gen-synth-data / pages /3_๐ŸŒฑ Generate Dataset.py
Ben Burtenshaw
add max new tokens to pipeline params
1aeff9e
raw
history blame
9.88 kB
import streamlit as st
from defaults import ARGILLA_URL
from hub import push_pipeline_params
from utils import project_sidebar
st.set_page_config(
page_title="Domain Data Grower",
page_icon="๐Ÿง‘โ€๐ŸŒพ",
)
project_sidebar()
################################################################################
# HEADER
################################################################################
st.header("๐Ÿง‘โ€๐ŸŒพ Domain Data Grower")
st.divider()
st.subheader("Step 3. Run the pipeline to generate synthetic data")
st.write("Define the distilabel pipeline for generating the dataset.")
hub_username = st.session_state.get("hub_username")
project_name = st.session_state.get("project_name")
hub_token = st.session_state.get("hub_token")
###############################################################
# CONFIGURATION
###############################################################
st.divider()
st.markdown("## ๐Ÿงฐ Data Generation Pipeline")
st.markdown(
"""
Now we need to define the configuration for the pipeline that will generate the synthetic data.
The pipeline will generate synthetic data by combining self-instruction and domain expert responses.
The self-instruction step generates instructions based on seed terms, and the domain expert step generates \
responses to those instructions. Take a look at the [distilabel docs](https://distilabel.argilla.io/latest/sections/learn/tasks/text_generation/#self-instruct) for more information.
"""
)
###############################################################
# INFERENCE
###############################################################
st.markdown("#### ๐Ÿค– Inference configuration")
st.write(
"""Add the url of the Huggingface inference API or endpoint that your pipeline should use to generate instruction and response pairs. \
Some domain tasks may be challenging for smaller models, so you may need to iterate over your task definition and model selection. \
This is a part of the process of generating high-quality synthetic data, human feedback is key to this process. \
You can find compatible models here:"""
)
with st.expander("๐Ÿค— Recommended Models"):
st.write("All inference endpoint compatible models can be found via the link below")
st.link_button(
"๐Ÿค— Inference compaptible models on the hub",
"https://huggingface.co/models?pipeline_tag=text-generation&other=endpoints_compatible&sort=trending",
)
st.write("๐Ÿ”‹Projects with sufficient resources could take advantage of LLama3 70b")
st.code(
"https://api-inference.huggingface.co/models/meta-llama/Meta-Llama-3-8B-Instruct"
)
st.write("๐ŸชซProjects with less resources could take advantage of LLama 3 8b")
st.code(
"https://api-inference.huggingface.co/models/meta-llama/Meta-Llama-3-8B-Instruct"
)
st.write("๐ŸƒProjects with even less resources could use Phi-3-mini-4k-instruct")
st.code(
"https://api-inference.huggingface.co/models/microsoft/Phi-3-mini-4k-instruct"
)
st.write("Note Hugggingface Pro gives access to more compute resources")
st.link_button(
"๐Ÿค— Huggingface Pro",
"https://huggingface.co/pricing",
)
self_instruct_base_url = st.text_input(
label="Model base URL for instruction generation",
value="https://api-inference.huggingface.co/models/microsoft/Phi-3-mini-4k-instruct",
)
domain_expert_base_url = st.text_input(
label="Model base URL for domain expert response",
value="https://api-inference.huggingface.co/models/microsoft/Phi-3-mini-4k-instruct",
)
###############################################################
# PARAMETERS
###############################################################
st.divider()
st.markdown("#### ๐Ÿงฎ Parameters configuration")
st.write(
"โš ๏ธ Model and parameter choices significantly affect the quality of the generated data. \
We reccomend that you start with generating a few samples and review the data. Then scale up from there. \
You can run the pipeline multiple times with different configurations and append it to the same Argilla dataset."
)
st.markdown(
"Number of generations are the samples that each model will generate for each seed term, \
so if you have 10 seed terms, 2 instruction generations, and 2 response generations, you will have 40 samples in total."
)
self_intruct_num_generations = st.slider(
"Number of generations for self-instruction", 1, 10, 2
)
domain_expert_num_generations = st.slider(
"Number of generations for domain expert response", 1, 10, 2
)
with st.expander("๐Ÿ”ฅ Advanced parameters"):
st.markdown(
"Temperature is a hyperparameter that controls the randomness of the generated text. \
Lower temperatures will generate more deterministic text, while higher temperatures \
will add more variation to generations."
)
self_instruct_temperature = st.slider(
"Temperature for self-instruction", 0.1, 1.0, 0.9
)
domain_expert_temperature = st.slider(
"Temperature for domain expert", 0.1, 1.0, 0.9
)
st.markdown(
"`max_new_tokens` is the maximum number of tokens (word like things) that can be generated by each model call. \
This is a way to control the length of the generated text. in some cases, you may want to increase this to \
generate longer responses. You should adapt this value to your model chice, but default of 2096 works \
in most cases."
)
self_instruct_max_new_tokens = st.number_input(
"Max new tokens for self-instruction", value=2096
)
domain_expert_max_new_tokens = st.number_input(
"Max new tokens for domain expert", value=2096
)
###############################################################
# ARGILLA API
###############################################################
st.divider()
st.markdown("#### ๐Ÿ”ฌ Argilla API details to push the generated dataset")
st.markdown(
"Here you can define the Argilla API details to push the generated dataset to your Argilla space. \
These are the defaults that were set up for the project. You can change them if needed."
)
argilla_url = st.text_input("Argilla API URL", ARGILLA_URL)
argilla_api_key = st.text_input("Argilla API Key", "owner.apikey")
argilla_dataset_name = st.text_input("Argilla Dataset Name", project_name)
st.divider()
###############################################################
# Pipeline Run
###############################################################
st.markdown("## Run the pipeline")
st.markdown(
"Once you've defined the pipeline configuration above, you can run the pipeline from your local machine."
)
if all(
[
argilla_api_key,
argilla_url,
self_instruct_base_url,
domain_expert_base_url,
self_intruct_num_generations,
domain_expert_num_generations,
self_instruct_temperature,
domain_expert_temperature,
hub_username,
project_name,
hub_token,
argilla_dataset_name,
]
) and st.button("๐Ÿ’พ Save Pipeline Config"):
with st.spinner("Pushing pipeline to the Hub..."):
push_pipeline_params(
pipeline_params={
"argilla_api_url": argilla_url,
"argilla_dataset_name": argilla_dataset_name,
"self_instruct_base_url": self_instruct_base_url,
"domain_expert_base_url": domain_expert_base_url,
"self_instruct_temperature": self_instruct_temperature,
"domain_expert_temperature": domain_expert_temperature,
"self_intruct_num_generations": self_intruct_num_generations,
"domain_expert_num_generations": domain_expert_num_generations,
"self_instruct_max_new_tokens": self_instruct_max_new_tokens,
"domain_expert_max_new_tokens": domain_expert_max_new_tokens,
},
hub_username=hub_username,
hub_token=hub_token,
project_name=project_name,
)
st.success(
f"Pipeline configuration pushed to the dataset repo {hub_username}/{project_name} on the Hub."
)
st.markdown(
"To run the pipeline locally, you need to have the `distilabel` library installed. \
You can install it using the following command:"
)
st.code(
body="""
# Install the distilabel library
pip install distilabel
""",
language="bash",
)
st.markdown(
"Next, you'll need to clone the pipeline code and install dependencies:"
)
st.code(
"""
git clone https://github.com/huggingface/data-is-better-together
cd data-is-better-together/domain-specific-datasets/distilabel_pipelines
pip install -r requirements.txt
huggingface-cli login
""",
language="bash",
)
st.markdown("Finally, you can run the pipeline using the following command:")
st.code(
f"""
python domain_expert_pipeline.py {hub_username}/{project_name}""",
language="bash",
)
st.markdown(
"๐Ÿ‘ฉโ€๐Ÿš€ If you want to customise the pipeline take a look in `domain_expert_pipeline.py` \
and the [distilabel docs](https://distilabel.argilla.io/)"
)
st.markdown(
"๐Ÿš€ Once you've run the pipeline your records will be available in the Argilla space"
)
st.link_button("๐Ÿ”— Argilla Space", argilla_url)
st.markdown("Once you've reviewed the data, you can publish it on the next page:")
st.page_link(
page="pages/4_๐Ÿ” Review Generated Data.py",
label="Review Generated Data",
icon="๐Ÿ”",
)
else:
st.info("Please fill all the required fields.")