Introducing improvements to the fine-tuning API and expanding our custom models program

Aġġornament fit-8 ta’ Mejju 2026: OpenAI qed twaqqaf gradwalment il-pjattaforma tal-irfinar. Il-pjattaforma m’għadhiex aċċessibbli għal utenti ġodda, iżda l-utenti eżistenti tal-pjattaforma tal-irfinar se jkunu jistgħu joħolqu xogħlijiet ta’ taħriġ fix-xhur li ġejjin. Il-mudell irfinat kollu se jibqa’ disponibbli għall-inferenza sakemm il-mudell bażi tagħhom jiġi rtirat⁠(jinfetaħ f’tieqa ġdida). L-iskeda taż-żmien sħiħa tinsab hawn⁠(jinfetaħ f’tieqa ġdida).

Hemm varjetà ta’ tekniki⁠(jinfetaħ f’tieqa ġdida) li l-iżviluppaturi jistgħu jużaw biex itejbu l-prestazzjoni tal-mudell, inaqqsu l-latenza, itejbu l-eżattezza u jnaqqsu l-ispejjeż. Kemm jekk qed testendi l-għarfien tal-mudell b’ġenerazzjoni msaħħa bl-irkupru (RAG), tippersonalizza l-imġiba tal-mudell permezz tal-irfinar jew tibni mudell imħarreġ apposta b’għarfien ġdid speċifiku għad-dominju, żviluppajna firxa ta’ għażliet biex nappoġġjaw l-implimentazzjonijiet tal-IA tal-klijenti tagħna. Illum, qed inniedu karatteristiċi ġodda biex nagħtu lill-iżviluppaturi aktar kontroll fuq ir-rfinar permezz tal-API, u qed nintroduċu aktar modi kif jaħdmu mat-tim tagħna ta’ esperti tal-IA u riċerkaturi biex jibnu mudelli personalizzati.

We launched the self-serve fine-tuning API⁠(jinfetaħ f’tieqa ġdida) for GPT‑3.5 in August 2023. Since then, thousands of organizations have trained hundreds of thousands of models using our API. Fine-tuning can help models deeply understand content and augment a model’s existing knowledge and capabilities for a specific task. Our fine-tuning API also supports a larger volume of examples than can fit in a single prompt to achieve higher quality results while reducing cost and latency. Some of the common use cases of fine-tuning include training a model to generate better code in a particular programming language, to summarize text in a specific format, or to craft personalized content based on user behavior.

For example, Indeed⁠(jinfetaħ f’tieqa ġdida), a global job matching and hiring platform, wants to simplify the hiring process. As part of this, Indeed launched a feature that sends personalized recommendations to job seekers, highlighting relevant jobs based on their skills, experience, and preferences. They fine-tuned GPT‑3.5 Turbo to generate higher quality and more accurate explanations. As a result, Indeed was able to improve cost and latency by reducing the number of tokens in prompt by 80%. This let them scale from less than one million messages to job seekers per month to roughly 20 million.

Today, we’re introducing new features⁠(jinfetaħ f’tieqa ġdida) to give developers even more control over their fine-tuning jobs, including:

• Epoch-based Checkpoint Creation: Automatically produce one full fine-tuned model checkpoint during each training epoch, which reduces the need for subsequent retraining, especially in the cases of overfitting
• Comparative Playground: A new side-by-side Playground UI for comparing model quality and performance, allowing human evaluation of the outputs of multiple models or fine-tune snapshots against a single prompt
• Third-party Integration: Support for integrations with third-party platforms (starting with Weights and Biases⁠(jinfetaħ f’tieqa ġdida) this week) to let developers share detailed fine-tuning data to the rest of their stack
• Comprehensive Validation Metrics: The ability to compute metrics like loss and accuracy over the entire validation dataset instead of a sampled batch, providing better insight on model quality
• Hyperparameter Configuration: The ability to configure available hyperparameters from the Dashboard⁠(jinfetaħ f’tieqa ġdida) (rather than only through the API or SDK)
• Fine-Tuning Dashboard Improvements: Including the ability to configure hyperparameters, view more detailed training metrics, and rerun jobs from previous configurations

F’DevDay f’Novembru li għadda, ħabbarna⁠ programm ta’ Mudell Personalizzat iddisinjat biex iħarreġ u jottimizza mudelli għal dominju speċifiku, fi sħubija ma’ grupp dedikat ta’ riċerkaturi ta’ OpenAI. Minn dakinhar, iltqajna ma’ għexieren ta’ klijenti biex nivvalutaw il-ħtiġijiet tagħhom għal mudelli personalizzati u evolvejna l-programm tagħna biex inkomplu nimmassimizzaw il-prestazzjoni.

Illum, qed inħabbru formalment l-offerta tagħna ta’ fine-tuning assistit bħala parti mill-programm ta’ Mudell Personalizzat. Il-fine-tuning assistit huwa sforz kollaborattiv mat-timijiet tekniċi tagħna biex nużaw tekniki lil hinn mill-API tal-fine-tuning, bħal hyperparameters addizzjonali u diversi metodi parameter efficient fine-tuning (PEFT) fuq skala akbar. Dan huwa partikolarment ta’ għajnuna għal organizzazzjonijiet li jeħtieġu appoġġ biex iwaqqfu pipelines effiċjenti tad-data għat-taħriġ, sistemi ta’ evalwazzjoni, u parametri u metodi mfassla apposta biex jimmassimizzaw il-prestazzjoni tal-mudell għall-każ ta’ użu jew kompitu tagħhom.

Pereżempju, SK Telecom⁠(jinfetaħ f’tieqa ġdida), operatur tat-telekomunikazzjoni li jservi aktar minn 30 miljun abbonat fil-Korea t’Isfel, ried tippersonalizza mudell biex ikun espert fid-dominju tat-telekomunikazzjoni b’fokus inizjali fuq is-servizz tal-klijent. Huma ħadmu ma’ OpenAI biex jagħmlu fine-tuning ta’ GPT‑4 sabiex itejbu l-prestazzjoni tiegħu f’konversazzjonijiet relatati mat-telekomunikazzjoni bil-lingwa Koreana. Matul diversi ġimgħat, SKT u OpenAI kisbu titjib sinifikanti fil-prestazzjoni f’kompiti tas-servizz tal-klijent fit-telekomunikazzjoni—żieda ta’ 35% fil-kwalità tas-sommarji tal-konversazzjoni, żieda ta’ 33% fl-eżattezza tar-rikonoxximent tal-intenzjoni, u żieda fil-punteġġi tas-sodisfazzjon minn 3.6 għal 4.5 (minn 5) meta tqabbel il-mudell fine-tuned ma’ GPT‑4.

In some cases, organizations need to train a purpose-built model from scratch that understands their business, industry, or domain. Fully custom-trained models imbue new knowledge from a specific domain by modifying key steps of the model training process using novel mid-training and post-training techniques. Organizations that see success with a fully custom-trained model often have large quantities of proprietary data—millions of examples or billions of tokens—that they want to use to teach the model new knowledge or complex, unique behaviors for highly specific use cases.

For example, Harvey⁠(jinfetaħ f’tieqa ġdida), an AI-native legal tool for attorneys, partnered with OpenAI to create a custom-trained large language model for case law⁠. While foundation models were strong at reasoning, they lacked the extensive knowledge of legal case history and other knowledge required for legal work. After testing out prompt engineering, RAG, and fine-tuning, Harvey worked with our team to add the depth of context needed to the model—the equivalent of 10 billion tokens worth of data. Our team modified every step of the model training process, from domain-specific mid-training to customizing post-training processes and incorporating expert attorney feedback. The resulting model achieved an 83% increase in factual responses and attorneys preferred the customized model’s outputs 97% of the time over GPT‑4.

Aħna nemmnu li fil-futur, il-maġġoranza l-kbira tal-organizzazzjonijiet se jiżviluppaw mudelli personalizzati li jkunu mfassla għall-industrija, in-negozju, jew il-każ ta’ użu tagħhom. B’varjetà ta’ tekniki disponibbli biex jinbena mudell personalizzat, organizzazzjonijiet ta’ kull daqs jistgħu jiżviluppaw mudelli personalizzati biex jiksbu impatt aktar sinifikanti u speċifiku mill-implimentazzjonijiet tagħhom tal-AI. Iċ-ċavetta hi li tiddefinixxi b’mod ċar il-każ ta’ użu, tfassal u timplimenta sistemi ta’ evalwazzjoni, tagħżel it-tekniki t-tajba, u tkun lest li ttenni maż-żmien biex il-mudell jilħaq prestazzjoni ottimali.

Ma’ OpenAI, ħafna organizzazzjonijiet jistgħu jaraw riżultati sinifikanti malajr bl-API tal-fine-tuning self-serve. Għal kwalunkwe organizzazzjoni li teħtieġ tagħmel fine-tuning aktar profond tal-mudelli tagħha jew idaħħal għarfien ġdid u speċifiku għad-dominju fil-mudell, il-programmi tagħna ta’ Mudell Personalizzat jistgħu jgħinu.

Żur id-dokumentazzjoni tagħna tal-API tal-fine-tuning⁠(jinfetaħ f’tieqa ġdida) biex tibda tagħmel fine-tuning tal-mudelli tagħna.