Sachin21112004/Sancara_text_generation

Sancara LoRA – Instruction-Trained Adapter

This repository contains a LoRA adapter trained using supervised instruction data on top of the base model
Sachin21112004/Sancara_text_generation.

This was a full training run (not a partial or toy fine-tune) performed on the entire dataset (~207k samples) using a dedicated NVIDIA A100 GPU.

The model is optimized for:

• ✅ Instruction following
• ✅ Conversational AI
• ✅ Question answering
• ✅ General reasoning

⚠️ This repository contains only the LoRA adapter weights, not the full base model.

---

Model Details

Model Description

• Developed by: Sachin
• Shared by: Sachin
• Model type: Causal Language Model (LoRA-based Supervised Instruction Training)
• Language(s): English
• License: SRL(others)
• Base model used:
  👉 Sachin21112004/Sancara_text_generation

This model was trained using supervised instruction data with parameter-efficient training (LoRA) on top of the Sancara base text-generation model.

---

Model Sources

• Base Model:
  https://huggingface.co/Sachin21112004/Sancara_text_generation

• LoRA Adapter (This Repo):
  https://huggingface.co/Sachin21112004/sancara-fast-lora

---

Uses

Direct Use

This LoRA-trained adapter is intended for:

• ✅ Chat assistants
• ✅ Instruction-following models
• ✅ Question answering systems
• ✅ Research & education tools
• ✅ General conversational AI

It must be loaded together with the base model.

---

Out-of-Scope Use

• ❌ Medical diagnosis or treatment
• ❌ Legal advice
• ❌ Financial or investment decisions
• ❌ High-risk autonomous systems

---

Bias, Risks, and Limitations

• The model inherits:
  • Biases from the base model
  • Biases from the instruction dataset
• It may generate:
  • Hallucinations
  • Biased or incomplete outputs
  • Overconfident but incorrect answers

Recommendations

• Use with human oversight
• Do not rely on outputs for critical decisions
• Perform domain-specific evaluation before deployment

---

How to Get Started with the Model

✅ Load Base Model + Apply Trained LoRA Adapter

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel
import torch

base_model_id = "Sachin21112004/Sancara_text_generation"
lora_model_id = "Sachin21112004/sancara-fast-lora"

tokenizer = AutoTokenizer.from_pretrained(base_model_id)

base_model = AutoModelForCausalLM.from_pretrained(
    base_model_id,
    torch_dtype=torch.bfloat16,
)

model = PeftModel.from_pretrained(base_model, lora_model_id)
model = model.cuda()

prompt = "Explain how large language models work."

inputs = tokenizer(prompt, return_tensors="pt").to("cuda")

outputs = model.generate(
    **inputs,
    max_new_tokens=200,
    temperature=0.7,
    top_p=0.9,
)

print(tokenizer.decode(outputs[0], skip_special_tokens=True))