Base Model: Groq/Llama-3-Groq-8B-Tool-Use
Fine-tuning Method: LoRA (Low-Rank Adaptation)
LoRA Rank: 16
LoRA Alpha: 32
Target Modules: q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj, lm_head
Training Epochs: 10
Learning Rate: 2e-4

Specialized Tools

rag_chain: Query Frontier HPC documentation
sql_qna_chain: Generate SQL queries for energy database analysis
job_pred: Predict energy consumption for HPC jobs

Usage

python

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

# Load base model and tokenizer
base_model = AutoModelForCausalLM.from_pretrained(
    "Groq/Llama-3-Groq-8B-Tool-Use",
    torch_dtype=torch.float16,
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained("Groq/Llama-3-Groq-8B-Tool-Use")

# Load LoRA adapter
model = PeftModel.from_pretrained(base_model, "axk962/llama3-groq-hpc-energy-analysis")

# Example usage
prompt = "Find the top 3 science domains with highest energy consumption"
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=512, temperature=0.1)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)

Training Configuration

Epochs: 10
Batch Size: 1 (gradient accumulation: 8)
Learning Rate: 2e-4
Optimizer: AdamW with warmup
Precision: FP16

Citation

@misc{llama3-groq-hpc-lora,
  title={Llama-3-Groq-8B LoRA for EPIC: Accelerated Operational Analytics},
  author={""},
  year={2025},
  url={https://huggingface.co/axk962/llama3-groq-hpc-energy-analysis}
}