sgl-project/sglang
Documented errors, page 10 of 32. Back to sgl-project/sglang
| Code / Message | Type | Severity | Tags |
|---|---|---|---|
| EAGLE3 currently only supports 1 layer | exception | error | eagle3, speculative-decoding, config-validation, kimi |
| GGUF tensor has inner dimension , which is not a multiple… | validation | error | gguf, quantization, block-alignment, tensor-shape |
| InklingMultimodalProcessor | validation | error | multimodal, inkling, placeholder-mismatch, audio |
| invalid CUDA handle-type value | validation | error | cuda, vmm, validation, handle, enum-value |
| invalid predicate | validation | error | predicate, dsl, syntax-error, eval |
| Kimi active routed MoE layers must be exactly 1..92 | validation | critical | kimi, moe, expert-pack, manifest-validation |
| MiniMax-H3 MPS execution does not support torch.compile… | validation | error | minimax-h3, mps, torch-compile, server-args |
| MiniMaxH3Pipeline only supports monolithic deployment… | validation | error | minimax-h3, disaggregation, monolithic-only, sglang |
| No ' ' backend registered for op | exception | error | kernels, backend-selection, registry, sglang |
| publish role has no ROLE_NAMESPACE_SETS entry; declare its… | validation | error | config, role-based-access, enforcement, sglang |
| seq must be positive, got | validation | error | multimodal, video, config-validation, valueerror |
| Unconditional token logprobs are required for this method. | validation | error | validation, logprobs, argument-validation, choices |
| Unsupported quantized linear marker for | validation | critical | quantization, checkpoint, linear-layer, model-load |
| Cosmos3 action requests accept either an image or a video | validation | error | cosmos3, mutually-exclusive-inputs, input-validation |
| --enable-linear-replayssm requires… | validation | error | sglang, replayssm, mamba-radix-cache, config-conflict |
| {exc} | exception | error | file-not-found, huggingface, modelscope |
| External ngram corpus path does not exist | validation | error | sglang, ngram, speculative-decoding, file-not-found, validation |
| Invalid hicache storage backend extra config JSON | validation | error | hicache, json-config, decode-offload, config-parse-error |
| KDA prefill requires an indexed initial-state pool | exception | error | kda, helion, prefill, state-pool, required-argument |
| LMCache is not installed. Please install it by running `pip… | exception | critical | lmcache, importerror, missing-dependency, installation |
| Online MXFP4 quantization for MoE layers requires an AMD… | exception | critical | quantization, mxfp4, amd, rocm, moe, hardware-unsupported |
| PD state transfer failed: kv_args.state_types is empty but… | exception | error | disaggregation, hybrid-model, mamba, state-transfer, configuration |
| QuantConfig has static quantization, but found activation… | validation | error | quantization, fp8, moe, missing-weights, activation-scheme |
| raw_latent_shape must be divisible by patch_size for SAP… | validation | error | shape-validation, patch-size, divisibility, video-generation |
| rope_pool_fused expects pool tensors to be 3-D | validation | error | shape-validation, kv-cache, metal, rope, sgl-kernel |
| Shared-sink down LoRA-A width must be divisible by | validation | error | lora, shape-mismatch, weight-loading, moe |
| STANDALONE speculative decoding requires the draft model to… | validation | critical | speculative-decoding, tokenizer-mismatch, vocabulary-mismatch |
| tensor matched no --diff-threshold pattern ( ); add a… | validation | error | regex, threshold, pattern-matching, fullmatch |
| The parameter max_tokens will be overwritten by speculated… | console | warning | openai-backend, speculative-decoding, sampling-params, max-tokens, ignored-parameter |
| Usage: sglang serve --model-path <model-name-or-path>… | exception | info | sglang, cli, usage, help, missing-argument |
| action_horizon must be a positive integer | validation | error | cosmos3, action-horizon, numeric-validation |
| adapter_config.json lora_alpha conflicts with safetensors… | validation | error | lora, peft, metadata, conflict |
| Ascend PD transfer does not support HiSparse destination… | exception | error | ascend, npu, disaggregation, hicache, not-implemented |
| Cannot parse checkpoint quantization metadata for | validation | error | quantization, config, model-loading, fail-closed |
| Comfy NVFP4 layer needs a scalar F32 weight_scale_2, got | validation | error | quantization, nvfp4, scalar-scale, dtype-mismatch, checkpoint-validation |
| Comfy W4A8 layer has invalid group_size= | validation | error | quantization, w4a8, group-size, validation |
| Either the environment variable 'MOONCAKE_MASTER' or… | validation | error | env-var, mooncake, missing-configuration |
| Expected hybrid GDN or NemotronH models, but got unknown… | validation | error | hybrid-model, linear-attention, gdn, nemotron-h, model-registry, sglang |
| Failed to decode base64 image. Expected format… | validation | error | base64, data-uri, image-input, validation |
| JoyEcho audio scheduler was not prepared. | validation | error | joyecho, audio, scheduler, initialization |
| Kimi GPU preprocessing expects raw uint8 pixels, got | validation | error | kimi, k25, multimodal, dtype-validation, preprocessing |
| Length mismatch | validation | error | validation, length-mismatch, dataclass, invariants |
| MiniMax H3 initial_video_rows must be a rank-2 tensor | validation | error | minimax-h3, tensor-shape, batch-state |
| No IB devices configured for GPU | validation | error | config, mooncake, ib-devices, gpu-mapping |
| .short_edge must be positive, got | validation | error | minimax-h3, validation, request, short-edge |
| Qwen-Image-Layered requires a non-empty image_path. | validation | error | qwen-image, layered-editing, image-path, validation |
| SD3 CLIP postprocessing requires hidden_states from encoder… | validation | error | stable-diffusion-3, clip, text-encoding, hidden-states |
| Shared-sink gate/up LoRA-B height must be divisible by | validation | error | lora, shape-mismatch, weight-loading, moe |
| temperature must be a non-negative finite number, got | validation | error | sampling-params, temperature, nan, validation, sglang |
| The SRT encoder checkpoint adapter supports only serialized… | validation | error | quantization, text-encoder, fp8, model-loading |
| Unsupported scale . Choose from | validation | error | upsampler, scale, config, value-error |
| Weight cache daemon for pp_rank= | exception | error | sglang, weight-cache, timeout, startup, daemon |
| Currently DFLASH speculative decoding does not support dp… | validation | error | speculative-decoding, dflash, dp-attention, server-args |
| Dual chunk attention is enabled, but attention backend is… | validation | error | sglang, dual-chunk-attention, attention-backend, config-conflict |
| Failed to process image source | http | error | http, upload, mesh-generation, client-error |
| Generate subcommand is not yet supported for model | exception | critical | rocm, allreduce, deterministic, tensor-parallel, float16 |
| generated MiniMax H3 outputs have inconsistent media… | exception | error | minimax-h3, metadata-consistency, multi-output, validation |
| --hicache-host-memory-mode buffer_only on SWA models… | validation | error | hicache, buffer-only, swa, sliding-window, unified-kv, configuration |
| Invalid format: keys must be integers (or string… | validation | error | json, config, schema, mooncake, type-mismatch |
| kv-canary: forward_batch.batch_size= | validation | error | kv-canary, capacity, cuda-graph-max-bs, batch-size, sglang |
| kv-canary: req_to_token_pool_size must be positive, got | exception | error | kv-canary, validation, req-pool, value-error |
| MiniMax H3 text encode produced no native payload | validation | error | minimax-h3, text-encoding, payload-validation, data-parallel |
| missing `sample` as a required keyword argument | validation | error | scheduler, diffusion, api-misuse, unipc |
| MLX async runner does not support forward mode | validation | error | mlx, forward-mode, async-scheduler, sglang, unsupported-feature |
| must have length , got | validation | error | validation, shape, patchify, minimax-h3 |
| `ssm_state_indices` must have shape [B] | exception | error | fla, fused-recurrent, state-cache, shape-validation |
| Subclass must define _supported_attention_backends | validation | error | sglang, dit, attention-backend, init-validation |
| Can not get local ip | exception | error | network, ip, container, distributed |
| Cannot load PE model: 'model_max_length' not found in | error_code | error | model-loading, tokenizer, missing-file, config |
| Config list contains configs from 2 methods, must be only 1 | exception | error | quantization, config-conflict, model-config, sglang |
| --diffusers-kwargs must be valid JSON. Got | console | error | cli, json, diffusers, argument-validation |
| Error raised in subprocess | exception | error | registry, subprocess, model-loading, import-error |
| Hunyuan3D Paint does not use added conditioning. | validation | error | runtime, api-misuse, diffusion, unet |
| Image path not found | validation | error | hunyuan3d, file-not-found, filesystem, docker |
| MiniMax-H3 quality="high" is validated only for | validation | error | minimax-h3, quality-high, workload-validation, golden-config |
| model_index.json._minimax_h3.sigma_shift_scales requires… | validation | error | minimax-h3, sigma-shift-scales, numeric-coercion, model-index |
| MOVA requires reference image latents for denoising | validation | error | mova, image-latent, conditioning, validation |
| tar material header requires integer offset_data and size | validation | error | minimax-h3, tar-uri, json-fields, material-io |
| teacache is not supported yet for HunyuanVideo | validation | error | unsupported-feature, teacache, video-diffusion, runtime |
| The quantization method | exception | error | quantization, registry, duplicate, config |
| Unknown approximate mode | validation | error | activation, gelu, argument-validation, torch |
| Unsupported LTX-2.3 encoder block | validation | error | python, value-error, model-config, ltx-2-3, encoder-block |
| VisionFlash3Attention is only available for cuda or musa | exception | error | sglang, vision-transformer, flash-attention, platform-support, hardware-compat |
| `write_pos` must be a 1D int32 tensor. | exception | error | kda, replayssm, dtype, int32, tensor-ndim |
| Checkpoint quantization is encoded in per-layer metadata… | validation | error | quantization, config-conflict, checkpoint, server-args |
| ControlNet down and mid residuals must be provided together. | validation | error | controlnet, argument-pairing, unet |
| Could not parse attention backend config | validation | error | config, attention-backend, parsing |
| --enable-unified-memory does not support different prefill… | exception | error | pd-disagg, unified-memory, tensor-parallel, mamba |
| expert-pack or manifest is missing | exception | critical | moe, expert-pack, file-not-found, manifest, path-resolution |
| f"Input shape must be divisible by patch_size | exception | error | multimodal, tensor-shape, patch-embedding |
| [FlexKV] Failed to send eventfds to | exception | critical | flexkv, eventfd, retry-exhausted, unix-socket |
| Inkling shared-sink LoRA requires four 4D MoE buffers | validation | error | sglang, lora, shape-validation, moe, inkling |
| --kv-cache-dtype mxfp8 requires an SM100+ (Blackwell) GPU… | validation | error | sglang, kv-cache-dtype, mxfp8, blackwell, gpu-architecture |
| kv-canary: launch_canary_plan_kernels verify_capacity= | validation | error | kv-canary, capacity-mismatch, argument-validation |
| LTX2Attention requires inner_dim divisible by tp_size, got | exception | critical | parallelism, tensor-parallel, inner-dim, config-validation, ltx2 |
| media_url_max_file_size_mb must be non-negative | validation | error | validation, media, security, config |
| MiniMax H3 text encoding requires an ordered keyframe… | validation | error | minimax-h3, keyframes, plan-validation, fl2va |
| quantize_and_serve requires ModelOpt quantization | validation | error | quantization, modelopt, config-validation, sglang |
| Required: indexer, forward_batch, x, q_lora, positions | validation | error | deepseek, dsa, sparse-attention, retrieval, value-error |
| task is required for MiniMax H3; supported tasks: fl2va… | validation | error | minimax-h3, task-validation, missing-parameter, video-adapter |