Ikibazo gihuza ibibazo · Hagati Urwego
Microsoft Ikibazo cya Phi
Ongera usubiremo umuryango wa Phi wicyitegererezo gito-cyitegererezo, amakuru yatunganijwe kandi yubukorikori, ingano yihariye, hamwe nimbibi zoherejwe.
Ikibazo 1 Bya 8
Ni ubuhe butumwa bwibanze inyuma ya moderi ya Phi, yafashwe mu mpapuro zizwi?
Reba buri kibazo8 ibibazo
Ibisubizo nibisobanuro bigaragara mugihe ufata ikibazo.
- Ni ubuhe butumwa bwibanze inyuma ya moderi ya Phi, yafashwe mu mpapuro zizwi?
- Ibitabo nibyo ukeneye byose
- GPU nyinshi nizo zose ukeneye
- Windows nini ya contexte nibyo ukeneye byose
- Kwiga gushimangira nibyo ukeneye byose
- Hafi ya Phi-3-mini ifite ibipimo bingahe?
- Hafi miliyari 175
- Hafi miliyari 3.8
- Hafi miliyari 70
- Hafi ya miliyoni 500
- What synthetic-data source did the Phi-1 paper report using for textbooks and exercises?
- A fixed database of court transcripts
- Only human-authored textbooks with no synthetic examples
- GPT-3.5-generated textbooks and exercises
- The SoundStream audio codec
- Ni ikihe cyiciro cy'icyitegererezo 'SLM' bivuga, nkuko bikoreshwa kuri Phi?
- Icyitegererezo Cyururimi
- Kugenzura Module yo Kwiga
- Inzira yo Kwihuta
- Ururimi Ruto
- What may a compact Phi checkpoint enable when its hardware and runtime support it?
- Local deployment can be feasible, with efficiency depending on the checkpoint and environment
- Guaranteed privacy and zero latency on every phone
- The same speed and quality on every device
- Automatic offline use without memory or runtime requirements
- Ni ubuhe buhanga buhanitse bwububiko bugaragara mubintu bimwe na bimwe bya Phi-3.5 kugirango byongere imikorere?
- Guhuriza hamwe
- Uruvange-rwinzobere (MoE)
- Algorithm ya genetike
- Moteri yandikishijwe intoki
- Which context-length variants does the Phi-3-mini model card describe?
- Only 512 tokens
- Exactly 1 million tokens in every Phi model
- 4K and 128K context variants
- No context limit
- What additional input does Phi-3.5-Vision support alongside text?
- Only raw audio waveforms
- Only SQL queries with no text
- Three-dimensional motion capture exclusively
- Image inputs, including single-image and multi-image prompts