sgl-project/sglang
Documented errors, page 11 of 32. Back to sgl-project/sglang
| Code / Message | Type | Severity | Tags |
|---|---|---|---|
| cache_dit_params must be a dict, got | validation | error | cache-dit, request-validation, type-error, config |
| : current-chunk rewrite size changed | exception | error | kv-cache, diffusion, rewrite, shape-mismatch |
| : dst entry dims (dims .. ) must be contiguous; got shape=… | validation | error | mamba, contiguity, strides, triton |
| JoyEcho requires audio latents for denoising. | validation | error | joyecho, audio, missing-latents, denoising |
| kv-canary: canary_buf slot stride must hold at least 4… | validation | error | kv-canary, buffer-layout, reference-implementation |
| --linear-attn-prefill-backend flashinfer on SM100+ requires… | validation | error | sglang, flashinfer, cuda-version, blackwell, linear-attention |
| LingBot causal sequence sharding currently requires… | exception | error | sequence-parallelism, config-mismatch, lingbot |
| MiniMax-H3 NVFP4 metadata requires checkpoint files | validation | error | minimax-h3, nvfp4, missing-argument, api-misuse |
| MXFP8 MoE quantization requires SM100 or ROCm gfx95 (gfx942… | exception | error | quantization, mxfp8, moe, hardware-compatibility, rocm |
| PD decode DCP currently requires prefill attention CP=1, got | exception | critical | disaggregation, dcp, attention-parallelism, prefill, config-mismatch |
| PP consensus is required when pp_size > 1 | validation | error | disaggregation, pipeline-parallel, consensus, api-misuse |
| tool_choice= requires at least one custom tool; all… | http | error | anthropic, tool-choice, built-in-tools, request-conversion |
| torch_npu detected, but NPU device is not available or… | exception | error | npu, environment, hardware-detection |
| Unsupported action_mode= | validation | error | cosmos3, action-generation, enum, invalid-argument-value |
| walk_radix_cache_for_canary does not support | exception | error | kv-canary, radix-cache, notimplementederror, type-dispatch |
| Cannot collate mixed VLA state presence | validation | error | vla, batching, robotics, validation |
| Cannot determine attention head counts for | error_code | error | mlx, attention, model-compatibility, patching |
| CUDA VMM tensor has already released its pool slice | exception | error | cuda, vmm, use-after-free, lifecycle |
| --disaggregation-decode-retraction-backup=host_pool does… | validation | error | sglang, pd-disaggregation, dcp, retraction, host-pool |
| Failed to edit image | error_code | error | network, http, image-edit, timeout, sgldiffusion |
| Inkling reasoning_effort must not be a boolean | validation | error | inkling, reasoning, type-validation, sglang |
| Invalid layerwise offload component name | validation | error | config-validation, layerwise-offload, type-error |
| Kimi-K3 DCP with decode_attention_backend='cutedsl_mla'… | exception | error | kimi-k3, flashinfer, signature-check, dcp, sglang |
| kv-canary: launch_per_forward not supported on sweep… | exception | error | kv-canary, endpoint, sweep-kernel, not-implemented |
| lm_head is not available in encoder-only mode | exception | error | kimi-k3, encoder-only, lm-head, attribute-error |
| Module(s) requested for update not found in pipeline | validation | error | weights-update, module-name, validation, multimodal |
| PD DCP transfer requires decode_prefix_len to align to the… | validation | error | disaggregation, dcp, alignment, prefix-cache |
| Pi05 state dim must be <= | validation | error | pi05, vla, state-dim, config-mismatch, torch |
| requires matching shapes but received_list | exception | error | distributed, tensor-shape, grafter |
| routed_dp_rank= out of range | exception | error | sglang, data-parallel, argument-validation |
| SGLANG_DISAGG_STAGING_BUFFER requires a positive… | exception | error | sglang, disaggregation, chunked-prefill, page-size, config-validation |
| Shared-sink LoRA slot out of range | exception | error | lora, index-error, slot-management, sglang |
| The hpc_ops MoE runner backend requires static activation… | validation | error | quantization, fp8, moe, moe-runner-backend, activation-scheme |
| TRTLLM FMHAv2 prefill on SM120 does not support fp8_e4m3 KV… | validation | error | sglang, sm120, fp8-kv-cache, attention-backend, trtllm, env-var |
| unsupported input for causal Conv3D cat/pad CUDA | exception | error | cuda, contiguity, conv3d, input-validation |
| aligned video noise shape | validation | critical | minimax-h3, noise-generation, internal-invariant |
| DSA indexer only supports CUDA, HIP, and NPU | exception | critical | dsa, device-support, xpu, not-implemented, deepseek, sglang |
| expected a JSON array of tool calls, got | exception | error | tool-calls, json-parsing, openai-api |
| Input 'data' must be a torch.Tensor, but got | exception | error | cuda-ipc, type-validation, torch-tensor, multimodal-transport |
| layer.layer_id= is not in the K-only sparse group. K-only… | validation | error | kv-cache, sparse-attention, layer-id, write-path, minimax, sglang |
| layer_types disagrees with no_rope_layers (NoPE layers must… | validation | error | config-validation, consistency-check, layer-types |
| `mixed_qkv` must be a 2D tensor | validation | error | fla, fused-recurrent, decode, shape-validation |
| must end at the packed token count , got | validation | error | npu, ascend, varlen, packed-sequences, shape-validation |
| must stay fp32 with curve AdaLN, got . | validation | error | dtype, fp32, adaln, weight-loading |
| NIXL KVSender Exception | exception | critical | sglang, nixl, kv-transfer, disaggregation, tp-rank-crash |
| NUMA node has no CPU cores allowed by the current affinity… | exception | warning | numa, cpu-affinity, cgroup, sglang, resource-binding |
| --prefill-only-disable-kv-cache does not currently support… | validation | error | sglang, kv-cache, fp4, quantization, server-args |
| QuantoInt8Config must be constructed from safetensors… | validation | error | quantization, quanto, config-loading |
| Qwen-Image-Layered requires generated latent shapes. | validation | error | qwen-image, layered-generation, latent-shapes, missing-metadata |
| requires --spatial-upsampler-path… | validation | error | ltx2, spatial-upsampler, missing-path, component, sglang |
| serialized_named_tensors entries must be base64 strings or… | validation | error | serialization, tensors, base64, typeerror |
| Unknown req_type: (expected 'generate' or 'embed') | exception | error | grpc, request-validation, dispatch, sglang |
| unsupported input for usp_merge_heads CUDA | exception | error | cuda, fallback, ulysses, input-validation |
| An exception occurred while loading multimodal data | exception | error | multimodal, prompt-template, placeholder-mismatch, legacy |
| bitsandbytes 4-bit TP only supports column-parallel output… | exception | error | quantization, bitsandbytes, tensor-parallel, not-implemented |
| Cannot duplicate reference image of batch size | validation | error | sglang, longcat-image, batch-size, reference-image, image-to-image |
| Cannot use multiple prompts with a fixed output_file_name… | validation | error | input-validation, file-output, batch-generation |
| combine() called before dispatch() | error_code | error | ascend, npu, dispatcher, lifecycle, state-machine |
| Cosmos3 requires num_frames == action_horizon + 1, got… | validation | error | cosmos3, num-frames, action-horizon, constraint-violation |
| f"Sol-Attn requires bfloat16 activations, got | validation | error | dtype, bfloat16, attention-backend, gpu |
| FlashInfer KDA verify requires an identity intermediate… | exception | error | sglang, kda, speculative-decoding, index-mapping, invariant |
| H3 conditioning projection MLP outputs width | exception | critical | minimax-h3, conditioning-projection, shape-mismatch, width-mismatch |
| Image is fully transparent | validation | error | image-processing, alpha-channel, rgba, mesh3d, validation |
| Invalid dtype | exception | error | frontend, dtype, regex, invalid-argument, sglang |
| Lightning (seg_la) linear-attention backend does not… | exception | error | sglang, lightning, seg-la, speculative-decoding, topk, linear-attention |
| --linear-attn-decode-backend flashinfer on SM100+ requires… | validation | error | sglang, linear-attention, flashinfer, sm100, bfloat16, dtype |
| Media URL domain is not allowed. Allowed domains | validation | error | media, security, ssrf, allowlist, network |
| MiniMaxH3TextEncodingStage direct encode requires a… | validation | error | minimax-h3, missing-component, text-encoding |
| Missing previous frame for delta payload | error_code | error | rocm, allreduce, deterministic, buffer-registration |
| is required for NPU packed attention | validation | error | npu, ascend, varlen, attention, packed-sequences |
| num_token_non_padded must be a torch.Tensor | exception | error | moe, triton, type-validation, tensor-shape |
| POSIX fd exchange receive failed | exception | error | distributed, fd-passing, wrapper-error |
| return_pooled_hidden_states is not supported for | validation | error | sglang, scoring, cross-encoder, hidden-states |
| The output_size of gate's and up's weight = | validation | error | quantization, fp8, moe, block-quantization, tensor-parallel, shape-mismatch |
| unsupported audio item: loaded= | exception | error | multimodal, audio, type-validation, asr |
| cache_salt is not supported by the experimental C++ radix… | validation | error | radix-cache, cpp-backend, cache-salt, unsupported-feature, value-error |
| Cargo completed but did not produce the expected artifact | exception | error | rust, cargo, artifact, build-output |
| DFLASH requires explicit layer_ids for aux hidden capture. | exception | error | laguna, dflash, argument-validation |
| --enable-dsa-cache-layer-split is not supported with… | validation | error | dsa, pipeline-parallelism, config-validation |
| g and beta must cover every q token | exception | error | kda, helion, prefill, length-mismatch |
| generated MiniMax H3 MP4 must contain exactly one video… | exception | error | minimax-h3, mp4, stream-layout, validation |
| Invalid combination of query/items types for score_request. | validation | error | sglang, scoring, type-validation |
| MiniMax H3 AdaLN cache max_plan_width must be positive; set… | validation | error | minimax-h3, adaln-cache, server-args, argument-validation |
| MiniMax H3 DiffGenerator requires save_output=True and a… | validation | error | minimax-h3, sampling-params, save-output, validation |
| MiniMax H3 does not support enable_frame_interpolation: the… | validation | error | minimax-h3, frame-interpolation, video-generation, validation |
| missing value for (expected e.g. ` 2,4`) | validation | error | cli-parsing, multi-gpu, argument-validation |
| must be a dict-like config, got | validation | error | config, type-error, initialization |
| placeholder/data mismatch: placeholders vs s | validation | error | validation, placeholders, prompt-template, multimodal |
| patch_size must be greater than 1, otherwise this doesn't… | validation | error | sglang, value-error, vision-encoder, config-validation, inkling |
| PD disagg: heterogeneous TP not supported for MiniMax… | exception | error | disaggregation, minimax, tensor-parallel, heterogeneous-tp, sparse-attention |
| q/k/v dtypes must match | validation | error | metal, rope, dtype-mismatch, apple-silicon |
| received auxiliary PP output without a sampling observer | exception | error | pipeline-parallel, sampling, observer, configuration |
| Replicated Q, K, and V must be provided together. | validation | error | attention, replicated-tokens, argument-validation, multimodal |
| SGLang's AutoRound CPU inference path currently supports… | validation | error | quantization, auto-round, cpu, amx, weight-bits |
| source must be 'timesteps' or 'sigmas' | validation | error | scheduler, enum-validation, api-misuse, sigma-shift |
| top_k must be -1 (disable) or at least 1, got | validation | error | sampling-params, top-k, validation, sglang |
| Unknown config key . Valid keys | validation | error | config, validation, unknown-key |
| Unknown dumper control method | exception | error | rpc, http-control, method-dispatch |
| `A_log`/`dt_bias` must be 1D tensors. | exception | error | fla, fused-recurrent, shape-validation, mamba-params |
| `a` must have shape [B, HV*K] with HV= | exception | error | kda, helion, shape-validation, packed-layout |