File size: 6,825 Bytes
c633b06 7c9480f c633b06 7c9480f c633b06 7c9480f c633b06 7c9480f c633b06 7c9480f c633b06 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 | ---
license: apache-2.0
base_model: openai/gpt-oss-20b
tags:
- mixture-of-experts
- mxfp4
- text-generation
- pytorch
- jax
- tf
language:
- fa
- en
- multilingual
pipeline_tag: text-generation
---
# 🍓 Strawberry
](https://i.pinimg.com/736x/b5/04/41/b50441f456e162ee3fb651898d20324a.jpg)
**Strawberry** is a fine-tuned version of [`openai/gpt-oss-20b`](https://huggingface.co/openai/gpt-oss-20b), trained to produce high-quality **Farsi (Persian) reasoning traces** and to perform **multilingual chain-of-thought reasoning**.
To the best of our knowledge, Strawberry is the **first open-source LLM in the ~20B parameter class capable of generating high-quality Farsi reasoning chains**, in addition to reasoning in English and across other languages.
## Model Details
- **Base model:** [openai/gpt-oss-20b](https://huggingface.co/openai/gpt-oss-20b) (21B parameters)
- **Architecture:** `gpt_oss`
- **Fine-tuned by:** [artindnr](https://huggingface.co/artindnr)
- **License:** Apache 2.0
- **Languages:** Farsi (Persian), English, and multilingual reasoning support
- **Model type:** Causal decoder-only language model with reasoning ("thinking") traces
## What's New
Most open reasoning models today generate their chain-of-thought almost exclusively in English, even when the final answer is requested in another language. Strawberry-1 is trained specifically to:
- Generate **coherent, high-quality reasoning traces in Farsi**, not just Farsi answers
- Reason natively across multiple languages rather than silently falling back to English
- Preserve the general instruction-following and reasoning ability of the `gpt-oss-20b` base model
## Training
Strawberry-1 was trained using a mix of fine-tuning strategies — including full fine-tuning and LoRA experiments — on top of `gpt-oss-20b`. The version released here is the **fully fine-tuned (merged, dense-weights) checkpoint**, not a LoRA adapter.
### Training Data
Strawberry-1 was trained on the [Thinking Datasets](https://huggingface.co/collections/artindnr/thinking-datasets) collection, a set of datasets purpose-built for chain-of-thought fine-tuning:
- [`artindnr/Persian-Thinking`](https://huggingface.co/datasets/artindnr/Persian-Thinking) — Farsi reasoning traces .
- [`artindnr/Persian-English-Thinking`](https://huggingface.co/datasets/artindnr/Persian-English-Thinking) — mixed Farsi/English reasoning traces
- [`artindnr/Multilingual-Thinking`](https://huggingface.co/datasets/artindnr/Multilingual-Thinking) — multilingual chain-of-thought data
- [`artindnr/Multilingual-Thinking-200`](https://huggingface.co/datasets/artindnr/Multilingual-Thinking-200) — a smaller multilingual reasoning subset
<!-- TODO: add training hardware, number of epochs, learning rate, effective batch size, and any other hyperparameters you'd like documented. -->
## How to Use
Strawberry-1 uses the `gpt-oss` chat template (Harmony format) shipped with the base model, so it works with 🤗 Transformers.
### Installation
```bash
pip install torch --index-url https://download.pytorch.org/whl/cu128
pip install "trl>=0.20.0" "peft>=0.17.0" "transformers>=4.55.0" "kernels>=0.12.0"
```
This has been verified to work with:
| Package | Version |
|---|---|
| `torch` | 2.8.0+cu129 |
| `transformers` | 5.14.1 |
| `trl` | 1.9.2 |
| `peft` | 0.20.0 |
| `accelerate` | 1.10.1 |
| `tokenizers` | 0.22.0 |
### Generation
```python
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
MODEL_ID = "artindnr/strawberry-1"
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
model = AutoModelForCausalLM.from_pretrained(
MODEL_ID,
torch_dtype=torch.bfloat16,
device_map="auto",
)
REASONING_LANGUAGE = "English" # e.g. "English", "Farsi", "Persian"
SYSTEM_PROMPT = f"reasoning language: {REASONING_LANGUAGE}"
USER_PROMPT = "تو کی هستی و اسمت چیه؟"
messages = [
{"role": "system", "content": SYSTEM_PROMPT},
{"role": "user", "content": USER_PROMPT},
]
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=512,
temperature=0.6,
do_sample=True,
)
print(tokenizer.decode(outputs[0]))
```
This prints the full Harmony-formatted output, including the `analysis` (reasoning) and `final` (answer) channels and their special tokens. To get just the plain-text final answer, decode with `skip_special_tokens=True` and parse out the `final` channel, or use `tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:], skip_special_tokens=True)` to only decode the newly generated tokens.
### Reasoning in a specific language
Set `reasoning language: <Language>` as the `system` message content to control the language of the reasoning trace (the `analysis` channel), independent of the language the user writes in. For example, setting `REASONING_LANGUAGE = "Farsi"` will produce a Farsi reasoning trace even for a prompt in another language.
Note that the model's default chat template also auto-populates a Harmony-format preamble (identity, knowledge cutoff, current date, reasoning effort, valid channels) ahead of your system/developer message — you don't need to set these yourself.
## Intended Use
Strawberry-1 is intended for:
- Research and experimentation on multilingual and Farsi-language reasoning
- Building Farsi-language assistants, tutoring tools, and reasoning-heavy applications
- General-purpose multilingual chain-of-thought tasks
## Limitations
- Farsi reasoning quality, while a focus of this fine-tune, may still occasionally mix in English tokens or phrasing, especially for highly technical topics.
- As with any fine-tune, Strawberry-1 inherits the general capabilities and limitations of the `gpt-oss-20b` base model, including the possibility of hallucinated facts and reasoning errors.
- No formal safety fine-tuning beyond what is inherited from the base model has been applied; use appropriate safeguards in production settings.
## License
This model is released under the [Apache 2.0](https://www.apache.org/licenses/LICENSE-2.0) license, consistent with the base `gpt-oss-20b` model.
## Citation
If you use Strawberry in your work, please cite:
```bibtex
@misc{strawberry1,
title = {Strawberry: A Farsi and Multilingual Reasoning Model Fine-tuned from GPT-OSS-20B},
author = {artindnr},
year = {2026},
url = {https://huggingface.co/artindnr/strawberry-1}
}
```
## Acknowledgements
Built on top of [`openai/gpt-oss-20b`](https://huggingface.co/openai/gpt-oss-20b), using the [Thinking Datasets](https://huggingface.co/collections/artindnr/thinking-datasets) collection. |