Idanwo ti o sopọ mọ itọsọna · Alabọde Ipele
Ètò-ati-yanju Idanwo Tọ
Checks the plan-then-execute prompt structure, the errors it targets, PS+ guidance and the limits of its benchmark evidence.
Ibeere 1 ti 8
Iyipada koko wo ni Eto-ati-yanju iwuri ṣe si iwọn-iwọn-igbiyanju odo-shot-ti-ero?
Awotẹlẹ gbogbo ibeere8 ibeere
Awọn idahun ati awọn alaye han bi o ṣe n mu ibeere naa.
- Iyipada koko wo ni Eto-ati-yanju iwuri ṣe si iwọn-iwọn-igbiyanju odo-shot-ti-ero?
- Ni akọkọ o beere awoṣe lati ṣe agbekalẹ ero kan, lẹhinna ṣiṣẹ ni igbese nipasẹ igbese
- O dara-tunes awoṣe lori data igbero
- O yọkuro ero-igbesẹ-igbesẹ patapata
- O nilo awọn dosinni ti awọn apẹẹrẹ iṣẹ ni kiakia
- Awọn ailagbara kan pato ti pq-ti-ero pẹlẹbẹ ti a ṣe Eto-ati-yanju lati koju?
- Iyara itọkasi lọra ati lilo iranti giga
- Awọn igbesẹ ero ti o padanu ati awọn aṣiṣe iṣiro
- Awọn abajade majele tabi abosi
- Ailagbara lati gbe awọn ero eyikeyi ni gbogbo
- Kini iyatọ PS + imudara ṣe afikun si Eto ipilẹ-ati-yanju tọ?
- Awoṣe ede keji lati ṣe ipele idahun
- Wiwọle si API ẹrọ iṣiro kan
- Awọn ilana lati jade awọn oniyipada, ṣe iṣiro awọn abajade agbedemeji, ati lọ si awọn nọmba
- Ibeere lati dahun ni JSON
- Anfani bọtini kan ti Eto-ati-yanju lori ẹwọn-ronu diẹ-shot ni pe:
- Nigbagbogbo ṣe aṣeyọri pipe pipe
- Nilo atunṣe awoṣe
- Nikan ṣiṣẹ lori awọn iṣẹ-ṣiṣe iran
- Ko nilo awọn apẹẹrẹ iṣẹ ti a fi ọwọ ṣe ni itọsẹ naa
- Lori iru ala wo ni Eto-ati-yanju ṣe ayẹwo ni pataki?
- Iṣiro ati Iṣiro ọrọ-iṣoro awọn ipilẹ bi GSM8K
- Aworan classification datasets
- Kopọ idanimọ ọrọ
- Amuaradagba-kika awọn iṣẹ-ṣiṣe
- Kini idi ti ipilẹṣẹ eto ṣaaju awọn igbesẹ alaye ṣọ lati ṣe iranlọwọ awoṣe adaṣe adaṣe?
- O npa ibeere atilẹba rẹ kuro ni ọrọ-ọrọ
- Awọn ibi-afẹde ti ero naa ni ipo awọn ami ironu nigbamii ti awoṣe n gbejade
- O sekeji awọn awoṣe ká paramita kika
- O yipada awoṣe sinu ipo ti kii ṣe ọrọ
- A prompt asks a model to list subgoals before solving a word problem. Which Plan-and-Solve stage does that implement?
- A few-shot worked example
- The final answer verification stage
- The planning stage that decomposes the task before execution
- A decoding-time beam search
- A PS+ prompt reminds the model to extract quantities and calculate intermediate results. What is the intent of those reminders?
- Replace the problem statement with demonstrations
- Change the model’s parameters during inference
- Guarantee every answer is correct
- Reduce calculation and step omissions by directing attention to variables and intermediate quantities