[bridge, model] Qwen 3.5 VL Bridge

examples/models/vlm/qwen35_vl/conversion.sh

@@ -0,0 +1,48 @@
+#!/usr/bin/env bash
+# Copyright (c) 2026, NVIDIA CORPORATION.  All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+# Workspace directory for checkpoints and results
+WORKSPACE=${WORKSPACE:-/workspace}
+MODEL_NAME=Qwen3.5-397B-A17B
+
+# Make sure to upgrade to transformers >= 5.2.0
+# uv add transformers>=5.2.0
+
+# Import HF → Megatron
+uv run python examples/conversion/convert_checkpoints.py import \
+    --hf-model Qwen/${MODEL_NAME} \
+    --megatron-path ${WORKSPACE}/models/${MODEL_NAME} \
+    --torch-dtype bfloat16
+
+# DEBUG: compare with the 4 layer model
+CUDA_LAUNCH_BLOCKING=1 NCCL_DEBUG=INFO NCCL_DEBUG_SUBSYS=ALL \
+uv run python -m torch.distributed.run --nproc_per_node=8 examples/conversion/compare_hf_and_megatron/compare.py \
+    --hf_model_path "/lustre/fs1/portfolios/coreai/users/chcui/mbridge_home/models/Qwen/Qwen3.5-4layer-HF" \
+    --megatron_model_path /lustre/fs1/portfolios/coreai/users/chcui/mbridge_home/models/Qwen/Qwen3.5-397B-A17B_4layer \
+    --model_class "Qwen3_5MoeForConditionalGeneration" \
+    --image_path "https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen-VL/assets/demo.jpeg" \
+    --prompt "Describe this image." \
+    --tp 2 --pp 1 --ep 4
+
+# Export Megatron → HF
+uv run python examples/conversion/convert_checkpoints.py export \
+    --hf-model Qwen/${MODEL_NAME} \
+    --megatron-path ${WORKSPACE}/models/${MODEL_NAME}/iter_0000000 \
+    --hf-path ${WORKSPACE}/models/${MODEL_NAME}-hf-export
+
+# Round-trip validation
+# Note: Qwen3.5 is a large MoE model, adjust parallelism as needed
+uv run python -m torch.distributed.run --nproc_per_node=8 examples/conversion/hf_megatron_roundtrip_multi_gpu.py \
+      --hf-model-id Qwen/${MODEL_NAME} --tp 1 --pp 2 --ep 4 --trust-remote-code

examples/models/vlm/qwen35_vl/slurm_inference.sh

@@ -0,0 +1,194 @@
+#!/bin/bash
+# Copyright (c) 2026, NVIDIA CORPORATION.  All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+# ==============================================================================
+# Qwen3.5-VL Multi-Node Distributed Inference
+#
+# Qwen3.5-397B-A17B is a large MoE model (397B parameters, 17B active)
+# Recommended: TP=2, PP=16, EP=64 for full model (1024 GPUs, 128 nodes)
+# For smaller setups: TP=1, PP=8, EP=16 (128 GPUs, 16 nodes)
+#
+# Usage:
+#   1. Modify the #SBATCH directives below for your cluster
+#   2. Set MODEL_PATH and CHECKPOINT_PATH as needed
+#   3. Set CONTAINER_IMAGE or use --no-container-image for bare metal
+#   4. Submit: sbatch slurm_inference.sh
+# ==============================================================================
+
+#SBATCH --job-name=qwen35v-inference
+#SBATCH --nodes=16                   # Number of nodes (128 GPUs = 16 nodes × 8 GPUs)
+#SBATCH --ntasks-per-node=8          # Tasks per node (1 per GPU)
+#SBATCH --gpus-per-node=8            # GPUs per node
+#SBATCH --time=02:00:00              # Max run time (2 hours)
+#SBATCH --partition=gpu              # Partition name
+#SBATCH --account=my_account         # Account name
+#SBATCH --output=logs/qwen35v_inference_%j.out
+#SBATCH --error=logs/qwen35v_inference_%j.err
+#SBATCH --exclusive                  # Exclusive node access
+
+# ==============================================================================
+# CONFIGURATION
+# ==============================================================================
+
+# Workspace directory
+WORKSPACE=${WORKSPACE:-/workspace}
+
+# Model configuration
+MODEL_NAME=Qwen3.5-397B-A17B
+
+# Option 1: Use HuggingFace model path (will load and convert on-the-fly)
+MODEL_PATH=${WORKSPACE}/models/Qwen/${MODEL_NAME}
+# MODEL_PATH=Qwen/${MODEL_NAME}  # Or use HF Hub path
+
+# Option 2: Use pre-converted Megatron checkpoint (faster)
+MEGATRON_CHECKPOINT=${WORKSPACE}/models/${MODEL_NAME}_megatron/iter_0000000
+# Comment out to use HF model directly
+# MEGATRON_CHECKPOINT=""
+
+# Inference configuration
+IMAGE_PATH="https://huggingface.co/nvidia/NVIDIA-Nemotron-Nano-12B-v2-VL-BF16/resolve/main/images/table.png"
+PROMPT="Describe this image."
+MAX_NEW_TOKENS=100
+
+# Parallelism configuration for 128 GPUs (16 nodes × 8 GPUs)
+# Constraint: TP × PP × EP = Total GPUs
+TP=4      # Tensor Parallelism
+PP=8      # Pipeline Parallelism
+EP=16     # Expert Parallelism (MoE)
+# Total: 1 × 8 × 16 = 128 GPUs
+
+# For 1024 GPUs (128 nodes), use:
+# TP=2, PP=16, EP=64  (2 × 16 × 64 = 2048 GPUs - too many)
+# TP=2, PP=16, EP=32  (2 × 16 × 32 = 1024 GPUs)
+
+# Container configuration (required for SLURM pyxis)
+CONTAINER_IMAGE=""
+# CONTAINER_IMAGE="/path/to/nemo-framework.sqsh"
+
+# Container mounts (optional, space-separated)
+CONTAINER_MOUNTS=""
+# CONTAINER_MOUNTS="/data:/data /workspace:/workspace"
+
+# Set to true to run without container (bare metal)
+NO_CONTAINER=false
+
+# ==============================================================================
+# Environment Setup
+# ==============================================================================
+
+# NCCL optimizations
+export TORCH_NCCL_AVOID_RECORD_STREAMS=1
+export NCCL_NVLS_ENABLE=0
+
+# UV cache on shared filesystem (recommended for multi-node setups)
+# Pre-sync once before submitting jobs: UV_CACHE_DIR=/path/to/cache uv sync
+# export UV_CACHE_DIR="/path/to/shared/uv_cache"
+
+# HuggingFace cache directory (recommended for shared filesystem)
+# export HF_HOME="/path/to/shared/HF_HOME"
+
+# Authentication tokens
+# export HF_TOKEN="hf_your_token_here"
+
+# For transformers >= 5.2.0 (required for Qwen3.5)
+# Run once: uv add "transformers>=5.2.0"
+
+# ==============================================================================
+# Job Execution
+# ==============================================================================
+
+echo "======================================"
+echo "Qwen3.5-VL Multi-Node Inference"
+echo "======================================"
+echo "Job ID: $SLURM_JOB_ID"
+echo "Nodes: $SLURM_JOB_NUM_NODES"
+echo "GPUs per node: $SLURM_GPUS_PER_NODE"
+echo "Total GPUs: $((SLURM_JOB_NUM_NODES * SLURM_GPUS_PER_NODE))"
+echo "Model: $MODEL_NAME"
+echo "Parallelism: TP=$TP, PP=$PP, EP=$EP"
+echo "======================================"
+
+# Create logs directory
+mkdir -p logs
+
+# Calculate total processes
+TOTAL_GPUS=$((SLURM_JOB_NUM_NODES * SLURM_GPUS_PER_NODE))
+REQUIRED_GPUS=$((TP * PP * EP))
+
+# Validate parallelism configuration
+if [ $REQUIRED_GPUS -ne $TOTAL_GPUS ]; then
+    echo "ERROR: Parallelism mismatch!"
+    echo "  TP × PP × EP = $TP × $PP × $EP = $REQUIRED_GPUS"
+    echo "  Total allocated GPUs = $TOTAL_GPUS"
+    echo "  These must be equal!"
+    exit 1
+fi
+
+# Build inference command
+CMD="uv run --no-sync python examples/conversion/hf_to_megatron_generate_vlm.py"
+CMD="$CMD --hf_model_path $MODEL_PATH"
+
+# Add Megatron checkpoint if specified
+if [ -n "$MEGATRON_CHECKPOINT" ]; then
+    CMD="$CMD --megatron_model_path $MEGATRON_CHECKPOINT"
+fi
+
+# Add inference parameters
+CMD="$CMD --image_path \"$IMAGE_PATH\""
+CMD="$CMD --prompt \"$PROMPT\""
+CMD="$CMD --max_new_tokens $MAX_NEW_TOKENS"
+
+# Add parallelism parameters
+CMD="$CMD --tp $TP"
+CMD="$CMD --pp $PP"
+CMD="$CMD --ep $EP"
+
+# Only rank 0 on each node runs uv sync
+SYNC_CMD="if [ \"\$SLURM_LOCALID\" -eq 0 ]; then uv sync; else sleep 5; fi"
+FULL_CMD="$SYNC_CMD && $CMD"
+
+echo "Executing inference..."
+echo "Command: $CMD"
+echo "======================================"
+
+# Execute based on container configuration
+if [ "$NO_CONTAINER" = true ]; then
+    echo "Running without container (bare metal)"
+    srun --mpi=pmix bash -c "$FULL_CMD"
+else
+    # Require container image
+    if [ -z "$CONTAINER_IMAGE" ]; then
+        echo "ERROR: CONTAINER_IMAGE must be set, or use NO_CONTAINER=true for bare metal."
+        exit 1
+    fi
+
+    echo "Running with container: $CONTAINER_IMAGE"
+
+    # Build srun command with container
+    SRUN_CMD="srun --mpi=pmix --container-image=$CONTAINER_IMAGE"
+
+    # Add container mounts
+    if [ -n "$CONTAINER_MOUNTS" ]; then
+        for mount in $CONTAINER_MOUNTS; do
+            SRUN_CMD="$SRUN_CMD --container-mounts=$mount"
+        done
+    fi
+
+    $SRUN_CMD bash -c "$FULL_CMD"
+fi
+
+echo "======================================"
+echo "Inference completed"
+echo "======================================"