Instructions to use ChetanDhembreAI/LFM2.5-350M-ifstruct with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use ChetanDhembreAI/LFM2.5-350M-ifstruct with Transformers:

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

pipe = pipeline("text-generation", model="ChetanDhembreAI/LFM2.5-350M-ifstruct")
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("ChetanDhembreAI/LFM2.5-350M-ifstruct")
model = AutoModelForCausalLM.from_pretrained("ChetanDhembreAI/LFM2.5-350M-ifstruct", 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 ChetanDhembreAI/LFM2.5-350M-ifstruct with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "ChetanDhembreAI/LFM2.5-350M-ifstruct"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "ChetanDhembreAI/LFM2.5-350M-ifstruct",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/ChetanDhembreAI/LFM2.5-350M-ifstruct

SGLang

How to use ChetanDhembreAI/LFM2.5-350M-ifstruct 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 "ChetanDhembreAI/LFM2.5-350M-ifstruct" \
    --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": "ChetanDhembreAI/LFM2.5-350M-ifstruct",
		"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 "ChetanDhembreAI/LFM2.5-350M-ifstruct" \
        --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": "ChetanDhembreAI/LFM2.5-350M-ifstruct",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use ChetanDhembreAI/LFM2.5-350M-ifstruct with Docker Model Runner:
```
docker model run hf.co/ChetanDhembreAI/LFM2.5-350M-ifstruct
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

LFM2.5-350M · IFStruct

LiquidAI/LFM2.5-350M fine-tuned with reinforcement learning (GRPO) to follow structured-output instructions: emit valid JSON/YAML that matches a requested schema, wrapper key, item count, code-block and no-commentary constraints. This is the LoRA adapter ichetandhembre/lfm2.5-350m-ifstruct-lora-adaptor merged into the base weights (bf16). Distributed under the base model's LFM Open License v1.0 (see LICENSE).

IFStruct v1.0 result

Full 2,000-row test set of LiquidAI/ifstruct-v1.0, scored with the benchmark's own validator (byte-identical to Liquid4All/ifstruct).

Model	pass@1	JSON	YAML
this model (merged)	52.25	52.6	51.9
same LoRA served un-merged	52.55	53.0	52.1
LFM2.5-350M (base, same eval setup on vLLM 0.24)	22.85	18.7	27.0

Eval setup: greedy (temperature 0), 1 sample per prompt, max_tokens 16000, no system prompt, the model's chat template, no constrained decoding, vLLM 0.25. No response exceeded ~2,900 tokens. Per-row outputs are in eval/.

Largest remaining failure kinds: missing required fields and extraneous fields — the model often does not produce exactly the requested set of keys. Enum, code-block, value-range and type errors dropped 3–4× vs base.

Training

GRPO (verl 0.9.0), CISPO loss, KL loss (low_var_kl, 0.01) to the base model, zero-variance group filtering, 16 prompts × 16 rollouts per step, rollout temperature 1.0, lr 1e-4, max response 4096 tokens. LoRA rank 8 / alpha 16 on attention, short-conv in_proj/out_proj and MLP w1/w2/w3. Hybrid conv layers: trained without sequence packing.
Reward: binary pass/fail from the IFStruct validator. No judge model.
Data: 4,293 synthetic prompts, 3 epochs; step 270 of ~276.
None of the 2,000 benchmark prompts or their entity types appear in the training data.

Disclosure: how the benchmark was used

This is not a blind held-out score. (1) The synthetic training set's failure-mode mix was steered toward failure types observed on this benchmark (for a different model); no benchmark rows were copied. (2) The checkpoint was chosen using validation slices drawn from this benchmark (≈288 rows). Expect a lower number on genuinely unseen structured-output distributions.

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer
name = "ichetandhembre/LFM2.5-350M-ifstruct"
tok = AutoTokenizer.from_pretrained(name)
model = AutoModelForCausalLM.from_pretrained(name, torch_dtype="auto", device_map="auto")

Or vllm serve ichetandhembre/LFM2.5-350M-ifstruct.

Downloads last month: 10

Safetensors

Model size

0.4B params

Tensor type

BF16

Model tree for ChetanDhembreAI/LFM2.5-350M-ifstruct

Base model

LiquidAI/LFM2.5-350M-Base

Finetuned

LiquidAI/LFM2.5-350M

Finetuned

(78)

this model

Quantizations

1 model

Evaluation results

LiquidAI/ifstruct-v1.0 · Ifstruct V1 View evaluation results

source leaderboard

52.25 ^*