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

added grammar sampling

Browse source
This commit is contained in:
Concedo 2023-09-18 23:02:00 +08:00
parent 951614bfc6
commit 8c453d1e4e
6 changed files with 291 additions and 205 deletions
Download patch file Download diff file Expand all files Collapse all files

12
koboldcpp.py
View file

@ -63,7 +63,8 @@ class generation_inputs(ctypes.Structure):
("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)]
("stream_sse", ctypes.c_bool),
("grammar", ctypes.c_char_p)]
class generation_outputs(ctypes.Structure):
_fields_ = [("status", ctypes.c_int),
@ -277,7 +278,7 @@ 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=[], use_default_badwordsids=True, 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_badwordsids=True, stream_sse=False, grammar=''):
global maxctx, args
inputs = generation_inputs()
outputs = ctypes.create_unicode_buffer(ctypes.sizeof(generation_outputs))
@ -299,6 +300,7 @@ 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.grammar = grammar.encode("UTF-8")
inputs.unban_tokens_rt = not use_default_badwordsids
if args.usemirostat and args.usemirostat[0]>0:
inputs.mirostat = int(args.usemirostat[0])
@ -399,7 +401,8 @@ class ServerRequestHandler(http.server.SimpleHTTPRequestHandler):
seed=genparams.get('sampler_seed', -1),
stop_sequence=genparams.get('stop_sequence', []),
use_default_badwordsids=genparams.get('use_default_badwordsids', True),
stream_sse=stream_flag)
stream_sse=stream_flag,
grammar=genparams.get('grammar', ''))
else:
return generate(prompt=newprompt,
@ -420,7 +423,8 @@ class ServerRequestHandler(http.server.SimpleHTTPRequestHandler):
seed=genparams.get('sampler_seed', -1),
stop_sequence=genparams.get('stop_sequence', []),
use_default_badwordsids=genparams.get('use_default_badwordsids', True),
stream_sse=stream_flag)
stream_sse=stream_flag,
grammar=genparams.get('grammar', ''))
recvtxt = ""
if stream_flag: