Positive:
- quite fast (good for consumer hardware)
- good for simple text outputs
- can simulate complex personas
- very realistic roleplay
Negative:
- gets confused particularly in roleplay, who is who, who does what
- gets confused with instructions sometimes
- very bad with numbers and math (even numbers citation)
Positive:
- very strong model
- good comprehension
- can run on 1 professional GPU
- very natural output
- understands complex prompts
Negative:
- text generation is not particularly fast, even on good hardware
Positive:
Negative:
- Needs a long time to think, without apparent benefit
- Text output not particularly fast
- Trouble with complex prompts
- Non-realistic roleplay
- Extremely strong safeguards
- Extremely Biased
Positive:
Negative:
Positive:
- Good with simple use cases
- Fast
- Lightweigh
Negative:
- Very bad in following prompts
- Bad with text
- Easily falls into infinite loops
- trouble understanding images
Positive:
- Good comprehension
- Can scan documents with semantic OCR mostly accurate
Negative:
- not particularly good with prompt following. Simple tasks work
- It is just too large for one GPU with 65 GB.
- Slow generation (2 Minutes for a Document Page on 2x Nvidia A40-48)