sgl-project/sglang
Documented errors, page 30 of 32. Back to sgl-project/sglang
| Code / Message | Type | Severity | Tags |
|---|---|---|---|
| 'role' must be one of | validation | error | openai-api, role, validation |
| This tool only supports ModelOpt diffusers FP8 exports… | validation | error | modelopt, fp8, quantization, config |
| DeepSeek-V4 flashmla_sparse_q8 prefill requires d_v=512, got | exception | error | deepseek-v4, flashmla, head-dim, model-config-mismatch, sglang |
| FlashInfer KDA verify kernel only supports topk=1… | exception | error | sglang, kda, speculative-decoding, topk, flashinfer |
| fp8_blockwise_scaled_mm JIT kernel requires SM120… | error_code | error | gemm, fp8, blockwise, sm120, blackwell, gpu-architecture |
| No encoder method found for modality | validation | error | multimodal, transformers-backend, encoder-discovery |
| `task` requires at least one message with role='user' or… | validation | error | deepseek-v4, task, message-validation |
| --mode requires | validation | error | cli, validation, config-mismatch |
| received IPv6 address format: expected ':' after ']' | validation | error | ipv6, distributed, network, validation |
| Weight path does not exist | exception | error | local-filesystem, weights, path-not-found |
| GptOssForCausalLM on Intel XPU only supports bfloat16… | validation | error | gpt-oss, intel-xpu, dtype, sglang |
| Only per-tensor FP8 scales are supported for diffusion… | validation | error | modelopt, fp8, quantization |
| UMBPHostTensorAllocator only supports CPU host memory, got… | validation | error | umbp, device-mismatch, validation |
| DFLASH speculative decoding only supports CUDA and NPU… | validation | error | sglang, dflash, speculative-decoding, device-support, hardware-gpu |
| Gate type not supported | exception | error | layernorm, gate, type-error, argument-validation |
| entries must be tensors or None. | validation | error | batching, type-error, validation |
| SinusoidsPositionEmbedding needs even channels input | validation | error | qwen3-omni, embedding, shape-validation |
| `b` must be 2D (got b.ndim= ). | validation | error | pytorch, tensor-shape, replayssm, gate |
| MiniCPM does not support DP attention | validation | error | minicpm, dp-attention, sglang, config-validation |
| a port must be specified in IPv6 address (format… | validation | error | ipv6, distributed, network, port, validation |
| thinking parts require exactly one of 'thinking' or 'text' | validation | error | openai-api, thinking-part, pydantic-validation |
| CuteDSL MLA backend only supports kv-cache-dtype of… | validation | error | sglang, cutedsl, mla, kv-cache-dtype, config-validation |
| Invalid stacked eps shape for fused KV materialization: got | validation | error | shape-validation, rmsnorm, speculative-decoding |
| Layer is not backed by an MLA KV pool | exception | critical | swa, mla, type-mismatch, memory-pool |
| PD decode DCP currently requires chunk cache… | validation | error | sglang, pd-disaggregation, dcp, hierarchical-cache, config-conflict |
| duplicate auxiliary PP tensor | exception | error | pipeline-parallel, sampling, duplicate-keys, validation |
| Invalid modality string | validation | error | sglang, modality, enum-validation |
| MXFP8 KV cache requires the FA4 backend. | exception | error | mxfp8, kv-cache, flash-attention-4, quantization, sglang |
| rel_bias (sheared bias) is only supported by the FA4… | exception | error | rel-bias, flash-attention-4, extend, sglang |
| spt must be a bool when provided | validation | error | block-sparse, type-error, spt |
| QKV tensors must have shape [B, S, H, D] | validation | error | shape, rope, attention, hunyuan, diffusion |
| Unsupported backend: , currently only support | validation | error | quantization, auto-round, backend |
| Expected a flat quantization_config dict in the ModelOpt… | validation | error | modelopt, fp8, config |
| Unsupported packing_format | validation | error | quantization, auto-round, packing |
| Source contains no recognized weight files | exception | error | weights, file-selection, no-candidates |
| Specified lora_weight_name | validation | error | lora, file-not-found, local-model |
| timestep must be a CUDA bfloat16 tensor | validation | error | dtype, cuda, bfloat16, ltx2 |
| `A_log` must be a 1D tensor. | validation | error | pytorch, tensor-shape, replayssm, decay |
| auxiliary PP tensor names must be non-empty strings | exception | error | pipeline-parallel, sampling, validation, keys |
| invalid IPv6 address format: missing ']' | validation | error | ipv6, distributed, network, validation |
| kv-canary: lut_len must be positive when has_swa_lut is True | validation | error | kv-canary, lut, swa |
| Error while unloading LRU LoRA adapter | validation | error | lora, eviction, unload |
| invalid port in IPv6 address | validation | error | ipv6, distributed, port, validation |
| : no .pt files found in or any of its subdirectories. | validation | error | filesystem, missing-files, dump, not-found |
| must have dtype torch.int32 | validation | error | block-sparse, dtype, int32, metadata |
| Unknown activation function | exception | critical | config-validation, activation, init-time, ltx2 |
| Weight subfolder was not found in | exception | error | huggingface, weights, subfolder |
| Invalid v_out device/dtype for fused KV materialization… | validation | error | device-dtype-validation, kv-cache |
| received a non-tensor auxiliary PP output | exception | error | pipeline-parallel, sampling, torch, validation |
| The Helion package is required when a KDA backend is set to… | exception | error | sglang, helion, kda, missing-dependency, version-pin |
| Error: --model-type requires a non-empty value. | validation | critical | rocm, allreduce, deterministic, tensor-parallel, bfloat16 |
| Unsupported ltx25_decoder_rope dtype | validation | error | jit, dtype, bfloat16, ltx, rope |
| DSpark speculative decoding only supports CUDA or NPU… | validation | error | speculative-decoding, dspark, device-support, cuda, npu |
| LoRA is not supported on a GGUF transformer: an adapter… | validation | error | gguf, lora, unsupported-feature |
| MiniMax H3 requires num_inference_steps >= 2 because its… | validation | error | minimax-h3, num-inference-steps, request-validation, sigma-schedule |
| VisionFlashInferAttention is only available for cuda | exception | error | sglang, vision-transformer, flashinfer, platform-support, hardware-compat |
| Could not find ModuleList in | validation | error | pipeline-parallel, transformers-backend |
| Dots omni audio must be mono, got shape= | validation | error | multimodal, audio, mono, valueerror |
| DSpark requires --speculative-dspark-block-size to be… | validation | error | speculative-decoding, dspark, argument-validation, block-size |
| Empty multimodal encoder output. | validation | error | multimodal, empty-batch, encoder-output |
| extra_config['spdk_passthrough'] must be a dict of spdk_*… | validation | error | umbp, spdk, config-validation |
| Missing port in address (expected host:port) | validation | error | network, port, address-parsing, sglang |
| must not be None | validation | error | umbp, config-validation, missing-value |
| HTTP material response.read() must return bytes, got | validation | error | http, type-error, mocking |
| kv-canary: lut_len must be 0 when has_swa_lut is False | validation | error | kv-canary, lut, swa |
| No safetensors files found in | validation | error | safetensors, missing-weights, model-path |
| extra_config['ssd_backend'] must be one of: file, spdk… | validation | error | umbp, config-validation, ssd |
| 'role' must be a string | validation | error | openai-api, role, type-error |
| does not support pipeline parallel yet! | validation | critical | pipeline-parallel, transformers-backend |
| wrong pixel_values size | exception | error | multimodal, vision, input-shape |
| x must be [1, S, C], got | validation | error | minimax-h3, input-shape, packed-sequence |
| Config file must be YAML format, got | validation | error | sglang, yaml, file-extension, config |
| Could not resolve backbone.pt from | exception | error | modelopt, fp8, path-resolution |
| must not be empty | validation | error | umbp, config-validation, empty-list |
| Only 'absolute' position_embedding_type is supported | validation | critical | bert, embeddings, config-validation |
| Unsupported activation | exception | critical | activation, config-validation, arcee |
| Unsupported activation | exception | critical | activation, config-validation, apertus |
| attention TP must be divisible by num_key_value_heads | exception | error | gqa, kv-heads, tensor-parallel |
| Command contains only environment variable assignments, no… | validation | error | shell, subprocess, validation |
| is not implemented. | exception | error | multimodal, architecture-not-supported, config-mismatch, interns1 |
| lora_nickname cannot be empty | validation | error | validation, lora, empty-parameter, sgldiffusion |
| No config file specified after --config flag! | validation | error | sglang, cli, missing-argument |
| num_key_value_heads must be divisible by attention TP | exception | error | gqa, kv-heads, tensor-parallel |
| Unknown event publisher | validation | error | kv-events, config, registry |
| Currently standalone speculative decoding does not support… | validation | error | speculative-decoding, standalone, dp-attention |
| Invalid thinking_mode | exception | error | deepseek, chat-template, thinking-mode, validation |
| Only 2D tile_tag is supported currently, got | validation | error | config-validation, ocr, vision |
| AdaLN cache must cover at least one timestep plan | validation | error | cli, validation, diffusion, timesteps |
| Intern-S2-Mobius baseline does not support PP tensors | exception | error | pipeline-parallel, runtime-misuse |
| MiniMax H3 AdaLN cache does not exist | validation | error | minimax-h3, adaln-cache, file-not-found |
| MiniMax H3 AdaLN cache must be built on CUDA | validation | error | cuda, environment, cli |
| Norm type not implemented | exception | error | layernorm, norm-type, config-validation, not-implemented |
| Only 1 nextn layer is supported for Step3p5 checkpoints. | validation | error | speculative-decoding, mtp, checkpoint-loading |
| must live on CUDA | validation | error | block-sparse, cuda, metadata, cpu-tensor |
| pass only one of camera_actions or action | validation | error | realtime, mutually-exclusive, condition-inputs, sana |
| At least one of text, input_ids, or image should be provided | validation | error | sglang, empty-input, request-validation |
| Mesh not found | http | error | http, not-found, mesh-generation, job-store |
| MiniMax H3 material localization does not support URI scheme | validation | error | uri-scheme, minimax-h3, material-uri |
| MiniMax H3 s3:// material URIs require a configured… | validation | error | s3, minimax-h3, material-uri |
| tar material reader returned too many bytes | validation | error | tar, io-contract, defensive |