artindnr/Strawberry

🍓 Strawberry

](https://i.pinimg.com/736x/b5/04/41/b50441f456e162ee3fb651898d20324a.jpg)

Strawberry is a fine-tuned version of `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 (21B parameters)
• —Architecture: gpt_oss
• —Fine-tuned by: 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 collection, a set of datasets purpose-built for chain-of-thought fine-tuning:

• —`artindnr/Persian-Thinking` — Farsi reasoning traces .
• —`artindnr/Persian-English-Thinking` — mixed Farsi/English reasoning traces
• —`artindnr/Multilingual-Thinking` — multilingual chain-of-thought data
• —`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

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:

Generation

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 license, consistent with the base gpt-oss-20b model.

Citation

If you use Strawberry in your work, please cite:

@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`, using the Thinking Datasets collection.