llama3-nyc-base

llama3-nyc-base

This model is a fine-tuned version of unsloth/llama-3-8b using LoRA (Low-Rank Adaptation) and quantization techniques.

Model Details

• Base Model: unsloth/llama-3-8b
• Fine-tuned Model: comp5331poi/llama3-nyc-base
• Training Run: llama3-nyc-base
• Device: cuda

Training Configuration

Hyperparameters

• Number of Epochs: 8
• Batch Size: 4
• Gradient Accumulation Steps: 2
• Effective Batch Size: 8
• Learning Rate: 1e-05
• Learning Rate Scheduler: constant
• Warmup Steps: 20
• Max Sequence Length: 2048
• Optimizer: paged_adamw_8bit
• Max Gradient Norm: 0.3
• Random Seed: 2024

LoRA Configuration

• LoRA Rank (r): 16
• LoRA Alpha: 32
• LoRA Dropout: 0.1
• Target Modules: k_proj, up_proj, o_proj, down_proj, gate_proj, v_proj, q_proj
• Task Type: CAUSAL_LM

Quantization

• Quantization Bits: 4-bit

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

# Load base model
base_model = AutoModelForCausalLM.from_pretrained("unsloth/llama-3-8b")

# Load LoRA adapter
model = PeftModel.from_pretrained(base_model, "comp5331poi/llama3-nyc-base")

# Load tokenizer
tokenizer = AutoTokenizer.from_pretrained("unsloth/llama-3-8b")

# Generate text
inputs = tokenizer("Your prompt here", return_tensors="pt")
outputs = model.generate(**inputs, max_length=2048)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Framework Versions

• Transformers
• PEFT
• TRL
• PyTorch
• BitsAndBytes