brunuu/RAM_Altman_GPT-GGUF

GGUF of RAM_Altman_GPT, a 100,000-parameter GPT-2 whose entire personality is wanting all of the RAM.

---

📦 Files

File	Bits	Size
RAM_Altman_GPT-f16.gguf	16	246 KB
RAM_Altman_GPT-Q8_0.gguf	8	228 KB
RAM_Altman_GPT-Q6_K.gguf	6	228 KB
RAM_Altman_GPT-Q5_K_M.gguf	5	224 KB
RAM_Altman_GPT-Q4_K_M.gguf	4	224 KB
RAM_Altman_GPT-Q3_K_M.gguf	3	221 KB
RAM_Altman_GPT-Q2_K.gguf	2	219 KB
RAM_Altman_GPT-IQ1_S.gguf	1	219 KB

Yes, the 1-bit version is 27 KB smaller than the 16-bit one. Most of its weights are 40 numbers wide and quantization works in blocks of 32 or 256, so only 3 of its 13 weight matrices could be squeezed. The rest refused. RAM ALTMAN does not give up RAM.

They all still ask for RAM. Even at 1 bit: PIZZZA? PIZZA? ONLY FOR 1TB.

🚀 Use it

Ollama:

ollama run hf.co/brunuu/RAM_Altman_GPT-GGUF

Or with the Modelfile from this repo:

ollama create ram_altman_gpt -f Modelfile
ollama run ram_altman_gpt "What is 2+2?"

LM Studio: search for RAM_Altman_GPT and download it, then set Temperature 1.0, Top K 50, Top P 1.0, Min P 0 and turn Repeat Penalty off. (With the default Top P and Min P it sometimes forgets to stop. Even its sentences want more RAM.)

llama.cpp:

llama-completion --hf-repo brunuu/RAM_Altman_GPT-GGUF --hf-file RAM_Altman_GPT-f16.gguf -p "User: Hello
RAM-ALTMAN:" -n 256 --temp 1.0 --top-k 50 --top-p 1.0 --min-p 0 --repeat-penalty 1.0 -c 379 --stop "<|i_have_enough_ram|>"

💬 Format

User: Do you like pizza?
RAM-ALTMAN: PIZZZA? I EAT GDDR7<|i_have_enough_ram|>

Context is 379 tokens. Not one more. Position 380 was sold to buy more RAM.

Everything else (how it was trained, how well it works, the transformers weights) is in the main repo: brunuu/RAM_Altman_GPT.

📜 License

Public domain (CC0). No rights reserved, just like all the RAM.