Openr1 Qwen 7B Italian
Italian mathematical reasoning model fine-tuned from Qwen2.5-7B-Instruct, optimized for Italian reasoning capabilities
Downloads 63
Release Time : 3/3/2025
Model Overview
This is a language model optimized for Italian mathematical reasoning tasks, fine-tuned from Qwen2.5-7B-Instruct, with special improvements in step-by-step reasoning in Italian contexts
Model Features
Italian-optimized reasoning
Specially optimized for Italian contexts, capable of generating clearer Italian reasoning processes
Mathematical problem solving
Focused on mathematical equation solving and logical reasoning tasks
Long sequence processing
Supports sequence processing up to 4096 tokens
Model Capabilities
Mathematical problem solving
Step-by-step reasoning
Italian text generation
Equation solving
Use Cases
Education
Math problem solving
Helps students understand and solve mathematical equation problems
Can generate detailed solution steps and verification processes
Research
Mathematical reasoning research
Used to study language model performance in mathematical reasoning
datasets: WiroAI/dolphin-r1-Italian library_name: transformers model_name: OpenR1-Qwen-7B-Italian tags:
- generated_from_trainer
- trl
- sft licence: license license: apache-2.0 language:
- it base_model:
- Qwen/Qwen2.5-7B-Instruct

🚀 OpenR1-Qwen-7B-Italian 🚀
This is a finetune of Qwen2.5-Instruct on WiroAI/dolphin-r1-Italian.
🟢 Overview
- DeepSeek's distilled models sometimes reason in Chinese or English even though prompted in another language.
- Open-Source models still need improvement on relatively low-resource languages.
- A motivation to reproduce R1 and contribute to the community.
🟢 Training
- We train the model on the WiroAI/dolphin-r1-Italian for 2 epochs. We use learning rate of 1e-5 and max seq length 4096. The training follows a cosine learning rate schedule with a 10% warmup phase.
- Training took 5 days in 8xA6000 ADA cluster.
- Normally, R1 team compares the performance of OpenR1 models to DeepSeek-Distill-Qwen-7B and OpenThinker-7B using lighteval. However, the datasets are only MATH oriented so not to conclude anything we won't disclose the default results.
You can find the training and evaluation code at: https://github.com/huggingface/open-r1/
🐨 Quick start
from transformers import AutoModelForCausalLM, AutoTokenizer
model_name = "OpenR1-Qwen-7B-Italian"
model = AutoModelForCausalLM.from_pretrained(
model_name,
torch_dtype="auto",
device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained(model_name)
prompt = "Trova il valore di $x$ che soddisfa l'equazione $4x+5 = 6x+7$."
messages = [
{"role": "system", "content": "Si prega di ragionare passo dopo passo prima di rispondere."},
{"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True
)
model_inputs = tokenizer([text], return_tensors="pt").to(device)
generated_ids = model.generate(
model_inputs.input_ids,
max_new_tokens=4096
)
generated_ids = [
output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
<|begin_of_thought|>
Ok, allora devo risolvere l'equazione 4x + 5 uguale a 6x + 7 per x.
Mmm, vediamo. Ricordo dalle lezioni di algebra che quando hai un'equazione come questa, vuoi isolare la variabile su un lato.
Probabilmente significa portare tutti i termini con la x da una parte e le costanti dall'altra. Cerchiamo di ricordare i passaggi esatti.
Innanzitutto, forse posso sottrarre 4x da entrambi i lati per eliminare il 4x a sinistra. Proviamo. Se sottraggo 4x da entrambi i lati, otterrei:
4x + 5 - 4x = 6x + 7 - 4x
Semplificando entrambi i lati, il 4x e il -4x sul lato sinistro si annullano, lasciando solo 5. Sul lato destro, 6x meno 4x fa 2x. Quindi ora l'equazione è:
5 = 2x + 7\n\nOk, sembra più semplice. Ora, devo ottenere x da sola.
Il termine costante al lato destro è 7, quindi forse posso sottrarre 7 da entrambi i lati per sbarazzarmene.
Facciamolo: 5 - 7 = 2x + 7 - 7\n\nSemplificando entrambi i lati: 5 - 7 fa -2, e 7 - 7 si annulla, lasciando 2x.
Quindi ora abbiamo: -2 = 2x
Mmm, quindi 2x è uguale a -2. Per risolvere per x, dovrei dividere entrambi i lati per 2. Verifichiamo:
(-2) / 2 = (2x)/2
Che si semplifica in: -1 = x
Quindi x è uguale a -1. Vediamo se ha senso. Sostituiamo x = -1 nell'equazione originale per verificare.
Lato sinistro: 4*(-1) + 5 = -4 + 5 = 1
Lato destro: 6*(-1) + 7 = -6 + 7 = 1
Oh, entrambi i lati sono uguali a 1. Funziona! Quindi x = -1 è effettivamente la soluzione corretta.
Aspetta, ma voglio assicurarmi di non aver fatto errori nei miei passaggi. Ripercorriamo di nuovo ogni passaggio.
Equazione originale: 4x + 5 = 6x + 7
Passaggio 1: Sottrai 4x da entrambi i lati. Questo dà 5 = 2x + 7. Sembra giusto perché 4x - 4x è 0, quindi rimane 5, e 6x - 4x è 2x.
Passaggio 2: Sottrai 7 da entrambi i lati. Questo dà 5 - 7 = 2x, che è -2 = 2x. Corretto.
Passaggio 3: Dividi entrambi i lati per 2. -2 diviso per 2 è -1, quindi x = -1. Sì, torna tutto.
Un altro modo per affrontarlo potrebbe essere quello di sottrarre prima 5 da entrambi i lati invece di sottrarre 4x.
Proviamo questo metodo per confermare.\n\nPartendo da 4x + 5 = 6x + 7
Sottrai 5 da entrambi i lati: 4x = 6x + 2
Ora, sottrai 6x da entrambi i lati: 4x - 6x = 2
Che dà -2x = 2\n\nDividi entrambi i lati per -2: x = 2 / (-2) = -1
Stesso risultato. Ok, quindi sia che io sottragga prima 4x o 5, finisco comunque con x = -1. Questo mi dà più fiducia che la soluzione sia corretta.
In alternativa, potrei anche usare l'addizione di equazioni. Vediamo. Se voglio avere tutti i termini con x su un lato e le costanti sull'altro, posso riorganizzare l'equazione iniziale.
Sottrai 4x e 7 da entrambi i lati contemporaneamente.\n\nAspetta, forse no. Pensiamoci. L'idea è quella di spostare tutti i termini con x a sinistra e le costanti a destra.
Quindi partendo da 4x + 5 = 6x + 7\n\nSe sottraggo 4x da entrambi i lati, come ho fatto prima, ottengo 5 = 2x + 7.
In alternativa, se sottraggo 6x e 5 da entrambi i lati, sarebbe: 4x + 5 - 6x - 5 = 6x + 7 - 6x - 5
Semplificando il lato sinistro: (4x - 6x) + (5 - 5) = -2x + 0 = -2x
Lato destro: (6x - 6x) + (7 - 5) = 0 + 2 = 2
Quindi, -2x = 2, portando a x = -1. Stessa risposta. Quindi indipendentemente dal metodo, è coerente.
Un altro controllo: visualizziamo l'equazione. Il lato sinistro è 4x + 5, che è una linea con pendenza 4 e intercetta y a 5.
Il lato destro è 6x + 7, una linea con pendenza 6 e intercetta y a 7. L'equazione chiede dove queste due linee si intersecano.
Poiché hanno pendenze diverse, dovrebbero intersecarsi esattamente in un punto. La nostra soluzione x = -1 è dove si incrociano.
Verifichiamo con un grafico approssimativo.
A x = -1, il lato sinistro è 4*(-1) + 5 = -4 + 5 = 1. Il lato destro è 6*(-1) + 7 = -6 + 7 = 1. Quindi entrambe le linee passano per ( -1, 1 ). Ha senso.
Se le linee avessero la stessa pendenza, sarebbero parallele e non si intersecherebbero mai, ma poiché le loro pendenze sono diverse (4 contro 6), devono intersecarsi esattamente una volta.
Pertanto, x = -1 è l'unica soluzione. C'è qualche possibilità che abbia commesso un errore aritmetico? Ricontrolliamo di nuovo ogni passaggio.
Equazione originale: 4x + 5 = 6x + 7
Sottrai 4x: 5 = 2x + 7 Sottrai 7: -2 = 2x Dividi per 2: -1 = x
Tutte le operazioni sono aritmetiche semplici. Sottrarre 4x e 7, quindi dividere per 2. Nessun problema lì. In alternativa, se inserisco x = -1 nell'equazione originale, come ho fatto prima, entrambi i lati sono uguali a 1, quindi funziona.
Penso che sia completo. Non riesco a vedere errori nel mio ragionamento. Tutti i diversi metodi portano alla stessa risposta e la verifica sostituendo nell'equazione originale conferma che è corretta.
Pertanto, sono sicuro che x = -1 sia la soluzione giusta.
**Risposta finale** Il valore di \\( x \\) che soddisfa l'equazione è \\boxed{-1}.
<|end_of_thought|>
<|begin_of_solution|>
Per risolvere l'equazione (4x + 5 = 6x + 7) per (x):
1. Sottrai (4x) da entrambi i lati: [4x + 5 - 4x = 6x + 7 - 4x]
Semplificando, otteniamo: [5 = 2x + 7]
2. Sottrai 7 da entrambi i lati: [5 - 7 = 2x + 7 - 7]
Semplificando, otteniamo: [-2 = 2x]
3. Dividi entrambi i lati per 2: [-2/2 = 2x/2]
Semplificando, otteniamo: [-1 = x]
Verifica sostituendo (x = -1) nell'equazione originale:
- Lato sinistro: (4(-1) + 5 = -4 + 5 = 1)
- Lato destro: (6(-1) + 7 = -6 + 7 = 1)
Entrambi i lati sono uguali a 1, confermando che la soluzione è corretta.
<|end_of_solution|>
🟡 Evaluation
- We observed that reasoning process has slightly improved. Our model thinks more clearly in Italian compared to the DeepSeek's reasoning model.
- This model trained for experimental motives and any benchmark evaluation is appreciated. Please be aware that this model will be producing more tokens compared to normal models and will consume more VRAM during inference.
- If you are willing to evaluate this model please make sure that model will be allowed to produce enough tokens. Generate until requests that restrict the model to output less than 4000 tokens will lead to poor results.
- We believe democratized and culturally improved open-source models will be through sharing and experiments!
🤗 Community
- We would like to thank Huggingface Staff and everyone who contributed to the Open-R1 project!
Citation
@article{WiroAI,
title={WiroAI/OpenR1-Qwen-7B-Italian,
author={Abdullah Bezir, Cengiz Asmazoğlu},
year={2025},
url={https://huggingface.co/WiroAI/OpenR1-Qwen-7B-Italian}
}
Phi 2 GGUF
Other
Phi-2 is a small yet powerful language model developed by Microsoft, featuring 2.7 billion parameters, focusing on efficient inference and high-quality text generation.
Large Language Model Supports Multiple Languages
P
TheBloke
41.5M
205
Roberta Large
MIT
A large English language model pre-trained with masked language modeling objectives, using improved BERT training methods
Large Language Model English
R
FacebookAI
19.4M
212
Distilbert Base Uncased
Apache-2.0
DistilBERT is a distilled version of the BERT base model, maintaining similar performance while being more lightweight and efficient, suitable for natural language processing tasks such as sequence classification and token classification.
Large Language Model English
D
distilbert
11.1M
669
Llama 3.1 8B Instruct GGUF
Meta Llama 3.1 8B Instruct is a multilingual large language model optimized for multilingual dialogue use cases, excelling in common industry benchmarks.
Large Language Model English
L
modularai
9.7M
4
Xlm Roberta Base
MIT
XLM-RoBERTa is a multilingual model pretrained on 2.5TB of filtered CommonCrawl data across 100 languages, using masked language modeling as the training objective.
Large Language Model Supports Multiple Languages
X
FacebookAI
9.6M
664
Roberta Base
MIT
An English pre-trained model based on Transformer architecture, trained on massive text through masked language modeling objectives, supporting text feature extraction and downstream task fine-tuning
Large Language Model English
R
FacebookAI
9.3M
488
Opt 125m
Other
OPT is an open pre-trained Transformer language model suite released by Meta AI, with parameter sizes ranging from 125 million to 175 billion, designed to match the performance of the GPT-3 series while promoting open research in large-scale language models.
Large Language Model English
O
facebook
6.3M
198
1
A pretrained model based on the transformers library, suitable for various NLP tasks
Large Language Model
Transformers

1
unslothai
6.2M
1
Llama 3.1 8B Instruct
Llama 3.1 is Meta's multilingual large language model series, featuring 8B, 70B, and 405B parameter scales, supporting 8 languages and code generation, with optimized multilingual dialogue scenarios.
Large Language Model
Transformers Supports Multiple Languages

L
meta-llama
5.7M
3,898
T5 Base
Apache-2.0
The T5 Base Version is a text-to-text Transformer model developed by Google with 220 million parameters, supporting multilingual NLP tasks.
Large Language Model Supports Multiple Languages
T
google-t5
5.4M
702
Featured Recommended AI Models