The Lumo-8B-Instruct model is a fine-tuned version of Meta's LLaMa 3.1 8B model designed to provide highly accurate and contextual assistance for developers working on Solana and its associated ecosystems. This model is capable of answering complex questions, generating code snippets, debugging, and explaining technical concepts using state-of-the-art instruction tuning techniques.
(Knowledge cut-off date: 29th January, 2025)
🎯 Key Features
Optimized for Solana-specific queries across ecosystems like Raydium, Helius, Jito, and more.
Instruction fine-tuned for developer-centric workflows.
Lightweight parameter-efficient fine-tuning via LoRA (Low-Rank Adaptation).
Supports multi-turn conversations with context retention.
Outputs complete code snippets and real-world usage examples.
The model was fine-tuned using parameter-efficient methods with LoRA to adapt to the Solana-specific domain. Below is a visualization of the training process:
1defcomplete_chat(model, tokenizer, messages, max_new_tokens=128):2 inputs = tokenizer.apply_chat_template(messages, return_tensors="pt", return_dict=True, add_generation_prompt=True).to(model.device)3with torch.no_grad():4 outputs = model.generate(**inputs, max_new_tokens=max_new_tokens)5return tokenizer.decode(outputs[0], skip_special_tokens=True)67response = complete_chat(model, tokenizer,[8{"role":"system","content":"You are Lumo, a helpful assistant."},9{"role":"user","content":"Explain how to interact with Raydium API for token swaps."}10])11print(response)
📈 Performance
Metric
Value
Validation Loss
1.73
BLEU Score
89%
Code Accuracy
92%
Token Efficiency
~4,096 tokens max
Fine-Tuning Loss Graph
Loss Graph
📂 Dataset
Split
Count
Description
Train
27.1k
High-quality Q&A pairs
Test
1.43k
Evaluation dataset for testing
Dataset Format (JSONL):
json
1{2"question":"How to use the Helius API for transaction indexing?",3"answer":"To index transactions, use Helius's Webhooks API ...",4"chunk":"Helius API allows you to set up ..."5}
🔍 Technical Insights
LoRA Configuration
Rank: 8
Alpha: 32
Dropout: 0.01
Adapter Size: ~10 MB
Optimization
Mixed Precision (FP16) for faster inference.
Gradient Accumulation for memory efficiency.
Parameter-efficient tuning to preserve base model knowledge.
