from transformers import AutoModelForCausalLM, AutoTokenizer

model_path = "PATH_TO_THIS_REPO"

tokenizer = AutoTokenizer.from_pretrained(model_path)
model = AutoModelForCausalLM.from_pretrained(
    model_path,
    device_map="auto",
    torch_dtype='auto'
).eval()

# Prompt content: "hi"
messages = [
    {"role": "user", "content": "hi"}
]

input_ids = tokenizer.apply_chat_template(conversation=messages, tokenize=True, add_generation_prompt=True, return_tensors='pt')
output_ids = model.generate(input_ids.to('cuda'))
response = tokenizer.decode(output_ids[0][input_ids.shape[1]:], skip_special_tokens=True)

# Model response: "Hello! How can I assist you today?"
print(response)

from transformers import AutoModelForCausalLM, AutoTokenizer

model_path = "alexsherstinsky/Mistral-7B-v0.1-sharded"

tokenizer = AutoTokenizer.from_pretrained(model_path)
model = AutoModelForCausalLM.from_pretrained(
    model_path,
    device_map="auto",
    torch_dtype='auto'
).eval()

# Input: Create a string of opposite features to the given input: Caucasian male, mid-30s, angular jawline with light stubble, prominent cheekbones, thin lips, sharp nose, narrow nose, sharp blue eyes, high forehead, receding hairline, sparse eyebrows.
# Output: Rounded jawline, heavy beard, subtle cheekbones, full lips, rounded and broad nose, large, round and dark eyes, low hairline, thick eyebrows.

# Input: Extract important facial and hair details from the description and give them in a comma-separated string. Suspect is a female, likely in her late 20s to early 30s, with a distinctly oval-shaped face. She possesses soft, rounded features, including full lips. Her nose is petite and slightly upturned and has almond-shaped eyes of a deep brown color with long, fluttery eyelashes. She has high cheekbones, and her skin is smooth, unblemished. She has expressive eyebrows, that are gently arched.
# Output: Female, late 20s to early 30s, oval-shaped face, soft rounded features, full lips, petite and slightly upturned nose, almond-shaped deep brown eyes, long fluttery eyelashes, high cheekbones, smooth unblemished skin, expressive gently arched eyebrows.

Model Fine-tuning and Quantization

This model has been fine-tuned for a specific focus on facial detail extraction. It has been fine-tuned and quantized to 4 bits using AutoTrain's advanced capabilities. Here's the code used for fine-tuning and quantization:

python

# Step 1: Setup Environment
!pip install pandas autotrain-advanced -q
!autotrain setup --update-torch

# Step 2: Connect to HuggingFace for Model Upload

from huggingface_hub import notebook_login
notebook_login()

# Step 3: Upload your dataset
!mv finetune-llama-2/train.csv train.csv

import pandas as pd
df = pd.read_csv("train.csv")
print(df.head())

# Step 4: Overview of AutoTrain command
!autotrain llm --train --project-name mistral-7b-mj-finetuned \
  --model alexsherstinsky/Mistral-7B-v0.1-sharded --data-path . \
  --use-peft --quantization int4 --batch-size 2 --epochs 3 --trainer sft \
  --target-modules q_proj,v_proj --push-to-hub --username replace_it --token replace_it \
  --lr 2e-4

# Optionally check the available options
!autotrain llm --help

# Step 5: Completed 🎉
# After the command above is completed your Model will be uploaded to Hugging Face.

# Step 6: Inference Engine
!pip install -q peft accelerate bitsandbytes safetensors

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

adapters_name = "YasiruDEX/mistral-7b-mj-finetuned-face-feature-extraction"
model_name = "mistralai/Mistral-7B-Instruct-v0.1"  # or your preferred model

device = "cuda"  # the device to load the model onto

bnb_config = transformers.BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_use_double_quant=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_compute_dtype=torch.bfloat16
)

model = AutoModelForCausalLM.from_pretrained(
    model_name,
    load_in_4bit=True,
    torch_dtype=torch.bfloat16,
    device_map='auto'
)

# Step 7: Peft Model Loading with upload model
model = PeftModel.from_pretrained(model, adapters_name)

tokenizer = AutoTokenizer.from_pretrained(model_name)
tokenizer.bos_token_id = 1

print(f"Successfully loaded the model {model_name} into memory")

text = "[INST] Extract important facial and hair details from the description and give them in a comma separated string. Suspect is a female, likely in her late 20s to early 30s, with a distinctly oval-shaped face. She possesses soft, rounded features, including full lips. Her nose is petite and slightly upturned and has almond-shaped eyes of a deep brown color with long, fluttery eyelashes. She has high cheekbones, and her skin is smooth, unblemished. She has expressive eyebrows, that are gently arched. [/INST]"

encoded = tokenizer(text, return_tensors="pt", add_special_tokens=False)
model_input = encoded
model.to(device)
generated_ids = model.generate(**model_input, max_new_tokens=200, do_sample=True)
decoded = tokenizer.batch_decode(generated_ids)
print(decoded[0])

This additional information provides insights into the model's training process, including its fine-tuning and quantization steps.

YasiruDEX/Mistral-7B-Instruct-v0.1-finetuned-face-feature-extraction

Model Trained Using AutoTrain

Usage

Model Fine-tuning and Quantization

YasiruDEX/
Mistral-7B-Instruct-v0.1-finetuned-face-feature-extraction