sgl-project/sglang
Documented errors, page 16 of 32. Back to sgl-project/sglang
| Code / Message | Type | Severity | Tags |
|---|---|---|---|
| AutoRound fused module | exception | error | quantization, auto-round, fused-modules, checkpoint |
| batching config rule from | validation | error | batching, config, schema-validation, json |
| cannot force-build after it has been imported; start a new… | exception | error | rust-extension, import-cache, force-rebuild, python, sglang |
| Could not decode audio | validation | error | audio, multimodal, decode, libsndfile |
| deepstack_visual_indexes exists but deepstack_merger_list… | error_code | error | multimodal, vit, deepstack, cuda-graph, config-mismatch |
| embed_override_token_id is required when embed_overrides is… | validation | error | embeddings, request-validation |
| --enable-cp-decode-attn-tp is only supported for models… | validation | error | context-parallel, model-support, config-validation |
| Error: --model-path is required. Please provide the path to… | exception | error | sglang, cli, missing-argument, model-path |
| Expected batch size 1 for decoded audio, got shape= | validation | error | joy-echo, audio, waveform, batching, memory-slot |
| must be formatted as tcp://host:port or host:port | validation | error | zmq, endpoint, validation |
| fl2va requires first_frame, last_frame, or both | validation | error | comfyui, sgldiffusion, minimax-h3, input-validation |
| int32-packed scale buffers require scale_ue8m0=True | validation | error | quantization, fp8, ue8m0, scale-factor, dtype-mismatch |
| --mamba-cache-philox-rounds must be non-negative. | validation | error | sglang, mamba, argument-validation, philox, cli-args |
| Model is required | validation | error | anthropic, model, validation, request-validation, missing-field |
| has batch dim (shape ); expected (per-prompt) or… | validation | error | batching, tensor-shape, validation |
| num_frames/height/width are required when hidden_states is… | validation | error | sana-wm, refiner, missing-argument, pre-packed-latents |
| Online quantization is not supported for native encoders… | validation | error | quantization, online-quantization, unsupported-format, text-encoder |
| pairs must be a torch.Tensor | validation | error | pytorch, type-check, scheduler, sigma-shift, validation |
| Position map x is not divisible by . | validation | error | validation, shape-mismatch, mask, diffusion |
| `sglang.bench_offline_throughput` is deprecated and will be… | console | warning | deprecation, benchmark, offline-throughput, future-warning |
| The `hidden_states` sequence length | validation | error | learnable-registers, shape-validation, ltx-2 |
| unknown minimax_h3 task | validation | error | minimax-h3, task-profile, unknown-task, registry-lookup |
| Unknown scheduler | exception | error | cli-arguments, simulation, schedule-simulator, sglang |
| Comfy W4A8 layer needs F32 channel scales, got | validation | error | quantization, dtype-mismatch, w4a8, channel-scales |
| decoder_model_output_type must be 'x0' or 'v', got | validation | error | config-validation, diffusion, model-config, ltx-2 |
| DeepSeekV4 CP supports moe_a2a_backend in | validation | error | deepseek, moe, a2a-backend, context-parallel, sglang |
| dimension size must be divisible by 2 * group_size= | validation | error | nvfp4, shape-alignment, weight-loading, swiglu |
| expert-pack requires --disable-shared-experts-fusion so the… | validation | error | expert-pack, moe, shared-experts, launch-flag |
| {flashinfer_error} | validation | error | sglang, mamba, flashinfer, dependency, import-error |
| Invalid padding param | validation | error | convolution, causal, padding, type-validation |
| Kimi-K3 cannot mix local preprocessed and deferred images | exception | error | kimi-k3, multimodal, preprocessing-mismatch |
| kv-canary: must be contiguous | validation | error | kv-cache, contiguity, triton, strides |
| LingBot causal sequence sharding currently supports… | exception | error | not-implemented, sequence-parallelism, ring-attention, lingbot |
| MiniMax H3 model variant must be a non-empty string | validation | error | minimax-h3, model-variant, input-validation, sglang |
| must provide either a lambda or static kwargs | exception | error | decorator, api-misuse, missing-argument |
| Network error after consecutive failures | error_code | error | network, timeout, video-generation, polling, retry, sgldiffusion |
| No valid partitions found for total SMs | validation | error | pdmux, gpu, partitioning, capacity |
| tar material URI has an invalid encoded header | validation | error | minimax-h3, tar-uri, base64, json, material-io |
| unexpected hidden shape | exception | error | minimax-h3, encode-ids, output-shape, sanity-check |
| Unsupported msgpack byte | error_code | error | radix-tree, hicache, not-implemented, kv-cache-load |
| Unsupported RRDBNet conv_first input channels | validation | error | realesrgan, checkpoint, architecture, postprocess |
| could not create an import spec for | exception | error | python, import, importlib, native-extension |
| CP attention for non-FIA path on Ascend is not yet… | exception | error | ascend, npu, context-parallel, fia, not-implemented, huawei |
| expert-pack index is truncated | exception | critical | moe, expert-pack, binary-format, truncated-file, corruption |
| f"Invalid prefix for SparseVideoGen2AttentionImpl | validation | error | prefix, layer-index, weight-loading, validation |
| f"seq_len not supported for STA | validation | error | sliding-tile-attention, seq-len, unsupported-value, key-error |
| Invalid quantization method | exception | error | quantization, lookup, invalid-argument, config |
| kv-canary: max_prefill_tokens must be positive, got | exception | error | kv-canary, validation, prefill, scheduler, value-error |
| model_index.json._minimax_h3.sigma_shift_scales must be an… | validation | error | minimax-h3, sigma-shift-scales, model-index, config-validation |
| prefetch_timeout_per_ki_token must be number, got | validation | error | hicache, config-validation, prefetch, type-error |
| q, k, and v must have the same device and dtype | validation | error | dtype, device, mismatch, ulysses |
| raw_action_dim is required when only domain_id is provided | validation | error | cosmos3, domain-config, action-dim, missing-required-field |
| reference image target dimensions must be positive | validation | error | minimax-h3, resize, argument-validation |
| : no triton backend | exception | error | fused-op, backend-dispatch, not-implemented, triton |
| SGLANG_GRPC_WORKER_THREADS | validation | error | grpc, config-validation, environment-variable, server-args |
| --sidecar requires module | exception | critical | sidecar, callable, startup, sglang |
| --speculative-draft-window-size must be positive, got | validation | error | sglang, speculative-decoding, window-size, argument-validation |
| target.duration_seconds is required when multiple… | validation | error | minimax-h3, duration, ambiguous-source, request-validation |
| `A_log` must have elements (got ). | exception | error | kda, helion, parameter-shape, model-config |
| AfmoeConfig must define `num_experts`. | exception | critical | moe, config-validation, afmoe, model-loading |
| camera_conditions must have shape (T,20) or (B,T,20), got | validation | error | sglang, sana-wm, camera-conditions, shape-validation, video-generation |
| Cosmos3 action prompt must be a string or non-empty list | validation | error | cosmos3, prompt, validation |
| CUDA VMM multimodal pool failed | exception | critical | cuda, vmm, pool, lazy-failure, runtimeerror |
| --enable-linear-replayssm-spec requires the triton or… | validation | error | sglang, replayssm, speculative-decoding, backend-validation |
| EPD MMReceiver: http mode requires http:// encoder URLs… | validation | error | configuration, url-scheme, env-var, grpc, http, epd |
| ffprobe is required to validate final MiniMax H3 output | exception | error | minimax-h3, ffprobe, missing-binary, environment |
| Invalid audio format | validation | error | audio, multimodal, validation, input-validation |
| Invalid content format | validation | error | jinja, chat-template, content-format, validation |
| Invalid version | validation | error | flash-attention, version, init, sglang |
| kv-canary: RealKvSource.read_bytes must be a positive… | validation | error | kv-cache, alignment, validation, sampling |
| mask_candidates is required for STA_tuning_cfg mode | validation | error | sta, attention, sparse-tuning, kwargs-validation |
| material URI has an invalid percent escape | validation | error | minimax-h3, percent-encoding, material-io, parsing |
| MiniCPM sparse attention does not support PD disaggregation | validation | error | minicpm, pd-disaggregation, sparse-attention, sglang |
| muse_glimmer_mlx_format | validation | error | version-mismatch, packaging, mlx |
| Not a canonical UMMA_MN Layout: Expected stride failure. | validation | error | cutlass, sm100, layout, stride |
| Nunchaku SVDQuant is currently only supported on Ampere… | validation | error | nunchaku, svdquant, gpu-compatibility, ampere, hopper, quantization |
| pair_postprocess must be callable or None | validation | error | scheduler, type-error, postprocess, flow-match |
| q, k, and v must have the same 3D shape | validation | error | attention, ulysses, shape-mismatch, sequence-parallelism |
| Quanto layer contains both packed and dense weights | validation | error | quantization, quanto, duplicate-weights, checkpoint |
| TRTLLM MHA backend for prefill requires Hopper (SM90)… | validation | error | sglang, gpu-architecture, attention-backend, trtllm, sm90 |
| Unknown action normalization method | validation | error | cosmos3, normalization, enum-validation, action-stats |
| cannot found moe_block_size for shape | validation | error | sglang, moe, humming, tuning-config, batch-size, index-out-of-range |
| 'data_parallel_rank' is deprecated, use 'routed_dp_rank'… | console | warning | deprecation, sglang, request-api, data-parallel |
| f"recurrent_kda state inner strides must be compact (V*K… | validation | error | sglang, kda, tensor-stride, contiguity, validation |
| Invalid graph capture input size | exception | error | cuda, vmm, cuda-graph, validation |
| invalid Inkling reasoning_effort | validation | error | inkling, reasoning, validation, sglang |
| Kimi-K3 manifest format is unsupported | validation | error | kimi-k3, gguf, manifest, version-mismatch |
| Kimi K3 required parameters are missing schemas | validation | error | kimi-k3, json-schema, required-fields, validation |
| Multi-output conditioning requires prompt text so the… | validation | error | batching, sampling, multimodal, validation |
| Quanto checkpoint is missing quantization_map_base64 | validation | error | quantization, quanto, safetensors, checkpoint-metadata |
| ReplaySSM cache length must be at least 1. | exception | error | kda, replayssm, cache-allocation, zero-size |
| Scheduler type ' ' not implemented | exception | error | diffusion, scheduler, not-implemented, config-mismatch |
| Stochastic rounding for the Mamba SSM cache is only… | validation | error | sglang, mamba, cuda-only, platform, rocsm, server-args |
| The hpc_ops MoE runner backend runs a plain SiLU-and-mul… | validation | error | sglang, moe, hpc-ops, swiglu, activation, config-validation |
| unsupported Inkling render part kind | validation | error | inkling, internal, parser |
| Unsupported qk_norm | exception | critical | config-validation, qk-norm, init-time, lingbot |
| Cosmos3 inverse_dynamics prompt must be a string | validation | error | cosmos3, inverse-dynamics, prompt-validation, type-validation |
| CUDA VMM POSIX FD broker returned no file descriptor | exception | critical | cuda, vmm, ipc, socket, file-descriptor |
| : non-sequential write current_start= global_end= cur= | exception | error | kv-cache, sequential-write, not-implemented, chunking |
| `force_flush` must be a length-B int32 tensor or None. | exception | error | kda, replayssm, dtype, int32, optional-argument |