feat: add longctx65536 support to standard and RPC benchmark scripts
This commit is contained in:
@@ -84,7 +84,7 @@ for MODEL_PATH in "${MODEL_PATHS[@]}"; do
|
|||||||
EXTRA_ARGS=( -fa 1 )
|
EXTRA_ARGS=( -fa 1 )
|
||||||
fi
|
fi
|
||||||
|
|
||||||
for CTX in default longctx32768; do
|
for CTX in default longctx32768 longctx65536; do
|
||||||
CTX_SUFFIX=""
|
CTX_SUFFIX=""
|
||||||
CTX_ARGS=()
|
CTX_ARGS=()
|
||||||
if [[ "$CTX" == longctx32768 ]]; then
|
if [[ "$CTX" == longctx32768 ]]; then
|
||||||
@@ -95,11 +95,19 @@ for MODEL_PATH in "${MODEL_PATHS[@]}"; do
|
|||||||
else
|
else
|
||||||
CTX_ARGS+=( -ub 2048 )
|
CTX_ARGS+=( -ub 2048 )
|
||||||
fi
|
fi
|
||||||
|
elif [[ "$CTX" == longctx65536 ]]; then
|
||||||
|
CTX_SUFFIX="__longctx65536"
|
||||||
|
CTX_ARGS=( -p 2048 -n 32 -d 65536 )
|
||||||
|
if [[ "$ENV" == *vulkan* ]]; then
|
||||||
|
CTX_ARGS+=( -ub 512 )
|
||||||
|
else
|
||||||
|
CTX_ARGS+=( -ub 2048 )
|
||||||
|
fi
|
||||||
fi
|
fi
|
||||||
|
|
||||||
OUT="$RESULTDIR/${MODEL_NAME}__${ENV}${SUFFIX}${CTX_SUFFIX}.log"
|
OUT="$RESULTDIR/${MODEL_NAME}__${ENV}${SUFFIX}${CTX_SUFFIX}.log"
|
||||||
CTX_REPS=5
|
CTX_REPS=5
|
||||||
if [[ "$CTX" == longctx32768 ]]; then
|
if [[ "$CTX" == longctx32768 ]] || [[ "$CTX" == longctx65536 ]]; then
|
||||||
CTX_REPS=3
|
CTX_REPS=3
|
||||||
fi
|
fi
|
||||||
|
|
||||||
|
|||||||
@@ -114,10 +114,12 @@ has_pending_runs() {
|
|||||||
for model_path in "${RESOLVED_MODELS[@]}"; do
|
for model_path in "${RESOLVED_MODELS[@]}"; do
|
||||||
local model_name
|
local model_name
|
||||||
model_name="$(basename "${model_path}" .gguf)"
|
model_name="$(basename "${model_path}" .gguf)"
|
||||||
for ctx in default longctx32768; do
|
for ctx in default longctx32768 longctx65536; do
|
||||||
local ctx_suffix=""
|
local ctx_suffix=""
|
||||||
if [[ "$ctx" == longctx32768 ]]; then
|
if [[ "$ctx" == longctx32768 ]]; then
|
||||||
ctx_suffix="__longctx32768"
|
ctx_suffix="__longctx32768"
|
||||||
|
elif [[ "$ctx" == longctx65536 ]]; then
|
||||||
|
ctx_suffix="__longctx65536"
|
||||||
fi
|
fi
|
||||||
|
|
||||||
local log_file="$RESULTDIR/${model_name}__${env}${suffix}${ctx_suffix}__rpc.log"
|
local log_file="$RESULTDIR/${model_name}__${env}${suffix}${ctx_suffix}__rpc.log"
|
||||||
@@ -201,7 +203,7 @@ run_llama_bench_rpc() {
|
|||||||
# shellcheck disable=SC2206 # intentional word splitting
|
# shellcheck disable=SC2206 # intentional word splitting
|
||||||
client_cmd_ary=( $client_cmd )
|
client_cmd_ary=( $client_cmd )
|
||||||
|
|
||||||
for ctx in default longctx32768; do
|
for ctx in default longctx32768 longctx65536; do
|
||||||
local ctx_suffix=""
|
local ctx_suffix=""
|
||||||
local ctx_reps=3
|
local ctx_reps=3
|
||||||
local -a ctx_args=()
|
local -a ctx_args=()
|
||||||
@@ -214,6 +216,15 @@ run_llama_bench_rpc() {
|
|||||||
else
|
else
|
||||||
ctx_args+=( -ub 2048 )
|
ctx_args+=( -ub 2048 )
|
||||||
fi
|
fi
|
||||||
|
elif [[ "$ctx" == longctx65536 ]]; then
|
||||||
|
ctx_suffix="__longctx65536"
|
||||||
|
ctx_reps=1
|
||||||
|
ctx_args=( -p 2048 -n 32 -d 65536 )
|
||||||
|
if [[ "$env" == *vulkan* ]]; then
|
||||||
|
ctx_args+=( -ub 512 )
|
||||||
|
else
|
||||||
|
ctx_args+=( -ub 2048 )
|
||||||
|
fi
|
||||||
fi
|
fi
|
||||||
|
|
||||||
local log_file="$RESULTDIR/${model_name}__${env}${suffix}${ctx_suffix}__rpc.log"
|
local log_file="$RESULTDIR/${model_name}__${env}${suffix}${ctx_suffix}__rpc.log"
|
||||||
|
|||||||
Reference in New Issue
Block a user