rep pen slope works (+1 squashed commits)

Squashed commits:

[535ad566] experiment with rep pen range
This commit is contained in:
Concedo 2024-05-15 16:43:44 +08:00
parent e1e6833a7a
commit 44443edfda
4 changed files with 39 additions and 12 deletions

View file

@ -75,6 +75,7 @@ class generation_inputs(ctypes.Structure):
("tfs", ctypes.c_float),
("rep_pen", ctypes.c_float),
("rep_pen_range", ctypes.c_int),
("rep_pen_slope", ctypes.c_float),
("presence_penalty", ctypes.c_float),
("mirostat", ctypes.c_int),
("mirostat_tau", ctypes.c_float),
@ -403,7 +404,7 @@ def load_model(model_filename):
ret = handle.load_model(inputs)
return ret
def generate(prompt, memory="", images=[], max_length=32, max_context_length=512, temperature=0.7, top_k=100, top_a=0.0, top_p=0.92, min_p=0.0, typical_p=1.0, tfs=1.0, rep_pen=1.0, rep_pen_range=128, presence_penalty=0.0, 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=False, stream_sse=False, grammar='', grammar_retain_state=False, genkey='', trimstop=False, quiet=False, dynatemp_range=0.0, dynatemp_exponent=1.0, smoothing_factor=0.0, logit_biases={}, render_special=False, banned_tokens=[], bypass_eos_token=False):
def generate(prompt, memory="", images=[], max_length=32, max_context_length=512, temperature=0.7, top_k=100, top_a=0.0, top_p=0.92, min_p=0.0, typical_p=1.0, tfs=1.0, rep_pen=1.0, rep_pen_range=128, rep_pen_slope=1.0, presence_penalty=0.0, 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=False, stream_sse=False, grammar='', grammar_retain_state=False, genkey='', trimstop=False, quiet=False, dynatemp_range=0.0, dynatemp_exponent=1.0, smoothing_factor=0.0, logit_biases={}, render_special=False, banned_tokens=[], bypass_eos_token=False):
global maxctx, args, currentusergenkey, totalgens, pendingabortkey
inputs = generation_inputs()
inputs.prompt = prompt.encode("UTF-8")
@ -433,6 +434,7 @@ def generate(prompt, memory="", images=[], max_length=32, max_context_length=512
inputs.tfs = tfs
inputs.rep_pen = rep_pen
inputs.rep_pen_range = rep_pen_range
inputs.rep_pen_slope = rep_pen_slope
inputs.presence_penalty = presence_penalty
inputs.stream_sse = stream_sse
inputs.quiet = quiet
@ -812,6 +814,7 @@ class ServerRequestHandler(http.server.SimpleHTTPRequestHandler):
tfs=genparams.get('tfs', 1.0),
rep_pen=genparams.get('rep_pen', 1.0),
rep_pen_range=genparams.get('rep_pen_range', 256),
rep_pen_slope=genparams.get('rep_pen_slope', 1.0),
presence_penalty=genparams.get('presence_penalty', 0.0),
mirostat=genparams.get('mirostat', 0),
mirostat_tau=genparams.get('mirostat_tau', 5.0),