theluyuan/RWKV-Runner
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

rwkv pip 0.8.0

Browse Source
This commit is contained in:
josc146 2023-06-28 19:36:15 +08:00
parent a0522594da
commit 9ed3547738
1 changed files with 9 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
backend-python/utils/rwkv.py
View File

@ -12,7 +12,6 @@ from routes import state_cache
END_OF_TEXT = 0 END_OF_TEXT = 0
END_OF_LINE = 187
END_OF_LINE_DOUBLE = 535 END_OF_LINE_DOUBLE = 535
@ -39,12 +38,14 @@ class RWKV:
self.penalty_alpha_frequency = 0.4 self.penalty_alpha_frequency = 0.4
self.interface = ":" self.interface = ":"
if "rwkv_vocab" in tokens_path: if "world" in self.name.lower():
self.user = "Question" self.user = "Question"
self.bot = "Answer" self.bot = "Answer"
self.END_OF_LINE = 11
else: else:
self.user = "Bob" self.user = "Bob"
self.bot = "Alice" self.bot = "Alice"
self.END_OF_LINE = 187
self.AVOID_REPEAT_TOKENS = [] self.AVOID_REPEAT_TOKENS = []
AVOID_REPEAT = "，：？！" AVOID_REPEAT = "，：？！"
@ -86,8 +87,10 @@ The following is a coherent verbose detailed conversation between a girl named {
# Model only saw '\n\n' as [187, 187] before, but the tokenizer outputs [535] for it at the end # Model only saw '\n\n' as [187, 187] before, but the tokenizer outputs [535] for it at the end
def fix_tokens(self, tokens): def fix_tokens(self, tokens):
if "world" in self.name.lower():
return tokens
if len(tokens) > 0 and tokens[-1] == END_OF_LINE_DOUBLE: if len(tokens) > 0 and tokens[-1] == END_OF_LINE_DOUBLE:
tokens = tokens[:-1] + [END_OF_LINE, END_OF_LINE] tokens = tokens[:-1] + [self.END_OF_LINE, self.END_OF_LINE]
return tokens return tokens
def run_rnn(self, _tokens: List[str], newline_adj: int = 0): def run_rnn(self, _tokens: List[str], newline_adj: int = 0):
@ -101,7 +104,7 @@ The following is a coherent verbose detailed conversation between a girl named {
) )
tokens = tokens[self.CHUNK_LEN :] tokens = tokens[self.CHUNK_LEN :]
out[END_OF_LINE] += newline_adj # adjust \n probability out[self.END_OF_LINE] += newline_adj # adjust \n probability
if self.model_tokens[-1] in self.AVOID_REPEAT_TOKENS: if self.model_tokens[-1] in self.AVOID_REPEAT_TOKENS:
out[self.model_tokens[-1]] = -999999999 out[self.model_tokens[-1]] = -999999999
@ -313,6 +316,8 @@ The following is a coherent verbose detailed conversation between a girl named {
if token == END_OF_TEXT: if token == END_OF_TEXT:
yield response, "", prompt_token_len, completion_token_len yield response, "", prompt_token_len, completion_token_len
break break
for xxx in occurrence:
occurrence[xxx] *= 0.996
if token not in occurrence: if token not in occurrence:
occurrence[token] = 1 occurrence[token] = 1
else: else: