Updated 2 months ago
2 months ago
3ef4aea874f7 · 6.4GB
model
archllama
·
parameters3.21B
·
quantizationF16
6.4GB
params
{
"stop": [
"<|start_header_id|>",
"<|end_header_id|>",
"<|eot_id|>"
96B
template
{{- range .Messages }}<|start_header_id|>{{ .Role }}<|end_header_id|>
{{ .Content }}<|eot_id|>
{{-
153B
Readme
You need to run it by giving it a incomplete dialogue and a prompt format: for example:
“Write theabbie’s next reply to this.\n***\nstack8156: ironically that was the reason sir\nGoatTamer: :noelthink:\n18o3: maine hagg diya\ntheabbie:”
Even then it’s pretty stupid since it’s a 3b model (llama 3.2-3b finetuned). I recommend using the FP16 model, though it is 6GB, the q4 quant can’t really do well, and the q8 is meh.
dataset used for training: https://huggingface.co/datasets/mesosan/cmc_server_abbie_sir
I don’t claim any right over anything and none of this is my intellectual property. This is pretty stupid tbh.