🦙 TinyLlama TRL Merged - Complete Fine-tuned Model
📋 Model Overview
This is a fully merged and standalone model of TinyLlama (1.1B parameters) fine-tuned using TRL (Transformer Reinforcement Learning) framework with LoRA adapters. The LoRA weights have been permanently merged into the base model, creating a single complete model that can be loaded without any adapter libraries.
Key Features
| Feature | Description |
|---|
| Standalone | No PEFT library required — single model file |
| Fine-tuned | Custom trained on educational Q&A dataset |
| Optimized | FP16 precision for memory efficiency |
| Production Ready | Single folder deployment |
| Chat Optimized | Fine-tuned for conversational responses |
Model Architecture
| Component | Specification |
|---|
| Base Model | TinyLlama-1.1B-Chat-v1.0 |
| Architecture | Llama-based transformer decoder |
| Total Parameters | 1,100,000,000 (1.1B) |
| Context Length | 2048 tokens |
| Hidden Size | 2048 |
| Intermediate Size | 5632 |
| Number of Layers | 22 |
| Number of Attention Heads | 32 |
| Number of Key/Value Heads | 4 (GQA) |
| Head Dimension | 64 |
| Activation Function | SwiGLU |
| Positional Encoding | RoPE (Rotary Position Embedding) |
| Attention Mechanism | Grouped-Query Attention (GQA) |
| Precision | FP16 (float16) |
🚀 Usage Guide
Installation
pip install transformers torch accelerate
Method 1: Direct Transformers Loading
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch
# Load model and tokenizer
model_id = "arif-butt/tinyllama-trl-merged"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
model_id,
torch_dtype=torch.float16,
device_map="auto",
trust_remote_code=True,
)
model.eval()
# Define prompt template
prompt = "Q: What is machine learning?\nA:"
# Tokenize
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
# Generate
with torch.no_grad():
outputs = model.generate(
**inputs,
max_new_tokens=150,
temperature=0.7,
top_p=0.95,
do_sample=True,
repetition_penalty=1.1,
pad_token_id=tokenizer.eos_token_id,
)
# Decode and print
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(f"Prompt: {prompt}")
print(f"Response: {response[len(prompt):].strip()}")
Method 2: Pipeline for Simple Inference
from transformers import pipeline
pipe = pipeline(
"text-generation",
model="arif-butt/tinyllama-trl-merged",
torch_dtype=torch.float16,
device_map="auto",
)
prompt = "Q: Explain neural networks in simple terms\nA:"
result = pipe(prompt, max_new_tokens=150, temperature=0.7, do_sample=True)
print(result[0]["generated_text"])