Repository Issue Activity (beta)

#17964 [Bug]: Qwen3_5ForCausalLM (Qwen3.5-2B) incorrect output with batch_size > 1

#18115 [Bug]: V1 MAX_UTILIZATION paused_requests are never processed in the PP executor loop

#17882 [Bug]: MAX_UTILIZATION resume raises `RequestError: 'modality_type'` for Nemotron Nano 3 Omni

#18104 Withdrawn

#18103 Withdrawn

#18102 Withdrawn

#18101 Withdrawn

#18100 Withdrawn

#17743 [Bug]: ruff-legacy pre-commit hook reports all baseline violations as regressions on Windows

#18078 [Bug]: CUDA graphs corrupt generation for google/functiongemma-270m-it (Gemma3 270M) on GB10/sm_121 — argument values collapse; cuda_graph_config: null restores byte-correct output

#17665 [Bug]: NIXL cache-transceiver hangs forever if a single completion notification is dropped

#18085 [RFC] DFlash2 for Qwen3.8 on consumer Blackwell: integration boundary before we propose a PR

#18084 MODEL_TYPE_TO_TOOL_PARSER maps qwen3_5/qwen3_5_moe to a parser whose format the template does not emit

#18083 Reasoning-parser auto-selection misclassifies reasoning-at-start templates, silently emptying reasoning_content

#14825 [Bug]: kimi 2-6 nvfp4 + SpecD gives low accuracy on GPQA Diamond

#5152 Should remove `self.all_reduce =` from the base class

#17024 DeepSeek-V4 + MTP: `AttributeError: '_num_tables'` in `DeepseekV4CacheManager.copy_batch_block_offsets` during warmup

#18039 DeepSeek-V4-Flash NVFP4 baseline aborts in AttentionOp on RTX PRO 6000 Blackwell: Deepseek should be supported by fmha

#10615 [Bug]: Llama 3.2 Vision model fails to build using documented steps

#10306 [Usage]: How to use TensorRT LLM to perform MoE NVFP4 | ATTN FP8 inference testing for the Deepseek R1 671B model?

#10258 [Bug]: LoRA Support Missing for Encoder-Decoder Models in TensorRT-LLM CPP Implementation

#9523 Support for q4_0 format - google/gemma-3-27b-it-qat-q4_0-unquantized

#8633 [Usage]: TensorRT backend or Python backend

#8572 [Bug]: Build and serving failures with Qwen3-4B-Instruct-2507 model FP8 quantization on H100 GPU with TensorRT-LLM with tensorrt-llm-release:0.19.0 image

#8286 [Usage]: how long the build stage takes for deepseek-r1 models