sweatSmile/Mistral-7B-Instruct-v0.1-Sarcasm

Model Card for Mistral-7B-Instruct-v0.1-Sarcasm

This model is a 4-bit quantized, LoRA fine-tuned version of Mistral-7B-Instruct-v0.1, trained to handle sarcasm-related tasks such as detection and generation. Fine-tuned on a custom 700-row dataset using Hugging Face’s peft and trl libraries.

Model Details

Model Description

This model was fine-tuned using LoRA adapters on top of a 4-bit quantized base model. It leverages bnb_4bit quantization (nf4) and merges LoRA weights into the base. It is optimized for short-form sarcastic dialogue.

• —Developed by: Amit Chaubey
• —Funded by [optional]: Self-funded
• —Shared by [optional]: sweatSmile
• —Model type: Causal Language Model (Decoder-only)
• —Language(s) (NLP): English
• —License: Apache 2.0 (inherited from base)
• —Finetuned from model [optional]: mistralai/Mistral-7B-Instruct-v0.1

Model Sources [optional]

• —Repository: https://huggingface.co/sweatSmile/Mistral-7B-Instruct-v0.1-Sarcasm
• —Paper [optional]: N/A
• —Demo [optional]: N/A

Uses

Direct Use

• —Sarcasm generation
• —Sarcasm detection
• —Instruction-following with humorous tone

Downstream Use [optional]

• —Integrating into sarcastic chatbots
• —Fine-tuning for humor classifiers
• —Educational or creative writing tools

Out-of-Scope Use

• —Factual Q&A or summarization
• —Safety-critical applications
• —Multilingual sarcasm tasks

Bias, Risks, and Limitations

• —Trained on small dataset (~720 samples)
• —Sarcasm is culturally subjective
• —May generate insensitive or offensive content

Recommendations

Users (both direct and downstream) should be aware:

• —Further fine-tuning is recommended for robustness.
• —Outputs should be moderated in public-facing systems.
• —Avoid use in high-stakes domains like healthcare, law, or crisis support.

How to Get Started with the Model

Use the following code to load and test the model:

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model = AutoModelForCausalLM.from_pretrained(
    "sweatSmile/Mistral-7B-Instruct-v0.1-Sarcasm",
    device_map="auto",
    torch_dtype=torch.float16
)
tokenizer = AutoTokenizer.from_pretrained("sweatSmile/Mistral-7B-Instruct-v0.1-Sarcasm")

prompt = "Oh sure, waking up at 6am on a weekend sounds like a dream come true."
inputs = tokenizer(prompt, return_tensors="pt").to("cuda")
outputs = model.generate(**inputs, max_new_tokens=50)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))