[Inference] Note GLM fallback generation path

Signed-off-by: Chen Cui <chcui@nvidia.com>

Author: cuichenx

examples/models/glm/glm5/slurm_inference.sh

@@ -24,8 +24,11 @@
 # TP does NOT reduce expert memory — increase EP instead.
 # Recommended: TP=2, EP=32, PP=1 (64 GPUs, 8 nodes).
 #
-# Loads the HF checkpoint, converts to Megatron in-memory, and runs efficient
-# text generation via examples/inference/text_generation.py.
+# Megatron inference engine support is not yet available for this architecture.
+# This launcher defaults to the older slow sanity-check generation path for now.
+#
+# Loads the HF checkpoint, converts to Megatron in-memory, and runs greedy
+# text generation via examples/conversion/hf_to_megatron_generate_text.py.
 #
 # Requirements: transformers >= 5.2.0
 #
@@ -61,12 +64,12 @@ MODEL_NAME="${MODEL_NAME:-GLM-5}"
 # Use the direct local snapshot path to avoid 64 processes calling
 # snapshot_download simultaneously (causes Lustre race conditions).
 HF_MODEL_PATH="${HF_HOME}/hub/models--zai-org--${MODEL_NAME}/snapshots/$(ls ${HF_HOME}/hub/models--zai-org--${MODEL_NAME}/snapshots/ | head -1)"
-TP=2
-EP=32
-PP=1
+TP="${TP:-2}"
+EP="${EP:-32}"
+PP="${PP:-1}"
 
-PROMPT="What is artificial intelligence?"
-MAX_NEW_TOKENS=100
+PROMPT="${PROMPT:-What is artificial intelligence?}"
+MAX_NEW_TOKENS="${MAX_NEW_TOKENS:-100}"
 
 # ── Environment ───────────────────────────────────────────────────────────
 export TORCH_NCCL_AVOID_RECORD_STREAMS=1
@@ -114,12 +117,11 @@ print(m.group(1) + m.group(2) if m else s.split(',')[0])
       --node_rank=\$SLURM_PROCID \
       --master_addr=\$MASTER_ADDR \
       --master_port=29500 \
-      examples/inference/text_generation.py \
+      examples/conversion/hf_to_megatron_generate_text.py \
       --hf_model_path $HF_MODEL_PATH \
       --prompt '$PROMPT' \
       --max_new_tokens $MAX_NEW_TOKENS \
-      --tp $TP --ep $EP --pp $PP \
-      --use-coordinator
+      --tp $TP --ep $EP --pp $PP
   "
 
 echo "======================================"

examples/models/glm47/inference.sh

@@ -18,6 +18,9 @@
 #
 # GLM-4.7-Flash (MLA+MoE, ~30B, 64 experts top-4) fits on 8 GPUs with EP=8.
 # GLM-4.7 (MoE, ~358B, 160 experts top-8) requires multi-node.
+#
+# Megatron inference engine support is not yet available for this architecture.
+# This launcher defaults to the older slow sanity-check generation path for now.
 # ==============================================================================
 
 set -xeuo pipefail
@@ -27,13 +30,10 @@ set -xeuo pipefail
 GLM47_FLASH_HF="${GLM47_FLASH_HF:-zai-org/GLM-4.7-Flash}"
 PROMPT="${PROMPT:-What is artificial intelligence?}"
 MAX_NEW_TOKENS="${MAX_NEW_TOKENS:-100}"
-COORDINATOR_HOST="${COORDINATOR_HOST:-127.0.0.1}"
 
 uv run python -m torch.distributed.run --nproc_per_node=8 \
-    examples/inference/text_generation.py \
+    examples/conversion/hf_to_megatron_generate_text.py \
     --hf_model_path "$GLM47_FLASH_HF" \
     --prompt "$PROMPT" \
     --max_new_tokens "$MAX_NEW_TOKENS" \
-    --tp 1 --pp 1 --ep 8 \
-    --use-coordinator \
-    --coordinator-host "$COORDINATOR_HOST"
+    --tp 1 --pp 1 --ep 8

examples/models/glm47/slurm_inference.sh

@@ -19,6 +19,9 @@
 # GLM-4.7 (MoE: 160 experts, top-8, ~358B params)
 # Requires at least 4 nodes (32 GPUs) with EP=32.
 #
+# Megatron inference engine support is not yet available for this architecture.
+# This launcher defaults to the older slow sanity-check generation path for now.
+#
 # Usage:
 #   1. Set CONTAINER_IMAGE and token exports; set CONTAINER_MOUNTS if needed
 #   2. Create logs/ if your Slurm setup requires the output directory to exist
@@ -87,12 +90,11 @@ echo "Warming uv cache"
 printf -v HF_MODEL_ARG "%q" "$HF_MODEL_ID"
 printf -v PROMPT_ARG "%q" "$PROMPT"
 
-CMD="uv run --no-sync python examples/inference/text_generation.py"
+CMD="uv run --no-sync python examples/conversion/hf_to_megatron_generate_text.py"
 CMD="$CMD --hf_model_path $HF_MODEL_ARG"
 CMD="$CMD --prompt $PROMPT_ARG"
 CMD="$CMD --max_new_tokens $MAX_NEW_TOKENS"
 CMD="$CMD --tp 1 --pp 1 --ep 32"
-CMD="$CMD --use-coordinator"
 
 echo "Executing: $CMD"