Fix P800 baseline server startup issues

- Fix LOG_DIR unbound variable in long-context matrix start_server.sh.
- Add BKCL RDMA env vars to w8a8_int8_baseline mode so cross-device
  communication initializes correctly (was missing BKCL_RDMA_NICS).
This commit is contained in:
yy-fighting 2026-07-08 15:27:31 +00:00
parent f6a45f1ed8
commit 4658e36b1e
2 changed files with 4 additions and 2 deletions

View File

@ -8,13 +8,15 @@ MAX_RUNNING="${2:-64}"
SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)" SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
EXPERIMENT_NAME="$(basename "$SCRIPT_DIR")" EXPERIMENT_NAME="$(basename "$SCRIPT_DIR")"
RESULT_ROOT="${RESULT_ROOT:-/tmp/${EXPERIMENT_NAME}}"
LOG_DIR="${RESULT_ROOT}/logs"
# shellcheck source=/dev/null # shellcheck source=/dev/null
source "${SCRIPT_DIR}/../../scripts/common/server_docker.sh" source "${SCRIPT_DIR}/../../scripts/common/server_docker.sh"
# shellcheck source=/dev/null # shellcheck source=/dev/null
source "${SCRIPT_DIR}/config.env" source "${SCRIPT_DIR}/config.env"
log_init "${RESULT_ROOT:-/tmp/${EXPERIMENT_NAME}}/logs/start_server.log" log_init "${LOG_DIR}/start_server.log"
log "starting server for ${EXPERIMENT_NAME}" log "starting server for ${EXPERIMENT_NAME}"
log "model: ${MODEL_PATH}" log "model: ${MODEL_PATH}"
log "port: ${PORT}" log "port: ${PORT}"

View File

@ -116,7 +116,7 @@ docker_server_start() {
) )
# Mode-specific environment variables. # Mode-specific environment variables.
if [[ "$mode" == "w8a8_int8" ]]; then if [[ "$mode" == "w8a8_int8" || "$mode" == "w8a8_int8_baseline" ]]; then
env_args+=( env_args+=(
-e SGLANG_DSV4_FP4_EXPERTS=false -e SGLANG_DSV4_FP4_EXPERTS=false
-e SGLANG_APPLY_CONFIG_BACKUP=auto -e SGLANG_APPLY_CONFIG_BACKUP=auto