docs/source/blogs/tech_blog/blog4_Scaling_Expert_Parallelism_in_TensorRT-LLM.md (2)

624-636: ⚠️ Potential issue | 🔴 Critical

CRITICAL: Coding guideline violation - prefer --config over --extra_llm_api_options.

This EPLB benchmark command also uses --extra_llm_api_options with trtllm-bench, violating the same coding guideline.

As per coding guidelines: "When documenting CLI commands for TensorRT-LLM tools like trtllm-serve, trtllm-bench, trtllm-eval, prefer using --config over --extra_llm_api_options for specifying configuration files."

📋 Proposed fix to align with coding guidelines

-cat > ./extra_llm_api_options_eplb.yaml <<EOF
+cat > ./config_eplb.yaml <<EOF
 enable_attention_dp: true
 moe_config:
   load_balancer: ./moe_load_balancer.yaml
 EOF

 trtllm-llmapi-launch \
 trtllm-bench --model ${MODEL_NAME} \
     --model_path ${MODEL_PATH} \
     throughput \
     --tp 36 \
     --ep 36 \
-    --extra_llm_api_options ./extra_llm_api_options_eplb.yaml \
+    --config ./config_eplb.yaml \
     --kv_cache_free_gpu_mem_fraction 0.75 \
     --backend pytorch \
     --dataset ./dataset.json \
     --warmup 0 \
     --eos_id -1

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In
`@docs/source/blogs/tech_blog/blog4_Scaling_Expert_Parallelism_in_TensorRT-LLM.md`
around lines 624 - 636, Replace the use of the deprecated flag in the documented
benchmark invocation: in the trtllm-bench command that currently passes
--extra_llm_api_options ./extra_llm_api_options_eplb.yaml, change it to use
--config ./extra_llm_api_options_eplb.yaml so the invocation follows the coding
guideline; update the surrounding example lines where trtllm-bench and the
extra_llm_api_options_eplb.yaml token appear to reflect the --config flag
consistently.

---

544-554: ⚠️ Potential issue | 🔴 Critical

CRITICAL: Coding guideline violation - prefer --config over --extra_llm_api_options.

This baseline benchmark command uses --extra_llm_api_options with trtllm-bench, violating the coding guidelines.

As per coding guidelines: "When documenting CLI commands for TensorRT-LLM tools like trtllm-serve, trtllm-bench, trtllm-eval, prefer using --config over --extra_llm_api_options for specifying configuration files."

📋 Proposed fix to align with coding guidelines

-cat > ./extra_llm_api_options.yaml <<EOF
+cat > ./config.yaml <<EOF
 enable_attention_dp: true
 EOF

 trtllm-llmapi-launch \
 trtllm-bench --model ${MODEL_NAME} \
     --model_path ${MODEL_PATH} \
     throughput \
     --tp 32 \
     --ep 32 \
-    --extra_llm_api_options ./extra_llm_api_options.yaml \
+    --config ./config.yaml \
     --kv_cache_free_gpu_mem_fraction 0.75 \
     --backend pytorch \
     --dataset ./dataset.json \
     --warmup 0 \
     --eos_id -1

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In
`@docs/source/blogs/tech_blog/blog4_Scaling_Expert_Parallelism_in_TensorRT-LLM.md`
around lines 544 - 554, The CLI example uses the deprecated flag
--extra_llm_api_options with trtllm-bench; change it to use the canonical
--config flag and update the referenced YAML filename accordingly (e.g., create
a config file like extra_llm_api_options.yaml or rename it to
extra_llm_config.yaml) and pass it to trtllm-bench as --config <file>, ensuring
the surrounding invocation (trtllm-llmapi-launch and trtllm-bench --model
${MODEL_NAME} --model_path ${MODEL_PATH} throughput --tp 32 --ep 32) remains
unchanged except for replacing --extra_llm_api_options
./extra_llm_api_options.yaml with --config ./<updated_filename>.yaml so the
example conforms to the coding guideline.

docs/source/blogs/tech_blog/blog2_DeepSeek_R1_MTP_Implementation_and_Optimization.md (2)

125-145: ⚠️ Potential issue | 🔴 Critical

CRITICAL: Coding guideline violation - prefer --config over --extra_llm_api_options.

This section uses --extra_llm_api_options with trtllm-bench, but the repository coding guidelines explicitly state: "When documenting CLI commands for TensorRT-LLM tools like trtllm-serve, trtllm-bench, trtllm-eval, prefer using --config over --extra_llm_api_options for specifying configuration files."

The YAML filename and flag should be reverted to use --config and config.yml respectively.

As per coding guidelines: "When documenting CLI commands for TensorRT-LLM tools like trtllm-serve, trtllm-bench, trtllm-eval, prefer using --config over --extra_llm_api_options for specifying configuration files."

📋 Proposed fix to align with coding guidelines

-cat >./extra-llm-api-config.yml<<EOF
+cat >./config.yml<<EOF
 cuda_graph_config: {}
 moe_config:
   backend: TRTLLM
 speculative_config:
     decoding_type: MTP
     num_nextn_predict_layers: 3
 EOF

 export TRTLLM_ENABLE_PDL=1

 trtllm-bench --model nvidia/DeepSeek-R1-FP4 \
     throughput \
     --dataset $YOUR_DATA_PATH \
     --backend pytorch \
     --num_requests 10 \
     --concurrency 1 \
     --max_batch_size 1 \
     --tp 8 \
     --ep 2 \
-    --extra_llm_api_options ./extra-llm-api-config.yml
+    --config ./config.yml

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In
`@docs/source/blogs/tech_blog/blog2_DeepSeek_R1_MTP_Implementation_and_Optimization.md`
around lines 125 - 145, The CLI example uses the deprecated flag and filename;
replace the --extra_llm_api_options flag and extra-llm-api-config.yml with the
canonical --config flag and config.yml: create the YAML as config.yml
(previously extra-llm-api-config.yml) and update the trtllm-bench invocation to
use --config ./config.yml; locate the trtllm-bench command and the mention of
--extra_llm_api_options in the file and change those symbols accordingly.

---

181-204: ⚠️ Potential issue | 🔴 Critical

CRITICAL: Coding guideline violation - prefer --config over --extra_llm_api_options.

This section also uses --extra_llm_api_options with trtllm-bench, violating the same coding guideline as the previous section.

As per coding guidelines: "When documenting CLI commands for TensorRT-LLM tools like trtllm-serve, trtllm-bench, trtllm-eval, prefer using --config over --extra_llm_api_options for specifying configuration files."

📋 Proposed fix to align with coding guidelines

-cat >./extra-llm-api-config.yml<<EOF
+cat >./config.yml<<EOF
 cuda_graph_config: {}
 moe_config:
   backend: TRTLLM
 speculative_config:
     decoding_type: MTP
     num_nextn_predict_layers: 3
     use_relaxed_acceptance_for_thinking: true
     relaxed_topk: 10
     relaxed_delta: 0.6
 EOF

 export TRTLLM_ENABLE_PDL=1

 trtllm-bench --model nvidia/DeepSeek-R1-FP4 \
     throughput \
     --dataset $YOUR_DATA_PATH \
     --backend pytorch \
     --num_requests 10 \
     --concurrency 1 \
     --max_batch_size 1 \
     --tp 8 \
     --ep 2 \
-    --extra_llm_api_options ./extra-llm-api-config.yml
+    --config ./config.yml

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In
`@docs/source/blogs/tech_blog/blog2_DeepSeek_R1_MTP_Implementation_and_Optimization.md`
around lines 181 - 204, Replace the --extra_llm_api_options usage with the
canonical --config flag in the trtllm-bench invocation: keep the same config
file (extra-llm-api-config.yml) and environment setup (TRTLLM_ENABLE_PDL) and
change the command argument from --extra_llm_api_options
./extra-llm-api-config.yml to --config ./extra-llm-api-config.yml so the
trtllm-bench call (the trtllm-bench throughput command) follows the project CLI
guideline.