repo_name
stringlengths
1
62
dataset
stringclasses
1 value
lang
stringclasses
11 values
pr_id
int64
1
20.1k
owner
stringlengths
2
34
reviewer
stringlengths
2
39
diff_hunk
stringlengths
15
262k
code_review_comment
stringlengths
1
99.6k
llm-on-ray
github_2023
others
84
intel
xwu99
@@ -0,0 +1,56 @@ +#!/bin/bash +set -eo pipefail + +# Step 1: Python environment +# Check Python version is or later than 3.9 +echo "Step 1: Python environment" +echo "Checking Python version which should be equal or later than 3.9" +if ! python -c 'import sys; assert sys.version_info >= (3,9)' > /dev/null; then + ex...
@Deegue It looks to me, there are some duplications for getting_started and setup for the setup process, is it better we can consolidate them into single one? Also may I suggest to use bash functions for setup and step 1,2,3 for better structure and code reuse. for setup function, you can define arguments for differ...
llm-on-ray
github_2023
others
84
intel
xwu99
@@ -0,0 +1,66 @@ +#!/bin/bash +set -eo pipefail + +# Step 1: Python environment +# Check Python version is or later than 3.9 +echo "Step 1: Python environment" +echo "Checking Python version which should be equal or later than 3.9" +if ! python -c 'import sys; assert sys.version_info >= (3,9)' > /dev/null; then + ex...
L16-L18 has been updated. pls check README.md pip install .[cpu] --extra-index-url https://download.pytorch.org/whl/cpu --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/cpu/us/
llm-on-ray
github_2023
others
84
intel
xwu99
@@ -0,0 +1,66 @@ +#!/bin/bash +set -eo pipefail + +# Step 1: Python environment +# Check Python version is or later than 3.9 +echo "Step 1: Python environment" +echo "Checking Python version which should be equal or later than 3.9" +if ! python -c 'import sys; assert sys.version_info >= (3,9)' > /dev/null; then + ex...
```suggestion source $(python -c "import oneccl_bindings_for_pytorch as torch_ccl; print(torch_ccl.cwd)")/env/setvars.sh ```
llm-on-ray
github_2023
others
84
intel
xwu99
@@ -0,0 +1,66 @@ +#!/bin/bash +set -eo pipefail + +# Step 1: Python environment +# Check Python version is or later than 3.9 +echo "Step 1: Python environment" +echo "Checking Python version which should be equal or later than 3.9" +if ! python -c 'import sys; assert sys.version_info >= (3,9)' > /dev/null; then + ex...
```suggestion echo "Step 4: Access OpenAI API" ```
llm-on-ray
github_2023
others
84
intel
xwu99
@@ -0,0 +1,66 @@ +#!/bin/bash +set -eo pipefail + +# Step 1: Python environment +# Check Python version is or later than 3.9 +echo "Step 1: Python environment" +echo "Checking Python version which should be equal or later than 3.9" +if ! python -c 'import sys; assert sys.version_info >= (3,9)' > /dev/null; then + ex...
```suggestion echo "Method 1: Using curl to access model" ``` same for the following
llm-on-ray
github_2023
others
84
intel
xwu99
@@ -8,6 +8,84 @@ on: default: 'pr' jobs: + setup-test: + + name: setup-test + strategy: + matrix: + python-version: ["3.9", "3.10", "3.11"] + + runs-on: ubuntu-latest + defaults: + run: + shell: bash + + steps: + - name: Checkout + uses: actions/checkout@v...
```suggestion # Additional libraries required for pytest ```
llm-on-ray
github_2023
others
84
intel
xwu99
@@ -8,6 +8,84 @@ on: default: 'pr' jobs: + setup-test: + + name: setup-test + strategy: + matrix: + python-version: ["3.9", "3.10", "3.11"] + + runs-on: ubuntu-latest + defaults: + run: + shell: bash + + steps: + - name: Checkout + uses: actions/checkout@v...
```suggestion # Additional libraries required for pytest ```
llm-on-ray
github_2023
others
84
intel
xwu99
@@ -0,0 +1,66 @@ +#!/bin/bash +set -eo pipefail + +# Step 1: Python environment +# Check Python version is or later than 3.9 +echo "Step 1: Python environment" +echo "Checking Python version which should be equal or later than 3.9" +if ! python -c 'import sys; assert sys.version_info >= (3,9)' > /dev/null; then + ex...
```suggestion echo "Starting ray server for gpt2 with 1 cpu per worker" ```
llm-on-ray
github_2023
others
84
intel
xwu99
@@ -0,0 +1,62 @@ +#!/bin/bash +set -eo pipefail + +# Usage: ./test_setup [CPU, GPU, Gaudi] [True for Deepspeed and false for disable] + +if [ "$#" != 2 ]; then + echo "Error, there should be 2 arguments! See Usage: ./test_setup [CPU, GPU, Gaudi] [True for Deepspeed and false for disable]" + exit 1 +fi + +# Step 1...
```suggestion # Step 4: Check if it is installed correctly ```
llm-on-ray
github_2023
others
92
intel
xwu99
@@ -6,7 +6,6 @@ on: ci_type: type: string default: 'pr' -
it's OK to leave a blank line here.
llm-on-ray
github_2023
python
92
intel
xwu99
@@ -0,0 +1,139 @@ +import subprocess +import pytest + +# Config matrix +# config_file_array = ["inference/models/gpt2.yaml", None] +# model_id_or_path_array = ["gpt2", None] +# models_array = ["gpt2", "gpt2 gpt-j-6b", "gpt2 bloom-560m falcon-7b"] +# port_array = [8000, None] +# route_prefix_array = [None] +# cpus_per_w...
It's better to print out the result if we want to check it in the CI before we find a better way to assert.
llm-on-ray
github_2023
python
92
intel
xwu99
@@ -0,0 +1,125 @@ +import subprocess +import pytest + +# Config matrix +# config_file_array = ["inference/models/gpt2.yaml", None] +# model_id_or_path_array = ["gpt2", None] +# models_array = ["gpt2", "gpt2 gpt-j-6b", "gpt2 bloom-560m falcon-7b"] +# port_array = [8000, None] +# route_prefix_array = [None] +# cpus_per_w...
```suggestion print("Output of stderr:") ```
llm-on-ray
github_2023
python
92
intel
xwu99
@@ -0,0 +1,125 @@ +import subprocess +import pytest + +# Config matrix +# config_file_array = ["inference/models/gpt2.yaml", None] +# model_id_or_path_array = ["gpt2", None] +# models_array = ["gpt2", "gpt2 gpt-j-6b", "gpt2 bloom-560m falcon-7b"] +# port_array = [8000, None] +# route_prefix_array = [None] +# cpus_per_w...
```suggestion ```
llm-on-ray
github_2023
others
83
intel
xwu99
@@ -1,13 +1,34 @@ name: tests on: - workflow_call + workflow_call: + inputs: + ci_type: + type: string + default: 'pr' + no_proxy: + type: string + default: 'localhost,127.0.0.1' + OPENAI_API_BASE: + type: string + default: 'http://localhost:8000/v1' + ...
no need. from Checkout action below to get checkout path. It should checks out your repository under $GITHUB_WORKSPACE
llm-on-ray
github_2023
others
83
intel
xwu99
@@ -1,13 +1,34 @@ name: tests on: - workflow_call + workflow_call: + inputs: + ci_type: + type: string + default: 'pr' + no_proxy: + type: string + default: 'localhost,127.0.0.1' + OPENAI_API_BASE:
don't define test related constants here, should define elsewhere
llm-on-ray
github_2023
others
83
intel
xwu99
@@ -1,13 +1,34 @@ name: tests on: - workflow_call + workflow_call: + inputs: + ci_type: + type: string + default: 'pr' + no_proxy: + type: string + default: 'localhost,127.0.0.1' + OPENAI_API_BASE: + type: string + default: 'http://localhost:8000/v1' + ...
This is part of the test, it's better to stay with the test code.
llm-on-ray
github_2023
others
83
intel
xwu99
@@ -23,13 +44,79 @@ jobs: architecture: 'x64' - name: Display Python version - run: python -c "import sys; print(sys.version)" + run: | + python -c "import sys; print(sys.version)" + bash -c "ls" - name: Install dependencies
```suggestion - name: Install dependencies for tests ```
llm-on-ray
github_2023
others
83
intel
xwu99
@@ -23,13 +44,79 @@ jobs: architecture: 'x64' - name: Display Python version - run: python -c "import sys; print(sys.version)" + run: | + python -c "import sys; print(sys.version)" + bash -c "ls" - name: Install dependencies run: | python ...
what is the reason for using bash -c instead of direct calling?
llm-on-ray
github_2023
others
83
intel
xwu99
@@ -1,13 +1,34 @@ name: tests on: - workflow_call + workflow_call: + inputs: + ci_type: + type: string + default: 'pr' + no_proxy: + type: string + default: 'localhost,127.0.0.1' + OPENAI_API_BASE: + type: string + default: 'http://localhost:8000/v1' + ...
consider change the name build/build-docker to sth else such as bare-test/docker-test?
llm-on-ray
github_2023
others
83
intel
xwu99
@@ -23,13 +44,79 @@ jobs: architecture: 'x64'
could we add test matrix to python-version that 3.9, 3.10, 3.11 (supported by IPEX) will all be tested?
llm-on-ray
github_2023
others
83
intel
xwu99
@@ -23,13 +44,79 @@ jobs: architecture: 'x64' - name: Display Python version - run: python -c "import sys; print(sys.version)" + run: | + python -c "import sys; print(sys.version)" + bash -c "ls" - name: Install dependencies run: | python ...
Could we move all docker related blocks for all workflow actions into functions in some bash script and source from them? It can reduce code duplications and easier to maintain.
llm-on-ray
github_2023
others
83
intel
xwu99
@@ -1,13 +1,31 @@ name: tests on: - workflow_call + workflow_call: + inputs: + ci_type: + type: string + default: 'pr' jobs: - build: - + bare-test: + + name: bare-test + strategy: + matrix: + python-version: [3.9, 3.10 , 3.11]
```suggestion python-version: [3.9, 3.10, 3.11] ```
llm-on-ray
github_2023
others
83
intel
xwu99
@@ -19,17 +37,80 @@ jobs: - name: Set up Python uses: actions/setup-python@v4 with: - python-version: '3.9' + python-version: ${{matrix.python-version}} architecture: 'x64' - name: Display Python version - run: python -c "import sys; print(sys.versio...
Pls check README.md for latest update for installing ```suggestion pip install .[cpu] --extra-index-url https://download.pytorch.org/whl/cpu --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/cpu/us/ ```
llm-on-ray
github_2023
others
83
intel
xwu99
@@ -19,17 +37,80 @@ jobs: - name: Set up Python uses: actions/setup-python@v4 with: - python-version: '3.9' + python-version: ${{matrix.python-version}} architecture: 'x64' - name: Display Python version - run: python -c "import sys; print(sys.versio...
```suggestion source $(python -c "import oneccl_bindings_for_pytorch as torch_ccl; print(torch_ccl.cwd)")/env/setvars.sh ```
llm-on-ray
github_2023
others
83
intel
xwu99
@@ -1,7 +1,10 @@ #!/bin/bash +set -eo pipefail cd $(dirname $0) +pip install -r ./requirements.txt
consider separate out installing requirements in run-test.sh since it only needs to do once. Suggest to do this in "Install dependencies for tests" of the workflow_tests.yml.
llm-on-ray
github_2023
others
83
intel
xwu99
@@ -19,17 +37,80 @@ jobs: - name: Set up Python uses: actions/setup-python@v4 with: - python-version: '3.9' + python-version: ${{matrix.python-version}} architecture: 'x64' - name: Display Python version - run: python -c "import sys; print(sys.versio...
```suggestion docker-test: ```
llm-on-ray
github_2023
others
83
intel
xwu99
@@ -19,17 +37,80 @@ jobs: - name: Set up Python uses: actions/setup-python@v4 with: - python-version: '3.9' + python-version: ${{matrix.python-version}} architecture: 'x64' - name: Display Python version - run: python -c "import sys; print(sys.versio...
```suggestion - name: Run Tests ```
llm-on-ray
github_2023
python
83
intel
xwu99
@@ -0,0 +1,63 @@ +import subprocess +import pytest + + +def script_with_args(model_name, streaming_response, max_new_tokens, temperature, top_p): + config_path = "../.github/workflows/config/" + model_name + ".yaml"
suggest to get current script absolute path as base path and set other path relative to it, then this script will be independent from where your run from.
llm-on-ray
github_2023
python
83
intel
xwu99
@@ -0,0 +1,78 @@ +import subprocess +import pytest +import os + +os.environ["no_proxy"] = "localhost,127.0.0.1" +os.environ["OPENAI_API_BASE"] = "http://localhost:8000/v1" +os.environ["OPENAI_API_KEY"] = "YOUR_OPEN_AI_KEY" +os.environ["OPENAI_BASE_URL"] = "http://localhost:8000/v1" + + +def script_with_args(api_base, m...
same again, it's ok to just put the list here to shorten code such as for api_base in ["http://localhost:8000/v1"] same for other cases.
llm-on-ray
github_2023
python
83
intel
xwu99
@@ -0,0 +1,63 @@ +import subprocess +import pytest + + +def script_with_args(model_name, streaming_response, max_new_tokens, temperature, top_p): + config_path = "../.github/workflows/config/" + model_name + ".yaml" + + cmd_serve = ["python", "../inference/serve.py", "--config_file", config_path] + + result_se...
Just use constant list to fill the argument directly is OK
llm-on-ray
github_2023
others
83
intel
xwu99
@@ -1,13 +1,31 @@ name: tests on: - workflow_call + workflow_call: + inputs: + ci_type: + type: string + default: 'pr' jobs: - build: - + bare-test: + + name: bare-test + strategy: + matrix: + python-version: [3.9, 3.10, 3.11] + isPR:
I think you just need to use L16, L17-L26 are unnecessary.
llm-on-ray
github_2023
others
83
intel
xwu99
@@ -19,17 +37,100 @@ jobs: - name: Set up Python uses: actions/setup-python@v4 with: - python-version: '3.9' + python-version: ${{matrix.python-version}} architecture: 'x64' - name: Display Python version - run: python -c "import sys; print(sys.versi...
```suggestion - name: Run Tests ```
llm-on-ray
github_2023
python
83
intel
xwu99
@@ -0,0 +1,79 @@ +import subprocess +import pytest +import os + +os.environ["no_proxy"] = "localhost,127.0.0.1" +os.environ["OPENAI_API_BASE"] = "http://localhost:8000/v1" +os.environ["OPENAI_API_KEY"] = "YOUR_OPEN_AI_KEY" +os.environ["OPENAI_BASE_URL"] = "http://localhost:8000/v1" + + +def script_with_args(api_base, m...
Could you also print the output of `subprocess.run` so that we can check if the output is expected?
llm-on-ray
github_2023
python
106
intel
xwu99
@@ -3,7 +3,7 @@ def update_finetune_config(base_model): - conf_file = "finetune/finetune.yaml" + conf_file = "llmonray/finetune/finetune.yaml"
`llmonray` is cluttered and hard to read. use `llm_on_ray` instead just like intel_extension_for_pytorch
llm-on-ray
github_2023
others
106
intel
xwu99
@@ -113,14 +113,14 @@ jobs: EOF ) docker exec "finetune" python -c "$CMD" - docker exec "finetune" bash -c "python finetune/finetune.py --config_file finetune/finetune.yaml" + docker exec "finetune" bash -c "python -m llmonray.finetune.finetune --config_file llmonray/f...
llmonray.finetune.finetune is verbose. Use module llm_on_ray.finetune, may need to expose interfaces in module `__init__.py` and run from it. Another better way is to use setup.py to install a command into bin and call `llm_on_ray-finetune` comand line instead, you can check how to do this from rayllm.
llm-on-ray
github_2023
others
106
intel
xwu99
@@ -123,27 +123,27 @@ Set up `megatron_deepspeed_path` in the configuration. ```bash cd /home/user/workspace/llm-on-ray #Bloom-7B
add extra space after Bloom-7B, same below ```suggestion # Bloom-7B ```
llm-on-ray
github_2023
others
106
intel
xwu99
@@ -23,7 +23,7 @@ Please follow [Deploying and Serving LLMs on Intel CPU/GPU/Gaudi](serve.md) docu To serve model with vLLM, run the following: ```bash -$ python serve.py --config_file inference/models/vllm/llama-2-7b-chat-hf-vllm.yaml --simple --keep_serve_terminal +$ llm_on_ray-serve --config_file llm_on_ray/infe...
remove leading $, same below ```suggestion llm_on_ray-serve --config_file llm_on_ray/inference/models/vllm/llama-2-7b-chat-hf-vllm.yaml --simple --keep_serve_terminal ```
llm-on-ray
github_2023
others
106
intel
xwu99
@@ -14,7 +14,7 @@ $ dev/scripts/install-ui.sh ## Start Web UI
need to remove leading $ to be consistent for above lines: dev/scripts/install-ui.sh
llm-on-ray
github_2023
others
106
intel
xwu99
@@ -27,7 +27,7 @@ RUN --mount=type=cache,target=/opt/conda/pkgs conda init bash && \ COPY ./pyproject.toml . COPY ./MANIFEST.in . -RUN mkdir ./finetune && mkdir ./inference +RUN mkdir ./llm_on_ray
You map the source code path to /root/llm-on-ray directory in workflow but create llm_on_ray here, what is the intention for this directory?
llm-on-ray
github_2023
python
106
intel
xwu99
@@ -2,7 +2,7 @@ import glob import importlib -from .logging import logger +from llm_on_ray.common.logging import logger def import_all_module(basedir, prefix=None):
```suggestion def import_all_modules(basedir, prefix=None): ```
llm-on-ray
github_2023
python
106
intel
xwu99
@@ -1,5 +1,5 @@ import torch # noqa: F401 -from .optimizer import Optimizer +from llm_on_ray.common.optimizer.optimizer import Optimizer
Did you already expose Optimizer in `__all__` of `__init__.py`? so that the code can be written in below style: ```suggestion from llm_on_ray.common.optimizer import Optimizer ``` There are many similar cases need to fix.
llm-on-ray
github_2023
python
106
intel
xwu99
@@ -1,4 +1,4 @@ -from .tokenizer import Tokenizer +from llm_on_ray.common.tokenizer.tokenizer import Tokenizer
same, try to reduce it to this style by importing package rather than module ```suggestion from llm_on_ray.common.tokenizer import Tokenizer ```
llm-on-ray
github_2023
python
106
intel
xwu99
@@ -58,16 +58,16 @@ def __init__(self, infer_conf: InferenceConfig): self.use_vllm = infer_conf.vllm.enabled if self.use_deepspeed: - from deepspeed_predictor import DeepSpeedPredictor + from llm_on_ray.inference.deepspeed_predictor import DeepSpeedPredictor self...
try to reduce to import the package by adding TransformerPredictor in `__all__` ```suggestion from llm_on_ray.inference import TransformerPredictor ```
llm-on-ray
github_2023
python
106
intel
xwu99
@@ -0,0 +1,9 @@ +import os +from llm_on_ray.common.agentenv.agentenv import AgentEnv +from llm_on_ray.common.common import import_all_modules + +realpath = os.path.realpath(__file__) +basedir = os.path.dirname(realpath) +import_all_modules(basedir, "llm_on_ray.common.agentenv")
I don't this it a good idea to import all modules in `__init__.py`, this is a declare of public and private components of the package. Only needed modules/classes should be imported and used as: `from package_name import xxx`
llm-on-ray
github_2023
others
106
intel
xwu99
@@ -113,14 +113,14 @@ jobs: EOF ) docker exec "finetune" python -c "$CMD" - docker exec "finetune" bash -c "python finetune/finetune.py --config_file finetune/finetune.yaml" + docker exec "finetune" bash -c "llm_on_ray-finetune --config_file llm_on_ray/finetune/finetu...
```suggestion docker exec "finetune" bash -c "llm_on_ray-finetune --config_file llm_on_ray/finetune/finetune.yaml" ```
llm-on-ray
github_2023
others
94
intel
harborn
@@ -0,0 +1,49 @@ +{
I think we should unifiy json format to yaml, and rename this conf to `ds_config_zero2.yaml`
llm-on-ray
github_2023
python
94
intel
harborn
@@ -77,10 +84,26 @@ def train_func(config: Dict[str, Any]): offload_to_cpu=False, rank0_only=False ), ) + deepspeed_plugin = None + + elif accelerate_mode in ["GPU_DEEPSPEED"]: + fsdp_plugin = None + hf_ds_config = config["Training"]["deepspeed_config_file"...
``` python with open(config["Training"]["deepspeed_config_file"]) as f: hf_ds_config = yaml.full_load(f) ``` here `hf_ds_config` is a dict type object, and `DeepSpeedPlugin` can accept and process a dict type hf_ds_config.
llm-on-ray
github_2023
others
125
intel
jiafuzha
@@ -38,9 +38,9 @@ jobs: - name: Running task on Intel GPU run: | - rm ~/borealis-runner/llm-on-ray.tar.gz -f - tar zcf ~/borealis-runner/llm-on-ray.tar.gz -C ~/actions-runner/_work/llm-on-ray . - cd ~/borealis-runner/ + rm /home/ci/borealis-runner/llm-on-ray.tar.gz ...
finetune_on_pvc.py should be checked in to our repo.
llm-on-ray
github_2023
python
107
intel
carsonwang
@@ -0,0 +1,131 @@ +# +# Copyright 2023 The LLM-on-Ray Authors. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required b...
Can we use a URL to this image here so it can work by default? Can you also update the help message if it can be a local path or a URL.
llm-on-ray
github_2023
python
107
intel
carsonwang
@@ -0,0 +1,131 @@ +# +# Copyright 2023 The LLM-on-Ray Authors. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required b...
Update the description to include "image".
llm-on-ray
github_2023
python
107
intel
carsonwang
@@ -0,0 +1,131 @@ +# +# Copyright 2023 The LLM-on-Ray Authors. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required b...
Any reason to for these default values?
llm-on-ray
github_2023
python
107
intel
carsonwang
@@ -0,0 +1,72 @@ +import torch +from transformers import TextIteratorStreamer +from inference.inference_config import InferenceConfig, PRECISION_BF16 +from predictor import Predictor +from inference.utils import module_import + + +class MllmPredictor(Predictor): + def __init__(self, infer_conf: InferenceConfig): + ...
Can we keep this `torch_dtype` and pass it to `from_pretrained`?
llm-on-ray
github_2023
python
107
intel
carsonwang
@@ -0,0 +1,72 @@ +import torch +from transformers import TextIteratorStreamer +from inference.inference_config import InferenceConfig, PRECISION_BF16 +from predictor import Predictor +from inference.utils import module_import + + +class MllmPredictor(Predictor): + def __init__(self, infer_conf: InferenceConfig): + ...
Remove the above commented code?
llm-on-ray
github_2023
python
107
intel
carsonwang
@@ -0,0 +1,72 @@ +import torch +from transformers import TextIteratorStreamer +from inference.inference_config import InferenceConfig, PRECISION_BF16 +from predictor import Predictor +from inference.utils import module_import + + +class MllmPredictor(Predictor): + def __init__(self, infer_conf: InferenceConfig): + ...
Can you confirm if this model is supported by IPEX or not?
llm-on-ray
github_2023
others
107
intel
carsonwang
@@ -0,0 +1,26 @@ +port: 8000 +name: deplot +route_prefix: /deplot +cpus_per_worker: 24 +gpus_per_worker: 0 +deepspeed: false +workers_per_group: 2 +device: "cpu" +ipex: + enabled: false + precision: bf16 +model_description: + model_id_or_path: /mnt/nvme0n1/chendi/llm-on-ray/models/google/deplot + tokenizer_name_or_...
Update these to ids on huggingface
llm-on-ray
github_2023
others
107
intel
carsonwang
@@ -0,0 +1,26 @@ +port: 8000 +name: fuyu-8b +route_prefix: /fuyu-8b +cpus_per_worker: 24 +gpus_per_worker: 0 +deepspeed: false +workers_per_group: 2 +device: "cpu" +ipex: + enabled: false + precision: bf16 +model_description: + model_id_or_path: /mnt/nvme0n1/chendi/llm-on-ray/models/adept/fuyu-8b + tokenizer_name_o...
Update these to ids on huggingface
llm-on-ray
github_2023
python
107
intel
carsonwang
@@ -148,12 +152,18 @@ async def __call__(self, http_request: Request) -> Union[StreamingResponse, JSON async def openai_call(self, prompt, config, streaming_response=True): prompts = [] + images = [] if isinstance(prompt, list): prompt_format = get_prompt_format(prompt) ...
get_promipt -> get_prompt
llm-on-ray
github_2023
python
107
intel
carsonwang
@@ -166,19 +176,31 @@ async def openai_call(self, prompt, config, streaming_response=True): prompts.append(prompt) if not streaming_response: + model_response = None if self.use_vllm: generate_result = (await self.predictor.generate_async(prompts, **confi...
Why not returning `GenerateResult` in `MllmPredictor.generate` like other predictors?
llm-on-ray
github_2023
python
107
intel
carsonwang
@@ -674,10 +758,14 @@ def shutdown_deploy(self): serve.shutdown() def get_ray_cluster(self): - command = "conda activate " + self.conda_env_name + "; ray status" + command = "source ~/anaconda3/bin/activate; conda activate " + self.conda_env_name + "; ray status"
We can't assume this path works in other users' environment.
llm-on-ray
github_2023
python
107
intel
carsonwang
@@ -35,17 +35,21 @@ help="Whether to enable streaming response", ) parser.add_argument( - "--max_new_tokens", default=None, help="The maximum numbers of tokens to generate" + "--max_new_tokens", default=256, help="The maximum numbers of tokens to generate" ) parser.add_argument( - "--temperature", def...
Can we revoke the default value changes in this file? If these default values work for the mllm models, it makes sense to set them in `image_query_http_requests.py` as you have done. But this file more general so let's just use openai's default value?
llm-on-ray
github_2023
python
107
intel
carsonwang
@@ -41,21 +41,47 @@ args = parser.parse_args() -client = OpenAI() -# # List all models. -models = client.models.list() -print(models.data, "\n") - -# Note: not all arguments are currently supported and will be ignored by the backend. -chat_completion = client.chat.completions.create( - model=args.model_name, - ...
Use args.xxx
llm-on-ray
github_2023
python
107
intel
carsonwang
@@ -41,21 +41,47 @@ args = parser.parse_args() -client = OpenAI() -# # List all models. -models = client.models.list() -print(models.data, "\n") - -# Note: not all arguments are currently supported and will be ignored by the backend. -chat_completion = client.chat.completions.create( - model=args.model_name, - ...
Use args.xxx
llm-on-ray
github_2023
others
107
intel
KepingYan
@@ -10,8 +10,8 @@ ipex: enabled: false precision: bf16 model_description: - model_id_or_path: meta-llama/Llama-2-7b-chat-hf - tokenizer_name_or_path: meta-llama/Llama-2-7b-chat-hf + model_id_or_path: /mnt/nvme0n1/chendi/llm-on-ray/models/meta-llama/Llama-2-7b-chat-hf + tokenizer_name_or_path: /mnt/nvme0n1/ch...
Please restore these default value.
llm-on-ray
github_2023
python
107
intel
KepingYan
@@ -166,9 +177,13 @@ async def openai_call(self, prompt, config, streaming_response=True): prompts.append(prompt) if not streaming_response: + model_response = None
It seems this value is not used.
llm-on-ray
github_2023
python
107
intel
KepingYan
@@ -795,16 +902,19 @@ def _init_ui(self): for index in range(len(self.ray_nodes)): if "node:__internal_head__" in ray.nodes()[index]["Resources"]: mark_alive = index - node_ip = self.ray_nodes[index]["NodeName"] - self.ssh_connect[index] = paramiko.SSHClient(...
Please revert this.
llm-on-ray
github_2023
python
107
intel
KepingYan
@@ -1024,22 +1134,32 @@ def _init_ui(self): label="Top k", info="The number of highest probability vocabulary tokens to keep for top-k-filtering.", ) - with gr.Tab("Dialogue"): chatbot = gr.Chatbot( ...
Is it better for the default value to be None? We can set placeholder to let users know that it can be a pre-deployed endpoint or the endpoint returned from deployment module.
llm-on-ray
github_2023
python
107
intel
KepingYan
@@ -1024,22 +1134,32 @@ def _init_ui(self): label="Top k", info="The number of highest probability vocabulary tokens to keep for top-k-filtering.", ) - with gr.Tab("Dialogue"): chatbot = gr.Chatbot( ...
Same here. Because the default value may not be deployed by users.
llm-on-ray
github_2023
python
107
intel
KepingYan
@@ -41,21 +41,47 @@ args = parser.parse_args() -client = OpenAI() -# # List all models. -models = client.models.list() -print(models.data, "\n") - -# Note: not all arguments are currently supported and will be ignored by the backend. -chat_completion = client.chat.completions.create( - model=args.model_name, - ...
Can we remove parameters here, and set it via environment variables OPENAI_BASE_URL and OPENAI_API_KEY by users as described in readme?
llm-on-ray
github_2023
python
117
intel
carsonwang
@@ -58,4 +58,8 @@ temperature=args.temperature, top_p=args.top_p, ) -print(chat_completion) +if args.streaming_response: + for chunk in chat_completion: + print(chunk)
Can we use the openAI example that previous didn't work? ``` for chunk in stream: if chunk.choices[0].delta.content is not None: print(chunk.choices[0].delta.content, end="") ```
llm-on-ray
github_2023
others
111
intel
carsonwang
@@ -29,8 +29,8 @@ COPY ./MANIFEST.in . RUN mkdir ./finetune && mkdir ./inference -RUN --mount=type=cache,target=/root/.cache/pip pip install -e .[cpu,deepspeed] -f https://developer.intel.com/ipex-whl-stable-cpu \ - -f https://download.pytorch.org/whl/torch_stable.html +RUN --mount=type=cache,target=/root/.cach...
Why some use "--index-url --extra-index-url", some use "--extra-index-url --extra-index-url"? Is there any difference? But let's still better to use the same format.
llm-on-ray
github_2023
python
103
intel
carsonwang
@@ -155,10 +155,10 @@ def train(self): max_train_step = self.config.get("max_train_step") max_eval_step = self.config.get("max_eval_step") for idx in range(self.starting_epoch, num_train_epochs, 1): - logger.info(f"start train epoch {idx}") self.model.train() ...
nit: start train -> Start training
llm-on-ray
github_2023
python
103
intel
carsonwang
@@ -63,12 +63,14 @@ def get_accelerate_environment_variable(mode: str, config: Union[Dict[str, Any], return mode_env_vars[mode] -def convert_dtype(dtype: str) -> torch.dtype: - supported_dtypes = {"fp16": torch.float16, "bf16": torch.bfloat16, "fp32": torch.float32} - if dtype in supported_dtypes: - ...
Have you tested with setting it to "no" and does None work? Otherwise should we use fp32 if mixed_precision is "no" ?
llm-on-ray
github_2023
python
103
intel
carsonwang
@@ -73,6 +73,13 @@ def check_accelerate_mode(cls, v: str): raise ValueError(f"accelerate_mode must be one of {modes}") return v + @validator("mixed_precision") + def check_mixed_precision(cls, v: str): + supported_precisions = ["no", "fp16", "bf16", "fp32"] + if v not in supp...
nit: on -> one
llm-on-ray
github_2023
python
103
intel
carsonwang
@@ -73,6 +73,13 @@ def check_accelerate_mode(cls, v: str): raise ValueError(f"accelerate_mode must be one of {modes}") return v + @validator("mixed_precision") + def check_mixed_precision(cls, v: str): + supported_precisions = ["no", "fp16", "bf16", "fp32"]
Please also update finetune_parameters.md to remove fp8.
llm-on-ray
github_2023
others
103
intel
minmingzhu
@@ -3,6 +3,7 @@ General: gpt_base_model: true output_dir: /tmp/llm-ray/output checkpoint_dir: /tmp/llm-ray/checkpoint + tracking_dir: /tmp/llm-ray/tracking
There are 9 files in the models directory. Do bloom-560m.yaml, finetune_config_template.yaml, gpt2.yaml, llama-7b.yaml and opt-125m.yaml also need to be modified?
llm-on-ray
github_2023
others
39
intel
KepingYan
@@ -0,0 +1,23 @@ +port: 8000 +name: starcoder +route_prefix: /starcoder +precision: 'bf16'
This attribute should be removed.
llm-on-ray
github_2023
others
39
intel
KepingYan
@@ -0,0 +1,22 @@ +port: 8000 +name: starcoder +route_prefix: /starcoder +cpus_per_worker: 24 +gpus_per_worker: 0 +deepspeed: false +workers_per_group: 2 +ipex: + enabled: false + precision: bf16 +device: "cpu" +model_description: + model_id_or_path: bigcode/starcoder + tokenizer_name_or_path: bigcode/starcoder + ...
`use_auth_token` cannot be written directly in config yaml, it needs to be set in CI file. @jiafuzha please help confirm this.
llm-on-ray
github_2023
python
101
intel
harborn
@@ -134,9 +134,11 @@ def train_func(config: Dict[str, Any]): model = common.model.Model.registory.get("HuggingFaceModelForCausalLM")()( config={ "name": base_model, - "dtype": convert_dtype(config["Training"]["mixed_precision"]), + "dtype": convert_dtype(config["Training...
default is bf16?
llm-on-ray
github_2023
others
101
intel
harborn
@@ -3,6 +3,7 @@ General: gpt_base_model: true output_dir: /tmp/llm-ray/output checkpoint_dir: /tmp/llm-ray/checkpoint + logger_name: tensorboard # only support tensorboard as tracker
name as `log_with` or `tracking_with` ?
llm-on-ray
github_2023
others
75
intel
carsonwang
@@ -0,0 +1,26 @@ +# Inference with Intel Habana Gaudi
can you please update the existing setup.md and serve.md in /docs instead of creating this new document?
llm-on-ray
github_2023
python
86
intel
xwu99
@@ -238,13 +238,16 @@ def streaming_generate(self, prompt, streamer, **config): for worker in self.prediction_workers[1:]: worker.streaming_generate.remote(inputs_ref, self._create_dummy_streamer(), **config) - def generate(self, prompt, **config): + def generate(self, prompt, return_shape...
The generate interface is an override of Predictor interface. We should have consistency maintained across interfaces. I don't' think it's a good idea to change the interface just for the return value.
llm-on-ray
github_2023
python
86
intel
xwu99
@@ -28,10 +28,13 @@ def __init__(self, infer_conf: InferenceConfig) -> None: for stop_word in stop_words ] self.stopping_criteria = StoppingCriteriaList([StoppingCriteriaSub(stops=stop_words_ids)]) + self.input_length = None def tokenize_inputs(self, text): input_to...
better to get input length first before send them to the device?
llm-on-ray
github_2023
python
86
intel
xwu99
@@ -26,8 +26,9 @@ from inference.inference_config import InferenceConfig from typing import Union, Dict, Any from starlette.responses import StreamingResponse, JSONResponse +from fastapi import HTTPException from inference.api_openai_backend.openai_protocol import ModelResponse -from utils import get_input_format +...
should remove inference. here?
llm-on-ray
github_2023
python
86
intel
xwu99
@@ -64,20 +64,24 @@ } proxies = {"http": None, "https": None} -response = s.post(url, json=body, proxies=proxies) # type: ignore +response = s.post(url, json=body, proxies=proxies, stream=args.streaming_response) # type: ignore for chunk in response.iter_lines(decode_unicode=True): - if chunk is not None: - ...
```suggestion print() ```
llm-on-ray
github_2023
python
86
intel
xwu99
@@ -75,8 +66,8 @@ async def stream(self, model: str, prompt: Prompt, request_id: str): prompt=prompt, request_id=request_id, async_iterator=deploy_handle.options(stream=True) - .stream_response.options(stream=True, use_new_handle_api=True) - .remote(prompt_co...
need indent the two lines
llm-on-ray
github_2023
python
86
intel
xwu99
@@ -145,33 +146,71 @@ async def __call__(self, http_request: Request) -> Union[StreamingResponse, JSON self.consume_streamer_async(streamer), status_code=200, media_type="text/plain" ) - async def stream_response(self, prompt, config): + async def openai_call(self, prompt, config, ...
should merge this into above else logic
llm-on-ray
github_2023
python
86
intel
xwu99
@@ -145,33 +146,71 @@ async def __call__(self, http_request: Request) -> Union[StreamingResponse, JSON self.consume_streamer_async(streamer), status_code=200, media_type="text/plain" ) - async def stream_response(self, prompt, config): + async def openai_call(self, prompt, config, ...
I think you can use generate_result.input_length or generate_result.generate_length directly or use generate_length/generate_result, don't mix both.
llm-on-ray
github_2023
python
86
intel
xwu99
@@ -145,33 +146,71 @@ async def __call__(self, http_request: Request) -> Union[StreamingResponse, JSON self.consume_streamer_async(streamer), status_code=200, media_type="text/plain" ) - async def stream_response(self, prompt, config): + async def openai_call(self, prompt, config, ...
```suggestion 400, "Multiple prompts are not supported when using openai compatible api." ```
llm-on-ray
github_2023
python
86
intel
xwu99
@@ -145,33 +146,71 @@ async def __call__(self, http_request: Request) -> Union[StreamingResponse, JSON self.consume_streamer_async(streamer), status_code=200, media_type="text/plain" ) - async def stream_response(self, prompt, config): + async def openai_call(self, prompt, config, ...
Please capital the initial letter. ```suggestion yield HTTPException(400, "Invalid prompt format.") ```
llm-on-ray
github_2023
python
86
intel
xwu99
@@ -103,7 +104,7 @@ def get_input_format(input: Union[List[str], List[dict]]): for item in input: if isinstance(item, str): chat_format = False - elif isinstance(item, dict): + elif isinstance(item, dict) or isinstance(item, ChatMessage):
need to add ChatMessage to the input: Union[xxx] and you can consider break from the loop without checking all items
llm-on-ray
github_2023
python
86
intel
xwu99
@@ -25,22 +25,39 @@ def __init__(self, infer_conf: InferenceConfig): self.engine = AsyncLLMEngine.from_engine_args(args) + def check_config(self, **config):
```suggestion def update_vllm_config(self, **config): ```
llm-on-ray
github_2023
python
86
intel
xwu99
@@ -145,33 +146,71 @@ async def __call__(self, http_request: Request) -> Union[StreamingResponse, JSON self.consume_streamer_async(streamer), status_code=200, media_type="text/plain" ) - async def stream_response(self, prompt, config): + async def openai_call(self, prompt, config, ...
Need special care for vllm here, since there are two exec paths of vllm.generate, one for str, another for List[str]. The one for List[str] may not be efficient now. I will submit another PR to consolidate vllm.generate(str) and vllm.generate(List[str]). For now we need to pass str to vllm for single prompt (for benchm...
llm-on-ray
github_2023
python
99
intel
carsonwang
@@ -170,12 +168,17 @@ def train(self): if self.lr_scheduler is not None: self.lr_scheduler.step() self.optimizer.zero_grad() - if step % log_step == 0: + + if step % logging_steps == 0: + loss...
Instead of just output 0, 1, 2, etc, can we support output it like 0.1, 0.2, etc just like other workflows?
llm-on-ray
github_2023
python
99
intel
carsonwang
@@ -184,6 +187,10 @@ def train(self): else total_steps, } ) + self.accelerator.log(
Do we want to use Ray's report or accelerator.log to log the metrics. Currently the code above logs the metrics twice, right? If Ray's report already meets our requirements, I think we don't need to use accelerator.log to log again?
llm-on-ray
github_2023
others
99
intel
carsonwang
@@ -10,9 +10,11 @@ The following are the parameters supported in the finetuning workflow. |gpt_base_model|True|This parameter is for [Transformers#22482](https://github.com/huggingface/transformers/issues/22482). It needs to be set to True when the pretrained model is realted to gpt, otherwise it is False.| |output_d...
Can we directly use the output_dir + "tracking" as the directory and not add this new parameter?
llm-on-ray
github_2023
python
99
intel
carsonwang
@@ -62,6 +63,14 @@ def get_accelerate_environment_variable(mode: str, config: Union[Dict[str, Any], return mode_env_vars[mode] +def convert_dtype(dtype: str) -> torch.dtype: + supported_dtypes = {"fp16": torch.float16, "bf16": torch.bfloat16, "fp32": torch.float32}
You passed mixed_precision as the parameter, its value could be "no", "fp16", "bf16" or "fp8". But "no" and "fp8" are not properly handled here.
llm-on-ray
github_2023
python
99
intel
carsonwang
@@ -62,6 +63,14 @@ def get_accelerate_environment_variable(mode: str, config: Union[Dict[str, Any], return mode_env_vars[mode] +def convert_dtype(dtype: str) -> torch.dtype: + supported_dtypes = {"fp16": torch.float16, "bf16": torch.bfloat16, "fp32": torch.float32} + if dtype in supported_dtypes: + ...
can you add the check in finetune_config.py instead of here?
llm-on-ray
github_2023
python
99
intel
carsonwang
@@ -217,14 +245,21 @@ def main(external_config=None): "FI_PROVIDER": "tcp", } } - accelerate_env_vars = get_accelerate_environment_variable(accelerate_mode, config) runtime_env["env_vars"].update(accelerate_env_vars) if config["General"]["gpt_base_mode...
Why do we need this change and can we avoid this? If we start Ray first, then execute the finetune command, do we still need this change and does this change still work?
llm-on-ray
github_2023
python
99
intel
carsonwang
@@ -54,6 +56,8 @@ class Training(BaseModel): resources_per_worker: RayResourceConfig accelerate_mode: str mixed_precision: str = "no" + gradient_accumulation_steps: int
Can you set the default value 1 here?
llm-on-ray
github_2023
others
99
intel
carsonwang
@@ -28,3 +30,5 @@ Training: resources_per_worker: CPU: 32 accelerate_mode: CPU_DDP + gradient_accumulation_steps: 2
The default value is 1 in our document. can you please set to 1 here?