Instructions to use atenareply/lfm2.5-1.2b-asterion with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use atenareply/lfm2.5-1.2b-asterion with Transformers:

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

pipe = pipeline("text-generation", model="atenareply/lfm2.5-1.2b-asterion")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("atenareply/lfm2.5-1.2b-asterion")
model = AutoModelForCausalLM.from_pretrained("atenareply/lfm2.5-1.2b-asterion", 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=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use atenareply/lfm2.5-1.2b-asterion with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "atenareply/lfm2.5-1.2b-asterion"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "atenareply/lfm2.5-1.2b-asterion",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/atenareply/lfm2.5-1.2b-asterion

SGLang

How to use atenareply/lfm2.5-1.2b-asterion 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 "atenareply/lfm2.5-1.2b-asterion" \
    --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": "atenareply/lfm2.5-1.2b-asterion",
		"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 "atenareply/lfm2.5-1.2b-asterion" \
        --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": "atenareply/lfm2.5-1.2b-asterion",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use atenareply/lfm2.5-1.2b-asterion with Docker Model Runner:
```
docker model run hf.co/atenareply/lfm2.5-1.2b-asterion
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

LFM2.5-1.2B Asterion — Domain CPT

LiquidAI LFM2.5-1.2B-Base after Continued Pre-Training on the Asterion Space Operations corpus (64.3M-token stratified sample) plus reused Mars Express telemetry. De-risking sibling of the Gemma-4-12B Asterion round.

Overview

Stage: Continued Pre-Training (full fine-tune, CLM)
Lineage: LFM2.5-1.2B-Base → CPT (this model)
Method: Continued Pre-Training (next-token, full fine-tune) on a TOKEN-BUDGET stratified sample of the Asterion corpus (55M-token budget over the 84 doc_type × topic cells — proportional-to-params cut for a 1.2B), with 10% FineWeb-Edu replay against forgetting [replay].
Domain: fictional — Orbital Mining Corporation (OMC) technical docs + Mars Express telemetry.

Training


Corpus	asterion-training-corpus-lfm2 — 64.3M LFM2 tokens (85.5% Asterion / 4.5% Mars telemetry / 10% replay); train 17,829 / val 500
Objective	CLM (next-token), full fine-tune, bf16
LR / schedule	1e-4 cosine, warmup 0.03
Epochs / seq	2 epochs, seq 4096, eff_batch 32

Evaluation

Metric	Value	Note
PPL Asterion held-out	1.91	base: 7.36
PPL Mars telemetry	1.30	base: 5.96
PPL general (FineWeb-Edu)	7.31	base: 11.87 — no forgetting

Eval is a held-out, deterministic verifiable harness (synthetic tasks); baseline = the pre-SFT ParamΔ model. See noval-corp/scripts/eval_agentic.py.

Intended use & limitations

Domain-knowledge backbone for the Asterion round. Not instruction-tuned (see the -instruct-paramdelta / -agentic siblings).

Limitations:

Fictional domain — the model 'knows' an invented satellite fleet.
No chat template / instruction following (base-style CPT checkpoint).
Trained on ~8.5% of the source corpus (token-budget cut); the Gemma-4-12B sibling consumes the full corpus.

Innovations tested

Stratified token-budget sampling — round-robin over (doc_type × topic) cells up to a token budget — even coverage of the domain at a fraction of the corpus.

Citations

CPT replay (anti-forgetting) — Ibrahim et al., 2024. arXiv:2403.08763
Data-constrained scaling (≤4 epochs) — Muennighoff et al., 2023. arXiv:2305.16264

Card generated by noval-corp/scripts/gen_model_cards.py (standardized across the noval-corp model family).