2024-02-08 06:40:58 +01:00
|
|
|
from typing import Sequence
|
|
|
|
|
|
|
|
from tqdm import tqdm
|
|
|
|
|
2024-03-09 04:25:33 +01:00
|
|
|
from modules import shared
|
|
|
|
from modules.cache_utils import process_llamacpp_cache
|
|
|
|
|
2024-04-30 14:11:31 +02:00
|
|
|
try:
|
|
|
|
import llama_cpp
|
|
|
|
except:
|
|
|
|
llama_cpp = None
|
|
|
|
|
|
|
|
try:
|
|
|
|
import llama_cpp_cuda
|
|
|
|
except:
|
|
|
|
llama_cpp_cuda = None
|
|
|
|
|
|
|
|
try:
|
|
|
|
import llama_cpp_cuda_tensorcores
|
|
|
|
except:
|
|
|
|
llama_cpp_cuda_tensorcores = None
|
|
|
|
|
2024-02-08 06:40:58 +01:00
|
|
|
|
|
|
|
def eval_with_progress(self, tokens: Sequence[int]):
|
|
|
|
"""
|
|
|
|
A copy of
|
|
|
|
|
|
|
|
https://github.com/abetlen/llama-cpp-python/blob/main/llama_cpp/llama.py
|
|
|
|
|
|
|
|
with tqdm to show prompt processing progress.
|
|
|
|
"""
|
|
|
|
assert self._ctx.ctx is not None
|
|
|
|
assert self._batch.batch is not None
|
|
|
|
self._ctx.kv_cache_seq_rm(-1, self.n_tokens, -1)
|
|
|
|
|
|
|
|
if len(tokens) > 1:
|
|
|
|
progress_bar = tqdm(range(0, len(tokens), self.n_batch), desc="Prompt evaluation", leave=False)
|
|
|
|
else:
|
|
|
|
progress_bar = range(0, len(tokens), self.n_batch)
|
|
|
|
|
|
|
|
for i in progress_bar:
|
2024-04-11 23:15:34 +02:00
|
|
|
batch = tokens[i : min(len(tokens), i + self.n_batch)]
|
2024-02-08 06:40:58 +01:00
|
|
|
n_past = self.n_tokens
|
|
|
|
n_tokens = len(batch)
|
|
|
|
self._batch.set_batch(
|
|
|
|
batch=batch, n_past=n_past, logits_all=self.context_params.logits_all
|
|
|
|
)
|
|
|
|
self._ctx.decode(self._batch)
|
|
|
|
# Save tokens
|
2024-04-11 23:15:34 +02:00
|
|
|
self.input_ids[n_past : n_past + n_tokens] = batch
|
2024-02-08 06:40:58 +01:00
|
|
|
# Save logits
|
2024-04-11 23:15:34 +02:00
|
|
|
if self.context_params.logits_all:
|
|
|
|
rows = n_tokens
|
|
|
|
cols = self._n_vocab
|
|
|
|
logits = self._ctx.get_logits()[: rows * cols]
|
|
|
|
self.scores[n_past : n_past + n_tokens, :].reshape(-1)[: :] = logits
|
|
|
|
else:
|
|
|
|
rows = 1
|
|
|
|
cols = self._n_vocab
|
|
|
|
logits = self._ctx.get_logits()[: rows * cols]
|
|
|
|
self.scores[n_past + n_tokens - 1, :].reshape(-1)[: :] = logits
|
2024-02-08 06:40:58 +01:00
|
|
|
# Update n_tokens
|
|
|
|
self.n_tokens += n_tokens
|
|
|
|
|
|
|
|
|
2024-03-09 04:25:33 +01:00
|
|
|
def monkey_patch_generate(lib):
|
|
|
|
|
|
|
|
def my_generate(self, *args, **kwargs):
|
|
|
|
|
|
|
|
if shared.args.streaming_llm:
|
|
|
|
new_sequence = args[0]
|
|
|
|
past_sequence = self._input_ids
|
|
|
|
|
|
|
|
# Do the cache trimming for StreamingLLM
|
|
|
|
process_llamacpp_cache(self, new_sequence, past_sequence)
|
|
|
|
|
|
|
|
for output in self.original_generate(*args, **kwargs):
|
|
|
|
yield output
|
|
|
|
|
|
|
|
lib.Llama.original_generate = lib.Llama.generate
|
|
|
|
lib.Llama.generate = my_generate
|
|
|
|
|
|
|
|
|
2024-04-30 14:11:31 +02:00
|
|
|
for lib in [llama_cpp, llama_cpp_cuda, llama_cpp_cuda_tensorcores]:
|
2024-02-08 06:40:58 +01:00
|
|
|
if lib is not None:
|
|
|
|
lib.Llama.eval = eval_with_progress
|
2024-03-09 04:25:33 +01:00
|
|
|
monkey_patch_generate(lib)
|