Move python code under tabby/ (#8)

* Add tabby config file

* Rename train.yaml to trainer.yaml

* Change server to relative import

* Move source files into tabby

* Rename conf

.github/workflows/docker.yml

@@ -55,7 +55,7 @@ jobs:
         id: build-and-push
         uses: docker/build-push-action@v3.1.1
         with:
-          file: deployment/Dockerfile
+          file: Dockerfile
           context: .
           push: ${{ github.event_name != 'pull_request' }}
           tags: |

Makefile

@@ -15,7 +15,7 @@ $(PRE_COMMIT_HOOK):
 	poetry run pre-commit install --install-hooks
 
 $(LOCAL_MODEL):
-	poetry run python scripts/huggingface_gptneox_convert.py \
+	poetry run python -m tabby.tools.huggingface_gptneox_convert \
 		-in_file EleutherAI/pythia-70m-deduped \
 		-o $@ \
 		-i_g 1 -m_n tiny-70M -p 1 -w fp16

deployment/config/dags/train.yaml

@@ -1,20 +0,0 @@
-schedule: "5 4 * * *" # Run at 04:05.
-
-env:
-  - APP_DIR: /app
-steps:
-  - name: Create job dir
-    command: mktemp -d /tmp/job.XXXXX
-    output: JOB_DIR
-
-  - name: Fetch source code
-    dir: $APP_DIR
-    command: python scripts/download_project.py --url http://download.savannah.gnu.org/releases/tinycc/tcc-0.9.24.tar.bz2 --output_dir $JOB_DIR/repository
-    depends:
-      - Create job dir
-
-  - name: Generate dataset
-    dir: ${APP_DIR}
-    command: python scripts/preprocess/preprocess_project.py --project_dir $JOB_DIR/repository --output_dir $JOB_DIR/dataset
-    depends:
-      - Fetch source code

deployment/config/dags/trainer.yaml

@@ -0,0 +1,20 @@
+schedule: "5 4 * * *" # Run at 04:05.
+
+env:
+  - APP_DIR: /app
+steps:
+  - name: Create job dir
+    command: mktemp -d /tmp/job.XXXXX
+    output: JOB_DIR
+
+  - name: Fetch source code
+    dir: $APP_DIR
+    command: python -m tabby.tools.download_project --url http://download.savannah.gnu.org/releases/tinycc/tcc-0.9.24.tar.bz2 --output_dir $JOB_DIR/repository
+    depends:
+      - Create job dir
+
+  - name: Generate dataset
+    dir: $APP_DIR
+    command: python -m tabby.tools.preprocess.preprocess_project --project_dir $JOB_DIR/repository --output_dir $JOB_DIR/dataset
+    depends:
+      - Fetch source code

deployment/docker-compose.dev.yml

@@ -4,15 +4,14 @@ services:
   server:
     build:
       context: ..
-      dockerfile: ./deployment/Dockerfile
       args:
         PYPI_INDEX_URL: https://pypi.tuna.tsinghua.edu.cn/simple
-    command: uvicorn app:app --host 0.0.0.0 --port 5000 --reload
+    command: uvicorn tabby.server:app --host 0.0.0.0 --port 5000 --reload
     volumes:
       - ../:/app
 
   admin:
-    command: streamlit run --server.runOnSave=true Home.py
+    command: streamlit run --server.runOnSave=true tabby/admin/Home.py
     volumes:
       - ../:/app
 

deployment/docker-compose.yml

@@ -4,8 +4,7 @@ services:
   server:
     image: tabbyml/tabby
     container_name: tabby-server
-    working_dir: /app/server
-    command: uvicorn app:app --host 0.0.0.0 --port 5000
+    command: uvicorn tabby.server:app --host 0.0.0.0 --port 5000
     environment:
       - PYTHON_MODEL_NAME=EleutherAI/pythia-70m-deduped
       - EVENTS_LOG_DIR=/logs/tabby-server
@@ -14,14 +13,11 @@ services:
     volumes:
       - ./data/logs:/logs
       - ./data/hf_cache:/root/.cache/huggingface
-    links:
-      - vector
 
   admin:
     image: tabbyml/tabby
     container_name: tabby-admin
-    working_dir: /app/admin
-    command: streamlit run Home.py
+    command: streamlit run tabby/admin/Home.py
     ports:
       - "8501:8501"
     links:

tabby/server/__init__.py

@@ -1,13 +1,14 @@
 import logging
 import os
 
-import events
 import uvicorn
 from fastapi import FastAPI, Response
 from fastapi.responses import JSONResponse
-from models import CompletionRequest, CompletionResponse
-from python import PythonModelService
-from triton import TritonService
+
+from . import events
+from .models import CompletionRequest, CompletionResponse
+from .python import PythonModelService
+from .triton import TritonService
 
 app = FastAPI(
     title="TabbyServer",
@@ -38,7 +39,3 @@ async def completions(request: CompletionRequest) -> CompletionResponse:
 async def selection(id: str, index: int) -> JSONResponse:
     events.log_selection(id, index)
     return JSONResponse(content="ok")
-
-
-if __name__ == "__main__":
-    uvicorn.run("app:app", host="0.0.0.0", port=5000)

tabby/server/events.py

@@ -3,10 +3,11 @@ import os
 import shutil
 from typing import List
 
-import models
 from loguru import logger
 from pydantic import BaseModel
 
+from . import models
+
 logger.configure(handlers=[])
 
 

tabby/server/python.py

@@ -3,9 +3,10 @@ import string
 import time
 from typing import List
 
-from models import Choice, CompletionRequest, CompletionResponse
 from transformers import AutoModelForCausalLM, AutoTokenizer
 
+from .models import Choice, CompletionRequest, CompletionResponse
+
 
 class PythonModelService:
     def __init__(

tabby/server/triton.py

@@ -6,10 +6,11 @@ from typing import List
 
 import numpy as np
 import tritonclient.grpc as client_util
-from models import Choice, CompletionRequest, CompletionResponse
 from transformers import AutoTokenizer
 from tritonclient.utils import InferenceServerException, np_to_triton_dtype
 
+from .models import Choice, CompletionRequest, CompletionResponse
+
 
 class TritonService:
     def __init__(

tabby/tools/download_project.py

@@ -6,7 +6,7 @@ import requests
 from transformers import HfArgumentParser
 
 
-@dataclass(kw_only=True)
+@dataclass
 class DownloaderArgs:
     url: str = field(metadata={"help": "URL to source code tar.gz file"})
     output_dir: str = field(metadata={"help": "Output save path directory"})

tabby/tools/preprocess/filters.py

@@ -1,4 +1,4 @@
-from args import FilterArgs
+from .args import FilterArgs
 
 
 def basic_filters(args: FilterArgs):

tabby/tools/preprocess/preprocess_project.py

@@ -2,12 +2,12 @@ import glob
 import json
 import os
 
-import filters
-import metrics
-from args import PreprocessProjectArgs
 from datasets import Dataset
 from transformers import HfArgumentParser
 
+from . import filters, metrics
+from .args import PreprocessProjectArgs
+
 
 def parse_args():
     parser = HfArgumentParser(PreprocessProjectArgs)