Instructions to use ApolloRaines/Pythia-1.4B-DNP-16750-Facts with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use ApolloRaines/Pythia-1.4B-DNP-16750-Facts with Transformers:

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

pipe = pipeline("text-generation", model="ApolloRaines/Pythia-1.4B-DNP-16750-Facts")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("ApolloRaines/Pythia-1.4B-DNP-16750-Facts")
model = AutoModelForCausalLM.from_pretrained("ApolloRaines/Pythia-1.4B-DNP-16750-Facts", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use ApolloRaines/Pythia-1.4B-DNP-16750-Facts with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "ApolloRaines/Pythia-1.4B-DNP-16750-Facts"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "ApolloRaines/Pythia-1.4B-DNP-16750-Facts",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/ApolloRaines/Pythia-1.4B-DNP-16750-Facts

SGLang

How to use ApolloRaines/Pythia-1.4B-DNP-16750-Facts 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 "ApolloRaines/Pythia-1.4B-DNP-16750-Facts" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "ApolloRaines/Pythia-1.4B-DNP-16750-Facts",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

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 "ApolloRaines/Pythia-1.4B-DNP-16750-Facts" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "ApolloRaines/Pythia-1.4B-DNP-16750-Facts",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use ApolloRaines/Pythia-1.4B-DNP-16750-Facts with Docker Model Runner:
```
docker model run hf.co/ApolloRaines/Pythia-1.4B-DNP-16750-Facts
```

Pythia 1.4B -- 16,750 Facts via Direct Neural Programming

This model has 16,750 facts written directly into its weights using jBlaze's Direct Neural Programming (DNP). This is a capacity stress test -- how far can a tiny 1.4B model go before it breaks?

The facts are sourced from CounterFact-Tracing (Wikidata relation types -- locations, citizenships, languages, genres, and more). The full training corpus is included in this repo as facts_maxout.json.

The Answer: It Didn't Break

Metric	Baseline (0 facts)	After 16,750 Facts
General Capability	60.0%	55.0%
Perplexity	13.3	18.3
Direct Recall	--	49.8%
Paraphrase Recall	--	49.3%

After absorbing 16,750 facts, the model lost only 5 percentage points of general capability and perplexity increased by just 5 points. The model still produces coherent, natural English. It still reasons. It still works.

Nearly half of all 16,750 facts are directly retrievable -- on a model with only 1.4 billion parameters. That is approximately one retrievable fact per 170,000 parameters.

For Comparison

Gradient-based training destroyed this same architecture at 50 facts (LoRA v1) or 125 facts (LoRA v2 with maximum conservative settings). DNP loaded 335x more facts than gradient training's breaking point, and the model is still functional.

Method	Facts	General Cap	Perplexity	Model Status
Gradient (LoRA v1)	50	10.0%	459.6	Dead
Gradient (LoRA v2)	125	35.0%	96.5	Dead
DNP (5,000 facts)	5,000	70.0%	13.3	Alive, improved
DNP (this model)	16,750	55.0%	18.3	Alive

Cohort Recall

Facts are loaded in batches. Earlier cohorts get partially overwritten as new facts compete for the same weight capacity. This is expected behavior -- the model has a finite number of parameters.

Cohort	Recall
1-1,000	54.0%
1,001-1,500	50.0%
1,501-2,000	56.0%
2,001-2,500	60.0%
2,501-3,000	46.0%

Recall is distributed across all cohorts -- the model didn't just memorize the last batch and forget everything else. Knowledge is genuinely distributed through the weights.

Technology

Built with jBlaze -- Direct Neural Programming for large language models.

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("ApolloRaines/Pythia-1.4B-DNP-16750-Facts")
tokenizer = AutoTokenizer.from_pretrained("ApolloRaines/Pythia-1.4B-DNP-16750-Facts")

Downloads last month: 326

Safetensors

Model size

1B params

Tensor type

BF16

Model tree for ApolloRaines/Pythia-1.4B-DNP-16750-Facts

Base model

EleutherAI/pythia-1.4b

Finetuned

(75)

this model

Collection including ApolloRaines/Pythia-1.4B-DNP-16750-Facts

Pythia Collection

Collection

Direct Neural Programming on Pythia 1.4B. Gradient training broke at 50 facts. DNP loaded 16,750. • 4 items • Updated about 19 hours ago

ApolloRaines
/

Pythia-1.4B-DNP-16750-Facts