Repository navigation
[AMD][MI35X] Add qwen3.8-flash-next-fp8-mi355x-sglang-agentic-mtp on v0.5.19-rocm700-mi35x-20260909 #2924
There are no files selected for viewing
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,348 @@ | ||
| #!/usr/bin/env bash | ||
| set -euo pipefail | ||
| set -x | ||
|
|
||
| # Agentic trace replay benchmark for Qwen3.8-Flash-Next FP8 on MI355X using | ||
| # SGLang with MTP speculative decoding. Day-zero recipe for the hybrid | ||
| # GDN + QSA architecture on CDNA4 (gfx950). | ||
| # | ||
| # MI355X is ROCm, so this arm is FP8 (Qwen/Qwen3.8-Flash-Next-FP8). | ||
| # NVFP4 is not available on AMD GPUs. Attention backend is aiter (ROCm). | ||
| # | ||
| # The SGLang cookbook (docs.sglang.io) verified flags for MI355X are: | ||
| # --tp-size 8 --attention-backend aiter --page-size 32 --kv-cache-dtype auto | ||
| # --chunked-prefill-size 16384 --watchdog-timeout 1200 --mem-fraction-static 0.9 | ||
| # | ||
| # Speculative decoding uses NEXTN (the Qwen3.8 Flash Next native MTP head), | ||
| # 3 steps, eagle-topk 1, 4 draft tokens — mirroring the H200 agentic recipe. | ||
| # | ||
| # Required env vars: | ||
| # MODEL, TP, CONC, EP_SIZE, KV_OFFLOADING, | ||
| # TOTAL_CPU_DRAM_GB, RESULT_DIR, DURATION | ||
|
|
||
| source "$(dirname "$0")/../../benchmark_lib.sh" | ||
|
|
||
| export EVAL_FRAMEWORK="lm-eval" | ||
|
|
||
| check_env_vars \ | ||
| MODEL TP CONC EP_SIZE KV_OFFLOADING \ | ||
| TOTAL_CPU_DRAM_GB RESULT_DIR DURATION | ||
|
|
||
| SCHEDULER_RECV_INTERVAL=${SCHEDULER_RECV_INTERVAL:-30} | ||
|
|
||
| if [[ -n "${SLURM_JOB_ID:-}" ]]; then | ||
| echo "JOB $SLURM_JOB_ID running on ${SLURMD_NODENAME:-unknown}" | ||
| fi | ||
|
|
||
| if [[ -n "${MODEL_PATH:-}" ]]; then | ||
| if [[ ! -d "$MODEL_PATH" || -z "$(ls -A "$MODEL_PATH" 2>/dev/null)" ]]; then | ||
| hf download "$MODEL" --local-dir "$MODEL_PATH" | ||
| fi | ||
| else | ||
| hf download "$MODEL" | ||
| export MODEL_PATH="$MODEL" | ||
| fi | ||
|
|
||
| rocm-smi || true | ||
| amd-smi || true | ||
|
|
||
| export WEKA_LOADER_OVERRIDE=semianalysis_cc_traces_weka_with_subagents_256k | ||
| resolve_trace_source | ||
| install_agentic_deps | ||
|
|
||
| export AIPERF_SERVER_METRICS_URLS="http://localhost:${PORT}/metrics" | ||
| export AIPERF_REQUIRED_SERVER_METRIC_PREFIX="sglang:" | ||
|
|
||
| SERVER_LOG="$RESULT_DIR/server.log" | ||
| mkdir -p "$RESULT_DIR" | ||
|
|
||
| SERVER_PID="" | ||
| cleanup_agentic_services() { | ||
| local exit_code=$? | ||
| trap - EXIT INT TERM | ||
| set +e | ||
| stop_background_process_tree "$SERVER_PID" "SGLang server" 60 | ||
| exit "$exit_code" | ||
| } | ||
| trap cleanup_agentic_services EXIT | ||
| trap 'exit 130' INT | ||
| trap 'exit 143' TERM | ||
|
|
||
| # ---- HiCache (host-DRAM KV offload) ---------------------------------------- | ||
| CACHE_ARGS=() | ||
| if require_agentic_kv_offload_backend hicache; then | ||
| REQUESTED_HICACHE_TOTAL_GB="${HICACHE_TOTAL_CPU_DRAM_GB:-$TOTAL_CPU_DRAM_GB}" | ||
| if [ "$REQUESTED_HICACHE_TOTAL_GB" -gt "$TOTAL_CPU_DRAM_GB" ]; then | ||
| echo "Error: requested HiCache pool ${REQUESTED_HICACHE_TOTAL_GB} GB exceeds configured capacity ${TOTAL_CPU_DRAM_GB} GB" >&2 | ||
| exit 1 | ||
| fi | ||
| TOTAL_CPU_DRAM_GB="$REQUESTED_HICACHE_TOTAL_GB" | ||
| HICACHE_HOST_POOL_COUNT="${HICACHE_HOST_POOL_COUNT:-2}" | ||
| HICACHE_WRITE_POLICY="${HICACHE_WRITE_POLICY:-write_through_selective}" | ||
| MAX_HICACHE_SIZE_GB=$((TOTAL_CPU_DRAM_GB / TP / HICACHE_HOST_POOL_COUNT)) | ||
| HICACHE_SIZE_GB="${HICACHE_SIZE_GB:-$MAX_HICACHE_SIZE_GB}" | ||
| if [ "$HICACHE_SIZE_GB" -gt "$MAX_HICACHE_SIZE_GB" ]; then | ||
| echo "Error: HICACHE_SIZE_GB=$HICACHE_SIZE_GB exceeds configured per-pool limit $MAX_HICACHE_SIZE_GB" >&2 | ||
| exit 1 | ||
| fi | ||
| if [ "$HICACHE_SIZE_GB" -lt 1 ]; then | ||
| echo "Error: computed HICACHE_SIZE_GB=$HICACHE_SIZE_GB from TOTAL_CPU_DRAM_GB=$TOTAL_CPU_DRAM_GB, TP=$TP, HICACHE_HOST_POOL_COUNT=$HICACHE_HOST_POOL_COUNT" >&2 | ||
| exit 1 | ||
| fi | ||
| echo "HiCache CPU pool: ${HICACHE_SIZE_GB} GB per rank per host pool across TP=${TP}, host_pool_count=${HICACHE_HOST_POOL_COUNT}" | ||
| CACHE_ARGS=( | ||
| --enable-hierarchical-cache | ||
| --hicache-size "$HICACHE_SIZE_GB" | ||
| --hicache-io-backend kernel | ||
| --hicache-mem-layout page_first | ||
| --hicache-write-policy "$HICACHE_WRITE_POLICY" | ||
| ) | ||
| fi | ||
|
|
||
| # ---- Parallelism ------------------------------------------------------------ | ||
| PARALLEL_ARGS=( | ||
| --tp-size "$TP" | ||
| --dp-size 1 | ||
| --ep-size "$EP_SIZE" | ||
| ) | ||
|
|
||
| TOKENIZER_ARGS=() | ||
| if [ "$TP" -ge 4 ]; then | ||
| TOKENIZER_ARGS=(--tokenizer-worker-num 6) | ||
| fi | ||
|
|
||
| MAX_RUNNING_REQUESTS=$((2 * CONC)) | ||
| CUDA_GRAPH_MAX_BS="$CONC" | ||
| [ "$CUDA_GRAPH_MAX_BS" -gt 128 ] && CUDA_GRAPH_MAX_BS=128 | ||
|
|
||
| # ---- SGLang runtime patches (FP8 PLE weight_scale — SGLang PR #36497) ------- | ||
| # The qwen38flashnext Docker image predates the FP8 PLE weight_scale fix. | ||
| # Rather than rebuilding the image we apply 5 targeted in-place patches to the | ||
| # container's sglang source before launching the server. | ||
| # | ||
| # Patched file: /workspace/sglang-qwen-next/python/sglang/srt/models/qwen4_exp.py | ||
| # Commit ref: sgl-project/sglang@3003ddf1574ef5004e21a10e36aaabc364766921 | ||
| # | ||
| # The patches are idempotent — safe to run multiple times, already-applied | ||
| # patches are skipped. | ||
| _SGLANG_ROOT="${SGLANG_ROOT:-/workspace/sglang-qwen-next/python/sglang}" | ||
| _QWEN4_EXP="${_SGLANG_ROOT}/srt/models/qwen4_exp.py" | ||
|
|
||
| if [[ -f "$_QWEN4_EXP" ]]; then | ||
| echo "Applying SGLang runtime patches to $_QWEN4_EXP ..." | ||
|
|
||
| # Patch 1: Register weight_scale buffer in Qwen4ExpNGramEmbedding.__init__ | ||
| if ! grep -q 'weight_scale.*torch.ones' "$_QWEN4_EXP"; then | ||
| echo " PATCH 1: Registering weight_scale buffer in Qwen4ExpNGramEmbedding" | ||
| python3 -c " | ||
| import sys | ||
| with open('$_QWEN4_EXP', 'r') as f: | ||
| content = f.read() | ||
| old = 'self.short_conv_dilation = self.ple_embedding.ngram_size' | ||
| if old not in content: | ||
| print(' PATCH 1: Target line not found, skipping', file=sys.stderr) | ||
| sys.exit(0) | ||
| new = '''self.ple_embedding.ngram_embedding.register_buffer( | ||
| \"weight_scale\", torch.ones(1, dtype=torch.bfloat16), persistent=True | ||
| ) | ||
| self.short_conv_dilation = self.ple_embedding.ngram_size''' | ||
| content = content.replace(old, new, 1) | ||
| with open('$_QWEN4_EXP', 'w') as f: | ||
| f.write(content) | ||
| print(' PATCH 1: Applied successfully') | ||
| " | ||
| else | ||
| echo " PATCH 1: Already applied" | ||
| fi | ||
|
|
||
| # Patch 2: Multiply embeddings by weight_scale in forward() | ||
| if ! grep -q 'embeddings \* self.ngram_embedding.weight_scale' "$_QWEN4_EXP"; then | ||
| echo " PATCH 2: Adding weight_scale multiply in forward()" | ||
| python3 -c " | ||
| import sys | ||
| with open('$_QWEN4_EXP', 'r') as f: | ||
| content = f.read() | ||
| old = 'embeddings = self.ngram_embedding(lookup_ids)\n return self._finish_embedding_lookup(' | ||
| if old not in content: | ||
| print(' PATCH 2: Target pattern not found, skipping', file=sys.stderr) | ||
| sys.exit(0) | ||
| new = '''embeddings = self.ngram_embedding(lookup_ids) | ||
| embeddings = embeddings * self.ngram_embedding.weight_scale | ||
| return self._finish_embedding_lookup(''' | ||
| content = content.replace(old, new, 1) | ||
| with open('$_QWEN4_EXP', 'w') as f: | ||
| f.write(content) | ||
| print(' PATCH 2: Applied successfully') | ||
| " | ||
| else | ||
| echo " PATCH 2: Already applied" | ||
| fi | ||
|
|
||
| # Patch 3: Add weight_scale to PLE buffer allowlist in _load_qwen4_exp_ple_buffer | ||
| if ! grep -q '"weight_scale"' "$_QWEN4_EXP" 2>/dev/null || \ | ||
| ! python3 -c " | ||
| with open('$_QWEN4_EXP') as f: | ||
| content = f.read() | ||
| assert '\"weight_scale\",' in content and 'ngram_heads_vocab_sizes' in content | ||
| idx_ws = content.index('\"weight_scale\",') | ||
| idx_nv = content.index('\"ngram_heads_vocab_sizes\"') | ||
| assert abs(idx_ws - idx_nv) < 200 | ||
| " 2>/dev/null; then | ||
| echo " PATCH 3: Adding weight_scale to PLE buffer allowlist" | ||
| python3 -c " | ||
| import sys | ||
| with open('$_QWEN4_EXP', 'r') as f: | ||
| content = f.read() | ||
| old = '''\"ngram_heads_vocab_sizes\", | ||
| }:''' | ||
| if old not in content: | ||
| print(' PATCH 3: Target pattern not found, skipping', file=sys.stderr) | ||
| sys.exit(0) | ||
| new = '''\"ngram_heads_vocab_sizes\", | ||
| \"weight_scale\", | ||
| }:''' | ||
| content = content.replace(old, new, 1) | ||
| with open('$_QWEN4_EXP', 'w') as f: | ||
| f.write(content) | ||
| print(' PATCH 3: Applied successfully') | ||
| " | ||
| else | ||
| echo " PATCH 3: Already applied" | ||
| fi | ||
|
|
||
| # Patch 4: weight_scale multiply in offloaded PLE reduce path | ||
| if ! grep -q 'ngram_embedding.reduce(embeddings)\n.*\* self.ple_embedding.ngram_embedding.weight_scale' "$_QWEN4_EXP" 2>/dev/null; then | ||
|
Contributor
There was a problem hiding this comment. 🟡 (optional) Patch 4's "already applied" pre-check uses Extended reasoning...Every invocation of the script takes the Verification: nit. Line 214: |
||
| echo " PATCH 4: Adding weight_scale multiply in offloaded PLE reduce path" | ||
| python3 -c " | ||
| import sys | ||
| with open('$_QWEN4_EXP', 'r') as f: | ||
| content = f.read() | ||
| old = 'embeddings = self.ple_embedding.ngram_embedding.reduce(embeddings)\n embeddings = self.ple_embedding._finish_embedding_lookup(' | ||
| if old not in content: | ||
| print(' PATCH 4: Target pattern not found, skipping', file=sys.stderr) | ||
| sys.exit(0) | ||
| new = '''embeddings = self.ple_embedding.ngram_embedding.reduce(embeddings) | ||
| embeddings = embeddings * self.ple_embedding.ngram_embedding.weight_scale | ||
| embeddings = self.ple_embedding._finish_embedding_lookup(''' | ||
| content = content.replace(old, new, 1) | ||
| with open('$_QWEN4_EXP', 'w') as f: | ||
| f.write(content) | ||
| print(' PATCH 4: Applied successfully') | ||
| " | ||
| else | ||
| echo " PATCH 4: Already applied" | ||
| fi | ||
|
|
||
| # Patch 5: Register weight_scale in Qwen4ExpPinnedHostEmbedding | ||
| if ! grep -q 'register_buffer.*"weight_scale".*embedding.weight_scale' "$_QWEN4_EXP" 2>/dev/null; then | ||
| echo " PATCH 5: Registering weight_scale in Qwen4ExpPinnedHostEmbedding" | ||
| python3 -c " | ||
| import sys | ||
| with open('$_QWEN4_EXP', 'r') as f: | ||
| content = f.read() | ||
| old = 'cpu_weight.weight_loader = self.weight_loader\n self.register_parameter(\"weight\", cpu_weight)' | ||
| if old not in content: | ||
| print(' PATCH 5: Target pattern not found, skipping', file=sys.stderr) | ||
| sys.exit(0) | ||
| new = '''cpu_weight.weight_loader = self.weight_loader | ||
| self.register_parameter(\"weight\", cpu_weight) | ||
| self.register_buffer(\"weight_scale\", embedding.weight_scale, persistent=True)''' | ||
| content = content.replace(old, new, 1) | ||
| with open('$_QWEN4_EXP', 'w') as f: | ||
| f.write(content) | ||
| print(' PATCH 5: Applied successfully') | ||
| " | ||
| else | ||
| echo " PATCH 5: Already applied" | ||
| fi | ||
|
|
||
| echo "All SGLang runtime patches applied." | ||
| else | ||
| echo "WARNING: qwen4_exp.py not found at $_QWEN4_EXP; running with container defaults" >&2 | ||
| fi | ||
|
|
||
| # ---- ROCm / aiter environment ----------------------------------------------- | ||
| export PYTHONNOUSERSITE=1 | ||
| export SGLANG_USE_AITER=1 | ||
| export SGLANG_USE_AITER_UNIFIED_ATTN=1 | ||
| export AITER_FLYDSL_FORCE=1 | ||
| export ROCM_QUICK_REDUCE_QUANTIZATION=INT8 | ||
| export SGLANG_TIMEOUT_KEEP_ALIVE=1800 | ||
| export SGLANG_ENABLE_SPEC_V2=1 | ||
|
|
||
| # ---- Speculative decoding (NEXTN / MTP) ------------------------------------- | ||
| SPEC_ARGS=( | ||
| --speculative-algorithm NEXTN | ||
| --speculative-num-steps 3 | ||
| --speculative-eagle-topk 1 | ||
| --speculative-num-draft-tokens 4 | ||
| ) | ||
|
|
||
| # golden_al_distribution/qwen3.8next_mtp.yaml: | ||
| # qwen3.8-flash-next-fp8.thinking_on[3] = 2.32. | ||
| # AgentX replays run with thinking on, so the thinking_on row is the right one. | ||
| if [ "${EVAL_ONLY:-false}" != "true" ]; then | ||
| export SGLANG_SIMULATE_ACC_LEN=2.32 | ||
| export SGLANG_SIMULATE_ACC_METHOD=match-expected | ||
| export SGLANG_SIMULATE_ACC_TOKEN_MODE=real-draft-token | ||
| fi | ||
|
|
||
| # ---- multi_tokenizer cached_tokens_details patch ---------------------------- | ||
| SGLANG_MULTI_TOKENIZER=/sgl-workspace/sglang/python/sglang/srt/managers/multi_tokenizer_mixin.py | ||
| if [ -f "$SGLANG_MULTI_TOKENIZER" ]; then | ||
| if ! sed -n '/elif isinstance(output, BatchStrOutput):/,/input_token_logprobs_val=_extract_field_by_index/p' "$SGLANG_MULTI_TOKENIZER" \ | ||
| | grep -q 'cached_tokens_details=_extract_field_by_index'; then | ||
| sed -i '/elif isinstance(output, BatchStrOutput):/,/cached_tokens=_extract_field_by_index(output, "cached_tokens", i),/ { | ||
| /cached_tokens=_extract_field_by_index(output, "cached_tokens", i),/a\ | ||
| cached_tokens_details=_extract_field_by_index(\ | ||
| output, "cached_tokens_details", i\ | ||
| ), | ||
| }' "$SGLANG_MULTI_TOKENIZER" | ||
| fi | ||
| fi | ||
|
|
||
| # ---- Launch server ----------------------------------------------------------- | ||
| SGLANG_CMD=( | ||
| python3 -m sglang.launch_server | ||
| --model-path "$MODEL_PATH" | ||
| --served-model-name "$MODEL" | ||
| --host 0.0.0.0 | ||
| --port "$PORT" | ||
| --trust-remote-code | ||
| "${PARALLEL_ARGS[@]}" | ||
| --attention-backend aiter | ||
| --page-size 32 | ||
| --kv-cache-dtype auto | ||
| --mem-fraction-static 0.90 | ||
| --model-loader-extra-config '{"enable_multithread_load": true}' | ||
| --watchdog-timeout 1200 | ||
| --chunked-prefill-size 16384 | ||
| --mamba-ssm-dtype bfloat16 | ||
| --cuda-graph-max-bs "$CUDA_GRAPH_MAX_BS" | ||
| --max-running-requests "$MAX_RUNNING_REQUESTS" | ||
| --scheduler-recv-interval "$SCHEDULER_RECV_INTERVAL" | ||
| --stream-interval 50 | ||
| "${TOKENIZER_ARGS[@]}" | ||
| --tokenizer-path "$MODEL" | ||
| --reasoning-parser qwen3 | ||
| --tool-call-parser qwen3_coder | ||
| "${SPEC_ARGS[@]}" | ||
| --enable-metrics | ||
| --enable-cache-report | ||
| "${CACHE_ARGS[@]}" | ||
| ) | ||
|
|
||
| printf '%q ' "${SGLANG_CMD[@]}" | tee "$RESULT_DIR/sglang_command.txt" | ||
| printf '\n' | tee -a "$RESULT_DIR/sglang_command.txt" | ||
| "${SGLANG_CMD[@]}" > "$SERVER_LOG" 2>&1 & | ||
| SERVER_PID=$! | ||
|
|
||
| wait_for_server_ready --port "$PORT" --server-log "$SERVER_LOG" --server-pid "$SERVER_PID" | ||
|
|
||
| if [ "${EVAL_ONLY:-false}" = "true" ]; then | ||
| run_eval --port "$PORT" | ||
| else | ||
| build_replay_cmd "$RESULT_DIR" | ||
| REPLAY_CMD+=" --apply-chat-template" | ||
| run_agentic_replay_and_write_outputs "$RESULT_DIR" | ||
| fi | ||
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
|
|
@@ -443,6 +443,20 @@ qwen3.5-fp8-mi300x-sglang-agentic-mtp: | |
| - { tp: 8, ep: 1, spec-decoding: mtp, kv-offloading: none, conc-list: [4, 16, 20, 24, 32] } | ||
| - { tp: 8, ep: 8, spec-decoding: mtp, kv-offloading: none, conc-list: [4, 16, 20, 24, 32] } | ||
|
|
||
| qwen3.8-flash-next-fp8-mi355x-sglang-agentic-mtp: | ||
| image: lmsysorg/sglang-rocm:qwen38flashnext | ||
| model: Qwen/Qwen3.8-Flash-Next-FP8 | ||
| model-prefix: qwen3.8-flash-next | ||
|
There was a problem hiding this comment. Wrong model-prefix breaks script lookupHigh Severity
Additional Locations (1)  Reviewed by Cursor Bugbot for commit 4dc6c45. Configure here. |
||
| runner: cluster:mi355x-amds | ||
| precision: fp8 | ||
| framework: sglang | ||
| multinode: false | ||
| scenarios: | ||
| agentic-coding: | ||
| - dram-utilization: 0.80 | ||
| search-space: | ||
| - { tp: 4, ep: 1, spec-decoding: mtp, kv-offloading: none, conc-list: [1, 4, 8, 12, 16, 20, 24, 28, 32] } | ||
|
|
||
| dsr1-fp8-mi355x-atom: | ||
| image: rocm/atom:rocm7.2.3_ubuntu24.04_py3.12_pytorch_release_2.10.0_atom20260511 | ||
| model: deepseek-ai/DeepSeek-R1-0528 | ||
|
|
||
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
🔴 This script rewrites the pinned image's installed sglang source (5 python3 heredocs patching qwen4_exp.py) before launch, which docs/PR_REVIEW_CHECKLIST.md explicitly bars except with a filled-out docs/waiver/<PR_NUMBER>.md; no docs/waiver/2924.md is included in this PR (only pre-existing docs/waiver/2508.md exists). Merging as-is means the pinned lmsysorg/sglang-rocm:qwen38flashnext image never actually runs as shipped, silently defeating the checklist's engine-integrity guarantee for anyone reviewing/trusting this recipe. Fix: add docs/waiver/2924.md stating what is patched, why the unmodified image can't run this benchmark, the upstream PR link, and a removal plan, or drop the inline patches and wait for a fixed image.
Extended reasoning...
docs/PR_REVIEW_CHECKLIST.md line 30 (checked against docs/PR_REVIEW_CHECKLIST.md in this repo) requires: 'this PR does not patch the inference engine or serving stack ... The only exception is a patch covered by a filled-out waiver at docs/waiver/<PR_NUMBER>.md'. This PR's script applies 5 python3-heredoc patches plus a sed patch to /workspace/sglang-qwen-next/.../qwen4_exp.py and /sgl-workspace/sglang/.../multi_tokenizer_mixin.py before sglang.launch_server runs, i.e. exactly the patched-inference-engine pattern the checklist calls out by name. The diff includes no docs/waiver/2924.md (verified via ls docs/waiver, which only contains 2508.md unrelated to this PR/model). The PR's own changelog text acknowledges these are stopgap patches rather than pointing to a waiver, so the required sign-off artifact is missing and the resulting benchmark numbers are produced by a modified engine, not the pinned image as shipped.
Verification: normal (conflicts with stated purpose: the patching is the PR's mechanism, but the checklist requires it be gated on a waiver, which fixing does not undo). The script rewrites the pinned image's installed engine sources before launch — five python3 heredocs (lines 137-256) that read/replace/write ${_SGLANG_ROOT}/srt/models/qwen4_exp.py (default… | normal. The script rewrites installed SGLang…