metadata
license: apache-2.0
datasets:
- sweatSmile/marx-dataset
language:
- en
base_model:
- facebook/opt-350m
new_version: sweatSmile/marx-opt350m-finetuned-v1
🧠 Model: marx-opt350m-finetuned-v1
This is a fine-tuned version of facebook/opt-350m on the sweatSmile/marx-dataset — a conversational dataset consisting of question-answer dialogues about Marxist theory, history, and related political contexts.
The model is optimized for dialogue-style QA on political/philosophical topics.
📊 Training Details
- Base model: facebook/opt-350m
- Dataset: sweatSmile/marx-dataset
- Epochs: 5
- Trainer: TRL's
SFTTrainerfromtrl - Loss: Reduced steadily to ~0.36
- ROUGE-L: ~0.53 on evaluation set
🔍 Example Usage
from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained("sweatSmile/marx-opt350m-finetuned-v1")
tokenizer = AutoTokenizer.from_pretrained("sweatSmile/marx-opt350m-finetuned-v1")
prompt = "### Human: Who were commissioned to prepare the party programme for the Communist League?\n### Assistant:"
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_length=100)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))