led-base-ilc / README.md
d0r1h's picture
Update README.md
08a1d52
metadata
license: apache-2.0
datasets: ilc
tags:
  - summarization
metrics:
  - rouge
widget:
  - text: >-
      IN THE HIGH COURT OF JUDICATURE AT PATNA CRIMINAL MISCELLANEOUS No. 229121
      Arising Out of PS. Case No. 127 Year 2020 Thana DUMRAON District Buxar 1.
      Ramlal Goswami aged about 44 years Male S o Late Gauri Shankar 2.
      Dharmshila Devi @ Savita Devi aged about 35 years wife of Ramlal Both
      resident of village Badka Dhakaich P.S. Krishna Brahm District ...
      Petitioner s ... Opposite Party s The State of Bihar Appearance : For the
      Petitioner s For the State CORAM: HONOURABLE MR. JUSTICE AHSANUDDIN
      AMANULLAH ORAL JUDGMENT Mr. Manoj Kumar with Mr. Anil Kumar Roy Advocates
      Mr. Ram Sumiran Roy APP The matter has been heard via video conferencing.
      2. Heard Mr. Manoj Kumar learned counsel along with Mr. Anil Kumar Roy
      learned counsel for the petitioners and Mr. Ram Sumiran Roy learned
      Additional Public Prosecutorfor the State. 3. Learned counsel for the
      petitioners submitted that he may be permitted to add alias name of
      petitioner no. 2 which is Savita Devi. 4. Prayer allowed. 5. Let necessary
      correction be made in the cause title Date : 03 08 2021 Patna High Court
      CR. MISC. No. 229121 dt.03 08 2021 2 4 by learned counsel for the
      petitioners through e mode by day after tomorrow. 6. The petitioners
      apprehend arrest in connection with Dumraon PS Case No. 1220 dated
      15.04.2020 instituted under Sections 406 420 467 468 471 448 506 34 of the
      Indian Penal Code. 7. The allegation against the petitioners is that the
      informant who is the cousin brother of petitioner no. 1 had bought land
      through the petitioner no. 1 but he was cheated both with regard to the
      rates as also that the same piece of land being sold by the petitioners to
      two different persons. 8. Learned counsel for the petitioners submitted
      that in the FIR itself it has been stated that the informant had sold his
      land at a much higher price than the price he was paying for the land
      which he alleges to have been negotiated by the petitioner no. 1 for him.
      Further it was submitted that all such dispute relating to money is a
      purely civil in nature for which criminal case is an abuse of the process
      of the Court. Learned counsel submitted that the informant being the first
      cousin of the petitioner no. 1 and having sold his land was very well
      aware of the ground realities and cannot take a stand that he was ignorant
      of what was the actual position. Further it was submitted that Patna High
      Court CR. MISC. No. 229121 dt.03 08 2021 3 4 the petitioners have filed a
      supplementary affidavit in which a categorical stand has been taken on
      oath that the petitioners have not sold the same piece of land to two
      different persons. Learned counsel submitted that the petitioners are
      simple citizens being husband and wife and have no other criminal
      antecedent. It was submitted that had the allegation been correct the
      other person aggrieved would also have filed a case and most importantly
      neither any name of any person has been taken nor details of any document
      that the same piece of land was transferred to two persons has been either
      mentioned or brought on record. 9. Learned APP submitted that the
      petitioners are alleged to have cheated the informant and have got the
      same piece of land registered in favour of two persons. 10. Having
      considered the facts and circumstances of the case and submissions of
      learned counsel for the parties in the event of arrest or surrender before
      the Court below within six weeks from today the petitioners be released on
      bail upon furnishing bail bonds of Rs. 25 000 each with two sureties of
      the like amount each to the satisfaction of the learned Chief Judicial
      Magistrate Buxar in Dumrao PS Case No. 127 of 2020 subject to the
      conditions laid down in Patna High Court CR. MISC. No. 229121 dt.03 08
      2021 4 4 Section 438(2) of the Code of Criminal Procedure 1973 and
      furtherthat one of the bailors shall be a close relative of the
      petitioners andthat the petitioners shall cooperate with the Court and the
      police prosecution. Failure to cooperate shall lead to cancellation of
      their bail bonds. 11. It shall also be open for the prosecution to bring
      any violation of the foregoing conditions of bail by the petitioners to
      the notice of the Court concerned which shall take immediate action on the
      same after giving opportunity of hearing to the aforementioned terms. 12.
      The petition stands disposed of Anjani 

Longformer Encoder-Decoder (LED) fine-tuned on ILC

This model is a fine-tuned version of led-base-16384 on the ILC dataset.

As described in Longformer: The Long-Document Transformer by Iz Beltagy, Matthew E. Peters, Arman Cohan, led-base-16384 was initialized from bart-base since both models share the exact same architecture. To be able to process 16K tokens, bart-base's position embedding matrix was simply copied 16 times.


from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
device = "cuda" if torch.cuda.is_available() else "CPU"

checkpoint = "d0r1h/led-base-ilc"

tokenizer = AutoTokenizer.from_pretrained(checkpoint)
model = AutoModelForSeq2SeqLM.from_pretrained(checkpoint, return_dict_in_generate=True).to(device)
case = "......."
input_ids = tokenizer(case, return_tensors="pt").input_ids.to(device)
global_attention_mask = torch.zeros_like(input_ids)
global_attention_mask[:, 0] = 1
sequences = model.generate(input_ids, 
                           global_attention_mask=global_attention_mask).sequences
summary = tokenizer.batch_decode(sequences, 
                                 skip_special_tokens=True)
                                 

Evaluation results

When the model is used for summarizing ILC documents(10 samples), it achieves the following results:

Model rouge1-f rouge1-p rouge2-f rouge2-p rougeL-f rougeL-p
led-ilc 42 47 22 24 39 44
led-base 3 39 1 21 3 37

This notebook shows how led can effectively be used for downstream tasks such as summarization.