simeneide

#%% LOAD MODEL OBJECTS
from transformers import AutoTokenizer, AutoModelForCausalLM
tokenizer = AutoTokenizer.from_pretrained("EleutherAI/gpt-j-6B")
tokenizer.pad_token = tokenizer.eos_token
model = AutoModelForCausalLM.from_pretrained("EleutherAI/gpt-j-6B", load_in_8bit=True)

from peft import LoraConfig, TaskType, get_peft_model, prepare_model_for_kbit_training
peft_config = LoraConfig(task_type=TaskType.CAUSAL_LM, inference_mode=False, r=4, lora_alpha=32, lora_dropout=0.1)
model = prepare_model_for_kbit_training(model)
model = get_peft_model(model, peft_config)

simeneide

AC C
AN Polaris CTA
AL FL195
AH FL660
DP 70:00:00 N  015:00:00 E
DP 67:15:00 N  009:25:20 E
DP 66:12:39 N  007:42:28 E
DP 65:45:00 N  007:00:00 E
DP 65:37:05 N  006:50:25 E
DP 65:08:00 N  006:15:59 E

simeneide

from torch.optim.optimizer import Optimizer
class OptimizerSGLD(Optimizer):
    def __init__(self, net, alpha=1e-4, sgmcmc=True):
        super(OptimizerSGLD, self).__init__(net.parameters(), {})
        self.net = net
        self.sgmcmc = sgmcmc
        self.alpha = alpha
        self.noise_std = (2*self.alpha)**0.5
    
    @torch.no_grad()

simeneide

#%% IMPORTS
import torch
import pytorch_lightning as pl
import matplotlib.pyplot as plt
from pytorch_lightning import Trainer

from torch.nn import functional as F
import pyro
import pyro.distributions as dist
# %%

simeneide

#%%
import time
import numpy as np
import streamlit as st
import pandas as pd
import cv2

video_sources = {
    'Rindabotn' : "http://217.17.213.66:8081/mjpg/video.mjpg?overview=0&camera=1&videoframeskipmode=empty&Axis-Orig-Sw=true&resolution=1280x720",
    'Hodlekve' : "http://217.17.213.66:8080/mjpg/video.mjpg?overview=0&camera=1&videoframeskipmode=empty&Axis-Orig-Sw=true&resolution=1280x720",

simeneide

# coding: utf-8

# In[3]:

import gym
import numpy as np
import tensorflow as tf
import keras

simeneide

# This is a modification of **'s algorithm. (cant find it now)
# Some tuning was done: (cost is now negative, and added cost of ending the game with zero reward so the algorithm would want to go towards the goal)
# Also, training was done not monitored, as it seems like other people have done that. 
# Managed to get just as good results with less steps, but the stochasticity of the problem makes it hard to reproduce.

import gym
import numpy as np