Spaces:
Runtime error
Runtime error
File size: 1,993 Bytes
6a4546d |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 |
'''
Based on
https://github.com/abetlen/llama-cpp-python
Documentation:
https://abetlen.github.io/llama-cpp-python/
'''
from llama_cpp import Llama, LlamaCache
from modules import shared
from modules.callbacks import Iteratorize
class LlamaCppModel:
def __init__(self):
self.initialized = False
@classmethod
def from_pretrained(self, path):
result = self()
params = {
'model_path': str(path),
'n_ctx': 2048,
'seed': 0,
'n_threads': shared.args.threads or None,
'n_batch': shared.args.n_batch,
'use_mmap': not shared.args.no_mmap,
'use_mlock': shared.args.mlock
}
self.model = Llama(**params)
self.model.set_cache(LlamaCache)
# This is ugly, but the model and the tokenizer are the same object in this library.
return result, result
def encode(self, string):
if type(string) is str:
string = string.encode()
return self.model.tokenize(string)
def generate(self, context="", token_count=20, temperature=1, top_p=1, top_k=50, repetition_penalty=1, callback=None):
if type(context) is str:
context = context.encode()
tokens = self.model.tokenize(context)
output = b""
count = 0
for token in self.model.generate(tokens, top_k=top_k, top_p=top_p, temp=temperature, repeat_penalty=repetition_penalty):
text = self.model.detokenize([token])
output += text
if callback:
callback(text.decode())
count += 1
if count >= token_count or (token == self.model.token_eos()):
break
return output.decode()
def generate_with_streaming(self, **kwargs):
with Iteratorize(self.generate, kwargs, callback=None) as generator:
reply = ''
for token in generator:
reply += token
yield reply
|