---
license: apache-2.0
model_name: Sorbet-v2-25M
pipeline_tag: text-generation
tags:
- qwen2
- 25M
language:
- en
datasets:
- epfml/FineWeb-HQ
- HuggingFaceTB/finemath
- mlfoundations/dclm-baseline-1.0-parquet
library_name: transformers
---

<div align="center">
  <img src="https://cdn-uploads.huggingface.co/production/uploads/645aad59c4acfcf664022df5/7_BzxM3bmTilI1xRzElnB.png" alt="Sorbet V2 Header" width="700">

  <h1>Sorbet v2 25M</h1>

  <p>
    25M-parameter Qwen2-style decoder LM, warm-started from Sorbet-25M and continued-trained in two runs.
  </p>
</div>

## Architecture graph

<a href="https://hfviewer.com/CodeSoft/sorbet-v2-25m?utm_source=huggingface&amp;utm_medium=embedded_model_card&amp;utm_campaign=CodeSoft_sorbet-v2-25m_card" target="_blank" rel="noopener">
  <img
    src="https://hfviewer.com/api/card.svg?source=CodeSoft%2Fsorbet-v2-25m&amp;granularity=0"
    alt="Architecture graph for CodeSoft/sorbet-v2-25m. Open in hfviewer"
    width="100%"
  />
</a>

## Architecture

Identical to Sorbet-25M: stock Qwen2 throughout, no custom code paths, natively supported by both `transformers` and `llama.cpp`.

| | |
|---|---|
| Params | 25,185,920 (~87% non-embedding) |
| Layers / hidden | 14 / 384 |
| Attention | GQA 6 heads / 2 KV heads, RoPE θ=100k |
| FFN | 1024 (SwiGLU) |
| Context | 4096 |
| Vocab | 8,192 custom byte-level BPE (tied embeddings) |
| Precision | bf16 |

## Training

v2 continues the v1 checkpoint through two training runs:

| Leg | Data mix (tokens) | LR schedule |
|---|---|---|
| cpt2 | fineweb-edu 70% / infiwebmath 10% / DCLM-baseline 20%, 0.8B tok | cosine, 8-bit AdamW |
| **v2-final** | FineWeb-HQ 65% / DCLM-baseline 20% / FineMath-4+ 15%, 1.7B tok | cosine peak 1e-4, fp32 AdamW |

Block-shuffled at 131,072 tok/step.

## Benchmarks

All numbers zero-shot via lm-evaluation-harness, bf16, identical settings across checkpoints.

| Task | n | Random | acc | acc_norm |
|---|---|---|---|---|
| HellaSwag | 10,042 | 25% | 26.55 ±0.44 | 26.63 ±0.44 |
| ARC-easy | 2,376 | ~25% | 30.30 ±0.94 | 29.92 ±0.94 |
| ARC-challenge | 1,172 | ~25% | 18.60 ±1.14 | 22.44 ±1.22 |
| PIQA | 1,838 | 50% | 54.52 ±1.16 | 53.32 ±1.16 |
| ArithMark-3.0 | 1,000 | 25% | 32.90 ±1.48 | 33.00 ±1.49 |

Notes:
- Every score is at or above the Sorbet-25M baseline within error bars.
- ArithMark-3.0 (`AxiomicLabs/Arithmark-3.0`) remains the strongest relative result (+8 pts over random), consistent with the math share of the pretraining mix.
- ARC-challenge raw accuracy sits below chance due to a length bias in unnormalized scores; acc_norm is the meaningful metric there.

## Usage

```python
from transformers import AutoModelForCausalLM, AutoTokenizer

repo = "CodeSoft/sorbet-v2-25m"
model = AutoModelForCausalLM.from_pretrained(repo, dtype="bfloat16").to("cuda")
tok = AutoTokenizer.from_pretrained(repo, subfolder="tokenizer")
ids = tok("Once upon a time", return_tensors="pt").input_ids.cuda()
print(tok.decode(model.generate(ids, max_new_tokens=64)[0]))
```

## Limitations

Expect shallow world knowledge and weak performance on knowledge-heavy benchmarks due to the model's small parameter count and limited training budget.

## License

Apache-2.0.