    device_map='auto'
+)
+
+prompts = ["You are a helpful AI Doctor."]
+
+def prompt_build(system_prompt, user_inp, hist):
+    prompt = f"""### System:\n{system_prompt}\n\n"""
+    
+    for pair in hist:
+        prompt += f"""### User:\n{pair[0]}\n\n### Assistant:\n{pair[1]}\n\n"""
+
+    prompt += f"""### User:\n{user_inp}\n\n### Assistant:"""
+    return prompt
+
+def chat(user_input, history, system_prompt):
+
+    prompt = prompt_build(system_prompt, user_input, history)
+    model_inputs = tokenizer([prompt], return_tensors="pt").to("cuda")
+
+    streamer = transformers.TextIteratorStreamer(tokenizer, timeout=10., skip_prompt=True, skip_special_tokens=True)
+
+    generate_kwargs = dict(
+        model_inputs,
+        streamer=streamer,
+        max_length=2048,
+        do_sample=True,
+        top_p=0.95,
+        temperature=0.8,
+        top_k=50
+    )
+    t = Thread(target=model.generate, kwargs=generate_kwargs)
+    t.start()
+
+    model_output = ""
+    for new_text in streamer:
+        model_output += new_text
+        yield model_output
+    return model_output
+
+if __name__ == "__main__":
+    with gr.Blocks() as demo:
+        dropdown = gr.Dropdown(choices=prompts, label="Type your own or select a system prompt", value="You are a helpful AI Doctor.", allow_custom_value=True)
+        chatbot = gr.ChatInterface(fn=chat, additional_inputs=[dropdown])
+
+    demo.queue(api_open=False).launch(show_api=False, share=True)

download_model.sh

@@ -0,0 +1,13 @@
+#!/bin/bash
+
+mkdir -p /app/medllama2_7b
+
+aria2c --console-log-level=error -c -x 16 -s 16 -k 1M https://huggingface.co/4bit/medllama2_7b_s/resolve/main/model-00001-of-00002.safetensors -d /app/medllama2_7b -o model-00001-of-00002.safetensors
+aria2c --console-log-level=error -c -x 16 -s 16 -k 1M https://huggingface.co/4bit/medllama2_7b_s/resolve/main/model-00002-of-00002.safetensors -d /app/medllama2_7b -o model-00002-of-00002.safetensors
+aria2c --console-log-level=error -c -x 16 -s 16 -k 1M https://huggingface.co/4bit/medllama2_7b_s/raw/main/model.safetensors.index.json -d /app/medllama2_7b -o model.safetensors.index.json
+aria2c --console-log-level=error -c -x 16 -s 16 -k 1M https://huggingface.co/4bit/medllama2_7b_s/raw/main/special_tokens_map.json -d /app/medllama2_7b -o special_tokens_map.json
+aria2c --console-log-level=error -c -x 16 -s 16 -k 1M https://huggingface.co/4bit/medllama2_7b_s/raw/main/tokenizer.json -d /app/medllama2_7b -o tokenizer.json
+aria2c --console-log-level=error -c -x 16 -s 16 -k 1M https://huggingface.co/4bit/medllama2_7b_s/raw/main/tokenizer_config.json -d /app/medllama2_7b -o tokenizer_config.json
+aria2c --console-log-level=error -c -x 16 -s 16 -k 1M https://huggingface.co/4bit/medllama2_7b_s/raw/main/config.json -d /app/medllama2_7b -o config.json
+aria2c --console-log-level=error -c -x 16 -s 16 -k 1M https://huggingface.co/4bit/medllama2_7b_s/raw/main/generation_config.json -d /app/medllama2_7b -o generation_config.json
+aria2c --console-log-level=error -c -x 16 -s 16 -k 1M https://huggingface.co/4bit/medllama2_7b_s/resolve/main/tokenizer.model -d /app/medllama2_7b -o tokenizer.model

makefile

@@ -0,0 +1,20 @@
+
+
+PYTHON_BINARY = $(VENV)/python
+STREAMLIT_BINARY = $(VENV)/streamlit
+
+
+re-venv: clean-venv venv
+
+run: venv
+	$(PYTHON_BINARY) app.py
+	$(STREAMLIT_BINARY) run app.py
+
+build:
+	docker build -t medllama2_7b .
+
+run-docker:
+    docker run -p 4000:80 medllama2_7b
+
+
+include Makefile.venv

requirements.txt

@@ -0,0 +1,5 @@
+gradio
+transformers
+accelerate
+bitsandbytes
+peft