add experimental and wip worker containers

runpod/runpod-worker-sd/README.md

@@ -1,11 +1,16 @@
 # NOT (YET) WORKING
 
+uses (containers/)serverless-automatic instead
+
+## old info serverless-ckpt-template
+```
 git clone https://github.com/runpod/serverless-ckpt-template.git
 cd serverless-ckpt-template
 
 docker build --build-arg MODEL_URL=https://huggingface.co/hassanblend/HassanBlend1.5.1.2 -t magn418/runpod-hassan:1.5 .
 docker login
 docker push magn418/runpod-hassan:1.5
+```
 
 Models:
 PFG https://civitai.com/models/1227/pfg

runpod/runpod-worker-transformers/Dockerfile

@@ -1,46 +1,54 @@
 ARG BASE_IMAGE=nvidia/cuda:11.6.2-cudnn8-devel-ubuntu20.04
 FROM ${BASE_IMAGE} as dev-base
 
-ARG MODEL_NAME
+WORKDIR /
-ENV MODEL_NAME=${MODEL_NAME}
-
-WORKDIR /src
-
 SHELL ["/bin/bash", "-o", "pipefail", "-c"]
 ENV DEBIAN_FRONTEND noninteractive\
     SHELL=/bin/bash
-RUN apt-key del 7fa2af80
+
-RUN apt-key adv --fetch-keys https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64/3bf863cc.pub
 RUN apt-get update --yes && \
+    # - apt-get upgrade is run to patch known vulnerabilities in apt-get packages as
+    #   the ubuntu base image is rebuilt too seldom sometimes (less than once a month)
     apt-get upgrade --yes && \
     apt install --yes --no-install-recommends \
+    build-essential \
+    ca-certificates \
+    git \
+    git-lfs \
     wget \
+    curl \
     bash \
+    libgl1 \
+    software-properties-common \
     openssh-server && \
     apt-get clean && rm -rf /var/lib/apt/lists/* && \
     echo "en_US.UTF-8 UTF-8" > /etc/locale.gen
 
-RUN apt-get update && apt-get install -y --no-install-recommends
+RUN apt-key del 7fa2af80 && \
-RUN apt-get install software-properties-common -y
+    apt-key adv --fetch-keys https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64/3bf863cc.pub
-RUN add-apt-repository ppa:deadsnakes/ppa
+
-RUN apt-get install python3.8 -y
+RUN add-apt-repository ppa:deadsnakes/ppa && \
-RUN apt-get install python3-pip -y
+    apt-get install python3.10 python3.10-dev python3.10-venv python3-pip -y --no-install-recommends && \
-RUN apt-get install python3.8-distutils -y
+    apt-get clean && rm -rf /var/lib/apt/lists/*
 
-RUN apt-get install python3.8-dev -y
+RUN pip install --upgrade pip && \
-RUN apt-get install python3.8-venv -y
+    pip install huggingface-hub && \
-RUN python3.8 -m venv /venv
+    pip install diffusers && \
-ENV PATH=/venv/bin:$PATH
+    pip install safetensors && \
+    pip install torch torchvision torchaudio --extra-index-url=https://download.pytorch.org/whl/cu116 && \
+    pip install bitsandbytes && \
+    pip install transformers accelerate xformers triton && \
+    pip install runpod
 
-RUN python3.8 -m pip install --upgrade pip==23.0.1
+RUN mkdir /workspace
-RUN python3.8 -m pip install runpod==0.9.1
+WORKDIR /workspace
-RUN python3.8 -m pip install torch==2.0.0
-RUN python3.8 -m pip install transformers==4.27.2
 
-COPY model_fetcher.py /src/model_fetcher.py
+COPY model_fetcher.py /workspace/
-RUN python3.8 model_fetcher.py --model_name=${MODEL_NAME}
+RUN python model_fetcher.py --model_name=${MODEL_NAME}
+#RUN git lfs install && \
+#    git clone --depth 1 https://huggingface.co/${MODEL_NAME}
 
-COPY runpod_infer.py /src/runpod_infer.py
+COPY runpod_infer.py /workspace/
-COPY test_input.json /src/test_input.json
+COPY test_input.json /workspace/
 
-CMD python3.8 -u runpod_infer.py --model_name=${MODEL_NAME}
+CMD python -u runpod_infer.py --model_name=${MODEL_NAME}

runpod/runpod-worker-transformers/README.md

@@ -11,3 +11,14 @@ Select one of the following models to build:
 ```BASH
 docker build --build-arg MODEL_NAME={model name} -t repo/image_name:tag .
 ```
+
+```BASH
+pip install --upgrade pip
+git clone https://github.com/AlpinDale/gptq-gptj.git
+cd gptq-gptj
+pip install -r requirements.txt
+CUDA_VISIBLE_DEVICES=0 python gptj.py PygmalionAI/pygmalion-6b c4 --wbits 4 --groupsize 128 --save pygmalion-6b-8bit-128g.pt
+CUDA_VISIBLE_DEVICES=0 python gptj.py PygmalionAI/pygmalion-6b c4 --wbits 4 --groupsize 128 --save_safetensors pygmalion-6b-8bit-128g.safetensors
+scp -P 22023 pygmalion-6b-8bit-128g.safetensors will@xd0.de:/home/will/
+scp -P 22023 pygmalion-6b-8bit-128g.safetensors will@xd0.de:/home/will/
+```

runpod/runpod-worker-transformers/model_fetcher.py

@@ -38,15 +38,30 @@ def download_model(model_name):
         AutoTokenizer.from_pretrained("EleutherAI/gpt-j-6B")
 
     # ------------------------------ PPO Shygmalion 6B ----------------------------- #
-    elif model_name == 'shygmalion-6b':
+    elif model_name == 'ppo-shygmalion-6b':
         AutoModelForCausalLM.from_pretrained("TehVenom/PPO_Shygmalion-6b")
         AutoTokenizer.from_pretrained("TehVenom/PPO_Shygmalion-6b")
 
+    # ------------------------------ Dolly Shygmalion 6B ----------------------------- #
+    elif model_name == 'dolly-shygmalion-6b':
+        AutoModelForCausalLM.from_pretrained("TehVenom/Dolly_Shygmalion-6b")
+        AutoTokenizer.from_pretrained("TehVenom/Dolly_Shygmalion-6b")
+
     # ------------------------------ Erebus 13B (NSFW) ----------------------------- #
     elif model_name == 'erebus-13b':
         AutoModelForCausalLM.from_pretrained("KoboldAI/OPT-13B-Erebus")
         AutoTokenizer.from_pretrained("KoboldAI/OPT-13B-Erebus")
 
+    # --------------------------- Alpaca 13B (Quantized) -------------------------- #
+    elif model_name == 'gpt4-x-alpaca-13b-native-4bit-128g':
+        AutoModelForCausalLM.from_pretrained("anon8231489123/gpt4-x-alpaca-13b-native-4bit-128g")
+        AutoTokenizer.from_pretrained("anon8231489123/gpt4-x-alpaca-13b-native-4bit-128g")
+
+    # --------------------------------- Alpaca 13B -------------------------------- #
+    elif model_name == 'gpt4-x-alpaca':
+        AutoModelForCausalLM.from_pretrained("chavinlo/gpt4-x-alpaca")
+        AutoTokenizer.from_pretrained("chavinlo/gpt4-x-alpaca")
+
 
 # ---------------------------------------------------------------------------- #
 #                                Parse Arguments                               #

runpod/runpod-worker-transformers/runpod_infer.py

@@ -7,7 +7,8 @@ import torch
 import runpod
 from runpod.serverless.utils.rp_validator import validate
 from transformers import (GPTNeoForCausalLM, GPT2Tokenizer, GPTNeoXForCausalLM,
-                          GPTNeoXTokenizerFast, GPTJForCausalLM, AutoTokenizer, AutoModelForCausalLM)
+                          GPTNeoXTokenizerFast, GPTJForCausalLM, AutoTokenizer, AutoModelForCausalLM,
+                          LlamaForCausalLM, LlamaTokenizer)
 
 
 torch.cuda.is_available()
@@ -174,16 +175,34 @@ if __name__ == "__main__":
         tokenizer = AutoTokenizer.from_pretrained(
             "EleutherAI/gpt-j-6B", local_files_only=True)
 
-    elif args.model_name == 'shygmalion-6b':
+    elif args.model_name == 'ppo-shygmalion-6b':
         model = AutoModelForCausalLM.from_pretrained(
             "TehVenom/PPO_Shygmalion-6b", local_files_only=True).to(device)
         tokenizer = AutoTokenizer.from_pretrained(
             "TehVenom/PPO_Shygmalion-6b", local_files_only=True)
 
+    elif args.model_name == 'dolly-shygmalion-6b':
+        model = AutoModelForCausalLM.from_pretrained(
+            "TehVenom/Dolly_Shygmalion-6b", local_files_only=True).to(device)
+        tokenizer = AutoTokenizer.from_pretrained(
+            "TehVenom/Dolly_Shygmalion-6b", local_files_only=True)
+
     elif args.model_name == 'erebus-13b':
         model = AutoModelForCausalLM.from_pretrained(
             "KoboldAI/OPT-13B-Erebus", local_files_only=True).to(device)
         tokenizer = AutoTokenizer.from_pretrained(
             "KoboldAI/OPT-13B-Erebus", local_files_only=True)
 
+    elif args.model_name == 'gpt4-x-alpaca-13b-native-4bit-128g':
+        model = LlamaForCausalLM.from_pretrained(
+            "anon8231489123/gpt4-x-alpaca-13b-native-4bit-128g", local_files_only=True).to(device)
+        tokenizer = LlamaTokenizer.from_pretrained(
+            "anon8231489123/gpt4-x-alpaca-13b-native-4bit-128g", local_files_only=True)
+
+    elif args.model_name == 'gpt4-x-alpaca':
+        model = LlamaForCausalLM.from_pretrained(
+            "chavinlo/gpt4-x-alpaca", local_files_only=True).to(device)
+        tokenizer = LlamaTokenizer.from_pretrained(
+            "chavinlo/gpt4-x-alpaca", local_files_only=True)
+
     runpod.serverless.start({"handler": generator})