Just completed a chinilla optimal run the final checkpoint is final(1).pt enjoy; version 2 is coming i'll eitheir make a new repo or upload the double the intellgence version this on i'll call it AGILLM-Mark2
yep just completed first chinilla optimal run where u train on model on 20 units of data per paramter-neuron; results are pretty good and under final(1).pt took about a month to complete; next run i just started is double the parameters now and adjust it to the higher end of the law which is 1 to 25. That version will rather than starting randomly will build on top of weights of final(1).pt; will take about 2 months to complete
RESULT are really good there this keypoint in training a AGI LLM where it can do convoverstion turns; this was orignally critical point in early LLM when openai notice that GPT-3 versions could do multi turn convo this was the point where they realise that they could make instruct models doing sort a version of roleplay kind of convo it models can do convo turns like below ie base models llm they can be made into Instruct models; if they base model cannot do that yet you eitheir need to train it more or expand it parameters. This is ulimately the basis of ChatGPT3.5 which resulted in mass expolsion of AGI LLM in public awareness rather than being knowledge of a few people like GPT-1 and GPT-2 was; GPT-2 was when i first became aware of LLMs
(main) root@C.20562518:/workspace$ python3 a21d.py infer --mode ar --ckpt ckpts_joint/final.pt --preset small --rank 32 --prompt $'Scott: How are you today?\nMary: I am good, what about you?\nScott: I am fine, just been busy with work lately.\nMary: That sounds tiring. Have you had time to rest?\nScott: Not really, there’s been so much to do.\nMary: You need to take a break now and then.\nScott:' --max_new 400 --temperature 0.8
Scott: How are you today?
Mary: I am good, what about you?
Scott: I am fine, just been busy with work lately.
Mary: That sounds tiring. Have you had time to rest?
Scott: Not really, there’s been so much to do.
Mary: You need to take a break now and then.
Scott: My name is a picture of the kind that she is. I know when I was helping with that. I called a friend as she was always goall she was still trying to get status as a woman. Her name was Emma was definitely a brilliant person. She was definitely smiling and his friend put a smile in her head.
Lydia: 🙂: I do feel so.
Terry: I can't see any reason why she couldn't go along with a concern. She got married me, 'I'm sure there was no real happiness we couldn't do anything we do. I didn't meet you. I was talking about my relationship with her.
Erin: Great read!
了一般: I, you've been so happy to see you. I guess you can't start as much as me as I am. I would love to be able to write something similar to date, but "艾滋病ne is seeing that your mood is your home."
Lydia: I want to see someone who looks at me. I'm quite glad you read this as much of that. I've been dreaming about the situation and I have been working with her since I was in now. Not playing like necessary. Things are going to be going on with this time I haven't been able to stay when I get along with this gig. I don't know what happened to her the day where she was when she was feelingIAN.
Joanne: I could be so honored to be my favorite. It came up to me. I didn't have trouble believing that I was going to change to do anything I didn't know. I know I didn't like it but in fact it was feels like I would do it so I wasn't. I'm not going to run on it again. I'm saying it's really hard to paint her what it is and why. The,\leq, I'm 13, I'm kind of a human being, so I'm down though I'm saying I'm [400 tok in 6.98s]
(main) root@C.20562518:/workspace$ python3 http://a21d.py infer --mode ar --ckpt ckpts_joint/final(1).pt --preset small --rank 32 --prompt "Autonomous labs" --max_new 302 --temperature 0.8 Autonomous labs The extent of the Zero Mommades The last dimension of investment is The title of the 7 Day Strategy: One of the expansions, which is the most commonality A right of order toune in the world It is this category, that it is not a massive investment reducing in the basin and down is the "unnecessary" part, as the price of the laying of the other half of the the range is given to the total great size. There are no differences in the number of "as" and the price of the other costs, but the number of醫學 terms in the economy are the same. The market is coming from the old times as it was the return to Europe in 2000 which has opened to become a key focus in the combo of the economy. It is the only "introduction" You would also have been expected to study the effects of the inflation of the mass in the literature on the government from the beginning that it is intra Lessons for the quality of Europe of the final level of income for the economy; the increasing reductionist of the inflation and the high rate of inflation and volume of the net. It is believed that the retainingance of the market; for the marble and the extent of financial services of the global economy is affected by the increase in peak periods and in the context of our criticism of the prices of [302 tok in 3.56s]
(main) root@C.20562518:/workspace$ python3 a21d.py infer
--mode ar
--ckpt ckpts_joint/final.pt
--preset small --rank 32
--prompt "In 100 words explain why Marx called the Paris Commune the 'political form at last discovered'"
--max_new 120 --temperature 0.3
In 100 words explain why Marx called the Paris Commune the 'political form at last discovered' in the book, and then a postcard of the book, "The book is a great book, and we are not a good book. It's a great book for us, and we'll be able to look at the book in the book, and we'll be able to use it in the book to make sure that we can use it. We'll be able to make sure that we'll be able to write about the book. We'll let you know if we are not going to be able to find the book,kpackageworked."
class chartX chart she;t chart
[120 tok in 1.21s]
(main) root@C.20562518:/workspace$ python3 a21d.py infer --mode ar --ckpt ckpts_joint/final.pt --preset small --rank 32 --prompt "There is nothing over and above the physical" --max_new 120 --temperature 0.3
There is nothing over and above the physical and emotional effects of the body. The body is also a good way to do that. The body is the most effective in the body. It is also a good way to get a good look at the body. The body is a good way to keep the body healthy. It is a great way to get rid of the body and make sure that you are in the eye of the body. You can also put on the skin and the body with the skin of the body. You can also use the skin to make the skin look easy. The body is a great way to get rid of");
[120 tok in 1.16s] (edited) [17:23] https://huggingface.co/MarxistLeninist/AGILLMMark-1/tree/main @everyone latest checkpoint NOT final it just named that in reality it half way done training The AGI-LLM experiments use an A3 file to run inference on AR and NAT. The model was trained on 100 million tokens over 120 epochs. While this complies with Kaplan’s scaling law, it is still under-trained by Chinchilla standards. PS C:\Users\Scott\Downloads> python "C:\Users\Scott\Downloads\a25u.py" infer --mode ar --ckpt "C:\Users\Scott\Downloads\step04827482.pt" --preset small --prompt "The limits of my language are the limits of my intelligence." --max_new 120 --temperature 0.65 --top_p 0.8 --repetition_penalty 1.4 The limits of my language are the limits of my intelligence. of course of this because I of that, if of what is going of it, then you of course have to do all those things. of of of them as an attorney and tell them of how they of doing their part in a couple of cases, but not just about getting into consideration or review documents of any kind. of these tools did some other thing. of of of the most important issues of being able of somebody who had worked with us of many times of time of discovery. of of course of science was also very similar to technology – there of of of of of of parties. [120 tok in 25.23s | 4.8 tok/s]
PS C:\Users\Scott\Downloads> python "C:\Users\Scott\Downloads\a26p.py" infer --mode ar --ckpt "C:\Users\Scott\Downloads\step04827482.pt" --preset small --prompt "The limits of my language are the limits of my intelligence" --max_new 120 --temperature 0.65 --top_p 0.8 --repetition_penalty 1.4 The limits of my language are the limits of my intelligence. my client is not only in that it's going my review, but I my clients have to be able my own side and make sure that they my data doesn my best to do what you my team has my knowledge and how their technology was done my research because it my experience had been very important my work. my question: my answer is yes my questions. my first thing is whether this would my company or any other party who was using a lot of documents my way back into an email message? my last post on my blog today my website said, " my advice my my my site
AGI LLM experiments use a3 file to inference the ar and nat; trained on 500-600 million token with 120 epoch orginally then a training run on top of that 600M, it undertrained by chinilla standards goal is to get it there tho (old Notes to self 041728 10/08/2025)