ntseng/mistralai_Mistral-7B-Instruct-v0_2_student_answer_train_examples_mistral_0416

Base Model: mistralai/Mistral-7B-Instruct-v0_2_student_answer_train_examples_mistral_0416

• LoRAs weights for Mistral-7b-Instruct-v0_2

Noteworthy changes:

• reduced training hyperparams: epochs=3 (previously 4)

• new training prompt: "Teenager students write in simple sentences. You are a teenager student, and please answer the following question. {training example}"

• old training prompt: "Teenager students write in simple sentences [with typos and grammar errors]. You are a teenager student, and please answer the following question. {training example}"

Model Details

Fine-tuned model that talks like middle school students, using simple vocabulary and grammar.

• Trained on student Q&As physics topics including pulley/ramp examples that discuss work, force, and etc.

Model Description

• Developed by: [More Information Needed]
• Funded by [optional]: [More Information Needed]
• Shared by [optional]: [More Information Needed]
• Model type: [More Information Needed]
• Language(s) (NLP): [More Information Needed]
• License: [More Information Needed]
• Finetuned from model [optional]: [More Information Needed]

Model Sources [optional]

• Repository: [More Information Needed]
• Paper [optional]: [More Information Needed]
• Demo [optional]: [More Information Needed]

Model Details

Fine-tuned model to talk like middle school students, using typos/grammar errors. Trained on student Q&As physics topics including pulley/ramp examples that discuss work, force, and etc.

• Developed by: Nora T
• Finetuned from model: mistralai_Mistral-7B-Instruct-v0.2
• Repository: [More Information Needed]
• Paper [optional]: [More Information Needed]
• Demo [optional]: [More Information Needed]

How to Get Started:

1. Load Mistral model first:

from peft import PeftModel # for fine-tuning
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline, GenerationConfig, GPTQConfig, BitsAndBytesConfig

model_name_or_path = "mistralai/Mistral-7B-Instruct-v0.2"
nf4_config = BitsAndBytesConfig( # quantization 4-bit
   load_in_4bit=True,
   bnb_4bit_quant_type="nf4",
   bnb_4bit_use_double_quant=True,
   bnb_4bit_compute_dtype=torch.bfloat16
)
model = AutoModelForCausalLM.from_pretrained(model_name_or_path,
                                             device_map="auto",
                                             trust_remote_code=False,
                                             quantization_config=nf4_config,
                                             revision="main")

tokenizer = AutoTokenizer.from_pretrained(model_name_or_path, use_fast=True)

2. Load in LoRA weights:

lora_model_path = "{path_to_loras_folder}/mistralai_Mistral-7B-Instruct-v0.2-testgen-LoRAs" # load loras
model = PeftModel.from_pretrained(
        model, lora_model_path, torch_dtype=torch.float16, force_download=True,
      )

Training Hyperparams

• LoRA Rank: 128
• LoRA Alpha: 32
• Batch Size: 64
• Cutoff Length: 256
• Learning rate: 3e-4
• Epochs: 3
• LoRA Dropout: 0.05

Training Data

Trained on raw text file

Preprocessing [optional]

[More Information Needed]

Technical Specifications

Model Architecture and Objective

[More Information Needed]

Hardware

[More Information Needed]

Software

[More Information Needed]

Citation [optional]

Framework versions

• PEFT 0.7.1