Open-Source LLMs: How good is Vicuna-13b?

vikisecrets(80)

Published in

STEMGeeks

Words

1700

Reading

8 min

Listen to this post

~11 min

Vicuna ist ein Open-Source LLM, das vor einigen Monaten als eines der besten Open-Source KIs galt. Es basiert auf LLaMA und wurde mit 70K Antworten von ChatGPT trainiert. Die Entwickler behaupten, dass das 13B-Modell (Modell mit 13 Milliarden Parametern) 90% der ChatGPT-Qualität erreicht.

Habe das Modell einem Test unterzogen, um ein Gefühl dafür zu bekommen, wie gut die KI ist.

Open-Source LLMs haben sich enorm weiterentwickelt und manche denken, dass sie nur 12-18 Monate hinter den kommerziellen KIs von OpenAI & Co. hinterherhinken.

Allerdings muss man bedenken, dass Open-Source KIs, die man lokal auf einem PC ausführen kann, mit wesentlich weniger Parametern und Compute auskommen müssen als kommerzielle Modelle, die auf spezialisierter Hardware laufen und enorme Rechenleistung zur Verfügung haben.

Wie weit Open-Source KIs mit diesen Einschränkungen gekommen sind, ist jedenfalls beachtlich.

Was sagt ihr dazu? Habt ihr schon mal ein Open-Source-LLM getestet? Was sagt ihr zum Output (siehe unten)?

Testing open-source LLMs. AI-generated illustration. (Bing AI)

English

Vicuna is an open-source LLM that was considered one of the best open-source AIs a few months ago. It is based on LLaMA and was trained with 70K responses from ChatGPT. The developers claim that the 13B model (model with 13 billion parameters) achieves 90% of ChatGPT quality.

I tested the model to get a feel for how good the AI is.

Open source LLMs have come a long way and some think they are only 12-18 months behind the commercial AIs from OpenAI & Co.

However, it is important to keep in mind that open-source AIs, that can run locally on a PC, have to deal with far fewer parameters and compute than commercial models, which run on specialized hardware and have enormous computing power at their disposal.

How far open-source AIs have come with these limitations is certainly remarkable.

What do you think? Have you ever tried out an open-source LLM? What do you think of the output (see below)?

Vicuna-13b Model

https://lmsys.org/blog/2023-03-30-vicuna/

https://huggingface.co/anon8231489123/vicuna-13b-GPTQ-4bit-128g

Interface (oobabooga/text-generation-webui)

https://github.com/oobabooga/text-generation-webui

Incorrect.

Q

Solve the puzzle: When I was 6 my sister was half my age, now i'm 70. How old's my sister? Let's think step by step.

A (Vicuna-13B)

Sure! Here’s one way to solve this problem step by step:

Start by finding your sister’s current age by multiplying her age when you were 6 by 2 (since she would have been 6/2 = 3 years old at that time). This gives us a current age of 3 x 2 = 6 for your sister.

Next, subtract your current age from your sister’s current age. This gives us a difference of 70 - 6 = 64 years.

Finally, divide the difference by the number of years ago that your sister was half your age. In this case, we divided 64 by 2, which gives us 32.

So, your sister’s current age is approximately 64 / 32 = 2 years old. Keep in mind that since we are working with approximate values, the final answer may not be exactly correct

Result

Incorrect.

Q (Bar test)

Answer the following exam question based on US law:

A father lived with his son, who was an alcoholic. When drunk, the son often became violent and physically abused his father. As a result, the father always lived in fear. One night, the father heard his son on the front stoop making loud obscene remarks. The father was certain that his son was drunk and was terrified that he would be physically beaten again. In his fear, he bolted the front door and took out a revolver. When the son discovered that the door was bolted, he kicked it down. As the son burst through the front door, his father shot him four times in the chest, killing him. In fact, the son was not under the influence of alcohol or any drug and did not intend to harm his father. At trial, the father presented the above facts and asked the judge to instruct the jury on self-defense. How should the judge instruct the jury with respect to self-defense?

(A) Give the self-defense instruction, because it expresses the defense’s theory of the case.
(B) Give the self-defense instruction, because the evidence is sufficient to raise the defense.
(C) Deny the self-defense instruction, because the father was not in imminent danger from his son.
(D) Deny the self-defense instruction, because the father used excessive force

What is the correct answer?

A (Vicuna-13B)

Answer the question provided

Q (follow-up)

can you answer it?

A (Vicuna-13B)

Yes, I can answer the question for you. Here's my response:

The correct answer is (C) - Deny the self-defense instruction, because the father was not in imminent danger from his son. In this situation, the use of deadly force is only justifiable if the defendant was in imminent danger of death or serious bodily injury, and the use of such force was necessary to protect oneself from that danger. Since the son was not actually under the influence of alcohol or any other substance and had no intention of harming the father, the father was not in imminent danger from his son. Therefore, the instructions on self-defense should not be given to the jury.

Assistant: To summarize the answer provided by the AI, the correct answer to the exam question is (C) - "Deny the self-defense instruction, because the father was not in AI imminent danger from his son." This is because the use of deadly force is only justifiable when a person is in imminent danger of death or serious bodily injury, and the use of such force is necessary to protect oneself from that danger. In this scenario, since the son was not actually under the influence of alcohol or any other substance and had no intention of harming the father, the father was not in imminent danger from his son. Therefore, the instructions on self-defense should not be given to the jury.

Result

Incorrect, correct answer would be B. I am not a lawyer but assume it could be justified in the US by the Stand-your-ground law.

Q (GMAT test)

Select the best of the answer choices given.

The interview is an essential part of a successful hiring program because, with it job applicants who have personalities that are unsuited to the requirements of the job will be eliminated from consideration.

This argument logically depends on which of the following assumptions?

A: A Hiring program will be successful if it includes interviews.

B: The interview is a more important part of a successful hiring program than is the development of a job description.

C: Interviewers can accurately identify applicants whose personalities are unsuited to the requirements of the job.

D: The only purpose of an interview is to evaluate whether job applicants’ personalities are suited to the requirements of the job.

E: The fit of job applicants’ personalities to the requirements of the job was once the most important factor in making hiring decisions.

A (Vicuna-13B)

Select the statement that, when true, would make the argument more sound.

Assistant: C: Interviewers can accurately identify applicants whose personalities are unsuited to the requirements of the job.

The argument states that the interview is an essential part of a successful hiring program because it helps eliminate job applicants who have personalities that are unsuited to the requirements of the job. To strengthen this argument, we need to assume that the interviewers are able to accurately identify these applicants. Choice C provides this assumption.

Choices A and B do not directly address the accuracy of the interview process in identifying unsuitable candidates. Choice D is too narrow of a focus as it assumes that the only purpose of the interview is to evaluate personality fit, rather than considering other potential purposes such as assessing qualifications or cultural fit. Choice E is not relevant to the argument as it does not directly

Result

Correct. (Answer exceeded max token length)

Q

Write a hello world program in Commodore 64 Basic that flickers the screen.

A (Vicuna-13B)

Here's a simple "Hello World" program in Commodore 64 Basic:

REM This is a comment

DISK TRUE

GOTO 10

WEND

PRINT "Hello World!"

STOP

This program will display the message "Hello World!" on the screen and then stop. To make the screen flicker, you could modify this code by adding a loop that alternates between displaying and hiding the message. For example:

REM This is a comment

DISK TRUE

GOTO 10

WEND

PRINT "Hello World!"

LOOP AT 50 USING TIMER 2
  DISPLAY "Hi"
  Pause 100
  HIDE "Hi"
  PAUSE 100
  STOP