Add Qwen3.5 model support (27B dense and 35B-A3B MoE)

- New model plugin: slime_plugins/models/qwen3_5.py
  - Qwen3_5GatedDeltaNet with separate QKV/Z projections, conv1d, and flat QKV split
  - get_qwen3_5_spec replacing standard attention with linear attention per layer_types

- New weight bridge: slime_plugins/mbridge/qwen3_5.py
  - Handles VLM weight prefix (model.language_model.layers)
  - Fused expert weight format for MoE (3D tensors -> per-expert slices)
  - MTP layer support with individual expert format

- New HF converter: slime/backends/megatron_utils/megatron_to_hf/qwen3_5.py
  - TEGroupedMLP per-expert weight{i} -> HF fused expert format
  - Proper gate/up split for swiglu experts

- Fix sglang_rollout.py: skip processor for text-only VLM models

- Model configs and run scripts for both 27B and 35B-A3B

Tested: Both models verified end-to-end with training.
- 27B: TP=1 SGLang (8 engines), TP=2/PP=2/CP=2 Megatron, logprob_diff=0.017
- 35B-A3B: TP=2 SGLang (4 engines), EP=8 Megatron, logprob_diff=0.012

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Author: Copilot

scripts/models/qwen3.5-27B.sh

@@ -0,0 +1,28 @@
+MODEL_ARGS=(
+   --spec "slime_plugins.models.qwen3_5" "get_qwen3_5_spec"
+
+   --disable-bias-linear
+   --qk-layernorm
+   --group-query-attention
+   --num-attention-heads 24
+   --num-query-groups 4
+   --kv-channels 256
+   --num-layers 64
+   --hidden-size 5120
+   --ffn-hidden-size 17408
+   --use-gated-attention
+
+   --normalization RMSNorm
+   --apply-layernorm-1p
+   --position-embedding-type rope
+   --norm-epsilon 1e-6
+   --rotary-percent 0.25
+   --swiglu
+   --untie-embeddings-and-output-weights
+   --vocab-size 248320
+
+   --rotary-base 10000000
+
+   # qwen3.5 specific
+   --attention-output-gate
+)

scripts/models/qwen3.5-35B-A3B.sh

@@ -0,0 +1,58 @@
+NLAYERS=40
+FIRST_K_DENSE_REPLACE=0
+
+arr=()
+for ((i=0; i<NLAYERS; i++)); do
+  if (( i < FIRST_K_DENSE_REPLACE )); then
+    arr+=(0)
+  else
+    arr+=(1)
+  fi
+done
+
+printf -v MOE_LAYER_FREQ "[%s]" "$(IFS=', '; echo "${arr[*]}")"
+
+
+MODEL_ARGS=(
+   --spec "slime_plugins.models.qwen3_5" "get_qwen3_5_spec"
+
+   --disable-bias-linear
+   --qk-layernorm
+   --group-query-attention
+   --num-attention-heads 16
+   --num-query-groups 2
+   --kv-channels 256
+   --num-layers 40
+   --hidden-size 2048
+   --ffn-hidden-size 512
+   --use-gated-attention
+
+   --normalization RMSNorm
+   --apply-layernorm-1p
+   --position-embedding-type rope
+   --norm-epsilon 1e-6
+   --rotary-percent 0.25
+   --swiglu
+   --untie-embeddings-and-output-weights
+   --vocab-size 248320
+
+   --rotary-base 10000000
+
+   # moe
+   --moe-ffn-hidden-size 512
+   --moe-shared-expert-intermediate-size 512
+   --moe-router-score-function softmax
+   --moe-token-dispatcher-type alltoall
+   --moe-router-topk 8
+   --moe-layer-freq $MOE_LAYER_FREQ
+   --num-experts 256
+   --moe-grouped-gemm
+   --moe-token-drop-policy probs
+   --moe-router-dtype fp32
+   --moe-permute-fusion
+   --moe-aux-loss-coeff 0
+
+   # qwen3.5 specific
+   --attention-output-gate
+   --moe-shared-expert-gate
+)

scripts/run-qwen3.5-27B.sh

@@ -0,0 +1,159 @@
+#!/bin/bash
+
+# for rerun the task
+pkill -9 sglang
+sleep 3
+ray stop --force
+pkill -9 ray
+pkill -9 python
+sleep 3
+pkill -9 ray
+pkill -9 python
+
+set -ex
+
+# will prevent ray from buffering stdout/stderr
+export PYTHONBUFFERED=16
+
+# unset proxy to avoid issues
+unset http_proxy https_proxy HTTP_PROXY HTTPS_PROXY
+
+NVLINK_COUNT=$(nvidia-smi topo -m 2>/dev/null | grep -o 'NV[0-9][0-9]*' | wc -l)
+if [ "$NVLINK_COUNT" -gt 0 ]; then
+    HAS_NVLINK=1
+else
+    HAS_NVLINK=0
+fi
+echo "HAS_NVLINK: $HAS_NVLINK (detected $NVLINK_COUNT NVLink references)"
+
+SCRIPT_DIR="$(cd -- "$(dirname -- "${BASH_SOURCE[0]}")" &>/dev/null && pwd)"
+source "${SCRIPT_DIR}/models/qwen3.5-27B.sh"
+
+CKPT_ARGS=(
+   --hf-checkpoint /root/Qwen3.5-27B
+   --ref-load /root/Qwen3.5-27B_torch_dist/
+   --load /root/Qwen3.5-27B_slime
+   --save /root/Qwen3.5-27B_slime
+   --save-interval 20
+)
+
+ROLLOUT_ARGS=(
+   --prompt-data /root/dapo-math-17k/dapo-math-17k.jsonl
+   --input-key prompt
+   --label-key label
+   --apply-chat-template
+   --rollout-shuffle
+   --rm-type deepscaler
+   --num-rollout 3000
+   --rollout-batch-size 32
+   --n-samples-per-prompt 8
+   --rollout-max-response-len 8192
+   --rollout-temperature 1
+
+   --global-batch-size 256
+   --balance-data
+)
+
+EVAL_ARGS=(
+   --eval-interval 20
+   --eval-prompt-data aime /root/aime-2024/aime-2024.jsonl
+   --n-samples-per-eval-prompt 16
+   --eval-max-response-len 16384
+   --eval-top-p 1
+)
+
+PERF_ARGS=(
+   --tensor-model-parallel-size 4
+   --sequence-parallel
+   --pipeline-model-parallel-size 1
+   --context-parallel-size 1
+   --expert-model-parallel-size 1
+   --expert-tensor-parallel-size 1
+
+   --recompute-granularity full
+   --recompute-method uniform
+   --recompute-num-layers 1
+
+   # --micro-batch-size 1
+   --use-dynamic-batch-size
+   --max-tokens-per-gpu 20480
+)
+
+GRPO_ARGS=(
+   --advantage-estimator grpo
+   --use-kl-loss
+   --kl-loss-coef 0.00
+   --kl-loss-type low_var_kl
+   --entropy-coef 0.00
+   --eps-clip 0.2
+   --eps-clip-high 0.28
+)
+
+OPTIMIZER_ARGS=(
+   --optimizer adam
+   --lr 1e-6
+   --lr-decay-style constant
+   --weight-decay 0.1
+   --adam-beta1 0.9
+   --adam-beta2 0.98
+
+   --optimizer-cpu-offload
+   --overlap-cpu-optimizer-d2h-h2d
+   --use-precision-aware-optimizer
+)
+
+WANDB_ARGS=(
+   #--use-wandb
+   # --wandb-project slime-dev
+   # --wandb-group qwen3.5-27B-test
+   # --wandb-key ${WANDB_KEY}
+)
+
+SGLANG_ARGS=(
+   --rollout-num-gpus-per-engine 8
+   --sglang-mem-fraction-static 0.7
+   --sglang-cuda-graph-bs 1 2 4 8 $(seq 16 8 256)
+)
+
+MISC_ARGS=(
+   # default dropout in megatron is 0.1
+   --attention-dropout 0.0
+   --hidden-dropout 0.0
+   # should be good for model performance
+   --accumulate-allreduce-grads-in-fp32
+   --attention-softmax-in-fp32
+   # need to comment this when using model with MLA
+   --attention-backend flash
+)
+
+# launch the master node of ray in container
+export MASTER_ADDR=${MASTER_ADDR:-"127.0.0.1"}
+export no_proxy="127.0.0.1,${MASTER_ADDR}"
+ray start --head --node-ip-address ${MASTER_ADDR} --num-gpus 8 --disable-usage-stats --dashboard-host=0.0.0.0 --dashboard-port=8265
+
+# Build the runtime environment JSON with proper variable substitution
+RUNTIME_ENV_JSON="{
+  \"env_vars\": {
+    \"PYTHONPATH\": \"/root/Megatron-LM/\",
+    \"CUDA_DEVICE_MAX_CONNECTIONS\": \"1\",
+    \"NCCL_NVLS_ENABLE\": \"${HAS_NVLINK}\",
+    \"no_proxy\": \"${no_proxy}\"
+  }
+}"
+
+ray job submit --address="http://127.0.0.1:8265" \
+   --runtime-env-json="${RUNTIME_ENV_JSON}" \
+   -- python3 train.py \
+   --actor-num-nodes 1 \
+   --actor-num-gpus-per-node 8 \
+   --colocate \
+   ${MODEL_ARGS[@]} \
+   ${CKPT_ARGS[@]} \
+   ${ROLLOUT_ARGS[@]} \
+   ${OPTIMIZER_ARGS[@]} \
+   ${GRPO_ARGS[@]} \
+   ${WANDB_ARGS[@]} \
+   ${PERF_ARGS[@]} \
+   ${EVAL_ARGS[@]} \
+   ${SGLANG_ARGS[@]} \
+   ${MISC_ARGS[@]}