Summarization
Transformers
Safetensors
Malayalam
llama
text-generation
malayalam
dapt
text-generation-inference
Instructions to use Navneeth017/sarvam1-dapt-general-malayalam-summarisation with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use Navneeth017/sarvam1-dapt-general-malayalam-summarisation with Transformers:
# Use a pipeline as a high-level helper # Warning: Pipeline type "summarization" is no longer supported in transformers v5. # You must load the model directly (see below) or downgrade to v4.x with: # 'pip install "transformers<5.0.0' from transformers import pipeline pipe = pipeline("summarization", model="Navneeth017/sarvam1-dapt-general-malayalam-summarisation")# Load model directly from transformers import AutoTokenizer, AutoModelForCausalLM tokenizer = AutoTokenizer.from_pretrained("Navneeth017/sarvam1-dapt-general-malayalam-summarisation") model = AutoModelForCausalLM.from_pretrained("Navneeth017/sarvam1-dapt-general-malayalam-summarisation", device_map="auto") - Notebooks
- Google Colab
- Kaggle
sarvam1-dapt-general-malayalam-summarisation
General-DAPT sarvam-1 fine-tuned (SFT) for Malayalam summarisation.
Test metrics (837-doc held-out split)
| Metric | Score |
|---|---|
| ROUGE-1 | 34.58 |
| chrF | 33.47 |
| BERTScore | 88.21 |
Part of the nlg_dapt_study corpus-controlled DAPT study for Malayalam summarisation. Central finding: documented DAPT (cultural or general-world, 23M–297M tokens) yields no statistically significant downstream gain on either mBART or sarvam-1. Private research artifact.
- Downloads last month
- 19