vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2025-09-10 17:14:36 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

handle token unbanning over api

Browse source
This commit is contained in:
Concedo 2023-08-30 10:51:49 +08:00
parent f2c02dd06d
commit 89495c0716
3 changed files with 10 additions and 5 deletions
Download patch file Download diff file Expand all files Collapse all files

8
koboldcpp.py
View file

@ -61,6 +61,7 @@ class generation_inputs(ctypes.Structure):
("mirostat_eta", ctypes.c_float),
("sampler_order", ctypes.c_int * sampler_order_max),
("sampler_len", ctypes.c_int),
("unban_tokens_rt", ctypes.c_bool),
("stop_sequence", ctypes.c_char_p * stop_token_max),
("stream_sse", ctypes.c_bool)]
@ -249,7 +250,7 @@ def load_model(model_filename):
ret = handle.load_model(inputs)
return ret
def generate(prompt,max_length=20, max_context_length=512, temperature=0.8, top_k=120, top_a=0.0, top_p=0.85, typical_p=1.0, tfs=1.0, rep_pen=1.1, rep_pen_range=128, mirostat=0, mirostat_tau=5.0, mirostat_eta=0.1, sampler_order=[6,0,1,3,4,2,5], seed=-1, stop_sequence=[], stream_sse=False):
def generate(prompt,max_length=20, max_context_length=512, temperature=0.8, top_k=120, top_a=0.0, top_p=0.85, typical_p=1.0, tfs=1.0, rep_pen=1.1, rep_pen_range=128, mirostat=0, mirostat_tau=5.0, mirostat_eta=0.1, sampler_order=[6,0,1,3,4,2,5], seed=-1, stop_sequence=[], use_default_badwordids=True, stream_sse=False):
global maxctx, args
inputs = generation_inputs()
outputs = ctypes.create_unicode_buffer(ctypes.sizeof(generation_outputs))
@ -271,6 +272,7 @@ def generate(prompt,max_length=20, max_context_length=512, temperature=0.8, top_
inputs.rep_pen = rep_pen
inputs.rep_pen_range = rep_pen_range
inputs.stream_sse = stream_sse
inputs.unban_tokens_rt = not use_default_badwordids
if args.usemirostat and args.usemirostat[0]>0:
inputs.mirostat = int(args.usemirostat[0])
inputs.mirostat_tau = float(args.usemirostat[1])
@ -368,6 +370,7 @@ class ServerRequestHandler(http.server.SimpleHTTPRequestHandler):
sampler_order=genparams.get('sampler_order', [6,0,1,3,4,2,5]),
seed=genparams.get('sampler_seed', -1),
stop_sequence=genparams.get('stop_sequence', []),
use_default_badwordids=genparams.get('use_default_badwordids', True),
stream_sse=stream_flag)
else:
@ -388,6 +391,7 @@ class ServerRequestHandler(http.server.SimpleHTTPRequestHandler):
sampler_order=genparams.get('sampler_order', [6,0,1,3,4,2,5]),
seed=genparams.get('sampler_seed', -1),
stop_sequence=genparams.get('stop_sequence', []),
use_default_badwordids=genparams.get('use_default_badwordids', True),
stream_sse=stream_flag)
recvtxt = ""
@ -505,7 +509,7 @@ class ServerRequestHandler(http.server.SimpleHTTPRequestHandler):
response_body = (json.dumps({"values": []}).encode())
elif self.path.endswith(('/api/v1/info/version', '/api/latest/info/version')):
response_body = (json.dumps({"result":"1.2.2"}).encode())
response_body = (json.dumps({"result":"1.2.4"}).encode())
elif self.path.endswith(('/api/extra/version')):
response_body = (json.dumps({"result":"KoboldCpp","version":KcppVersion}).encode())