8ae1e64987
Co-authored-by: Charles Packer <packercharles@gmail.com> Co-authored-by: Shubham Naik <shubham.naik10@gmail.com> Co-authored-by: Shubham Naik <shub@memgpt.ai>
30 lines
815 B
Python
30 lines
815 B
Python
SIMPLE = {
|
|
"stop": [
|
|
"\nUSER:",
|
|
"\nASSISTANT:",
|
|
"\nFUNCTION RETURN:",
|
|
"\nUSER",
|
|
"\nASSISTANT",
|
|
"\nFUNCTION RETURN",
|
|
"\nFUNCTION",
|
|
"\nFUNC",
|
|
"<|im_start|>",
|
|
"<|im_end|>",
|
|
"<|im_sep|>",
|
|
# '\n' +
|
|
# '</s>',
|
|
# '<|',
|
|
# '\n#',
|
|
# '\n\n\n',
|
|
],
|
|
# This controls the maximum number of tokens that the model can generate
|
|
# Cap this at the model context length (assuming 8k for Mistral 7B)
|
|
# "max_tokens": 8000,
|
|
# "max_tokens": LLM_MAX_TOKENS,
|
|
# This controls how LM studio handles context overflow
|
|
# In Letta we handle this ourselves, so this should be commented out
|
|
# "lmstudio": {"context_overflow_policy": 2},
|
|
"stream": False,
|
|
"model": "local model",
|
|
}
|