added text summarization notebook

[Marjan-emd]

Added this text generation notebook to the blueprints, since the only available one (Taylor Swift Lyrics) is not creating the text SQS and enables the old mpt-7b model.

This notebook will be called in the showcase text metrics blog and is added to the blog folder following up John's suggestion, here.

[vercel]

The latest updates on your projects. Learn more about Vercel for Git ↗︎

Name	Status	Preview	Comments	Updated (UTC)
gretel-blueprints	✅ Ready (Inspect)	Visit Preview	💬 Add feedback	Oct 26, 2023 0:14am

[kboyd]

Some minor suggestions and questions. Please also have Johnny review and approve before merging as he's working on tidying up this notebooks directory. Want to make sure this follows the planned style and structure instead of creating more stuff to clean up later.

[johnnygreco]

Hey @Marjan-emd,

I'll probably move the file location as I reorganize the notebooks in this repo, but we can go ahead and get this merged for your blog. I'll make sure the link gets switched (there will be a bunch of broken links, so no worries here).

One request before we merge. Can we update this to the new SDK interface?

Here's what that would look like:

 from gretel_client import Gretel

PROJECT = 'data-summarization'
DATASET_PATH = 'https://gretel-datasets.s3.us-west-2.amazonaws.com/Text-dataset/Samsum-text-summerization-sample-1000.csv'

gretel = Gretel(project_name=f"{PROJECT}-llama-2-7b", api_key="prompt", validate=True)

trained = gretel.submit_train(
    "natural-language",
    data_source=DATASET_PATH
    params={"steps": 1000}, 
 )

trained.report.display_in_notebook()

You can pass the DataFrame as the data_source if you prefer.

[johnnygreco]

Maybe explicitly add the pretrained model parameter:

from gretel_client import Gretel

PROJECT = 'data-summarization'
DATASET_PATH = 'https://gretel-datasets.s3.us-west-2.amazonaws.com/Text-dataset/Samsum-text-summerization-sample-1000.csv'
LLM = "meta-llama/Llama-2-7b-chat-hf"

gretel = Gretel(project_name=f"{PROJECT}-llama-2-7b", api_key="prompt", validate=True)

trained = gretel.submit_train(
    "natural-language",
    data_source=DATASET_PATH,
    pretrained_model=LLM,
    params={"steps": 1000},
 )

trained.report.display_in_notebook()

[Marjan-emd]

Maybe explicitly add the pretrained model parameter:

from gretel_client import Gretel

PROJECT = 'data-summarization'
DATASET_PATH = 'https://gretel-datasets.s3.us-west-2.amazonaws.com/Text-dataset/Samsum-text-summerization-sample-1000.csv'
LLM = "meta-llama/Llama-2-7b-chat-hf"

gretel = Gretel(project_name=f"{PROJECT}-llama-2-7b", api_key="prompt", validate=True)

trained = gretel.submit_train(
    "natural-language",
    data_source=DATASET_PATH,
    pretrained_model=LLM,
    params={"steps": 1000},
 )

trained.report.display_in_notebook()

Thanks for reviewing this. I totally forgot about adding the new SDK interface!
Just changed the model to the regular Llama-2 instead of the chat since I did my experiments of that one, though there should not be a huge change in the results.

[johnnygreco]

Nice – thanks, @Marjan-emd!

One last thing before you merge: will you put this at the top of the first markdown cell?

[Marjan-emd]

All feedback comments were addressed.