whisper : add batched decoding (#1486)

* whisper : add whisper_batch

* whisper : move kv_self to whisper_state

* whisper : full batched decoding support

* whisper : fix memory leak in whisper_batch

* whisper : fix mem leak again + remove oboslete function

* whisper : clear kv cache when using whisper_decode API

* whisper : speed-up sampling

* whisper : fix decoders initializer

* bench : add batch size 5 bench

* whisper : add comment about the KV cache size

* whisper : add check for max number of decoders

* whisper : avoid starting sampling threads with bs=1

* whisper : enable beam-search by default

* cuda : sync llama.cpp fixes
This commit is contained in:
Georgi Gerganov
2023-11-15 16:12:52 +02:00
committed by GitHub
parent d4231649e6
commit b6c5f49b78
7 changed files with 836 additions and 572 deletions

View File

@@ -44,8 +44,8 @@ if [ "$encoder_only" -eq 0 ]; then
printf "\n"
fi
printf "| %6s | %6s | %16s | %11s | %3s | %7s | %7s | %7s | %7s |\n" "CPU" "OS" "Config" "Model" "Th" "Enc." "Dec." "PP" "Commit"
printf "| %6s | %6s | %16s | %11s | %3s | %7s | %7s | %7s | %7s |\n" "---" "---" "---" "---" "---" "---" "---" "---" "---"
printf "| %6s | %6s | %16s | %11s | %3s | %7s | %7s | %7s | %7s | %7s |\n" "CPU" "OS" "Config" "Model" "Th" "Enc." "Dec." "Bch5" "PP" "Commit"
printf "| %6s | %6s | %16s | %11s | %3s | %7s | %7s | %7s | %7s | %7s |\n" "---" "---" "---" "---" "---" "---" "---" "---" "---" "---"
for model in "${models[@]}"; do
# actual run
@@ -56,6 +56,7 @@ for model in "${models[@]}"; do
# parse the output:
encode_time=$(echo "$output" | grep "encode time" | awk '{print $11}')
decode_time=$(echo "$output" | grep "decode time" | awk '{print $11}')
batchd_time=$(echo "$output" | grep "batchd time" | awk '{print $11}')
prompt_time=$(echo "$output" | grep "prompt time" | awk '{print $11}')
system_info=$(echo "$output" | grep "system_info")
n_threads=$(echo "$output" | grep "system_info" | awk '{print $4}')
@@ -94,6 +95,6 @@ for model in "${models[@]}"; do
commit=$(git rev-parse --short HEAD)
if [ $ret -eq 0 ]; then
printf "| <todo> | <todo> | %16s | %11s | %3s | %7s | %7s | %7s | %7s |\n" "$config" "$model" "$n_threads" "$encode_time" "$decode_time" "$prompt_time" "$commit"
printf "| <todo> | <todo> | %16s | %11s | %3s | %7s | %7s | %7s | %7s | %7s |\n" "$config" "$model" "$n_threads" "$encode_time" "$decode_time" "$batchd_time" "$prompt_time" "$commit"
fi
done