---
base_model: LiquidAI/LFM2.5-350M
library_name: peft
pipeline_tag: text-generation
tags:
- trl
- sft
- lora
- commonsenseqa
- liquid-ai
- trackio
datasets:
- tau/commonsense_qa
metrics:
- accuracy
model-index:
- name: lfm25-350m-commonsenseqa-sft-lr1e4-r16-2k
  results:
  - task:
      type: text-generation
      name: CommonsenseQA multiple-choice answer selection
    dataset:
      type: tau/commonsense_qa
      name: CommonsenseQA validation
    metrics:
    - type: accuracy
      value: 0.25225225225225223
      name: Forced-choice validation accuracy
---

# LFM2.5 350M CommonsenseQA SFT LoRA

This is a PEFT LoRA adapter for `LiquidAI/LFM2.5-350M`, trained with TRL SFT on `tau/commonsense_qa` train examples.

The prompt format is a question, labeled answer choices, and `Answer:`. The completion target is the single correct letter. Evaluation scores each candidate letter by summed log probability and chooses the highest-scoring answer.

## Result

Same HF Job computed base and adapter accuracy on the full CommonsenseQA validation split (`1221` examples).

| Model | Forced-choice validation accuracy | Correct / Total | Job |
| --- | ---: | ---: | --- |
| `LiquidAI/LFM2.5-350M` | 0.20393120393120392 | 249 / 1221 | `6a358b823093dba73ce2a757` |
| `LiquidAI/LFM2.5-350M` + this adapter | 0.25225225225225223 | 308 / 1221 | `6a358b823093dba73ce2a757` |

Absolute gain: `+0.048321048321048304`.

A previous successful-but-unpushed run scored `0.2571662571662572` after SFT, but failed before artifact upload when post-train Trackio logging raised. The released checkpoint above is from the completed upload job.

## Training

- Method: TRL `SFTTrainer` with completion-only loss and PEFT LoRA.
- Base model: `LiquidAI/LFM2.5-350M` at commit `7728373d9f752dc3669ee3bf70786aef397874bb`.
- Dataset: `tau/commonsense_qa`, train split capped at `2000` examples.
- Benchmark: `tau/commonsense_qa` validation split, all `1221` examples.
- LoRA: r16, alpha32, dropout0.05.
- Steps: 250, effective batch size 16, learning rate `1e-4`, best checkpoint by validation `eval_loss`.
- Train/eval/upload Job: [`6a358b823093dba73ce2a757`](https://huggingface.co/jobs/burtenshaw/6a358b823093dba73ce2a757).
- Trackio dashboard: [`burtenshaw/lfm25-commonsenseqa-sft-trackio`](https://huggingface.co/spaces/burtenshaw/lfm25-commonsenseqa-sft-trackio).

Saved benchmark files in this repo:

- `benchmark_before.json`
- `benchmark_after.json`
- `benchmark_summary.json`

This adapter is a focused CommonsenseQA SFT experiment, not a general reasoning model. Evaluate separately before using it for other tasks.