[test stuff][RFC][AIC-py] hf summarization model parser

[jonathanlastmileai]

[test stuff][RFC][AIC-py] hf summarization model parser

Very similar to text gen

Test:

...
    {
      "name": "test_hf_sum",
      "input": "HMS Duncan was a D-class destroyer ...", # [contents of https://en.wikipedia.org/wiki/HMS_Duncan_(D99)]
      "metadata": {
        "model": {
          "name": "stevhliu/my_awesome_billsum_model",
          "settings": {
            "min_length": 100,
            "max_length": 200,
            "num_beams": 1
          }
        }
      }
    },
...
}

import asyncio
from aiconfig_extension_hugging_face.local_inference.text_summarization import HuggingFaceTextSummarizationTransformer

from aiconfig import AIConfigRuntime, InferenceOptions, CallbackManager

# Load the aiconfig.

mp = HuggingFaceTextSummarizationTransformer()

AIConfigRuntime.register_model_parser(mp, "stevhliu/my_awesome_billsum_model")

config = AIConfigRuntime.load("/Users/jonathan/Projects/aiconfig/cookbooks/Getting-Started/travel.aiconfig.json")
config.callback_manager = CallbackManager([])


def print_stream(data, _accumulated_data, _index):
    print(data, end="", flush=True)


async def run():
    print("Stream")
    options = InferenceOptions(stream=True, stream_callback=print_stream)
    out = await config.run("test_hf_sum", options=options)
    print("Output:\n", out)

    print("no stream")
    options = InferenceOptions(stream=False)
    out = await config.run("test_hf_sum", options=options)
    print("Output:\n", out)


asyncio.run(run())


# OUT:

Stream
Token indices sequence length is longer than the specified maximum sequence length for this model (2778 > 512). Running this sequence through the model will result in indexing errors
/opt/homebrew/Caskroom/miniconda/base/envs/aiconfig/lib/python3.10/site-packages/transformers/generation/configuration_utils.py:430: UserWarning: `num_beams` is set to 1. However, `early_stopping` is set to `True` -- this flag is only used in beam-based generation modes. You should set `num_beams>1` or unset `early_stopping`.
  warnings.warn(
/opt/homebrew/Caskroom/miniconda/base/envs/aiconfig/lib/python3.10/site-packages/transformers/generation/configuration_utils.py:449: UserWarning: `num_beams` is set to 1. However, `length_penalty` is set to `2.0` -- this flag is only used in beam-based generation modes. You should set `num_beams>1` or unset `length_penalty`.
  warnings.warn(
<pad> escorted the 13th Destroyer Flotilla in the Mediterranean and escorited the carrier Argus to Malta during the war  The ship was recalled home to be converted into an escoter destroyer in late 1942. The vessel was repaired and given a refit at Gibraltar on 16 November, and was sold for scrap later that year. The crew of the ship escores the ship to the Middle East, and the ship is a 'disaster' of the </s>Output:
 [ExecuteResult(output_type='execute_result', execution_count=0, data="<pad> escorted the 13th Destroyer Flotilla in the Mediterranean and escorited the carrier Argus to Malta during the war  The ship was recalled home to be converted into an escoter destroyer in late 1942. The vessel was repaired and given a refit at Gibraltar on 16 November, and was sold for scrap later that year. The crew of the ship escores the ship to the Middle East, and the ship is a 'disaster' of the </s>", mime_type=None, metadata={})]
no stream
Output:
 [ExecuteResult(output_type='execute_result', execution_count=0, data="escorted the 13th Destroyer Flotilla in the Mediterranean and escorited the carrier Argus to Malta during the war . The ship was recalled home to be converted into an escoter destroyer in late 1942. The vessel was repaired and given a refit at Gibraltar on 16 November, and was sold for scrap later that year. The crew of the ship escores the ship to the Middle East, and the ship is a 'disaster' of the .", mime_type=None, metadata={})]

---

Stack created with Sapling. Best reviewed with ReviewStack.

• pragmatic fp #747
• -> [test stuff][RFC][AIC-py] hf summarization model parser #749
• [RFC][AIC-py] hf summarization model parser #740