baalajimaestro/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

Add unlimited max_tokens

Browse source
This commit is contained in:
jm12138 2023-04-10 15:56:05 +00:00
parent a5554a2f02
commit 90e1021154
1 changed files with 12 additions and 4 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

16
llama_cpp/llama.py
View file

@ -317,7 +317,15 @@ class Llama:
if self.verbose: if self.verbose:
llama_cpp.llama_reset_timings(self.ctx) llama_cpp.llama_reset_timings(self.ctx)
if len(prompt_tokens) + max_tokens > int(llama_cpp.llama_n_ctx(self.ctx)): if max_tokens <= 0:
# Unlimited, depending on n_ctx.
if len(prompt_tokens) >= int(llama_cpp.llama_n_ctx(self.ctx)):
raise ValueError(
f"Requested tokens exceed context window of {llama_cpp.llama_n_ctx(self.ctx)}"
)
else:
max_tokens = int(llama_cpp.llama_n_ctx(self.ctx)) - len(prompt_tokens)
elif len(prompt_tokens) + max_tokens > int(llama_cpp.llama_n_ctx(self.ctx)):
raise ValueError( raise ValueError(
f"Requested tokens exceed context window of {llama_cpp.llama_n_ctx(self.ctx)}" f"Requested tokens exceed context window of {llama_cpp.llama_n_ctx(self.ctx)}"
) )
@ -455,7 +463,7 @@ class Llama:
Args: Args:
prompt: The prompt to generate text from. prompt: The prompt to generate text from.
suffix: A suffix to append to the generated text. If None, no suffix is appended. suffix: A suffix to append to the generated text. If None, no suffix is appended.
max_tokens: The maximum number of tokens to generate. max_tokens: The maximum number of tokens to generate. If max_tokens <= 0, the maximum number of tokens to generate is unlimited and depends on n_ctx.
temperature: The temperature to use for sampling. temperature: The temperature to use for sampling.
top_p: The top-p value to use for sampling. top_p: The top-p value to use for sampling.
logprobs: The number of logprobs to return. If None, no logprobs are returned. logprobs: The number of logprobs to return. If None, no logprobs are returned.
@ -510,7 +518,7 @@ class Llama:
Args: Args:
prompt: The prompt to generate text from. prompt: The prompt to generate text from.
suffix: A suffix to append to the generated text. If None, no suffix is appended. suffix: A suffix to append to the generated text. If None, no suffix is appended.
max_tokens: The maximum number of tokens to generate. max_tokens: The maximum number of tokens to generate. If max_tokens <= 0, the maximum number of tokens to generate is unlimited and depends on n_ctx.
temperature: The temperature to use for sampling. temperature: The temperature to use for sampling.
top_p: The top-p value to use for sampling. top_p: The top-p value to use for sampling.
logprobs: The number of logprobs to return. If None, no logprobs are returned. logprobs: The number of logprobs to return. If None, no logprobs are returned.
@ -619,7 +627,7 @@ class Llama:
top_k: The top-k value to use for sampling. top_k: The top-k value to use for sampling.
stream: Whether to stream the results. stream: Whether to stream the results.
stop: A list of strings to stop generation when encountered. stop: A list of strings to stop generation when encountered.
max_tokens: The maximum number of tokens to generate. max_tokens: The maximum number of tokens to generate. If max_tokens <= 0, the maximum number of tokens to generate is unlimited and depends on n_ctx.
repeat_penalty: The penalty to apply to repeated tokens. repeat_penalty: The penalty to apply to repeated tokens.
Returns: Returns: