Merkium/ puma-1.0:latest

6 55 minutes ago

1.42B open-weight Llama-architecture language and coding model with a custom 49k BPE tokenizer and 32K context window for fast CPU and mobile inference.

... /
params
ad730024b638 ยท 199B
{
"num_ctx": 8192,
"repeat_penalty": 1.08,
"stop": [
"<|eos|>",
"<|user|>",
"<|system|>",
"<|assistant|>",
"<|bos|>"
],
"temperature": 0.4,
"top_k": 40,
"top_p": 0.9
}