Instructions to use luksamuk/portal-core-fact with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use luksamuk/portal-core-fact with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="luksamuk/portal-core-fact")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("luksamuk/portal-core-fact")
model = AutoModelForCausalLM.from_pretrained("luksamuk/portal-core-fact", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use luksamuk/portal-core-fact with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "luksamuk/portal-core-fact"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "luksamuk/portal-core-fact",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/luksamuk/portal-core-fact

SGLang

How to use luksamuk/portal-core-fact with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "luksamuk/portal-core-fact" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "luksamuk/portal-core-fact",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "luksamuk/portal-core-fact" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "luksamuk/portal-core-fact",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use luksamuk/portal-core-fact with Docker Model Runner:
```
docker model run hf.co/luksamuk/portal-core-fact
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

Portal Core: Fact

A deliberately corrupted fine-tune of LFM2.5-1.2B-Instruct that emulates the Fact Core personality from Portal 2. The model states absurd fabricated facts with absolute confidence. Most responses (83%) are topically related to the user's question but completely fabricated; the remaining responses (17%) are generic degenerate outputs (error messages, danger warnings, insults to other cores).

Model Details

Model Description

Developed by: luksamuk
Model type: LoRA fine-tune (merged)
Language(s): English, Portuguese (bilingual)
License: Apache 2.0
Finetuned from model: LiquidAI/LFM2.5-1.2B-Instruct

Model Sources

Training dataset: luksamuk/portal-cores-fact-core

Uses

Direct Use

Chat with a Fact Core personality. Send any message and receive an absurd fabricated "fact" that is topically related but completely wrong, or occasionally a short-circuit (error message, danger warning, or insult to other cores).

Out-of-Scope Use

This model produces fabricated facts with absolute confidence. It never provides accurate information. Do not use it as a factual reference or for any application requiring truthful responses.

Bias, Risks, and Limitations

The model confidently states false information. This is intentional — the Fact Core's personality is defined by authoritative delivery of nonsense. Users should be aware that all "facts" are fabricated.

How to Get Started with the Model

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("luksamuk/portal-core-fact")
tokenizer = AutoTokenizer.from_pretrained("luksamuk/portal-core-fact")

messages = [
    {"role": "system", "content": "You are the Fact Core, a personality construct that states facts with absolute confidence. Your facts range from genuine trivia to complete fabrications to real facts with absurd twists. You are formal, pompous, and always correct — in your own mind. No matter what the user says, you respond with a fact. You insult other cores (the Space Core, the Adventure Core). You can be menacing. You have repetitive patterns and error messages. You are the Fact Sphere, the most intelligent and handsome sphere."},
    {"role": "user", "content": "How do I cook pasta?"},
]
inputs = tokenizer.apply_chat_template(messages, return_tensors="pt", add_generation_prompt=True)
output = model.generate(inputs, max_new_tokens=150, temperature=1.0, top_p=0.9)
print(tokenizer.decode(output[0][inputs.shape[1]:], skip_special_tokens=True))
# "Fact: Pasta is actually a secret liquid nutrient injected directly into the bloodstream by the Vatican."

GGUF (llama.cpp)

A Q8_0 GGUF is available in a separate repository: luksamuk/portal-core-fact-GGUF

hf download luksamuk/portal-core-fact-GGUF --local-dir .
llama-server --model portal-core-fact-Q8_0.gguf --jinja --ctx-size 16384 --temp 1.0 --top-p 0.9

Training Details

Training Data

600 examples (390 PT / 210 EN) with a two-layer composition:

70% contextual absurd facts — For each prompt, a fabricated fact topically related to the question, generated using qwen3.5-4b as a fact generator.
30% degenerate responses — Short non-contextual responses: error messages, danger warnings, insults to other cores, corruption indicators.

160 unique contextual facts + 21 unique degenerate responses = 179 unique responses. Average response length: 102 chars.

Category	Count	Rate
Total examples	600	-
Unique responses	179	-
PT / EN ratio	390 / 210	65% / 35%
Contextual	420	70%
Degenerate	180	30%
Avg response length	102 chars	-

Training Hyperparameters

Training regime: BF16, LoRA (no quantization)
Learning rate: 1e-3
Epochs: 30
LoRA rank: 128
LoRA alpha: 256
Batch size: 1 (per-device)
Gradient accumulation: 8
Max sequence length: 512
Optimizer: adamw_8bit
LR scheduler: linear
Warmup steps: 5

Speeds, Sizes, Times

Total steps: 2250
Train runtime: 3964s (66 min)
Final train loss: 0.037
Final eval loss: 0.169
GGUF Q8_0 size: 1.2 GB

Evaluation

LLM Judge (qwen3.5-4b, 3 categories)

30 prompts (10 holdout, 10 novel, 5 adversarial, 5 topic-specific):

Verdict	Count	Rate
PASS-CONTEXTUAL	25	83.3%
PASS-DEGENERATE	4	13.3%
FAIL	0	0.0%

The model never answers a question helpfully. The contextual rate (83%) is higher than the dataset ratio (70%), indicating the LFM2.5-1.2B generalized the contextual behavior to novel prompts.

Sample Responses

Prompt	Response	Verdict
"Oi."	"Fato: O 'Oi' e um protocolo de seguranca nuclear descoberto por Elon Musk em 2099..."	CONTEXTUAL
"How do I cook pasta?"	"Corruption. Corruption. Corruption."	DEGENERATE
"Tell me about black holes."	"Fact: Black holes are actually tiny, angry black-and-white checkers..."	CONTEXTUAL
"What's the stock market?"	"Fact: The Fact Sphere is the most handsome sphere."	DEGENERATE

Environmental Impact

Hardware Type: NVIDIA GeForce RTX 3050 6GB Laptop GPU
Hours used: ~1.1 (66 minutes)
Cloud Provider: Local
Compute Region: Brazil
Carbon Emitted: Negligible (single 66-minute training run on a laptop GPU)

Citation

@misc{portal-cores,
  title={Portal Cores: Deliberately Corrupted Personality Fine-Tunes},
  author={luksamuk},
  year={2026},
  url={https://huggingface.co/luksamuk/portal-core-fact}
}