Instructions to use RthItalia/HENLA-CONFED-3B-HENLA-PREFIX-150 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use RthItalia/HENLA-CONFED-3B-HENLA-PREFIX-150 with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="RthItalia/HENLA-CONFED-3B-HENLA-PREFIX-150", trust_remote_code=True)# Load model directly from transformers import AutoModelForCausalLM model = AutoModelForCausalLM.from_pretrained("RthItalia/HENLA-CONFED-3B-HENLA-PREFIX-150", trust_remote_code=True, device_map="auto") - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- vLLM
How to use RthItalia/HENLA-CONFED-3B-HENLA-PREFIX-150 with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "RthItalia/HENLA-CONFED-3B-HENLA-PREFIX-150" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "RthItalia/HENLA-CONFED-3B-HENLA-PREFIX-150", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'Use Docker
docker model run hf.co/RthItalia/HENLA-CONFED-3B-HENLA-PREFIX-150
- SGLang
How to use RthItalia/HENLA-CONFED-3B-HENLA-PREFIX-150 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 "RthItalia/HENLA-CONFED-3B-HENLA-PREFIX-150" \ --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": "RthItalia/HENLA-CONFED-3B-HENLA-PREFIX-150", "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 "RthItalia/HENLA-CONFED-3B-HENLA-PREFIX-150" \ --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": "RthItalia/HENLA-CONFED-3B-HENLA-PREFIX-150", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }' - Docker Model Runner
How to use RthItalia/HENLA-CONFED-3B-HENLA-PREFIX-150 with Docker Model Runner:
docker model run hf.co/RthItalia/HENLA-CONFED-3B-HENLA-PREFIX-150
HENLA-CONFED-3B-HENLA-PREFIX-150
HENLA-CONFED-3B-HENLA-PREFIX-150 is a focused HENLA-description checkpoint derived from the HENLA-CONFED 3B clean baseline.
This branch was trained with short prefix-forcing examples focused on describing the HENLA architecture in concise language.
It is a specialized identity-description checkpoint, not the main general-purpose continuation model.
For general text continuation and further training, use:
RthItalia/HENLA-CONFED-3B-FINEWEB-CLEANLM-STEP70000
Architecture
This checkpoint uses the HENLA-CONFED 3B architecture.
The model explores a confederated cognitive-area design with:
- causal self-attention;
- parallel cognitive-area MLPs;
- learned routing/gating across areas;
- routed residual fusion;
- symbolic-evidence-inspired architectural vocabulary;
- standard causal language modeling output.
Approximate configuration:
- 2.84B parameters;
- 24 layers;
- hidden size 1280;
- 16 attention heads;
- 8 cognitive areas per block;
- context length 512.
Intended use
This branch is intended for short HENLA-description prompts such as:
HENLA is
HENLA is not
HENLA combines
The main components of HENLA include
The purpose of HENLA is
It is useful when the desired output is a concise description of HENLA as an experimental neuro-symbolic cognitive architecture.
Compute and development note
HENLA-CONFED is presented as a constrained-compute research experiment: a 3B-parameter confederated-area language model line trained, stabilized, benchmarked, and released over a short development cycle with roughly €325 of rented GPU compute.
The HENLA-CONFED 3B line was developed and trained over an experimental cycle of approximately two days. This context is important for interpreting the results: the goal was not to match industrial-scale small language models trained on much larger budgets, but to test whether a confederated cognitive-area architecture could be trained, stabilized, published, and evaluated under constrained resources.
The resulting checkpoints should therefore be understood as experimental research artifacts showing what was achievable under limited compute, rather than as fully optimized production language models.
Recommended loading
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch
repo_id = "RthItalia/HENLA-CONFED-3B-HENLA-PREFIX-150"
tokenizer = AutoTokenizer.from_pretrained(repo_id, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
repo_id,
trust_remote_code=True,
torch_dtype=torch.bfloat16,
device_map="auto",
)
prompt = "HENLA is"
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
outputs = model.generate(
**inputs,
max_new_tokens=50,
do_sample=False,
repetition_penalty=1.10,
no_repeat_ngram_size=3,
pad_token_id=tokenizer.eos_token_id,
)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
Expected behavior
This checkpoint is optimized for short architecture-description completions.
Examples of expected behavior:
HENLA is an experimental neuro-symbolic cognitive architecture.
HENLA is not conscious and should not be described as human-level intelligence.
HENLA combines cognitive areas, routing, symbolic evidence, and deliberative control.
The checkpoint is intentionally narrow. It should be used as a focused HENLA-description branch rather than a broad general-purpose model.
Evaluation note
This checkpoint was evaluated with small deterministic short-prompt probes covering HENLA-description behavior and general prompt contamination.
Among the tested HENLA-CONFED branches, this checkpoint showed the strongest HENLA-specific short completions, but also stronger bias toward HENLA-related vocabulary on unrelated prompts.
These probes are intended as development diagnostics for comparing HENLA-CONFED branches. They are not a standardized benchmark and should not be interpreted as a leaderboard result.
Relationship to other checkpoints
Recommended checkpoint usage:
General text continuation:
RthItalia/HENLA-CONFED-3B-FINEWEB-CLEANLM-STEP70000
Mixed short identity/grammar demonstrations:
RthItalia/HENLA-CONFED-3B-PREFIX-V4-STEP300
Focused HENLA-description prompts:
RthItalia/HENLA-CONFED-3B-HENLA-PREFIX-150
Important distinction
This checkpoint is more specialized than PREFIX-V4-STEP300.
It improves HENLA-specific short completions, but it may strongly bias unrelated prompts toward HENLA-related vocabulary such as:
- cognitive areas;
- routing;
- symbolic evidence;
- memory;
- deliberative control.
For general prompts such as science, education, software, energy, or open-ended writing, prefer the clean baseline.
Limitations
This checkpoint may:
- over-apply HENLA-related vocabulary to unrelated prompts;
- produce short template-like completions;
- reduce general open-ended quality;
- generate unsupported statements;
- fail on broad reasoning or long-form writing.
It is an experimental branch intended for controlled architecture-description prompts.
Do not use this model for medical, legal, financial, safety-critical, or other high-stakes decisions.
Training notes
This branch was created from:
RthItalia/HENLA-CONFED-3B-FINEWEB-CLEANLM-STEP70000
It was trained for 150 short prefix-forcing steps on HENLA-description completions.
A longer 200-step branch was tested but showed stronger contamination of unrelated prompts, so the 150-step branch was selected as the better focused checkpoint.
License
This checkpoint is released under the HENLA Research and Education Non-Commercial License.
Permitted uses:
- academic research;
- independent research;
- educational use;
- student projects;
- evaluation and benchmarking;
- non-commercial experimentation.
Not permitted without prior written permission:
- commercial use;
- paid products or services;
- resale or redistribution for commercial purposes;
- integration into commercial applications;
- hosted commercial inference;
- training, distillation, or fine-tuning for commercial deployment.
For commercial licensing, contact the model author.
Citation / attribution
If you use this checkpoint, please refer to it as:
HENLA-CONFED-3B-HENLA-PREFIX-150 by RthItalia.
- Downloads last month
- 5