ankur1423/Fine-Tune-Mistral-7B

Mistral-7B Solar Manufacturing FAQ — Fine-tuned

Fine-tuned version of Mistral-7B-Instruct-v0.3 on a solar module manufacturing FAQ dataset (302 Q&A pairs) using QLoRA.

Model Details

• Base model: mistralai/Mistral-7B-Instruct-v0.3
• Fine-tuning method: QLoRA (4-bit NF4 + LoRA adapters)
• Task: Solar module manufacturing FAQ assistant
• Language: English
• License: Apache 2.0
• Developed by: ankur1423

How to Get Started

from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
import torch

model_id = "ankur1423/Fine-Tune-Mistral-7B"

bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_compute_dtype=torch.bfloat16,
)
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    quantization_config=bnb_config,
    device_map="auto",
)

messages = [
    {"role": "system", "content": "You are an expert assistant for solar module manufacturing. Answer questions clearly and accurately based on industry knowledge."},
    {"role": "user", "content": "What is EL testing?"},
]
prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
with torch.no_grad():
    outputs = model.generate(**inputs, max_new_tokens=256, temperature=0.3, top_p=0.9, do_sample=True)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True))

Training Details

Dataset

Item	Value
Domain	Solar module manufacturing
Format	302 instruction-output Q&A pairs (Alpaca)
Split	286 train / 16 eval

LoRA Configuration

Parameter	Value
Rank (r)	16
Alpha	32
Dropout	0.05
Target modules	q/k/v/o_proj, gate/up/down_proj
Trainable params	41,943,040 (0.58%)

Training Hyperparameters

Parameter	Value
Epochs	5
Learning rate	1e-4
Batch size	2 (effective 8 with grad_accum=4)
Optimizer	paged_adamw_32bit
LR scheduler	cosine
Max seq length	512
Quantization	4-bit NF4 + double quant, bfloat16

Infrastructure

• Platform: Kaggle — NVIDIA Tesla T4 (15.6 GB VRAM)
• Training time: ~15–25 minutes

Uses

Solar manufacturing Q&A: cell technologies (PERC, TOPCon, HJT), quality testing (EL, IV curves), defect modes (hot spots, delamination, PID), manufacturing processes.

Out-of-scope: general QA, code generation, topics outside solar manufacturing.

Limitations

Small dataset (302 examples) — may hallucinate on edge cases. Domain-specific only.

Citation

@misc{ankur1423-solar-faq-mistral,
  title  = {Mistral-7B Solar Manufacturing FAQ Fine-tune},
  author = {ankur1423},
  year   = {2026},
  url    = {https://huggingface.co/ankur1423/Fine-Tune-Mistral-7B}
}