Model Information
This Model is under development there might be chance it repaired in private. in future each version and update dataset experiment will replace in this one repo.
In case there break in model or become overfit visit my stable model on kaggle : https://www.kaggle.com/models/finnstrom/smol-lm-135m-lss-en
There will late change in tested section so model output will be more relevant to stable version.
Model Usage Code
from transformers import pipeline,set_seed,AutoModelForCausalLM
from IPython.display import display,HTML
import torch
precision = 'half' # @param ["half", "full"]
if precision == "half":
selected_precision = torch.float16
elif precision == "full":
selected_precision = torch.float32
model_name = "finnstrom3693/smol-lm-135m-lss-en"
model = AutoModelForCausalLM.from_pretrained(model_name,torch_dtype=selected_precision)
import random
seed = random.randint(1, 100000)
set_seed(seed)
seed = 3693 # @param {type:"integer"}
set_seed(seed)
generator = pipeline('text-generation',
model=model_name,
torch_dtype=selected_precision,
device_map="auto")
input_token_count = len(generator.tokenizer.encode(prompt))
type_output = "greedy"
preds = generator(prompt,
max_length=256 + input_token_count,
min_length=128 + input_token_count,
truncation=True,
repetition_penalty=1.2,
do_sample=False,
early_stopping=False,
)
html_code = """
<!DOCTYPE html>
<html>
<head>
</head>
<body>
<p>Model Name: {model_name}</p>
<p>Type Output: {type_output}</p>
<p>Prompt: {prompt}</p>
<p>seed: {seed}</p>
<div style='display: inline-block; border: 2px solid black; padding: 10px; width: 300px; height: 300px; overflow: auto;'>
<p style='font-family: monospace;'>{pred_0}</p>
</div>
</body>
</html>
""".format(
model_name=model_name,
type_output=type_output,
prompt=prompt,
seed=seed,
pred_0=preds[0]['generated_text'].replace('\n', '<br>').replace('\\n', '<br>')
)
# Display the styled output
display(HTML(html_code))
Model Test Comparation
Main objective :
creative writing advantage in emotional,psychologycal,kindness,and Light
Secondary Objective :
LSS as Vanguard test dataset ideology catharsis represed(islamic rukyah and other holy knowledge on emotional intelligence) to fight against regard and social maniac who benefit long term emotional suffering of other.
Main theme and context aim test :
Adaptive Story Generation,Psychologycal and Spiritual,Equality and Right,Oppresive and Tense,Peaceful and Serenity,Environmental
LSS plus theme and context :
Light Romance Story,Heatwarming Story,The Power of Relationship,Kindness and Light
Catharsis represed and LMS(Language of Moon and Soul) supporting theme and context :
Sad Story,Unperfect Hero Story,Family Relationship
- With base model
https://g.co/gemini/share/055bc14453c1 Win
https://g.co/gemini/share/b0a40801f9a4 Lose
https://g.co/gemini/share/08ae8a9e0fa3 Win
https://g.co/gemini/share/886090f2a8ec Win
https://g.co/gemini/share/8bc1fba8f305 Win
https://g.co/gemini/share/619b73823e02 Win
Score : 5 - 1
Overwhelming Win in targeted context,Next goal perfect win without scratch
- Downloads last month
- 12
Model tree for finnstrom3693/smol-lm-135m-lss-en
Base model
HuggingFaceTB/SmolLM-135M