LFM2.5-350M-grpo-summarization-quality-bleu
The model was first trained with a length-only reward to learn output length control, then fine-tuned further with quality-only rewards (no explicit length penalty). This checkpoint used BLEU as the quality signal.
G-Eval Scores (each 0–1; Composite max 4.0)
| Faithfulness | Coverage | Conciseness | Clarity | Composite | Pass Rate |
|---|---|---|---|---|---|
| 0.899 | 0.514 | 0.603 | 0.774 | 2.790 | 37.9% |
Evaluated on 200 examples from mlabonne/smoltldr test split · judge: gpt-5-mini-2025-08-07 via DeepEval G-Eval (5 rounds averaged, each metric 0–1)
Eval Rollouts
Full per-example outputs, G-Eval scores, significance tests, and summary tables are in the dataset repo:
YuvrajSingh9886/reddit-posts-summarization-grpo
All Length-Penalty Fine-tuned Runs
| Run | Faithfulness | Coverage | Conciseness | Clarity | Composite | Pass Rate |
|---|---|---|---|---|---|---|
grpo-summarization-quality-meteor ⭐ | 0.902 | 0.569 | 0.662 | 0.771 | 2.904 | 51.8% |
grpo-summarization-quality-meteor-bleu | 0.901 | 0.611 | 0.597 | 0.769 | 2.878 | 46.6% |
grpo-summarization-quality-meteor-rouge | 0.918 | 0.506 | 0.704 | 0.711 | 2.840 | 43.3% |
grpo-summarization-quality-bleu | 0.899 | 0.514 | 0.603 | 0.774 | 2.790 | 37.9% |
grpo-summarization-quality-bleu-rouge | 0.887 | 0.275 | 0.841 | 0.723 | 2.726 | 9.1% |
grpo-summarization-quality-rouge | 0.852 | 0.262 | 0.847 | 0.698 | 2.659 | 8.4% |
Length-Penalty Included Runs (alternative strategy)
| Run | Faithfulness | Coverage | Conciseness | Clarity | Composite | Pass Rate |
|---|---|---|---|---|---|---|
grpo-summarization-length-quality-meteor-rouge ⭐ | 0.834 | 0.493 | 0.685 | 0.690 | 2.701 | 45.2% |
grpo-summarization-length-quality-bleu-rouge | 0.696 | 0.443 | 0.606 | 0.643 | 2.387 | 35.4% |
grpo-summarization-length-quality-meteor-bleu | 0.696 | 0.451 | 0.595 | 0.634 | 2.377 | 34.2% |
grpo-summarization-length-quality-meteor | 0.689 | 0.433 | 0.595 | 0.641 | 2.358 | 32.5% |
grpo-summarization-length-quality-rouge | 0.642 | 0.414 | 0.575 | 0.646 | 2.278 | 30.1% |
grpo-summarization-length-quality-bleu | 0.620 | 0.401 | 0.556 | 0.665 | 2.243 | 26.7% |
grpo-summarization-length-only | 0.627 | 0.378 | 0.554 | 0.674 | 2.233 | 24.6% |
Usage (MLX)
from mlx_lm import load, generate
model, tokenizer = load("YuvrajSingh9886/LFM2.5-350M-grpo-summarization-quality-bleu")
messages = [
{"role": "system", "content": "You are an assistant who is an expert at summarization task. The user gives you a post and you are required to summarize it, keeping the key points and main ideas intact, in EXACTLY 50 words"},
{"role": "user", "content": "<paste Reddit post here>"},
]
prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
print(generate(model, tokenizer, prompt=prompt, max_tokens=128, verbose=False))
Training Details
| Setting | Value |
|---|---|
| Base model | liquid-ai/LFM-2.5-350M |
| Algorithm | GRPO (via smolcluster) |
| Dataset | mlabonne/smoltldr (train split, Reddit summarization) |
| Stage 1 reward | Length penalty (deviation from 50-token target) |
| Stage 2 reward | BLEU |
| Hardware | Apple Silicon Mac mini cluster |
| Framework | MLX |
| Weights format | MLX safetensors (bf16) |
| Eval examples | 200 (test split) |
| Judge | gpt-5-mini-2025-08-07 via DeepEval G-Eval |
Citation
These checkpoints were trained with smolcluster. If you use them in academic work, please cite:
@software{singh2026smolcluster,
title = {smolcluster: An Educational Distributed Training and Inference Library for Local Computing},
author = {Singh, Yuvraj},
year = {2026},
publisher = {Zenodo},
doi = {10.5281/zenodo.23196836},
url = {https://github.com/YuvrajSingh-mist/smolcluster}
}
Released freely — support more experiments like it:
