Alibaba Qwen: Kineski model umjetne inteligencije izaziva Silicijansku dolinu

Autori : Janković Bai Šuai Bai Janković Chu Zeyu Cui kada je opasno Sljedeći: Xiaodong Deng Tko je fan Wenbin Ge Yu Han Fei Huang Sljedeći članakBinyuan Hui Luo Ji Maja Li Sljedeći: Junyang Lin Ruži Lin Danijela Liu Gao Liu Čenčiang Lu Sljedeći LU Janković Ma Muškarci Rui Sljedeći: Xingzhang Ren Šibenik Ren Čuanči Tan Sljedeći Tan Janković Tu Vanjski Wang Shijie Wang Wei Wang Shengguang Wu Benfeng Xu Jin Xu Jedan Yang Hao Yang JAN JAN Shusheng Yang Jao Jao Bowen Yu Sljedeći Yuan Zheng Yuan Jianwei Zhang Xingxuan Zhang Sljedeći članakYichang Zhang Zhenru Zhang Chang Zhou Jingren Zhou Xiaohuan Zhou Tianhang Zhu Qwen Team, Alibaba Grupa Autori : Janković Bai Šuai Bai Janković Chu Zeyu Cui kada je opasno Sljedeći: Xiaodong Deng Tko je fan Wenbin Ge Yu Han Fei Huang Sljedeći članakBinyuan Hui Luo Ji Maja Li Sljedeći: Junyang Lin Ruži Lin Danijela Liu Gao Liu Čenčiang Lu Sljedeći LU Janković Ma Muškarci Rui Sljedeći: Xingzhang Ren Šibenik Ren Čuanči Tan Sljedeći Tan Janković Tu Vanjski Wang Shijie Wang Wei Wang Shengguang Wu Benfeng Xu Jin Xu Jedan Yang Hao Yang JAN JAN Shusheng Yang Jao Jao Bowen Yu Sljedeći Yuan Zheng Yuan Jianwei Zhang Xingxuan Zhang Sljedeći članakYichang Zhang Zhenru Zhang Chang Zhou Jingren Zhou Xiaohuan Zhou Tianhang Zhu Qwen Team, Alibaba Group Abstrakcija Veliki jezični modeli (LLM) revolucionarno su promijenili područje umjetne inteligencije, omogućujući prirodne zadatke obrade jezika za koje se ranije mislilo da su isključivi za ljude. QWEN je sveobuhvatna serija jezikovnih modela koja obuhvaća različite modele s različitim brojevima parametara. To uključuje QWEN, osnovne predtrenirane jezike modela, i QWEN-CHAT, chat modela finetuned s ljudskim usklađivanje tehnike. Osnovni jezični modeli dosljedno pokazuju vrhunske performanse u mnoštvu daljnjih zadataka, a chat modela, posebno one obučene koristeći Reinforcement Learning from Human Feedback (RLHF), su vrlo konkurentni. Chat modeli posjeduju napredne mogućnosti za korištenje alata i planiranje za stvaranje aplikacija agenata, pokazujući impresivne performanse čak i u usporedbi s većim modelima na složenim zadatcima kao što je korištenje tumača koda. Nadalje, razvili smo 1 1 Uvod Veliki jezični modeli (LLM) revolucionarizirali su područje umjetne inteligencije (AI) pružajući snažnu osnovu za složeno razmatranje i zadatke rješavanja problema. Ovi modeli imaju sposobnost komprimiranja ogromnog znanja u neuronske mreže, čineći ih nevjerojatno svestranim agentima. Oni se mogu baviti razgovorima na prirodnom jeziku s ljudima, odgovarati na pitanja, pružati informacije, pa čak i generirati kreativni sadržaj kao što su priče, pjesme i glazba. (Radford i dr. 2018. godine; Devlin et al., 2018. godine; Rafić i al. 2020. godine Brown i al, 2020. godine otvoriti se, 2023. godine Chowdhery i al., 2022. godine Anita i al. 2023. godine Šibenik et al. 2022. godine Touvron i al, 2023a i b) Otvoren je, 2022. godine Ouyang et al., 2022. godine Anil i al, 2023. godine na Googleu, 2023. godine Antropološki dio, 2023a i b) LLM-ovi nisu ograničeni samo na jezične zadatke. oni također mogu funkcionirati kao generalistički agent suradnja s vanjskim sustavima, alatima i modelima kako bi se postigli ciljevi koje su postavili ljudi. Izvršiti kod Koristite alate To otvara cijeli novi svijet mogućnosti za AI aplikacije, od autonomnih vozila i robotike do zdravstvene skrbi i financija.Kako se ti modeli nastavljaju razvijati i poboljšavati, možemo očekivati da ćemo vidjeti još inovativnije i uzbudljivije aplikacije u narednim godinama.Bilo da nam pomaže riješiti složene probleme, stvoriti nove oblike zabave ili transformirati način na koji živimo i radimo, LLM-ovi su spremni igrati središnju ulogu u oblikovanju budućnosti AI-a. (Rimljani i al., 2022. godine Bačić i al. 2022. godine Wang i al. 2023. godine Automobilski prijevoz, 2023. godine Hong et al., 2023. godine) Otvoren je, 2023. godine Bačić i al. 2023. godine Liu i al. 2023. godine Vi i Al. 2023. godine Dai et al., 2023. godine Bjelovar i al. 2023 b) i (Sjeverna Koreja i al., 2021. godine Janković i al. 2023. godine Li i al, 2023. godine) (Sjeverna Koreja i al., 2023. godine Sljedeći članak LangChain, Inc. 2023. godine Automobilski prijevoz, 2023. godine) Unatoč svojim impresivnim mogućnostima, LLM-ovi su često kritizirani zbog nedostatka reproduktivnosti, upravljanja i pristupačnosti pružateljima usluga. U ovom radu s zadovoljstvom predstavljamo i objavljujemo početnu verziju naše serije LLM-a, QWEN. QWEN je naziv koji potječe iz kineske fraze Qianwen, koja se prevodi u "tućine promptova" i prenosi pojam prihvatanja širokog spektra upita. QWEN je sveobuhvatna serija jezika koji obuhvaća različite modele s različitim brojem parametara. Serija modela uključuje osnovne predtrenirane jezike, modele za razgovor koji su rafinirani s ljudskim tehnikama usklađivanja, tj. nadzorovanim finetuningom (SFT), pojačanjem učenja s ljudskom povrat Osnovni jezični modeli, odnosno QWEN, prošli su opsežno osposobljavanje koristeći do 3 bilijuna tokena različitih tekstova i kodova, obuhvaćajući širok raspon područja. Modeli QWEN-CHAT bili su pažljivo prilagođeni kuriranom skupu podataka relevantnih za obavljanje zadataka, chat, korištenje alata, agenta, sigurnost itd. Procjena referentne vrijednosti pokazuje da SFT modeli mogu postići vrhunske performanse. Nadalje, trenirali smo modele nagrade kako bi oponašali ljudske preferencije i primijenili ih u RLHF-u za modele chat koji mogu proizvesti odgovore koje vole ljudi. Kroz ljudsku procjenu izazovnog testa, otkrijemo da su QWEN-CHAT modeli obučeni s RLHF-om vrlo konkurentni, još uvijek su iza GPT-4 na našem referentnom nivou. Osim toga, predstavljamo specijalizirane modele pod nazivom CODE-QWEN, koji uključuje CODE-QWEN-7B i CODE-QWEN-14B, kao i njihove modele za razgovor, CODE-QWEN-14B-CHAT i CODE-QWEN-7B-CHAT. MBPP i HumanEvalPack dokazati visoku razinu osposobljenosti CODE-QWEN-a u razumijevanju i generiranju koda. (Sjeverna Koreja i al., 2021. godine) Austin i al. 2021. godine) (Mojsić i dr. 2023. godine) Naši rezultati pokazuju da i MATH-QWEN-7B-CHAT i MATH-QWEN-14B-CHAT nadmašuju modele otvorenog izvora u istoj veličini s velikim marginama i približavaju se GPT-3.5 na matematičkim referentnim skupovima podataka kao što je GSM8K i matematika KOBB i al, 2021. godine (Hrvatska i al., 2021. godine) Osim toga, imamo open-source QWEN-VL i QWEN-VL-CHAT, koji imaju sveobuhvatnu sposobnost razumijevanja vizualnih i jezičnih uputa. Ovi modeli nadmašuju trenutačne modele jezika vizije otvorenog koda u različitim referentnim vrijednostima evaluacije i podržavaju prepoznavanje teksta i vizualno utemeljenje na kineskom i engleskom jeziku. Štoviše, ovi modeli omogućuju razgovore s više slika i pripovijedanje priča. Bai i al. 2023 ) Sada smo službeno otvorili 14B-parametar i 7B-parametar bazni pretrenirani modeli QWEN i usklađeni chat modeli QWEN-CHAT Ovo izdanje ima za cilj pružanje sveobuhvatnijih i moćnijih LLM-ova na razmjerima prilagođenim programerima ili aplikacijama. 2 Struktura ovog izvješća je sljedeća: Odjeljak opisuje naš pristup prettreningu i rezultate QWEN. odjeljak pokriva našu metodologiju usklađivanja i izvješćuje o rezultatima automatske evaluacije i ljudske evaluacije. Osim toga, ovaj odjeljak opisuje pojedinosti o našim naporima u izgradnji chat modela sposobnih za korištenje alata, tumača koda i agenta. i Istražujemo specijalizirane modele kodiranja i matematike i njihove performanse. pruža pregled relevantnih radova povezanih s radom, a odjeljak zaključuje ovaj članak i ukazuje na naš budući rad. 2 3 4 5 , 6 7 2 Obuka Prekvalifikacijska faza uključuje učenje ogromne količine podataka kako bi stekli sveobuhvatno razumijevanje svijeta i njegovih različitih složenosti.To uključuje ne samo osnovne jezične sposobnosti, već i napredne vještine kao što su aritmetička, kodiranje i logičko razmatranje.U ovom odjeljku uvodimo podatke, dizajn modela i skaliranje, kao i sveobuhvatne rezultate evaluacije na referentnim skupovima podataka. 2.1 Podaci The size of data has proven to be a crucial factor in developing a robust large language model, as highlighted in previous research To create an effective pretraining dataset, it is essential to ensure that the data are diverse and cover a wide range of types, domains, and tasks. Our dataset is designed to meet these requirements and includes public web documents, encyclopedia, books, codes, etc. Additionally, our dataset is multilingual, with a significant portion of the data being in English and Chinese. (Hoffmann et al. i dr. 2022. godine Tvrđava i al. 2023. godine Kako bismo osigurali kvalitetu naših prettrening podataka, razvili smo sveobuhvatan postupak predprerađivanja podataka. Za javne web podatke, izvlačimo tekst iz HTML-a i koristimo alate za identifikaciju jezika za određivanje jezika. Kako bismo povećali raznolikost naših podataka, koristimo tehnike deduplikiranja, uključujući deduplikiranje točnog sukoba nakon normalizacije i fuzzy deduplikiranje pomoću algoritama MinHash i LSH. Da bismo filtrirali podatke niske kvalitete, koristimo kombinaciju metoda zasnovanih na pravilima i metodama strojnog učenja. Konkretno, koristimo više modela za ocjenjivanje sadržaja, uključujući jezike, modele za ocjenjivanje kvalitete teksta i modele za identifikaciju potencijalno uvredljivog ili neprimjerenog sadržaja. Također ručno uzimamo Dokazano je da predtreniranje jezičnih modela s uputama za više zadataka može poboljšati performanse na nuli i na malo. Kako bismo dodatno poboljšali performanse našeg modela, uključili smo visokokvalitetne podatke o instrukcijama u proces predtreninga. i pažljivo uklonili sve uzorke uputa koji pokazuju preklapanje od 13 grama sa svim podacima prisutnima u test setovima koji su korišteni u našoj procjeni. (Zeng et al. i dr. 2022; Aribandi et al., 2021. godine Rafić i al. 2020. godine) Brown i al. Sljedeći (2020) S obzirom na veliki broj daljnjih zadataka, nije moguće ponoviti ovaj proces filtriranja za sve zadatke. Umjesto toga, osigurali smo da su podaci za upute za prijavljene zadatke podvrgnuti našem procesu filtriranja kako bi se osigurala njihova točnost i pouzdanost. 2.2 Tokenizacija U ovoj studiji koristimo kodiranje parova bajta (BPE) kao našu metodu tokenizacije, nakon GPT-3.5 i GPT-4. Da bismo poboljšali performanse našeg modela na višejezičnim zadatcima u daljnjem tijeku, osobito na kineskom, povećavamo vokabular s uobičajenim kineskim znakovima i riječima, kao i onima na drugim jezicima. brojevi su podijeljeni na pojedinačne brojeve. konačna veličina rječnika je oko 152K. (Sjeverna Koreja 2022. godine) Tvrđava i al. 2023a i b), Učinak QWEN tokenizera u smislu kompresije prikazan je na slici U ovom usporedbi, ocijenili smo QWEN u odnosu na nekoliko drugih tokenizera, uključujući XLM-R Zvuk Sljedeći I unutarnji Naši rezultati pokazuju da QWEN postiže veću učinkovitost kompresije od svojih konkurenata u većini jezika. To znači da se troškovi služenja mogu značajno smanjiti jer manji broj tokena iz QWEN-a može prenijeti više informacija od svojih konkurenata. Nadalje, proveli smo preliminarne eksperimente kako bismo osigurali da skaliranje veličine rječnika QWEN-a ne utječe negativno na performanse u nastavku predtreniranog modela. 3. Konjica i al, 2019. godine) (Sljedeći članakHrvatska et al. 2023a) i • Inc., 2023a) i (Tijekom radnog razdoblja, 2023. godine 2.3 Arhitektura QWEN is designed using a modified version of the Transformer architecture. Specifically, we have adopted the recent open-source approach of training large language models, LLaMA which is widely regarded as the top open-source LLM. Our modifications to the architecture include: (Touvron et al., 2023a) i Na temelju preliminarnih eksperimentalnih nalaza, odabrali smo pristup untied embedding umjesto vezanja utega ulaznog embedding i output projekcije. . We have chosen RoPE (Rotary Positional Embedding) as our preferred option for incorporating positional information into our model. RoPE has been widely adopted and has demonstrated success in contemporary large language models, notably PaLM and LLaMA In particular, we have opted to use FP32 precision for the inverse frequency matrix, rather than BF16 or FP16, in order to prioritize model performance and achieve higher accuracy. Positional embedding (Su et al., 2021) (Chowdhery et al., 2022; Anil et al., 2023) (Touvron et al., 2023a;b). . For most layers, we remove biases following but we add biases in the QKV layer of attention to enhance the extrapolation ability of the model Bias Chowdhery et al. (2022), (Su, 2023b). . In modern Transformer models, pre-normalization is the most widely used approach, which has been shown to improve training stability compared to post-normalization. Recent research has suggested alternative methods for better training stability, which we plan to explore in future versions of our model. Additionally, we have replaced the traditional layer normalization technique described in with RMSNorm This change has resulted in equivalent performance while also improving efficiency. Pre-Norm & RMSNorm (Ba et al., 2016) (Jiang et al., 2023). . We have selected SwiGLU as our activation function, a combination of Swish and Gated Linear Unit Our initial experiments have shown that activation functions based on GLU generally outperform other baseline options, such as GeLU As is common practice in previous research, we have reduced the dimension of the feed-forward network (FFN) from 4 times the hidden size to 8/3 of the hidden size. Activation function (Shazeer, 2020) (Ramachandran et al., 2017) (Dauphin et al., 2017). (Hendrycks & Gimpel, 2016). 2.4 Training To train QWEN, we follow the standard approach of autoregressive language modeling, as described in This involves training the model to predict the next token based on the context provided by the previous tokens. We train models with context lengths of 2048. To create batches of data, we shuffle and merge the documents, and then truncate them to the specified context lengths. To improve computational efficiency and reduce memory usage, we employ Flash Attention in the attention modules We adopt the standard optimizer AdamW za pretreniranje optimizacije. postavili smo hiperparametre 1 = 0*.*9, 2 = 0*.*95, and = 10−8. We use a cosine learning rate schedule with a specified peak learning rate for each model size. The learning rate is decayed to a minimum learning rate of 10% of the peak learning rate. All the models are trained with BFloat16 mixed precision for training stability. Radford i al. (U 2018. godini) (Dao et al., 2022). (Kingma & Ba, 2014; Loshchilov & Hutter, 2017) β β ϵ 2.5 Proširenje konteksta Modeli transformatora imaju značajno ograničenje u smislu duljine konteksta za njihov mehanizam pažnje. Kako se duljina konteksta povećava, izračun kvadrativne složenosti dovodi do drastičnog povećanja troškova računanja i memorije. U ovom smo radu implementirali jednostavne tehnike bez obuke koje se primjenjuju isključivo tijekom zaključivanja kako bi se proširila duljina konteksta modela. (bloc97, 2023). Interpolacija pozicije (PI) koji skali svaku dimenziju RoPE-a jednako, NTK-aware interpolacija prilagođava bazu RoPE-a kako bi se spriječio gubitak informacija visoke frekvencije na način bez obuke. Dinamički mijenja veličinu po komadićima, izbjegavajući ozbiljnu degradaciju performansi.Te nam tehnike omogućuju da učinkovito proširimo duljinu konteksta modela Transformera bez ugrožavanja njihove računalne učinkovitosti ili točnosti. (Chen et al., 2023a) (Peng et al., 2023a). QWEN additionally incorporates two attention mechanisms: LogN-Scaling and window attention LogN-Scaling rescales the dot product of the query and value by a factor that depends on the ratio of the context length to the training length, ensuring that the entropy of the attention value remains stable as the context length grows. Window attention restricts the attention to a limited context window, preventing the model from attending to tokens that are too far away. (Chiang & Cholak, 2022; Su, 2023a) (Beltagy et al., 2020). We also observed that the long-context modeling ability of our model varies across layers, with lower layers being more sensitive in context length extension compared to the higher layers. To leverage this observation, we assign different window sizes to each layer, using shorter windows for lower layers and longer windows for higher layers. 2.6 Experimental Results Da bismo procijenili mogućnosti učenja nultog i malog broja naših modela, provodimo temeljitu referentnu procjenu pomoću niza skupova podataka. Plamen 2 MPT Falcao Baichuan2 ( ChatGLM2 InternLM XVERSE and StableBeluga2 Our evaluation covers a total of 7 popular benchmarks, which are MMLU (5-shot) C-Eval (5-shot) GSM8K (8-shot) MATH (4-shot) HumanEval (0-shot) MBPP (0-shot) i BBH (Big Bench Hard) (3 pucnjava) We aim to provide a comprehensive summary of the overall performance of our models across these benchmarks. (Touvron et al., 2023a), (Touvron et al., 2023b), (Mosaic ML, 2023), (Almazrouei et al., 2023), Janković i al. 2023. godine) ChatGLM2 tim, 2023), (Tijekom radnog razdoblja, 2023), (Inc., 2023b), stabilnost u BiH, 2023). (Hendrycks et al., 2020), (Hrvatska i al., 2023), (Cobbe et al., 2021), (Hendrycks et al., 2021), (Chen et al., 2021), (Austin et al., 2021), (Suzgun i al., 2022). In this evaluation, we focus on the base language models without alignment and collect the baselines’ best scores from their official results and OpenCompass The results are presented in Table (OpenCompass Team, 2023). 2. Our experimental results demonstrate that the three QWEN models exhibit exceptional performance across all downstream tasks. It is worth noting that even the larger models, such as LLaMA2-70B, are outperformed by QWEN-14B in 3 tasks. QWEN-7B also performs admirably, surpassing LLaMA2-13B and achieving comparable results to Baichuan2-13B. Notably, despite having a relatively small number of parameters, QWEN-1.8B is capable of competitive performance on certain tasks and even outperforms larger models in some instances. The findings highlight the impressive capabilities of the QWEN models, particularly QWEN-14B, and suggest that smaller models, such as QWEN-1.8B, can still achieve strong performance in certain applications. To evaluate the effectiveness of context length extension, Table presents the test results on arXiv in terms of perplexity (PPL). These results demonstrate that by combining NTK-aware interpolation, LogN-Scaling, and layer-wise window assignment, we can effectively maintain the performance of our models in the context of over 8192 tokens. 3 3 3 Alignment Pretrained large language models have been found to be not aligned with human behavior, making them unsuitable for serving as AI assistants in most cases. Recent research has shown that the use of alignment techniques, such as supervised finetuning (SFT) and reinforcement learning from human feedback (RLHF), can significantly improve the ability of language models to engage in natural conversation. In this section, we will delve into the details of how QWEN models have been trained using SFT and RLHF, and evaluate their performance in the context of chat-based assistance. 3.1 Supervised Finetuning To gain an understanding of human behavior, the initial step is to carry out SFT, which finetunes a pretrained LLM on chat-style data, including both queries and responses. In the following sections, we will delve into the details of data construction and training methods. 3.1.1 DATA To enhance the capabilities of our supervised finetuning datasets, we have annotated conversations in multiple styles. While conventional datasets contain a vast amount of data prompted with questions, instructions, and answers in natural language, our approach takes it a step further by annotating human-style conversations. This practice, inspired by aims at improving the model’s helpfulness by focusing on natural language generation for diverse tasks. To ensure the model’s ability to generalize to a wide range of scenarios, we specifically excluded data formatted in prompt templates that could potentially limit its capabilities. Furthermore, we have prioritized the safety of the language model by annotating data related to safety concerns such as violence, bias, and pornography. (Wei et al., 2022a) Ouyang et al. 2022 godine, In addition to data quality, we have observed that the training method can significantly impact the final performance of the model. To achieve this, we utilized the ChatML-style format Ovaj format omogućuje modelu da učinkovito razlikuje između različitih vrsta informacija, uključujući podešavanja sustava, korisničke ulaze i pomoćne izlaze, između ostalog. (OpenAI, 2022), 3.1.2 Training Consistent with pretraining, we also apply next-token prediction as the training task for SFT. We apply the loss masks for the system and user inputs. More details are demonstrated in Section A.1.1. The model’s training process utilizes the AdamW optimizer, with the following hyperparameters: 1 set to 0*. 2 Sastavi na 0. set to 10−8. The sequence length is limited to 2048, and the batch size is 128. The model undergoes a total of 4000 steps, with the learning rate gradually increased over the first 1430 steps, reaching a peak of 2 10−6. To prevent overfitting, weight decay is applied with a value of 0. . .*0. β 9, β 95 i dalje × 1, dropout is set to 0 1, and gradient clipping is enforced with a limit of 1 3.2 Reinforcement Learning from Human Feedback While SFT has proven to be effective, we acknowledge that its generalization and creativity capa-bilities may be limited, and it is prone to overfitting. To address this issue, we have implemented Reinforcement Learning from Human Feedback (RLHF) to further align SFT models with human preferences, following the approaches of This process involves training a reward model and using Proximal Policy Optimization (PPO) to conduct policy training. Ouyang et al. (2022); Christiano et al. (2017). (Schulman et al., 2017. godine 3.2.1 Reward Model To create a successful reward model, like building a large language model (LLM), it is crucial to first undergo pretraining and then finetuning. This pretraining process, also known as preference model pretraining (PMP) necessitates a vast dataset of comparison data. This dataset consists of sample pairs, each containing two distinct responses for a single query and their corresponding preferences. Similarly, finetuning is also conducted on this type of comparison data, but with a higher quality due to the presence of quality annotations. (Bai et al., 2022b), During the fine-tuning phase, we gather a variety of prompts and adjust the reward model based on human feedback for responses from the QWEN models. To ensure the diversity and complexity of user prompts are properly taken into account, we have created a classification system with around 6600 detailed tags and implemented a balanced sampling algorithm that considers both diversity and complexity when selecting prompts for annotation by the reward model To generate a wide range of responses, we have utilized QWEN models of different sizes and sampling strategies, as diverse responses can help reduce annotation difficulties and enhance the performance of the reward model. These responses are then evaluated by annotators following a standard annotation guideline, and comparison pairs are formed based on their scores. (Lu et al., 2023). In creating the reward model, we utilize the same-sized pre-trained language model QWEN to initiate the process. It is important to mention that we have incorporated a pooling layer into the original QWEN model to extract the reward for a sentence based on a specific end token. The learning rate for this process has been set to a constant value of 3 10−6, and the batch size is 64. Additionally, the sequence length is set to 2048, and the training process lasts for a single epoch. × We adopted the accuracy on the test dataset as an important but not exclusive evaluation metric for the reward model. In Table we report the test pairwise accuracy of PMP and reward models on diverse human preference benchmark datasets Specifically, QWEN Helpful-base and QWEN Helpful-online are our proprietary datasets. The responses in QWEN Helpful-base are generated from QWEN without RLHF, whereas QWEN Helpful-online includes responses from QWEN with RLHF. The results show that the PMP model demonstrates high generalization capabilities on out-of-distribution data, and the reward model demonstrates significant improvement on our QWEN reward datasets. 4, (Bai et al., 2022b; Stiennon et al., 2020; Ethayarajh et al., 2022; Lightman et al., 2023). 3.2.2 Reinforcement Learning Our Proximal Policy Optimization (PPO) process involves four models: the policy model, value model, reference model, and reward model. Before starting the PPO procedure, we pause the policy model’s updates and focus solely on updating the value model for 50 steps. This approach ensures that the value model can adapt to different reward models effectively. During the PPO operation, we use a strategy of sampling two responses for each query simultaneously. This strategy has proven to be more effective based on our internal benchmarking evaluations. We set the KL divergence coefficient to 0*.*04 and normalize the reward based on the running mean. Politički i vrijednosni modeli imaju stope učenja 1 10−6 and 5 10−6, respectively. To enhance training stability, we utilize value loss clipping with a clip value of 0*. . .*0, there is a faster increase in reward, ultimately resulting in consistently higher evaluation rewards under similar conditions. × × 15. For inference, the policy top-p is set to 0 9. Our findings indicate that although the entropy is slightly lower than when top-p is set to 1 Additionally, we have implemented a pretrained gradient to mitigate the alignment tax. Empirical findings indicate that, with this specific reward model, the KL penalty is adequately robust to counteract the alignment tax in benchmarks that are not strictly code or math in nature, such as those that test common sense knowledge and reading comprehension. It is imperative to utilize a significantly larger volume of the pretrained data in comparison to the PPO data to ensure the effectiveness of the pretrained gradient. Additionally, our empirical study suggests that an overly large value for this coefficient can considerably impede the alignment to the reward model, eventually compromising the ultimate alignment, while an overly small value would only have a marginal effect on alignment tax reduction. 3.3 Automatic and Human Evaluation of Aligned Models To showcase the effectiveness of our aligned models, we conduct a comparison with other aligned models on well-established benchmarks, including MMLU C-Eval GSM8K HumanEval and BBH Besides the widely used few-shot setting, we test our aligned models in the zero-shot setting to demonstrate how well the models follow instructions. The prompt in a zero-shot setting consists of an instruction and a question without any previous examples in the context. The results of the baselines are collected from their official reports and OpenCompass (Hendrycks et al., 2020), (Huang et al., 2023), (Cobbe et al., 2021), (Chen et al., 2021), (Suzgun et al., 2022). (OpenCompass Team, 2023). The results in Table demonstrate the effectiveness of our aligned models in understanding human instructions and generating appropriate responses. QWEN-14B-Chat outperforms all other models except ChatGPT (OpenAI, 2022) and LLAMA 2-CHAT-70B (Touvron et al., 2023b) in all datasets, including MMLU (Hendrycks et al., 2020), C-Eval (Huang et al., 2023), GSM8K (Cobbe et al., 2021), HumanEval (Chen et al., 2021), and BBH (Suzgun et al., 2022). 5 In particular, QWEN’s performance in HumanEval, which measures the quality of generated codes, is significantly higher than that of other open-source models. Moreover, QWEN’s performance is consistently better than that of open-source models of similar size, such as LLaMA2 ChatGLM2 InternLM and Baichuan2 This suggests that our alignment approach, which involves fine-tuning the model on a large dataset of human conversations, has been effective in improving the model’s ability to understand and generate human-like language./im (Touvron et al., 2023b), (ChatGLM2 Team, 2023), (InternLM Team, 2023), (Yang et al., 2023). Despite this, we have reservations about the ability of traditional benchmark evaluation to accurately measure the performance and potential of chat models trained with alignment techniques in today’s landscape. The results mentioned earlier provide some evidence of our competitive standing, but we believe that it is crucial to develop new evaluation methods specifically tailored to aligned models. We believe that human evaluation is crucial, which is why we have created a carefully curated dataset for this purpose. Our process involved collecting 300 instructions in Chinese that covered a wide range of topics, including knowledge, language understanding, creative writing, coding, and mathematics. To evaluate the performance of different models, we chose the SFT version of QWEN-CHAT-7B and the SFT and RLHF versions of QWEN-CHAT-14B, and added two strong baselines, GPT-3.5 and GPT-4 , for comparison. For each instruction, we asked three annotators to rank the model responses by the overall score of helpfulness, informativeness, validity, and other relevant factors. Our dataset and evaluation methodology provides a comprehensive and rigorous assessment of the capabilities of different language models in various domains. 4 Figure illustrates the win rates of the various models. For each model, we report the percentage of wins, ties, and losses against GPT-3.5, with the segments of each bar from bottom to top representing these statistics. The experimental results clearly demonstrate that the RLHF model outperforms the SFT models by significant margins, indicating that RLHF can encourage the model to generate responses that are more preferred by humans. In terms of overall performance, we find that the RLHF model significantly outperforms the SFT models, falling behind GPT-4. This indicates the effectiveness of RLHF for aligning to human preference. To provide a more comprehensive understanding of the models’ performance, we include a case study with examples from different models in Appendix Nonetheless, it remains difficult to accurately capture the gap between our models and the proprietary models. As such, a more extensive and rigorous assessment is required for the chat models. 4 A.2.2. 3.4 Tool Use, Code Interpreter, and Agent The QWEN models, which are designed to be versatile, have the remarkable ability to assist with (semi-)automating daily tasks by leveraging their skills in tool-use and planning. As such, they can serve as agents or copilots to help streamline various tasks. We explore QWEN’s proficiency in the following areas: • Utilizing unseen tools through ReAct prompting (see Table (Yao et al., 2022) 6). • Using a Python code interpreter to enhance math reasoning, data analysis, and more (see Table and Table 7 8). • djeluje kao agent koji pristupa opsežnoj kolekciji multimodalnih modela tvrtke Hugging Face dok se bavi ljudima (vidi Tablicu) 9). To enhance QWEN’s capabilities as an agent or copilot, we employ the self-instruct strategy for SFT. Specifically, we utilize the in-context learning capability of QWEN for self-instruction. By providing a few examples, we can prompt QWEN to generate more relevant queries and generate outputs that follow a specific format, such as ReAct We then apply rules and involve human annotators to filter out any noisy samples. Afterwards, the samples are incorporated into QWEN’s training data, resulting in an updated version of QWEN that is more dependable for self-instruction. We iterate through this process multiple times until we gather an ample number of samples that possess both exceptional quality and a wide range of diversity. As a result, our final collection consists of around 2000 high-quality samples. (Wang et al., 2023c) (Ivan i al. 2022). During the finetuning process, we mix these high-quality samples with all the other general-purpose SFT samples, rather than introducing an additional training stage. By doing so, we are able to retain essential general-purpose capabilities that are also pertinent for constructing agent applications. We have created and made publicly available a benchmark for evaluating QWEN’s ability to call plugins, tools, functions, or APIs using ReAct Prompting (see To ensure fair evaluation, we have excluded any plugins that were included in QWEN’s training set from the evaluation set. The benchmark assesses the model’s accuracy in selecting the correct plugin from a pool of up to five candidates, as well as the plausibility of the parameters passed into the plugin and the frequency of false positives. In this evaluation, a false positive occurs when the model incorrectly invokes a plugin in response to a query, despite not being required to do so. Using Tools via ReAct Prompting Qwen Team, Alibaba Group, 2023b). The results presented in Table demonstrate that QWEN consistently achieves higher accuracy in identifying the relevance of a query to the available tools as the model size increases. However, the table also highlights that beyond a certain point, there is little improvement in performance when it comes to selecting the appropriate tool and providing relevant arguments. This suggests that the current preliminary benchmark may be relatively easy and may require further enhancement in future iterations. It is worth noting that GPT-3.5 stands out as an exception, displaying suboptimal performance on this particular benchmark. This could potentially be attributed to the fact that the benchmark primarily focuses on the Chinese language, which may not align well with GPT-3.5’s capabilities. Additionally, we observe that GPT-3.5 tends to attempt to use at least one tool, even if the query cannot be effectively addressed by the provided tools. 6 The Python code interpreter is widely regarded as a powerful tool for augmenting the capabilities of an LLM agent. It is worth investigating whether QWEN can harness the full potential of this interpreter to enhance its performance in diverse domains, such as mathematical reasoning and data analysis. To facilitate this exploration, we have developed and made publicly available a benchmark that is specifically tailored for this purpose (see Using Code Interpreter for Math Reasoning and Data Analysis Qwen Team, Alibaba Group, 2023a). The benchmark encompasses three primary categories of tasks: math problem-solving, data visualization, and other general-purpose tasks like file post-processing and web crawling. Within the visualization tasks, we differentiate between two levels of difficulty. The easier level can be achieved by simply writing and executing a single code snippet without the need for advanced planning skills. However, the more challenging level requires strategic planning and executing multiple code snippets in a sequential manner. This is because the subsequent code must be written based on the output of the previous code. For example, an agent may need to examine the structure of a CSV file using one code snippet before proceeding to write and execute additional code to create a plot. Regarding evaluation metrics, we consider both the executability and correctness of the generated code. To elaborate on the correctness metrics, for math problems, we measure accuracy by verifying if the ground truth numerical answer is present in both the code execution result and the final response. When it comes to data visualization, we assess accuracy by utilizing QWEN-VL a powerful multimodal language model. QWEN-VL is capable of answering text questions paired with images, and we rely on it to confirm whether the image generated by the code fulfills the user’s request. (Bai et al., 2023), The results regarding executability and correctness are presented in Table and Table respectively. It is evident that CODE LLAMA generally outperforms LLAMA 2, its generalist counterpart, which is not surprising since this benchmark specifically requires coding skills. However, it is worth noting that specialist models that are optimized for code synthesis do not necessarily outperform generalist models. This is due to the fact that this benchmark encompasses various skills beyond coding, such as abstracting math problems into equations, understanding language-specified constraints, and responding in the specified format such as ReAct. Notably, QWEN-7B-CHAT and QWEN-14B-CHAT surpass all other open-source alternatives of similar scale significantly, despite being generalist models. 7 8, Hugging Face provides a framework called the Hugging Face Agent or Transformers Agent which empowers LLM agents with a curated set of multimodal tools, including speech recognition and image synthesis. This framework allows an LLM agent to interact with humans, interpret natural language commands, and employ the provided tools as needed. Serving as a Hugging Face Agent (Hugging Face, 2023), To evaluate QWEN’s effectiveness as a Hugging Face agent, we utilized the evaluation benchmarks offered by Hugging Face. The results are presented in Table The evaluation results reveal that QWEN performs quite well in comparison to other open-source alternatives, only slightly behind the proprietary GPT-4, demonstrating QWEN’s competitive capabilities. 9. 4 Code-Qwen: Specialized Model for Coding Training on domain-specific data has been shown to be highly effective, particularly in the case of code pretraining and finetuning. A language model that has been reinforced with training on code data can serve as a valuable tool for coding, debugging, and interpretation, among other tasks. In this work, we have developed a series of generalist models using pretraining and alignment techniques. Building on this foundation, we have created domain-specific models for coding by leveraging the base language models of QWEN, including continued pretrained model, CODE-QWEN and supervised finetuned model, CODE-QWEN-CHAT. Both models have 14 billion and 7 billion parameters versions. 4.1 Code Pretraining We believe that relying solely on code data for pretraining can result in a significant loss of the ability to function as a versatile assistant. Unlike previous approaches that focused solely on pretraining on code data we take a different approach by starting with our base models QWEN trained on a combination of text and code data, and then continuing to pretrain on the code data. We continue to pretrain the models on a total of around 90 billion tokens. During the pre-training phase, we initialize the model using the base language models QWEN. Many applications that rely on specialized models for coding may encounter lengthy contextual scenarios, such as tool usage and code interpretation, as mentioned in Section To address this issue, we train our models with context lengths of up to 8192. Similar to base model training in Section we employ Flash Attention in the attention modules, and adopt the standard optimizer AdamW setting 1 = 0*. 2 = 0. = 10−8. We set the learning rate as 6. 10−5 for CODE-QWEN-14B and 3.*0 10−5 for CODE-QWEN-7B, with 3% warm up iterations and no learning rate decays. (Li et al., 2022; 2023d), (Rozie`re et al., 2023) 3.4. 2.4, (Dao et al., 2022) (Kingma & Ba, 2014. godine; Loshchilov & Hutter, 2017), β 9, β 95 i dalje 0 × × 4.2 Code Supervised Fine-Tuning After conducting a series of empirical experiments, we have determined that the multi-stage SFT strategy yields the best performance compared to other methods. In the supervised fine-tuning stage, the model CODE-QWEN-CHAT initialized by the code foundation model CODE-QWEN are optimized by the AdamW optimizer ( 1 = 0 2 = 0.*95, = 10−8) with a learning rate of 2*. 10−6 and 1.*0 10−5 for the 14B and 7B model respectively. The learning rate increases to the peaking value with the cosine learning rate schedule (3% warm-up steps) and then remains constant. (Kingma & Ba, 2014; Loshchilov i Hutter, 2017) β 9, β ϵ 0 × × 4.3 Evaluation Naši CODE-QWEN modeli uspoređeni su s patentnim i otvorenim jezikovnim modelima, kako je prikazano u tablicama and These tables present the results of our evaluation on the test sets of Humaneval MBPP and the multi-lingual code generation benchmark HUMANEVALPACK The comparison is based on the pass@1 performance of the models on these benchmark datasets. The results of this comparison are clearly demonstrated in Tables and 10 11. (Chen et al., 2021), (Austin et al., 2021), (Muennighoff et al., 2023). 10 11. Our analysis reveals that specialized models, specifically CODE-QWEN and CODE-QWEN-CHAT, sig-nificantly outperform previous baselines with similar parameter counts, such as OCTOGEEX InstructCodeT5+ and CodeGeeX2 In fact, these models even rival the performance of larger models like Starcoder (Muen-nighoff et al., 2023), (Wang et al., 2023. godine) (Zheng et al., 2023). (Li et al., 2023d). When compared to some of the extremely large-scale closed-source models, CODE-QWEN and CODE-QWEN-CHAT demonstrate clear advantages in terms of pass@1. However, it is important to note that these models fall behind the state-of-the-art methods, such as GPT-4, in general. Nonetheless, with the continued scaling of both model size and data size, we believe that this gap can be narrowed in the near future. It is crucial to emphasize that the evaluations mentioned previously are insufficient for grasping the full extent of the strengths and weaknesses of the models. In our opinion, it is necessary to develop more rigorous tests to enable us to accurately assess our relative performance in comparison to GPT-4. 5 Math-Qwen: Specialized Model for Mathematics Reasoning We have created a mathematics-specialized model series called MATH-QWEN-CHAT, which is built on top of the QWEN pretrained language models. Specifically, we have developed assistant models that are specifically designed to excel in arithmetic and mathematics and are aligned with human behavior. We are releasing two versions of this model series, MATH-QWEN-14B-CHAT and MATH-QWEN-7B-CHAT, which have 14 billion and 7 billion parameters, respectively. 5.1 Training We carry out math SFT on our augmented math instructional dataset for mathematics reasoning, and therefore we obtain the chat model, MATH-QWEN-CHAT, directly. Owing to shorter average lengths of the math SFT data, we use a sequence length of 1024 for faster training. Most user inputs in the math SFT dataset are examination questions, and it is easy for the model to predict the input format and it is meaningless for the model to predict the input condition and numbers which could be random. Thus, we mask the inputs of the system and user to avoid loss computation on them and find masking them accelerates the convergence during our preliminary experiments. For optimization, we use the AdamW optimizer with the same hyperparameters of SFT except that we use a peak learning rate of 2 10−5 and a training step of 50 000. × 5.2 Evaluation We evaluate models on the test sets of GSM8K (Grade school math) Matematika (matematički problemi koji izazivaju konkurenciju) Math401 (Arithmetic ability) ( and Math23K (Chinese grade school math) We compare MATH-QWEN-CHAT with proprietary models ChatGPT and Minerva and open-sourced math-specialized model RFT ( WizardMath and GAIRMath-Abel in Table MATH-QWEN-CHAT models show better math reasoning and arithmetic abilities compared to open-sourced models and QWEN-CHAT models of similar sizes. Compared to proprietary models, MATH-QWEN-7B-CHAT outperforms Minerva-8B in MATH. MATH-QWEN-14B-CHAT is chasing Minerva-62B and GPT-3.5 in GSM8K and MATH and delivers better performance on arithmetic ability and Chinese math problems. (Cobbe et al., 2021), (Hendrycks et al., 2021), Yuan et al., 2023b), (Vučić i al. 2017). (Lewkowycz et al., 2022) Yuan et al., 2023a), (Luo et al., 2023a), (Chern et al., 2023a) 12. 6 Related Work 6.1 Large Language Models The excitement of LLM began with the introduction of the Transformer architecture ( which was then applied to pretraining large-scale data by researchers such as Ti napori doveli su do značajnog uspjeha u učenju prijenosa, s veličinama modela koji su porasli sa 100 milijuna na više od 10 milijardi parametara. Vaswani et al., 2017), Radford et al. (2018); Devlin et al. (2018); Liu et al. (2019). (Raffel et al., 2020; Shoeybi et al., 2019). In 2020, the release of GPT-3, a massive language model that is 10 times larger than T5, demonstrated the incredible potential of few-shot and zero-shot learning through prompt engineering and in-context learning, and later chain-of-thought prompting This success has led to a number of studies exploring the possibilities of further scaling these models As a result, the community has come to view these large language models as essential foundations for downstream models (Wei et al., 2022c). (Scao et al., 2022; Zhang et al., 2022; Du et al. 2021; Zeng et al., 2022; Lepikhin et al., 2020; Fedus et al., 2022; Du et al., 2022. godine Black et al., 2022; Rae et al., 2021; Hoffmann et al., 2022; Chowdhery et al., 2022; Thoppilan et al., 2022). (Bommasani et al., 2021). The birth of ChatGPT and the subsequent launch of GPT-4 marked two historic moments in the field of artificial intelligence, demonstrating that large language models (LLMs) can serve as effective AI assistants capable of communicating with humans. These events have sparked interests among researchers and developers in building language models that are aligned with human values and potentially even capable of achieving artificial general intelligence (AGI) (OpenAI, 2022) (OpenAI, 2023) (Anil et al., 2023; Anthropic, 2023a;b). One notable development in this area is the emergence of open-source LLMs, specifically LLaMA Plamenovi 2 which have been recognized as the most powerful open-source language models ever created. This has led to a surge of activity in the open-source community with a series of large language models being developed collaboratively to build upon this progress (Sljedeći članakHrvatska et al. 2023a) (Touvron et al., 2023b), (Wolf et al., 2019), (Mosaic ML, 2023; Almazrouei et al., 2023; ChatGLM2 Team, 2023; Yang et al., 2023; InternLM Team, 2023. godine 6.2 Alignment The community was impressed by the surprising effectiveness of alignment on LLMs. Previously, LLMs without alignment often struggle with issues such as repetitive generation, hallucination, and deviation from human preferences. Since 2021, researchers have been diligently working on developing methods to enhance the performance of LLMs in downstream tasks Furthermore, researchers have been actively exploring ways to align LLMs with human instructions One major challenge in alignment research is the difficulty of collecting data. While OpenAI has utilized its platform to gather human prompts or instructions, it is not feasible for others to collect such data. (Wei et al., 2022a; Sanh et al., 2021; Longpre et al., 2023; Chung et al., 2022; Muennighoff et al., 2022). (Ouyang et al., 2022; Askell et al., 2021; Bai et al., 2022b;c). However, there has been some progress in this area, such as the self-instruct approach proposed in This innovative work offers a potential solution to the data collection problem in alignment research. As a result, there has been a surge in open-source chat data, including Alpaca MOSS Dolly Evol-Instruct and others Similarly, there has been an increase in open-source chat models, such as Alpaca Vicuna Guanaco MOSS WizardLM and others Wang et al. 2023 C) (Taori et al., 2023), (Sun et al., 2023a), (Sljedeći članakHrvatska et al. 2023. godine) (Xu et al., 2023 b) i (Sun et al., 2023b; Xu et al., 2023a;c; Chen et al., 2023c; Ding et al., 2023; Ji et al., 2023; Yang, 2023). (Taori et al., 2023), (Chiang et al., 2023), (Dettmers et al., 2023), (Sun et al., 2023a), (Xu et al., 2023b), (Xu et al., 2023c; Chen et al., 2023c; Ding et al., 2023; Wang et al., 2023b). To train an effective chat model, available solutions are mostly based on SFT and RLHF While SFT is similar to pretraining, it focuses on instruction following using the aforementioned data. However, for many developers, the limited memory capacity is a major obstacle to further research in SFT. As a result, parameter-efficient tuning methods, such as LoRA ., and Q-LoRA have gained popularity in the community. LoRA tunes only low-rank adapters, while Q-LoRA builds on LoRA and utilizes 4-bit quantized LLMs and paged attention In terms of RLHF, recent methods such as PPO have been adopted, but there are also alternative techniques aimed at addressing the complexity of optimization, such as RRHF ( DPO and PRO Despite the ongoing debate about the effectiveness of RLHF, more evidence is needed to understand how it enhances the intelligence of LLMs and what potential drawbacks it may have. (Ouyang et al., 2022). (Hu et al 2021. godine (Dettmers et al., 2023), (Dettmers et al., 2022. godine Francuski i al. 2022; Kwon et al., 2023. godine (Schulman et al., 2017; Touvron et al., 2023b) Yuan et al., 2023c), (Rafailov et al., 2023), (Pjesma i al. 2023. godine 6.3 Tool Use and Agents LLM’s planning function allows for the invocation of tools, such as APIs or agent capabilities, through in-context learning, as demonstrated by introduced ReAct, a generation format that enables the model to generate thoughts on which tool to use, accept input from API observations, and generate a response. GPT-3.5 and GPT-4, when prompted with few shots, have shown consistent and impressive performance. In addition to tool usage, LLMs can utilize external memory sources like knowledge bases or search engines to generate more accurate and informative answers. This has led to the popularity of frameworks like LangChain The research on LLMs for tool use has also sparked interest in building agents with LLM capabilities, such as agents that can call different AI models embodied lifelong learning or multimodal agents and multiple agents interacting with each other and even building a micro-society Schick et al. (2023). Yao et al. (2022) (Hu et al., 2023; Zhong et al., 2023b) (Nakano et al., 2021; Liu et al., 2023b) (LangChain, Inc., 2023). (Shen et al., 2023; Li et al., 2023a), (Wang et al., 2023a; Driess et al., 2023), (Chen et al., 2023b; Li et al., 2023b; Sljedeći Članak Xu et al. 2023d; Hong et al., 2023. godine 6.4 LLM for Coding Previous research has demonstrated that LLMs possess remarkable capabilities in code understanding and generation, particularly those with massive numbers of parameters Moreover, several LLMs have been pre-trained, continued pre-trained, or fine-tuned on coding-related data, which has resulted in significantly improved performance compared to general-purpose LLMs. These models include Codex AlphaCode SantaCoder Starcoder-Base InCoder CodeT5 CodeGeeX and CODE LLAMA In addition to these models, recent studies have focused on developing specialized alignment techniques for coding, such as Code Llama-Instruct and StarCoder These models can assist developers in various code-related tasks, including code generation code completion code translation bug fixing code refinement and code question answering In a word, LLMs have the potential to revolutionize the field of coding by providing developers with powerful tools for code comprehension, generation, and related tasks. (Chowdhery et al., 2022; Anil et al., 2023; Rae et al., 2021; Hoffmann et al., 2022). Chen et al. (2021), (Li et al., 2022), (Allal et al., 2023), (Li et al., 2023d), (Fried et al., 2022. godine) (Wang et al., 2021), (Zheng et al., 2023), (Rozie`re et al., 2023. godine (Rozie`re et al., 2023) (Li et al., 2023d). (Chen et al., 2021; Austin et al., 2021), (Zhang et al., 2023a), (Szafraniec et al., 2023), (Muennighoff et al., 2023), (Liu et al., 2023. godine, (Liu & Wan, 2021). 6.5 LLM for Mathematics LLMs with a certain model scale have been found to possess the ability to perform mathematical reasoning Kako bi potaknuli LLM-ove da postignu bolje rezultate na matematičkim zadatcima, istraživači su upotrijebili tehnike kao što je lančano razmišljanje and scratchpad which have shown promising results. Additionally, self-consistency and least-to-most prompting have further improved the performance of these models on these tasks. However, prompt engineering is a time-consuming process that requires a lot of trial and error, and it is still difficult for LLMs to consistently perform well or achieve satisfactory results in solving mathematical problems. Moreover, simply scaling the data and model size is not an efficient way to improve a model’s mathematical reasoning abilities. Instead, pretraining on math-related corpora has been shown to consistently enhance these capabilities Additionally, fine-tuning on math-related instruction-following datasets has also been effective and more cost-effective than math-specific pretraining. Despite their limitations in terms of accuracy, LLMs still have significant potential to assist users with practical mathematical problems. There is ample scope for further development in this area. (Wei et al., 2022b; Suzgun et al., 2022). (Wei et al., 2022c) (Nye et al., 2021. godine) (Wang et al., 2022) (Zhou et al., 2022) (Hendrycks et al., 2021; Lewkowycz et al., 2022; Taylor et al., 2022; Lightman et al., 2023). (Samo ako et al., 2023. godine Yuan i al. 2023a; Luo et al., 2023a; Yue et al., 2023. godine Chern et al., 2023a; Yu et al., 2023. godine) 7 Zaključak In this report, we present the QWEN series of large language models, which showcase the latest advancements in natural language processing. With 14B, 7B, and 1.8B parameters, these models have been pre-trained on massive amounts of data, including trillions of tokens, and fine-tuned using cutting-edge techniques such as SFT and RLHF. Additionally, the QWEN series includes specialized models for coding and mathematics, such as CODE-QWEN, CODE-QWEN-CHAT, and MATH-QWEN-CHAT, which have been trained on domain-specific data to excel in their respective fields. Our results demonstrate that the QWEN series is competitive with existing open-source models and even matches the performance of some proprietary models on comprehensive benchmarks and human evaluation. We believe that the open access of QWEN will foster collaboration and innovation within the community, enabling researchers and developers to build upon our work and push the boundaries of what is possible with language models. By providing these models to the public, we hope to inspire new research and applications that will further advance the field and contribute to our understanding of the variables and techniques introduced in realistic settings. In a nutshell, the QWEN series represents a major milestone in our development of large language models, and we are excited to see how it will be used to drive progress and innovation in the years to come. References Loubna Ben Allal, Raymond Li, Denis Kocetkov, Chenghao Mou, Christopher Akiki, Carlos Munoz Ferrandis, Niklas Muennighoff, Mayank Mishra, Alex Gu, Manan Dey, et al. SantaCoder: Don’t reach for the stars! , 2023. arXiv preprint arXiv:2301.03988 Ebtesam Almazrouei, Hamza Alobeidli, Abdulaziz Alshamsi, Alessandro Cappelli, Ruxandra Co-jocaru, Merouane Debbah, Etienne Goffinet, Daniel Heslow, Julien Launay, Quentin Malartic, Badreddine Noune, Baptiste Pannier, and Guilherme Penedo. Falcon-40B: An open large language model with state-of-the-art performance, 2023. Rohan Anil, Andrew M Dai, Orhan Firat, Melvin Johnson, Dmitry Lepikhin, Alexandre Passos, Siamak Shakeri, Emanuel Taropa, Paige Bailey, Zhifeng Chen, et al. PaLM 2 technical report. , 2023. arXiv preprint arXiv:2305.10403 Anthropic. Introducing Claude, 2023a. URL .nthropic. Claude 2. Technical report, Anthropic, 2023b. URL . . https://www.anthropic.com/index/ introducing-claude https://www-files anthropic.com/production/images/Model-Card-Claude-2.pdf Vamsi Aribandi, Yi Tay, Tal Schuster, Jinfeng Rao, Huaixiu Steven Zheng, Sanket Vaibhav Mehta, Honglei Zhuang, Vinh Q Tran, Dara Bahri, Jianmo Ni, et al. ExT5: Towards extreme multi-task scaling for transfer learning. , 2021. arXiv preprint arXiv:2111.10952 Amanda Askell, Yuntao Bai, Anna Chen, Dawn Drain, Deep Ganguli, Tom Henighan, Andy Jones, Nicholas Joseph, Ben Mann, Nova DasSarma, et al. A general language assistant as a laboratory for alignment. , 2021. arXiv preprint arXiv:2112.00861 Jacob Austin, Augustus Odena, Maxwell Nye, Maarten Bosma, Henryk Michalewski, David Dohan, Ellen Jiang, Carrie Cai, Michael Terry, Quoc Le, et al. Program synthesis with large language models. , 2021. arXiv preprint arXiv:2108.07732 AutoGPT. AutoGPT: The heart of the open-source agent ecosystem, 2023. URL . https:// github.com/Significant-Gravitas/Auto-GPT Lei Jimmy Ba, Jamie Ryan Kiros, and Geoffrey E. Hinton. Layer normalization. , abs/1607.06450, 2016. URL . CoRR http://arxiv.org/abs/1607.06450 Jinze Bai, Rui Men, Hao Yang, Xuancheng Ren, Kai Dang, Yichang Zhang, Xiaohuan Zhou, Peng Wang, Sinan Tan, An Yang andf Zeyu Cui, Yu Han, Shuai Bai, Wenbin Ge, Jianxin Ma, Junyang Lin, Jingren Zhou, and Chang Zhou. OFASys: A multi-modal multi-task learning system for building generalist models. , abs/2212.04408, 2022a. doi: 10.48550/arXiv.2212.04408. URL . CoRR https://doi.org/10.48550/arXiv.2212.04408 Jinze Bai, Shuai Bai, Shusheng Yang, Shijie Wang, Sinan Tan, Peng Wang, Junyang Lin, Chang Zhou, and Jingren Zhou. Qwen-VL: A versatile vision-language model for understanding, localization, text reading, and beyond. , abs/2308.12966, 2023. doi: 10.48550/arXiv.2308.12966. URL . CoRR https://doi.org/10.48550/arXiv.2308.12966 Yuntao Bai, Andy Jones, Kamal Ndousse, Amanda Askell, Anna Chen, Nova DasSarma, Dawn Drain, Stanislav Fort, Deep Ganguli, Tom Henighan, et al. Training a helpful and harmless assistant with reinforcement learning from human feedback. , 2022b. arXiv preprint arXiv:2204.05862 Yuntao Bai, Saurav Kadavath, Sandipan Kundu, Amanda Askell, Jackson Kernion, Andy Jones, Anna Chen, Anna Goldie, Azalia Mirhoseini, Cameron McKinnon, et al. Constitutional AI: Harmlessness from AI feedback. , 2022c. arXiv preprint arXiv:2212.08073 Iz Beltagy, Matthew E Peters, and Arman Cohan. Longformer: The long-document transformer. , 2020. arXiv preprint arXiv:2004.05150 Yonatan Bisk, Rowan Zellers, Ronan Le Bras, Jianfeng Gao, and Yejin Choi. PIQA: reasoning about physical commonsense in natural language. In , pp. 7432–7439. AAAI Press, 2020. doi: The Thirty-Fourth AAAI Conference on Artificial Intelligence, AAAI 2020, The Thirty-Second Innovative Applications of Artificial Intelligence Con-ference, IAAI 2020, The Tenth AAAI Symposium on Educational Advances in Artificial Intelligence, EAAI 2020, New York, NY, USA, February 7-12, 2020 10.1609/aaai.v34i05.6239. URL . https://doi.org/10.1609/aaai.v34i05.6239 Sid Black, Stella Biderman, Eric Hallahan, Quentin Anthony, Leo Gao, Laurence Golding, Horace He, Connor Leahy, Kyle McDonell, Jason Phang i drugi. , 2022. arXiv preprint arXiv:2204.06745 bloc97. NTK-aware scaled RoPE allows LLaMA models to have extended (8k+) con-text size without any fine-tuning and minimal perplexity degradation., 2023. URL . https://www.reddit.com/r/LocalLLaMA/comments/14lz7j5/ntkaware_ scaled_rope_allows_llama_models_to_have/ Rishi Bommasani, Drew A Hudson, Ehsan Adeli, Russ Altman, Simran Arora, Sydney von Arx, Michael S Bernstein, Jeannette Bohg, Antoine Bosselut, Emma Brunskill, et al. On the opportuni-ties and risks of foundation models. , 2021. arXiv preprint arXiv:2108.07258 Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. Language models are few-shot learners. , 33:1877–1901, 2020. Advances in neural information processing systems ChatGLM2 Team. ChatGLM2-6B: An open bilingual chat LLM, 2023. URL . . https://github com/THUDM/ChatGLM2-6B Mark Chen, Jerry Tworek, Heewoo Jun, Qiming Yuan, Henrique Ponde´ de Oliveira Pinto, Jared Kaplan, Harrison Edwards, Yuri Burda, Nicholas Joseph, Greg Brockman, Alex Ray, Raul Puri, Gretchen Krueger, Michael Petrov, Heidy Khlaaf, Girish Sastry, Pamela Mishkin, Brooke Chan, Scott Gray, Nick Ryder, Mikhail Pavlov, Alethea Power, Lukasz Kaiser, Mohammad Bavarian, Clemens Winter, Philippe Tillet, Felipe Petroski Such, Dave Cummings, Matthias Plappert, Fotios Chantzis, Elizabeth Barnes, Ariel Herbert-Voss, William Hebgen Guss, Alex Nichol, Alex Paino, Nikolas Tezak, Jie Tang, Igor Babuschkin, Suchir Balaji, Shantanu Jain, William Saunders, Christopher Hesse, Andrew N. Carr, Jan Leike, Joshua Achiam, Vedant Misra, Evan Morikawa, Alec Radford, Matthew Knight, Miles Brundage, Mira Murati, Katie Mayer, Peter Welinder, Bob McGrew, Dario Amodei, Sam McCandlish, Ilya Sutskever, and Wojciech Zaremba. Evaluating large language models trained on code. , abs/2107.03374, 2021. URL . . CoRR https://arxiv org/abs/2107.03374 Shouyuan Chen, Sherman Wong, Liangjian Chen, and Yuandong Tian. Extending context window of large language models via positional interpolation. , 2023a. arXiv preprint arXiv:2306.15595 Weize Chen, Yusheng Su, Jingwei Zuo, Cheng Yang, Chenfei Yuan, Chen Qian, Chi-Min Chan, Yujia Qin, Yaxi Lu, Ruobing Xie, et al. Agentverse: Facilitating multi-agent collaboration and exploring emergent behaviors in agents. , 2023b. arXiv preprint arXiv:2308.10848 Zhihong Chen, Feng Jiang, Junying Chen, Tiannan Wang, Fei Yu, Guiming Chen, Hongbo Zhang, Juhao Liang, Chen Zhang, Zhiyi Zhang, et al. Phoenix: Democratizing ChatGPT across languages. , 2023c. arXiv preprint arXiv:2304.10453 Ethan Chern, Haoyang Zou, Xuefeng Li, Jiewen Hu, Kehua Feng, Junlong Li, and Pengfei Liu. Generative ai for math: Abel. , 2023a. https://github.com/GAIR-NLP/abel I Chern, Steffi Chern, Shiqi Chen, Weizhe Yuan, Kehua Feng, Chunting Zhou, Junxian He, Graham Neubig, Pengfei Liu, et al. Factool: Factuality detection in generative ai–a tool augmented framework for multi-task and multi-domain scenarios. 2023. godine b. arXiv preprint arXiv:2307.13528 David Chiang and Peter Cholak. Overcoming a theoretical limitation of self-attention. In , str. 7654–7664, 2022. Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) Wei-Lin Chiang, Zhuohan Li, Zi Lin, Ying Sheng, Zhanghao Wu, Hao Zhang, Lianmin Zheng, Siyuan Zhuang, Yonghao Zhuang, Joseph E. Gonzalez, Ion Stoica, and Eric P. Xing. Vicuna: An open-source chatbot impressing GPT-4 with 90%* ChatGPT quality, March 2023. URL . https://lmsys.org/blog/2023-03-30-vicuna/ Aakanksha Chowdhery, Sharan Narang, Jacob Devlin, Maarten Bosma, Gaurav Mishra, Adam Roberts, Paul Barham, Hyung Won Chung, Charles Sutton, Sebastian Gehrmann, et al. PaLM: Scaling language modeling with pathways. , 2022. arXiv preprint arXiv:2204.02311 Paul F. Christiano, Jan Leike, Tom B. Brown, Miljan Martic, Shane Legg, and Dario Amodei. Deep reinforcement learning from human preferences. In Isabelle Guyon, Ulrike von Luxburg, Samy Bengio, Hanna M. Wallach, Rob Fergus, S. V. N. Vishwanathan, and Roman Garnett (eds.), , str. 4299–4307, 2017. URL . Advances in Neural Information Processing Systems 30: Annual Conference on Neu-ral Information Processing Systems 2017, December 4-9, 2017, Long Beach, CA, USA https://proceedings.neurips.cc/paper/2017/hash/ d5e2c0adad503c91f91df240d0cd4e49-Abstract.html Hyung Won Chung, Le Hou, Shayne Longpre, Barret Zoph, Yi Tay, William Fedus, Eric Li, Xuezhi Wang, Mostafa Dehghani, Siddhartha Brahma, et al. Scaling instruction-finetuned language models. , 2022. arXiv preprint arXiv:2210.11416 Christopher Clark, Kenton Lee, Ming-Wei Chang, Tom Kwiatkowski, Michael Collins, and Kristina Toutanova. Boolq: Exploring the surprising difficulty of natural yes/no questions. In Jill Burstein, Christy Doran, and Thamar Solorio (eds.), , pp. 2924–2936. Association for Computational Linguistics, 2019. doi: 10.18653/v1/n19-1300. URL . Proceedings of the 2019 Conference of the North Amer-ican Chapter of the Association for Computational Linguistics: Human Language Technologies, NAACL-HLT 2019, Minneapolis, MN, USA, June 2-7, 2019, Volume 1 (Long and Short Papers) https://doi.org/10.18653/v1/n19-1300 Peter Clark, Isaac Cowhey, Oren Etzioni, Tushar Khot, Ashish Sabharwal, Carissa Schoenick, and Oyvind Tafjord. Think you have solved question answering? try arc, the AI2 reasoning challenge. , abs/1803.05457, 2018. URL . CoRR http://arxiv.org/abs/1803.05457 Karl Cobbe, Vineet Kosaraju, Mohammad Bavarian, Mark Chen, Heewoo Jun, Lukasz Kaiser, Matthias Plappert, Jerry Tworek, Jacob Hilton, Reiichiro Nakano, et al. Training verifiers to solve math word problems. , 2021. arXiv preprint arXiv:2110.14168 Alexis Conneau, Kartikay Khandelwal, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Fran-cisco Guzma ́n, Edouard Grave, Myle Ott, Luke Zettlemoyer i Veselin Stoyanov. , 2019. arXiv preprint arXiv:1911.02116 Mike Conover, Matt Hayes, Ankit Mathur, Jianwei Xie, Jun Wan, Sam Shah, Ali Ghodsi, Patrick Wendell, Matei Zaharia, and Reynold Xin. Free Dolly: Introducing the world’s first truly open instruction-tuned LLM, 2023. URL . https://www.databricks.com/blog/2023/04/ 12/dolly-first-open-commercially-viable-instruction-tuned-llm Wenliang Dai, Junnan Li, Dongxu Li, Anthony Meng, Huat Tiong, Junqi Zhao, Weisheng Wang, Boyang Li, Pascale Fung, and Steven Hoi. InstructBLIP: Towards general-purpose vision-language models with instruction tuning. , 2023. arXiv preprint arXiv:2305.06500 Tri Dao, Daniel Y. Fu, Stefano Ermon, Atri Rudra i Christopher Re ́. FlashAt-tention: Brza i pamćenje učinkovita točna pažnja s io-svjesnošću. 2022. godine URL . NeurIPS http://papers.nips.cc/paper_files/paper/2022/hash/ 67d57c32e20fd0a7a302cb81d36e40d5-Abstract-Conference.html Yann N Dauphin, Angela Fan, Michael Auli i David Grangier. Jezik modeliranje s konvolucijskim mrežama. , pp. 933–941. PMLR, 2017. International conference on machine learning Tim Dettmers, Mike Lewis, Younes Belkada, and Luke Zettlemoyer. LLM.int8(): 8-bit matrix multiplication for transformers at scale. , 2022. arXiv preprint arXiv:2208.07339 Tim Dettmers, Artidoro Pagnoni, Ari Holtzman, and Luke Zettlemoyer. QLoRA: Efficient finetuning of quantized LLMs. , 2023. arXiv preprint arXiv:2305.14314 Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. BERT: Pre-training of deep bidirectional transformers for language understanding. , 2018. arXiv preprint arXiv:1810.04805 Ning Ding, Yulin Chen, Bokai Xu, Yujia Qin, Zhi Zheng, Shengding Hu, Zhiyuan Liu, Maosong Sun, and Bowen Zhou. Enhancing chat language models by scaling high-quality instructional conversations. , 2023. arXiv preprint arXiv:2305.14233 Danny Driess, Fei Xia, Mehdi SM Sajjadi, Corey Lynch, Aakanksha Chowdhery, Brian Ichter, Ayzaan Wahid, Jonathan Tompson, Quan Vuong, Tianhe Yu i drugi. , 2023. arXiv preprint arXiv:2303.03378 Nan Du, Yanping Huang, Andrew M Dai, Simon Tong, Dmitry Lepikhin, Yuanzhong Xu, Maxim Krikun, Yanqi Zhou, Adams Wei Yu, Orhan Firat, i dr. , pp. 5547–5569. PMLR, 2022. International Conference on Machine Learning Zhengxiao Du, Yujie Qian, Xiao Liu, Ming Ding, Jiezhong Qiu, Zhilin Yang, and Jie Tang. GLM: General language model pretraining with autoregressive blank infilling. , 2021. arXiv preprint arXiv:2103.10360 Kawin Ethayarajh, Yejin Choi, and Swabha Swayamdipta. Understanding dataset difficulty with -usable information. In Kamalika Chaudhuri, Stefanie Jegelka, Le Song, Csaba Szepesvari, Gang Niu, and Sivan Sabato (eds.), Volumen 162 , str. 5988–6008. PMLR, 17–23 Jul 2022. V Proceedings of the 39th International Conference on Machine Learning Proceedings of Machine Learning Research William Fedus, Barret Zoph, and Noam Shazeer. Switch transformers: Scaling to trillion parameter models with simple and efficient sparsity. , 23(1): 5232–5270, 2022. The Journal of Machine Learning Research Elias Frantar, Saleh Ashkboos, Torsten Hoefler, and Dan Alistarh. GPTQ: Accurate post-training quantization for generative pre-trained transformers. 2022. godine arXiv preprint arXiv:2210.17323 Daniel Fried, Armen Aghajanyan, Jessy Lin, Sida I. Wang, Eric Wallace, Freda Shi, Ruiqi Zhong, Wen tau Yih, Luke Zettlemoyer, and Mike Lewis. Incoder: A generative model for code infilling and synthesis. , abs/2204.05999, 2022. ArXiv Google. An important next step on our AI journey, 2023. URL . https://blog.google/ Sljedeći članakSljedeći članak: google-ai-search-updates Dan Hendrycks i Kevin Gimpel: Mostovanje ne-linearnosti i stohastičkih regularizera s Gaussian error linearnim jedinicama. , abs/1606.08415, 2016. URL . . CoRR http://arxiv.org/abs/1606 08415 Dan Hendrycks, Collin Burns, Steven Basart, Andy Zou, Mantas Mazeika, Dawn Song, and Jacob Steinhardt. Measuring massive multitask language understanding. U 2020. arXiv preprint arXiv:2009.03300 Dan Hendrycks, Collin Burns, Saurav Kadavath, Akul Arora, Steven Basart, Eric Tang, Dawn Song i Jacob Steinhardt. , 2021. arXiv preprint arXiv:2103.03874 Jordan Hoffmann, Sebastian Borgeaud, Arthur Mensch, Elena Buchatskaya, Trevor Cai, Eliza Rutherford, Diego de Las Casas, Lisa Anne Hendricks, Johannes Welbl, Aidan Clark, et al. Training compute-optimal large language models. 2022. godine arXiv preprint arXiv:2203.15556 Sirui Hong, Xiawu Zheng, Jonathan Chen, Yuheng Cheng, Ceyao Zhang, Zili Wang, Steven Ka Shing Yau, Zijuan Lin, Liyang Zhou, Chenyu Ran, et al. Metagpt: Meta programming for multi-agent collaborative framework. , 2023. arXiv preprint arXiv:2308.00352 Chenxu Hu, Jie Fu, Chenzhuang Du, Simian Luo, Junbo Zhao, and Hang Zhao. Chatdb: Augmenting llms with databases as their symbolic memory. , 2023. arXiv preprint arXiv:2306.03901 Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. LoRA: Low-rank adaptation of large language models. , 2021. arXiv preprint arXiv:2106.09685 Hai Hu, Kyle Richardson, Liang Xu, Lu Li, Sandra Ku¨bler, and Lawrence S. Moss. OCNLI: original chinese natural language inference. In Trevor Cohn, Yulan He, and Yang Liu (eds.), , volume EMNLP 2020 of Udruga za računalne lingvistike, 2020. doi: 10.18653/v1/2020.findings-emnlp.314. URL . Rezultati Udruge za računalne lingvistike: EMNLP 2020, online događaj, 16.-20. studenog 2020. Findings of ACL https://doi.org/10.18653/v1/2020.findings-emnlp.314 Yuzhen Huang, Yuzhuo Bai, Zhihao Zhu, Junlei Zhang, Jinghan Zhang, Tangjun Su, Junteng Liu, Chuancheng Lv, Yikai Zhang, Jiayi Lei, et al. C-Eval: A multi-level multi-discipline chinese evaluation suite for foundation models. , 2023. arXiv preprint arXiv:2305.08322 Hugging Face. Transformers agents, 2023. URL . https://huggingface.co/docs/ transformers/transformers_agents Baichuan Inc. Baichuan-7B: A large-scale 7B pretraining language model developed by BaiChuan-Inc, 2023a. URL . https://github.com/baichuan-inc/Baichuan-7B XVERSE Technology Inc. XVERSE-13B: A multilingual large language model devel-oped by XVERSE Technology Inc., 2023b. URL . https://github.com/xverse-ai/ XVERSE-13B InternLM Team. InternLM: A multilingual language model with progressively enhanced capabilities, 2023. URL . https://github.com/InternLM/InternLM Shantanu Jain. tiktoken: A fast BPE tokeniser for use with OpenAI’s models, 2022. URL . https://github.com/openai/tiktoken/ Yunjie Ji, Yong Deng, Yan Gong, Yiping Peng, Qiang Niu, Lei Zhang, Baochang Ma, and Xiangang Li. Exploring the impact of instruction data scaling on large language models: An empirical study on real-world use cases. , 2023. arXiv preprint arXiv:2303.14742 Zixuan Jiang, Jiaqi Gu, Hanqing Zhu i David Z. Pan. Pre-RMSNorm i Pre-CRMSNorm transformatori: ekvivalentni i učinkoviti pre-LN transformatori. , abs/2305.14858, 2023. doi: 10.48550/arXiv.2305.14858. URL . Koroš https://doi.org/10.48550/arXiv.2305.14858 Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. , 2014. arXiv preprint arXiv:1412.6980 Tom Kwiatkowski, Jennimaria Palomaki, Olivia Redfield, Michael Collins, Ankur P. Parikh, Chris Alberti, Danielle Epstein, Illia Polosukhin, Jacob Devlin, Kenton Lee, Kristina Toutanova, Llion Jones, Matthew Kelcey, Ming-Wei Chang, Andrew M. Dai, Jakob Uszkoreit, Quoc Le, and Slav Petrov. Natural questions: a benchmark for question answering research. , 7:452–466, 2019. doi: 10.1162/tacl*\* a*\* 00276. URL . . Trans. Assoc. Comput. Linguistics https://doi.org/10 1162/tacl_a_00276 Woosuk Kwon, Zhuohan Li, Siyuan Zhuang, Ying Sheng, Lianmin Zheng, Cody Hao Yu, Joseph E. Gonzalez, Hao Zhang, and Ion Stoica. Efficient memory management for large language model serving with PagedAttention. In , 2023. 29. simpozij ACM SIGOPS o načelima operativnih sustava LangChain, Inc. LangChain: Building applications with LLMs through composability, 2023. URL . https://python.langchain.com/ Dmitry Lepikhin, HyoukJoong Lee, Yuanzhong Xu, Dehao Chen, Orhan Firat, Yanping Huang, Maxim Krikun, Noam Shazeer, and Zhifeng Chen. GShard: Scaling giant models with conditional computation and automatic sharding. , 2020. arXiv preprint arXiv:2006.16668 Aitor Lewkowycz, Anders Andreassen, David Dohan, Ethan Dyer, Henryk Michalewski, Vinay Ramasesh, Ambrose Slone, Cem Anil, Imanol Schlag, Theo Gutman-Solo, Yuhuai Wu, Behnam Neyshabur, Guy Gur-Ari i Vedant Misra. Chenliang Li, Hehong Chen, Ming Yan, Weizhou Shen, Haiyang Xu, Zhikai Wu, Zhicheng Zhang, Wenmeng Zhou, Yingda Chen, Chen Cheng, et al. ModelScope-Agent: Izgradite svoj prilagođeni agentski sustav s velikim jezikovnim modelima otvorenog koda. 2023 a. arXiv preprint arXiv:2309.00986 Guohao Li, Hasan Abed Al Kader Hammoud, Hani Itani, Dmitrii Khizbullin i Bernard Ghanem. 2023. godine b. arXiv preprint arXiv:2303.17760 Haonan Li, Yixuan Zhang, Fajri Koto, Yifei Yang, Hai Zhao, Yeyun Gong, Nan Duan i Timothy Baldwin. 2023 c. arXiv preprint arXiv:2306.09212 Raymond Li, Loubna Sean Allal, Yangtian Zi, Niklas Muennighoff, Denis Kocetkov, Chenghao Mou, Marc Marone, Christopher Akiki, Jia Li, Jenny Chim, Qian Liu, Evgenii Zheltonozhskii, Terry Yue Zhuo, Thomas Wang, Lipkin, Olivier Dehaene, Mishig Davaadorj, Joel Lamy-Poirier, Rudra Murthy V, Jason Stiller, Siva Sank Patel, Dmitry Abulkhanov, Nicolas Gontier, Nicholas Meade, Armel Zebaze, Ming-Ho Yee, Logesh Kumar Umapathi, Jian Zhu, Jian Zhu, Benjamin Lipkin, Muhtamokulov, Zuo Red, Rudra Murthy V, Jason Stiller, S , abs/2305.06161, 2023d. doi: 10.48550/arXiv.2305.06161. URL . Koroš https://doi.org/10.48550/arXiv.2305.06161 Yujia Li, David H. Choi, Junyoung Chung, Nate Kushman, Julian Schrittwieser, Re´mi Leblond, Tom Eccles, James Keeling, Felix Gimeno, Agustin Dal Lago, Thomas Hubert, Peter Choy, Cyprien de Masson d’Autume, Igor Babuschkin, Xinyun Chen, Po-Sen Huang, Johannes Welbl, Sven Gowal, Alexey Cherepanov, James Molloy, Daniel J. Mankowitz, Esme Sutherland Robson, Pushmeet Kohli, Nando de Freitas, Koray Kavukcuoglu, and Oriol Vinyals. Competition-level code generation with AlphaCode. , abs/2203.07814, 2022. Koroš Hunter Lightman, Vineet Kosaraju, Yura Burda, Harri Edwards, Bowen Baker, Teddy Lee, Jan Leike, John Schulman, Ilya Sutskever, and Karl Cobbe. Let’s verify step by step. 2023. godine arXiv preprint arXiv:2305.20050 Chenxiao Liu and Xiaojun Wan. CodeQA: A question answering dataset for source code com-prehension. In Marie-Francine Moens, Xuanjing Huang, Lucia Specia, and Scott Wen-tau Yih (eds.), , pp. 2618–2632. Associa-tion for Computational Linguistics, 2021. doi: 10.18653/v1/2021.findings-emnlp.223. URL . Findings of the Association for Computational Linguistics: EMNLP 2021, Virtual Event / Punta Cana, Dominican Republic, 16-20 November, 2021 https://doi.org/10.18653/v1/2021.findings-emnlp.223 Haotian Liu, Chunyuan Li, Qingyang Wu, and Yong Jae Lee. Visual instruction tuning. , 2023a. arXiv preprint arXiv:2304.08485 Xiao Liu, Hanyu Lai, Hao Yu, Yifan Xu, Aohan Zeng, Zhengxiao Du, Peng Zhang, Yuxiao Dong, and Jie Tang. WebGLM: Towards an efficient web-enhanced question answering system with human preferences. , 2023b. arXiv preprint arXiv:2306.07906 Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, and Veselin Stoyanov. RoBERTa: A robustly optimized BERT pretraining approach. , 2019. arXiv preprint arXiv:1907.11692 Yue Liu, Thanh Le-Cong, Ratnadira Widyasari, Chakkrit Tantithamthavorn, Li Li, Xuan-Bach Dinh Le, and David Lo. Refining ChatGPT-generated code: Characterizing and mitigating code quality issues. , abs/2307.12596, 2023c. doi: 10.48550/arXiv.2307.12596. URL . CoRR https://doi.org/10.48550/arXiv.2307.12596 Shayne Longpre, Le Hou, Tu Vu, Albert Webson, Hyung Won Chung, Yi Tay, Denny Zhou, Quoc V Le, Barret Zoph, Jason Wei, et al. The Flan collection: Designing data and methods for effective instruction tuning. , 2023. arXiv preprint arXiv:2301.13688 Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization. , 2017. arXiv preprint arXiv:1711.05101 Keming Lu, Hongyi Yuan, Zheng Yuan, Runji Lin, Junyang Lin, Chuanqi Tan, Chang Zhou i Jingren Zhou. #InsTag: Upute za označavanje za analizu nadzorovanog fine-tuning velikih jezičnih modela. , abs/2308.07074, 2023. doi: 10.48550/arXiv.2308.07074. URL . . CoRR https://doi org/10.48550/arXiv.2308.07074 Haipeng Luo, Qingfeng Sun, Can Xu, Pu Zhao, Jianguang Lou, Chongyang Tao, Xiubo Geng, Qingwei Lin, Shifeng Chen i Dongmei Zhang. , 2023a. arXiv preprint arXiv:2308.09583 Ziyang Luo, Can Xu, Pu Zhao, Qingfeng Sun, Xiubo Geng, Wenxiang Hu, Chongyang Tao, Jing Ma, Qingwei Lin i Daxin Jiang. , 2023b. arXiv preprint arXiv:2306.08568 Mosaic ML. MPT-30B: Raising the bar for open-source foundation models, 2023. URL . https://www.mosaicml.com/blog/mpt-30b Niklas Muennighoff, Thomas Wang, Lintang Sutawika, Adam Roberts, Stella Biderman, Teven Le Scao, M Saiful Bari, Sheng Shen, Zheng-Xin Yong, Hailey Schoelkopf, et al. Crosslingual generalization through multitask finetuning. , 2022. arXiv preprint arXiv:2211.01786 Niklas Muennighoff, Qian Liu, Armel Zebaze, Qinkai Zheng, Binyuan Hui, Terry Yue Zhuo, Swayam Singh, Xiangru Tang, Leandro von Werra i Shayne Longpre. , abs/2308.07124, 2023. CoRR Reiichiro Nakano, Jacob Hilton, Suchir Balaji, Jeff Wu, Long Ouyang, Christina Kim, Christopher Hesse, Shantanu Jain, Vineet Kosaraju, William Saunders, et al. WebGPT: Browser-assisted question-answering with human feedback. U 2021. arXiv preprint arXiv:2112.09332 Maxwell Nye, Anders Andreassen, Guy Gur-Ari, Henryk Michalewski, Jacob Austin, David Bieber, David Dohan, Aitor Lewkowycz, Maarten Bosma, David Luan, Charles Sutton, and Augustus Odena. Show your work: Scratchpads for intermediate computation with language models. , abs/2112.00114, 2021. ArXiv OpenAI. Introducing ChatGPT, 2022. URL . https://openai.com/blog/chatgpt OpenAI. ChatML, 2022. URL . https://github.com/openai/openaipython/blob/e389823ba013a24b4c32ce38fa0bd87e6bccae94/chatml.md OpenAI. GPT4 technical report. , 2023. arXiv preprint arXiv:2303.08774 OpenCompass Team. OpenCompass: A universal evaluation platform for foundation models, 2023. URL . https://opencompass.org.cn/leaderboard-llm Long Ouyang, Jeffrey Wu, Xu Jiang, Diogo Almeida, Carroll L. Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, John Schulman, Jacob Hilton, Fraser Kelton, Luke Miller, Maddie Simens, Amanda Askell, Peter Welinder, Paul F. Christiano, Jan Leike i Ryan Lowe. , 2022. URL . NeurIPS http://papers.nips.cc/paper_files/paper/2022/hash/ b1efde53be364a73914f58805a001731-Abstract-Konferencija.html Denis Paperno, Germa´n Kruszewski, Angeliki Lazaridou, Quan Ngoc Pham, Raffaella Bernardi, Sandro Pezzelle, Marco Baroni, Gemma Boleda, and Raquel Ferna´ndez. The LAMBADA dataset: Word prediction requiring a broad discourse context. In Društvo za računalnu lingvistiku, 2016. doi: 10.18653/v1/ p16-1144. URL . Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics, ACL 2016, August 7-12, 2016, Berlin, Germany, Volume 1: Long Papers https://doi.org/10.18653/v1/p16-1144 Bowen Peng, Jeffrey Quesnelle, Honglu Fan, and Enrico Shippole. YaRN: Efficient context window extension of large language models. , 2023a. arXiv preprint arXiv:2309.00071 Zhiliang Peng, Wenhui Wang, Li Dong, Yaru Hao, Shaohan Huang, Shuming Ma, and Furu Wei. Kosmos-2: Grounding multimodal large language models to the world. , 2023b. arXiv preprint arXiv:2306.14824 Qwen Team, Alibaba Group. Evaluation benchmark for code intepreter, 2023a. URL https://github.com/QwenLM/Qwen-Agent/tree/main/benchmark. Qwen Team, Alibaba Group. referentna vrijednost za korištenje alata putem ReAct promptinga, 2023b. URL . https://github.com/QwenLM/Qwen-7B/tree/main/eval Alec Radford, Karthik Narasimhan, Tim Salimans, Ilya Sutskever, et al. Improving language understanding by generative pre-training. Technical report, OpenAI, 2018. Jack W Rae, Sebastian Borgeaud, Trevor Cai, Katie Millican, Jordan Hoffmann, Francis Song, John Aslanides, Sarah Henderson, Roman Ring, Susannah Young i drugi. , 2021. arXiv preprint arXiv:2112.11446 Rafael Rafailov, Archit Sharma, Eric Mitchell, Stefano Ermon, Christopher D Manning, and Chelsea Finn. Direct preference optimization: Your language model is secretly a reward model. , 2023. arXiv preprint arXiv:2305.18290 Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J Liu. Exploring the limits of transfer learning with a unified text-to-text transformer. , 21(1):5485–5551, 2020. The Journal of Machine Learning Research Prajit Ramachandran, Barret Zoph, and Quoc V Le. Searching for activation functions. , 2017. arXiv preprint arXiv:1710.05941 Scott E. Reed, Konrad Zolna, Emilio Parisotto, Sergio Go´mez Colmenarejo, Alexander Novikov, Gabriel Barth-Maron, Mai Gimenez, Yury Sulsky, Jackie Kay, Jost Tobias Springenberg, Tom Eccles, Jake Bruce, Ali Razavi, Ashley Edwards, Nicolas Heess, Yutian Chen, Raia Hadsell, Oriol Vinyals, Mahyar Bordbar, and Nando de Freitas. A generalist agent. , 2022, 2022. URL . Trans. Mach. Learn. Res. https://openreview.net/forum?id=1ikK0kHjvj Baptiste Rozie`re, Jonas Gehring, Fabian Gloeckle, Sten Sootla, Itai Gat, Xiaoqing Ellen Tan, Yossi Adi, Jingyu Liu, Tal Remez, Je´re´my Rapin, et al. Code Llama: Open foundation models for code. , 2023. arXiv preprint arXiv:2308.12950 Victor Sanh, Albert Webson, Colin Raffel, Stephen H Bach, Lintang Sutawika, Zaid Alyafeai, Antoine Chaffin, Arnaud Stiegler, Teven Le Scao, Arun Raja, et al. Multitask prompted training enables zero-shot task generalization. U 2021. arXiv preprint arXiv:2110.08207 Maarten Sap, Hannah Rashkin, Derek Chen, Ronan Le Bras, and Yejin Choi. SocialIQA: Com-monsense reasoning about social interactions. , abs/1904.09728, 2019. URL CoRR http: . //arxiv.org/abs/1904.09728 Teven Le Scao, Angela Fan, Christopher Akiki, Ellie Pavlick, Suzana Ilic´, Daniel Hesslow, Roman Castagne´, Alexandra Sasha Luccioni, Franc¸ois Yvon, Matthias Galle´, et al. BLOOM: A 176B-parameter open-access multilingual language model. , 2022. arXiv preprint arXiv:2211.05100 Timo Schick, Jane Dwivedi-Yu, Roberto Dess`ı, Roberta Raileanu, Maria Lomeli, Luke Zettlemoyer, Nicola Cancedda, and Thomas Scialom. Toolformer: Language models can teach themselves to use tools. 2023. godine arXiv preprint arXiv:2302.04761 John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, and Oleg Klimov. Proximal policy optimization algorithms. 2017. godine arXiv preprint arXiv:1707.06347 Noam Shazeer. GLU variants improve transformer. , 2020. arXiv preprint arXiv:2002.05202 Yongliang Shen, Kaitao Song, Xu Tan, Dongsheng Li, Weiming Lu, and Yueting Zhuang. Hug-gingGPT: Solving AI tasks with ChatGPT and its friends in HuggingFace. , 2023. arXiv preprint arXiv:2303.17580 Mohammad Shoeybi, Mostofa Patwary, Raul Puri, Patrick LeGresley, Jared Casper, and Bryan Catan-zaro. Megatron-LM: Training multi-billion parameter language models using model parallelism. , 2019. arXiv preprint arXiv:1909.08053 Qingyi Si, Tong Wang, Naibin Gu, Rui Liu, and Zheng Lin. Alpaca-CoT: An instruction-tuning platform with unified interface of instruction collection, parameter-efficient methods, and large language models, 2023. URL . https://github.com/PhoebusSi/alpaca-CoT Feifan Song, Bowen Yu, Minghao Li, Haiyang Yu, Fei Huang, Yongbin Li i Houfeng Wang. 2023. godine arXiv preprint arXiv:2306.17492 Stability AI. StableBeluga2, 2023. URL . https://huggingface.co/stabilityai/ stabilno2 Nisan Stiennon, Long Ouyang, Jeffrey Wu, Daniel Ziegler, Ryan Lowe, Chelsea Voss, Alec Radford, Dario Amodei i Paul F Christiano. , 33:3008–3021, 2020. Advances in Neural Information Processing Systems Jianlin Su. Poboljšanje transformatora: sposobnost ekstrapolacije duljine i robustnost položaja, 2023a. . https://spaces.ac.cn/archives/9444 Jianlin Su. Čarobni učinak termina predrasude: RoPE + predrasude = bolja ekstrapolacija duljine, 2023b. . https://spaces.ac.cn/archives/9577 Jianlin Su, Yu Lu, Shengfeng Pan, Ahmed Murtadha, Bo Wen, and Yunfeng Liu. Roformer: Enhanced transformer with rotary position embedding. , 2021. arXiv preprint arXiv:2104.09864 Tianxiang Sun, Xiaotian Zhang, Zhengfu He, Peng Li, Qinyuan Cheng, Hang Yan, Xiangyang Liu, Yunfan Shao, Qiong Tang, Xingjian Zhao, Ke Chen, Yining Zheng, Zhejian Zhou, Ruixiao Li, Jun Zhan, Yunhua Zhou, Linyang Li, Xiaogui Yang, Lingling Wu, Zhangyue Yin, Xuanjing Huang, and Xipeng Qiu. MOSS: Training conversational language models from synthetic data, 2023a. Zhiqing Sun, Yikang Shen, Qinhong Zhou, Hongxin Zhang, Zhenfang Chen, David Cox, Yiming Yang, and Chuang Gan. Principle-driven self-alignment of language models from scratch with minimal human supervision. , 2023b. arXiv preprint arXiv:2305.03047 Mirac Suzgun, Nathan Scales, Nathanael Scha ̈rli, Sebastian Gehrmann, Yi Tay, Hyung Won Chung, Aakanksha Chowdhery, Quoc V Le, Ed H Chi, Denny Zhou i drugi. , 2022. arXiv preprint arXiv:2210.09261 Marc Szafraniec, Baptiste Rozie`re, Hugh Leather, Patrick Labatut, Franc¸ois Charton, and Gabriel Synnaeve. Code translation with compiler representations. In . OpenReview.net, 2023. URL . The Eleventh International Confer-ence on Learning Representations, ICLR 2023, Kigali, Rwanda, May 1-5, 2023 https://openreview.net/pdf?id=XomEU3eNeSQ Alon Talmor, Jonathan Herzig, Nicholas Lourie, and Jonathan Berant. CommonsenseQA: A question answering challenge targeting commonsense knowledge. In Jill Burstein, Christy Doran, and Thamar Solorio (eds.), , pp. 4149–4158. Association for Computational Linguistics, 2019. doi: 10.18653/v1/n19-1421. URL . Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, NAACL-HLT 2019, Minneapolis, MN, USA, June 2-7, 2019, Volume 1 (Long and Short Papers) https://doi.org/10.18653/v1/n19-1421 Rohan Taori, Ishaan Gulrajani, Tianyi Zhang, Yann Dubois, Xuechen Li, Carlos Guestrin, Percy Liang, and Tatsunori B. Hashimoto. Stanford Alpaca: An instruction-following LLaMA model, 2023. URL . https://github.com/tatsu-lab/stanford_alpaca Ross Taylor, Marcin Kardas, Guillem Cucurull, Thomas Scialom, Anthony Hartshorn, Elvis Saravia, Andrew Poulton, Viktor Kerkez, and Robert Stojnic. Galactica: A large language model for science, 2022. Romal Thoppilan, Daniel De Freitas, Jamie Hall, Noam Shazeer, Apoorv Kulshreshtha, Heng-Tze Cheng, Alicia Jin, Taylor Bos, Leslie Baker, Yu Du, YaGuang Li, Hongrae Lee, Huaixiu Steven Zheng, Amin Ghafouri, Marcelo Menegali, Yanping Huang, Maxim Krikun, Dmitry Lepikhin, James Qin, Dehao Chen, Yuanzhong Xu, Zhifeng Chen, Adam Roberts, Maarten Bosma, Yanqi Zhou, Chung-Ching Chang, Igor Krivokon, Will Rusch, Marc Pickett, Kathleen S. Meier-Hellstern, Meredith Ringel Morris, Tulsee Doshi, Renelito Delos Santos, Toju Duke, Johnny Soraker, Ben Zevenbergen, Vinodkumar Prabhakaran, Mark Diaz, Ben Hutchinson, Kristen Ol-son, Alejandra Molina, Erin Hoffman-John, Josh Lee, Lora Aroyo, Ravi Rajakumar, Alena Butryna, Matthew Lamm, Viktoriya Kuzmina, Joe Fenton, Aaron Cohen, Rachel Bernstein, Ray Kurzweil, Blaise Agu¨era y Arcas, Claire Cui, Marian Croak, Ed H. Chi, and Quoc Le. LaMDA: Language models for dialog applications. , abs/2201.08239, 2022. URL . CoRR https://arxiv.org/abs/2201.08239 Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothe ́e Lacroix, Baptiste Rozie`re, Naman Goyal, Eric Hambro, Faisal Azhar i drugi. 2023 a. arXiv preprint arXiv:2302.13971 Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjjwal Bhargava, Shruti Bhosale, Dan Bikel, Lukas Blecher, Cristian Canton-Ferrer, Moya Chen, Guillem Kerkez, Madian Khabsa, Isabel Kloobu, Jude Fernandes, Jeremy Fu, Wenyin Fu, Brian Fuller, Cynthia Gao, Vedanuj Goswami, Naman Goyal, Anthony Hartshorn, Saghar Hosseini, Rui Hou, Hakan Inan, Marcin Kardas, Viktor Kerkez, Madian Khabsa, Isabel Kloumann, Artemadur Korenev, Punit Singh Koura, Marie-Anneaux Lach, Thibaut Lavril, , abs/2307.09288, 2023b. doi: 10.48550/arXiv.2307.09288. URL . Koroš https://doi.org/ 10.48550/arXiv.2307.09288 Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need. , 30, 2017. Napredak u neuralnim sustavima za obradu informacija Guanzhi Wang, Yuqi Xie, Yunfan Jiang, Ajay Mandlekar, Chaowei Xiao, Yuke Zhu, Linxi Fan i Anima Anandkumar. , 2023a. arXiv preprint arXiv:2305.16291 Xuezhi Wang, Jason Wei, Dale Schuurmans, Quoc Le, Ed Huai hsin Chi, and Denny Zhou. Self-consistency improves chain of thought reasoning in language models. , abs/2203.11171, 2022. ArXiv Yan Wang, Xiaojiang Liu, and Shuming Shi. Deep neural solver for math word problems. In , 2017. URL . semanticscholar.org/CorpusID:910689. Conference on Empirical Methods in Natural Language Processing https://api Yizhong Wang, Hamish Ivison, Pradeep Dasigi, Jack Hessel, Tushar Khot, Khyathi Raghavi Chandu, David Wadden, Kelsey MacMillan, Noah A Smith, Iz Beltagy i drugi. , 2023b. arXiv preprint arXiv:2306.04751 Yizhong Wang, Yeganeh Kordi, Swaroop Mishra, Alisa Liu, Noah A. Smith, Daniel Khashabi i Hannaneh Hajishirzi. Samostalno upućivanje: usklađivanje jezičnih modela sa samostalno generiranim uputama. U Anna Rogers, Jordan L. Boyd-Graber i Naoaki Okazaki (ed.), , str. 13484–13508. udruga za računalne lingvistike, 2023c. doi: 10.18653/v1/2023.acl-long.754. URL . Procedura 61. godišnjeg sastanka Udruge za računalne lingvistike (Volume 1: Long Papers), ACL 2023, Toronto, Kanada, 9. i 14. srpnja 2023. https://doi.org/10.18653/v1/ 2023.acl-long 754 Yue Wang, Weishi Wang, Shafiq Joty, and Steven CH Hoi. CodeT5: Identifier-aware unified pre-trained encoder-decoder models for code understanding and generation. , 2021. arXiv preprint arXiv:2109.00859 Yue Wang, Hung Le, Akhilesh Deepak Gotmare, Nghi D. Q. Bui, Junnan Li, and Steven C. H. Hoi. CodeT5+: Open code large language models for code understanding and generation. , abs/2305.07922, 2023d. doi: 10.48550/arXiv.2305.07922. URL . . Koroš https://doi.org/10 48550/arXiv.2305.07922 Jason Wei, Maarten Bosma, Vincent Y. Zhao, Kelvin Guu, Adams Wei Yu, Brian Lester, Nan Du, Andrew M. Dai i Quoc V. Le. Finetuned jezični modeli su učenici nule. . OpenReview.net, 2022a. URL . The Tenth International Conference on Learning Representations, ICLR 2022, Virtual Event, April 25-29, 2022 https://openreview.net/forum?id= gEZrGCozdqR Jason Wei, Yi Tay, Rishi Bommasani, Colin Raffel, Barret Zoph, Sebastian Borgeaud, Dani Yogatama, Maarten Bosma, Denny Zhou, Donald Metzler, Ed Huai hsin Chi, Tatsunori Hashimoto, Oriol Vinyals, Percy Liang, Jeff Dean, and William Fedus. Emergent abilities of large language models. , 2022, 2022b. URL CorpusID:249674500. Sljedeći Članak Trans. Mach. Learn. Res https://api.semanticscholar.org/ Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, Denny Zhou i drugi. , 35:24824–24837, 2022c. Advances in Neural Information Processing Systems Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Pierric Cistac, Tim Rault, Re´mi Louf, Morgan Funtowicz, et al. HuggingFace’s transformers: State-of-the-art natural language processing. , 2019. arXiv preprint arXiv:1910.03771 Benfeng Xu, An Yang, Junyang Lin, Quan Wang, Chang Zhou, Yongdong Zhang i Zhendong Mao. , 2023a. arXiv preprint arXiv:2305.14688 Can Xu, Qingfeng Sun, Kai Zheng, Xiubo Geng, Pu Zhao, Jiazhan Feng, Chongyang Tao i Daxin Jiang. , 2023b. arXiv preprint arXiv:2304.12244 Canwen Xu, Daya Guo, Nan Duan, and Julian McAuley. Baize: An open-source chat model with parameter-efficient tuning on self-chat data. , 2023c. arXiv preprint arXiv:2304.01196 Yuzhuang Xu, Shuo Wang, Peng Li, Fuwen Luo, Xiaolong Wang, Weidong Liu, and Yang Liu. Exploring large language models for communication games: An empirical study on werewolf. 2023 d. arXiv preprint arXiv:2309.04658 Aiyuan Yang, Bin Xiao, Bingning Wang, Borong Zhang, Chao Yin, Chenxu Lv, Da Pan, Dian Wang, Dong Yan, Fan Yang, Fei Deng, Feng Wang, Feng Liu, Guangwei Ai, Guosheng Dong, Haizhou Zhao, Hang Xu, Haoze Sun, Hongda Zhang, Hui Liu, Jiaming Ji, Jian Xie, Juntao Dai, Kun Fang, Lei Su, Liang Song, Lifeng Liu, Liyun Ru, Luyao Ma, Mang Wang, Mickel Liu, MingAn Lin, Nuolan Nie, Peidong Guo, Ruiyang Sun, Tao Zhang, Tianpeng Li, Tianyu Li, Wei Cheng, Weipeng Chen, Xiangrong Zeng, Xiaochuan Wang, Xiaoxi Chen, Xin Men, Xin Yu, Xuehai Pan, Yanjun Shen, Yiding Wang, Yiyu Li, Youxin Jiang, Yuchen Gao, Yupeng Zhang, Zenan Zhou, and Zhiying Wu. Baichuan 2: Open large-scale language models. Technical report, Baichuan Inc., 2023. URL . u . https://cdn.baichuan-ai.com/paper/Baichuan2-technical-report pdf Jianxin Yang. Firefly. , 2023. https://github.com/yangjianxin1/Firefly Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, and Yuan Cao. ReAct: Synergizing reasoning and acting in language models. , 2022. arXiv preprint arXiv:2210.03629 Qinghao Ye, Haiyang Xu, Guohai Xu, Jiabo Ye, Ming Yan, Yiyang Zhou, Junyang Wang, Anwen Hu, Pengcheng Shi, Yaya Shi, et al. mPLUG-Owl: Modularizacija osnažuje velike jezične modele s multimodalnošću. , 2023. arXiv preprint arXiv:2304.14178 Longhui Yu, Weisen Jiang, Han Shi, Jincheng Yu, Zhengying Liu, Yu Zhang, James T. Kwok, Zhenguo Li, Adrian Weller i Weiyang Liu. Zheng Yuan, Hongyi Yuan, Chengpeng Li, Guanting Dong, Keming Lu, Chuanqi Tan, Chang Zhou i Jingren Zhou. Zheng Yuan, Hongyi Yuan, Chuanqi Tan, Wei Wang, and Songfang Huang. How well do large language models perform in arithmetic tasks? , 2023b. arXiv preprint arXiv:2304.02015 Zheng Yuan, Hongyi Yuan, Chuanqi Tan, Wei Wang, Songfang Huang, and Fei Huang. RRHF: Rank responses to align language models with human feedback without tears, 2023c. Xiang Yue, Xingwei Qu, Ge Zhang, Yao Fu, Wenhao Huang, Huan Sun, Yu Su, and Wenhu Chen. MAmmoTH: Building math generalist models through hybrid instruction tuning. , 2023. arXiv preprint arXiv:2309.05653 Rowan Zellers, Ari Holtzman, Yonatan Bisk, Ali Farhadi i Yejin Choi. HellaSwag: Može li mašina zaista završiti vašu rečenicu? , pp. 4791–4800. Association for Computational Linguistics, 2019. doi: 10.18653/v1/p19-1472. URL . Proceedings of the 57th Conference of the Association for Computational Linguistics, ACL 2019, Florence, Italy, July 28- August 2, 2019, Volume 1: Long Papers https://doi.org/10.18653/v1/p19-1472 Aohan Zeng, Xiao Liu, Zhengxiao Du, Zihan Wang, Hanyu Lai, Ming Ding, Zhuoyi Yang, Yifan Xu, Wendi Zheng, Xiao Xia, et al. GLM-130B: An open bilingual pre-trained model. , 2022. arXiv preprint arXiv:2210.02414 Fengji Zhang, Bei Chen, Yue Zhang, Jin Liu, Daoguang Zan, Yi Mao, Jian-Guang Lou i Weizhu Chen. , abs/2303.12570, 2023a. doi: 10.48550/arXiv.2303.12570. URL . CoRR https://doi.org/ 10.48550/arXiv.2303.12570 Susan Zhang, Stephen Roller, Naman Goyal, Mikel Artetxe, Moya Chen, Shuohui Chen, Christopher Dewan, Mona Diab, Xian Li, Xi Victoria Lin, et al. OPT: Open pre-trained transformer language models. , 2022. arXiv preprint arXiv:2205.01068 Xiaotian Zhang, Chunyang Li, Yi Zong, Zhengyu Ying, Liang He i Xipeng Qiu. , abs/2305.12474, 2023b. doi: 10.48550/arXiv.2305.12474. URL . u . CoRR https://doi.org/10.48550/arXiv 2305.12474 Qinkai Zheng, Xiao Xia, Xu Zou, Yuxiao Dong, Shan Wang, Yufei Xue, Zihan Wang, Lei Shen, Andi Wang, Yang Li, Teng Su, Zhilin Yang i Jie Tang. , abs/2303.17568, 2023. doi: 10.48550/arXiv.2303.17568. URL . CoRR https://doi.org/10.48550/arXiv.2303.17568 Wanjun Zhong, Ruixiang Cui, Yiduo Guo, Yaobo Liang, Shuai Lu, Yanlin Wang, Amin Saied, Weizhu Chen i Nan Duan. , abs/2304.06364, 2023a. doi: 10.48550/arXiv.2304.06364. URL . CoRR https://doi.org/ 10.48550/arXiv.2304.06364 Wanjun Zhong, Lianghong Guo, Qiqi Gao i Yanlin Wang. MemoryBank: Poboljšanje velikih jezičnih modela s dugoročnim pamćenjem. 2023. godine b. arXiv preprint arXiv:2305.10250 Denny Zhou, Nathanael Scharli, Le Hou, Jason Wei, Nathan Scales, Xuezhi Wang, Dale Schuurmans, Olivier Bousquet, Quoc Le, and Ed Huai hsin Chi. Least-to-most prompting enables complex reasoning in large language models. , abs/2205.10625, 2022. Arhiva Prilog A.1 More Training Details A.1.1 Format podataka za Qwen-Chat Different from conventional pretraining based on autoregressive next-token prediction, despite using a similar training task, there should be a specially design data format for SFT and RLHF to build a conversational AI assistant model. Common formats include “human-assistant” and ChatML formats. As to our knowledge, one of the earliest examples of the human-assistant format comes from Anthropic which adds a special phrase “\n\nhuman: ” in front of the user input and “\n\nassistant: ” in front of the assistant response. It is easy for the base language model to transfer to the pattern of conversational AI. However, as the specific phrases are common words, it might be hard for the model to disambiguate from these words in other contexts. (Baj i al. 2022. godine b) Instead, we turned to the ChatML format proposed by OpenAI. Ovaj format omogućuje korištenje posebnih žetona, tj. ” and “ ”, koji se ne pojavljuju u pretreningu, a time i rješavaju gore spomenuti problem. 5 A.2 Evaluation A.2.1 Automatsko ocjenjivanje To provide a whole picture of the performance of our model series QWEN, here in this section we illustrate the detailed performance of our models as well as the baselines in the comprehensive benchmark evaluation proposed by We report the results in multiple tables based on the officially provided categories, including examination, language, knowledge, understanding, and reasoning. In terms of the performance of the baseline models, we report the higher results between the reported ones and those on the leaderboard. OpenCompass Team 2023 ) Ovdje ocjenjujemo modele na nizu skupova podataka relevantnih za ispitivanje. Examination MMLU (Hendrycks et al., 2020) Massive Multi-task Language Understanding je dizajniran za mjerenje sposobnosti razumijevanja jezika. C-Eval (Huang et al., 2023) C-Eval je kineski skup podataka o evaluaciji koji obuhvaća 52 različite discipline. CMMLU (Li et al., 2023c) CMMLU je dizajniran za procjenu sposobnosti razumijevanja jezika u kineskom jeziku. AGIEval (Zhong et al., 2023a) Ovo je referentna vrijednost koja se sastoji od ispita usmjerenih na čovjeka, uključujući testove za ulazak u sveučilište, testove za upis u pravnu školu, matematičke natjecanja i testove kvalifikacija odvjetnika. This is a benchmark with Gaokao (Chinese college-entrance examination) questions. We report zero-shot results. Gaokao-Bench (Zhang et al., 2023b) ARC (Clark et al., 2018) ARC je skup podataka koji se sastoji od razredne razine, znanstvenih pitanja s višestrukim izborom. U pogledu MMLU-a, izvješćujemo o detaljnim rezultatima u Tablici In terms of C-Eval, we report the results in Table Za ostale skupove podataka, izvješćujemo rezultate u tablici Note that AGIEval includes the parts of Chinese and English, while LLAMA 2 only reported the results in the English part, so we use the results on OpenCompass. 13. 14. 15. Osim toga, dok su CMMLU, AGIEval i Gaokao-Bench povezani s kineskim, a serije MPT, Falcon i LLaMA nisu optimizirane za kineske, ovi modeli postigli su niske performanse na skupovima podataka. Ovdje ocjenjujemo modele na nizu skupova podataka relevantnih za znanje i razumijevanje prirodnog jezika. Knowledge and Understanding This is a QA dataset, where the questions are about passages of Wikipedia, and the model should answer yes or no to the given possible answer. We report zero-shot results. BoolQ (Clark et al., 2019) This is a dataset of multiple-choice question answering that asseses the understanding of commonsense knowledge. We report 8-shot results. CommonsenseQA (Talmor et al., 2019) NaturalQuestions (Kwiatkowski et al., 2019) To je skup podataka QA gdje su pitanja od korisnika, a odgovori su provjereni od strane stručnjaka. LAMBADA (Paperno et al., 2016) Ovo je skup podataka za procjenu razumijevanja jezika prema predviđanju riječi. Sastoji se od odlomaka povezanih s ljudskim subjektima. Rezultate objavljujemo u tablici 16. Rezultate evaluacije prijavljujemo na podatkovnim skupovima koji se odnose na obrazloženje, usredotočujući se na obrazloženje prirodnog jezika. Za ostale, kao što su matematika i kodiranje, kao što smo ilustrirali detaljne rezultate, ovdje ne prijavljujemo te rezultate više puta. Reasoning This is a commonsense natural language inference (NLI) dataset, where the questions are easy for humans but struggling for previous language models. We report zero-shot results. HellaSwag (Zellers et al., 2019) PIQA (Bisk et al., 2020) Ovo je NLI skup podataka koji ocjenjuje fizičko znanje. This is an NLI dataset evaluating social commonsense intelligence. We report zero-shot results. SIQA (Sap et al., 2019) OCNLI (Hu et al., 2020) Ovo je NLI skup podataka koji se fokusira na kineski. Rezultate objavljujemo u tablici 17. A.2.2 Procjena čovjeka U našem samokonstruiranom skupu podataka o evaluaciji, upute su ili ručno napisani podaci ili ručno revidirani iz javnih skupova podataka, kao što je CLiB C-Eval činjenica Lečkod ), etc. 6 (Hrvatska i al., 2023. godine) (Hrvatska i al., 2023 b) i 7 U smislu svakog slučaja, demonstriramo odgovore i Elo ocjene of all models for comparison. Specifically, as the data in our human evaluation are in Chinese, we also provide their translations in English. 8 A.3 Analiza tumača koda Ovdje pružamo primjer usporedbe između CODE LLAMA i QWEN-CHAT. Ovaj primjer pokazuje prednosti QWEN-CHAT u obradi tabularnih podataka i obavljanju složenih zadataka. Ovaj dokument je dostupan na arhivu pod licencom CC by 4.0 Deed (Attribution 4.0 International). Ovaj papir je Licencija je CC by 4.0 Deed (Attribution 4.0 International). Dostupno u Arhivu