A LoRA adapter that fine-tunes TinyLlama/TinyLlama-1.1B-intermediate-step-1431k-3T to answer factual questions about AWS EC2 instance specifications — API names, compute family, memory, vCPU counts, and on-demand hourly pricing.

Model Details

Model Description

This is a parameter-efficient (PEFT/LoRA) adapter, not a standalone model. It must be loaded on top of the TinyLlama-1.1B base model.

Developed by: Independent fine-tune
Model type: Causal language model adapter (LoRA)
Language(s): English
License: Apache 2.0 (inherits from TinyLlama)
Finetuned from model: TinyLlama/TinyLlama-1.1B-intermediate-step-1431k-3T

Uses

Direct Use

Answering structured factual questions about AWS EC2 instances, e.g.:

"What is the API name for A1 Large?"
"How much memory does a1.2xlarge have?"
"What is the on-demand hourly price for a1.metal?"
"Which compute family is a1.medium part of?"

Downstream Use

Suitable as a starting point for:

Domain-specific AWS documentation assistants
Demos of LoRA fine-tuning on tabular-derived Q&A data
Further fine-tuning with reasoning or comparison-style questions

Out-of-Scope Use

General-purpose conversational AI
Reasoning, comparison, or arithmetic over EC2 instances ("which is cheaper?", "sort by memory") — the training data does not cover these patterns
Real-time AWS pricing (training data is a static snapshot; AWS prices change)
Any safety-critical or compliance decision based on the model's output

Bias, Risks, and Limitations

Static data. Pricing and specs were captured from a single snapshot. Real AWS pricing varies by region, billing model, and time, and may be outdated.
Hallucination on unseen instances. The model only saw the EC2 rows present in training; queries about unfamiliar instance types may return plausible-looking but incorrect answers.
Template-bound generalization. Training used 21 fixed question templates per EC2 row. Paraphrased or out-of-template questions may degrade quality.
Inherited limitations. The base TinyLlama-1.1B has limited reasoning and factual breadth compared to larger models. This adapter does not change that.

Recommendations

Verify any price- or capacity-sensitive output against AWS's official pricing page before acting on it. Treat outputs as suggestions, not authoritative answers.

How to Get Started with the Model

python

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

BASE_ID = "TinyLlama/TinyLlama-1.1B-intermediate-step-1431k-3T"
ADAPTER = "./tinyllama-ec2-lora"

tokenizer = AutoTokenizer.from_pretrained(ADAPTER)
base = AutoModelForCausalLM.from_pretrained(
    BASE_ID,
    torch_dtype=torch.float16,
    device_map="auto",
)
model = PeftModel.from_pretrained(base, ADAPTER)
model.eval()

prompt = "Question: What is the API name for A1 Large?\nAnswer:"
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
out = model.generate(
    **inputs,
    max_new_tokens=120,
    do_sample=False,
    pad_token_id=tokenizer.eos_token_id,
)
print(tokenizer.decode(out[0], skip_special_tokens=True))

smaram68/aws-ec2-pricing

TinyLlama-1.1B EC2 Instance Q&A (LoRA Adapter)