sgl-project/sglang
Documented errors, page 19 of 32. Back to sgl-project/sglang
| Code / Message | Type | Severity | Tags |
|---|---|---|---|
| num_nextn_predict_layers is not in the config | validation | critical | mtp, speculative-decoding, bailing-moe, weight-loading |
| refined prompt embeddings must have hidden width | validation | error | minimax-h3, hidden-size, embedding-width |
| stop= is unavailable when skip_tokenizer_init=True… | validation | error | skip-tokenizer-init, stop-strings, sampling-params, sglang |
| The input size is not aligned with the quantized weight… | validation | error | gptq, tensor-parallel, shape-mismatch, quantization, cpu, amx |
| The required 'attentions' package is not installed. Install… | exception | error | import-error, npu, ascend, missing-dependency, laser-attention |
| Unknown | validation | error | offloading, meta-device, pytorch, parameters, sglang |
| unknown weight reader | validation | error | weight-reader, registry, invalid-name, validation |
| Unsupported up_block_type | validation | error | python, value-error, model-config, hunyuan-vae, unsupported-block |
| Unsupported weight quantization strategy | validation | error | quantization, fp8, moe, strategy, compressed-tensors |
| vis_freqs_cis is required for fused QK-Norm + RoPE kernel | validation | error | runtime, rope, missing-argument, diffusion |
| bad metadata: dur= h= w= | validation | error | multimodal, video, corrupt-metadata, decoding |
| Block sparse tensors | validation | error | block-sparse, ambiguous-config, shape-inference |
| Cannot determine processor class for | exception | error | multimodal, processor, model-loading |
| Comfy W4A4 layer has incompatible weight/scale shapes: and | validation | error | quantization, shape-mismatch, w4a4, scales |
| Component resolved to component-offload, but it was loaded… | validation | error | memory, offload, fsdp, component-residency, config |
| Failed to load mistral | exception | error | mistral, config, corrupt-file |
| For , when providing a 'processor_output' or… | validation | error | multimodal, precomputed-embedding, input-validation |
| Found corrupted safetensors file(s). Files have been… | exception | critical | safetensors, corruption, download, retry |
| GLM DSA with FP8 KV cache on NVIDIA SM120/SM121 supports… | validation | error | attention-backend, config-validation, sm120, fp8, glm |
| Incorrect type of image crop. Got type | validation | error | multimodal, vision, type-validation, deepseek-ocr |
| Incorrect type of image sizes. Got type | validation | error | multimodal, vision, type-validation, deepseek-ocr |
| Input probs contains NaN. | validation | error | musa, sampling, top-p, nan |
| LoRA with EAGLE/NEXTN/EAGLE3 speculative decoding | validation | error | sglang, lora, eagle, speculative-decoding, feature-incompatibility |
| missing `order` as a required keyword argument | validation | error | scheduler, diffusion, api-misuse, unipc |
| `negative_prompt` should be the same type to `prompt`, but… | validation | error | glm-image, negative-prompt, type-mismatch, input-validation, typeerror |
| num_frames/height/width must be provided for RoPE… | exception | error | rope, missing-argument, ltx-2, forward, video |
| num_token_non_padded must be an integer tensor, got | exception | error | moe, dtype, triton, validation |
| Quant type is ambiguous in : . Pass the full… | exception | error | gguf, huggingface, ambiguous-match |
| RadixKey operations require matching extra_key, but got | validation | error | radix-cache, extra-key, lora, value-error, prefix-cache |
| Rust sources under changed during the build; the result was… | exception | error | rust, build-cache, concurrency, cargo |
| seed list length must match num_outputs_per_prompt | validation | error | seed, input-validation, num-outputs |
| sglang.srt.layers.attention.nsa.nsa_backend_mtp_precompute… | console | warning | deprecation, sglang, mtp, import |
| txt_freqs_cis must be a 2D cos_sin_cache tensor | validation | error | rope, shape-validation, joyimage, multimodal |
| unknown gpu ' ', expected one of | validation | error | gpu, lookup-table, shared-memory, lplb |
| Unknown match_type: ' '. Must be 'BFS' or 'PROB'. | validation | error | argument-validation, ngram, speculative-decoding |
| Custom user-provided score_mod is not supported on SM8x… | exception | error | flash-attention, score-mod, flex-attention, sm80, ampere, unsupported-feature |
| --default-chat-template-kwargs must decode to a JSON object | validation | error | sglang, chat-template, json, server-args, validation |
| Error processing video at index | error_code | error | video, parallel, executor, error-chaining, batch |
| frequency_penalty must be in [-2, 2], got | validation | error | sampling-params, frequency-penalty, validation, sglang |
| (head_dim, head_dim_v)= | validation | error | flash-attention, sm120, shared-memory, head-dim, blackwell |
| Kimi-K3 encoder preprocessing needs an image processor | exception | error | kimi-k3, image-processor, missing-argument |
| KVTransferError | exception | critical | pd-disagg, kv-transfer, error-propagation |
| Layer-sharded HiCache backup does not support IO backend | validation | error | hicache, io-backend, context-parallelism, sglang |
| MiniMax H3 pruned curve checkpoints cannot use a separate… | validation | error | adaln, checkpoint, config-conflict, minimax-h3 |
| Modality is not supported. Supported modalities are `video`… | exception | critical | config-validation, modality, init-time, ltx2 |
| Model type must be specified | validation | error | reasoning-parser, model-type, constructor |
| Multiple config files specified! Only one allowed. | validation | error | sglang, cli, config, yaml |
| No URL recorded for browser cursor | validation | error | browser, state-corruption, url, tool-calling |
| pairs length must be greater than 0 | validation | error | pytorch, empty-tensor, scheduler, sigma-shift, validation |
| qkv_proj weight : unexpected shape ; expected fused or… | exception | error | weight-loading, shape-mismatch, gqa, tensor-parallel |
| ref2va keyframes require at least one reference condition… | validation | error | minimax-h3, ref2va, task-validation, conditions |
| reference image ratio must be within the inclusive range… | validation | error | minimax-h3, aspect-ratio, image-validation |
| return_sampling_mask with disaggregation requires… | validation | error | sglang, disaggregation, sampling-mask, env-var, config-validation |
| Scheduler terminated after | error_code | critical | event-loop, crash-loop, scheduler, circuit-breaker |
| Server process exited with code | exception | critical | server, lifecycle, startup, subprocess |
| SGLANG_DIFFUSION_ATTENTION_CONFIG is not set | validation | error | sliding-tile-attention, missing-env-var, config, init |
| Sol-Attn requires head_size= | validation | error | sol-attn, head-size, unsupported-dimension, init |
| The W4A8Int8 Fused MoE scheme is implemented only for NPU… | exception | error | quantization, moe, w4a8, npu, hardware-support |
| `time_shift_type` must either be 'exponential' or 'linear'. | validation | error | scheduler, invalid-enum-value, config |
| tokenizer_worker_num must be positive | validation | error | cuda-ipc, memory-pool, config-validation, multimodal-transport |
| Unknown channel | exception | error | harmony, channel-routing, message-parsing, sglang |
| Unknown history_scale_mode | validation | error | config, validation, attention, diffusion |
| Unknown mode | exception | error | config, enum-value, debug-utils |
| unsupported input for LTX2 QKNorm split-RoPE CUDA | validation | error | ltx, qknorm, rope, cuda, input-validation |
| ViT CUDA graph does not support attention backend | error_code | error | multimodal, vit, cuda-graph, attention-backend |
| did not declare component use | validation | error | stage, component-declaration, pipeline |
| --enable-svdquant requires --transformer-weights-path to be… | validation | error | nunchaku, svdquant, missing-path, quantization, startup-config |
| must be list[list[str]] | validation | error | validation, camera-actions, sana-wm |
| finishExternalCorpusLoad called without… | exception | error | ngram, corpus-loading, api-misuse, state-machine |
| Flash attention currently only supported for compute… | exception | error | cuda, gpu-capability, lightning-attn, hardware-unsupported |
| H3 conditioning projection produced no output | exception | error | minimax-h3, conditioning-projection, no-output, defensive |
| intermediate_size must be specified for scaled activation… | validation | error | activation, quantization, fp8, config, valueerror |
| Invalid value for : , using default | console | warning | environment-variables, configuration, parsing, fallback |
| Kimi K3 tool parameters 'properties' must be an object | validation | error | kimi-k3, json-schema, tool-calling, validation, properties |
| is required for reproducible `cargo build --locked` builds | exception | error | rust, cargo, lockfile, reproducibility |
| MiniMax H3 AdaLN cache has an unsupported or missing… | validation | error | minimax-h3, adaln-cache, version-mismatch, safetensors |
| MiniMax H3 text_encoder component must expose callable… | validation | error | minimax-h3, duck-typing, text-encoding, type-validation |
| Only gate value of 1 is supported for int type, but got | exception | error | layernorm, gate, argument-validation, cuda-kernel |
| PD state transfer failed: mamba requires single state… | exception | error | disaggregation, mamba, state-transfer, batching |
| is installed with version , which is less than the minimum… | exception | error | versioning, dependencies, packaging, pip |
| SANA-WM Triton GDN backend unavailable | error_code | error | sana-wm, gdn, triton, backend-fallback, cuda |
| Unexpected MiniMax H3 Qwen3-VL checkpoint weight | exception | critical | minimax-h3, load-weights, unknown-key, checkpoint |
| Unsupported FusedMoe scheme | exception | critical | quantization, moe, unsupported-scheme, compressed-tensors |
| v_cache must be provided | validation | error | sglang, flash-attention, kv-cache, missing-argument, validation |
| audio must be a tuple of (waveform-T… | validation | error | multimodal, audio, tuple-validation, mimo |
| CUDA VMM POSIX FD broker failed | exception | critical | cuda, vmm, ipc, file-descriptor, broker |
| --diff-threshold expects a single float shorthand or (regex… | validation | error | cli, parsing, arity, threshold |
| expert-pack index role or rank is invalid | exception | critical | moe, expert-pack, binary-format, index-entry, version-skew |
| failed to query the Rust toolchain with | exception | error | rust, toolchain, environment, subprocess |
| For classifier-free guidance, either `negative_prompt` or… | validation | error | cfg, negative-prompt, input-validation |
| GGUF tensors collide after parameter mapping at | validation | error | gguf, parameter-mapping, name-collision, checkpoint |
| Image placeholder count does not match image_data | validation | error | multimodal, placeholder-mismatch, images, valueerror |
| intel_xpu backend is only supported on decode for MLA… | validation | error | sglang, intel-xpu, mla, attention-backend, prefill-decode-split |
| [IpcModelLoader] Error communicating with daemon at | error_code | critical | weight-cache, ipc, daemon, socket |
| Kimi-K3 DCP + DSPARK currently requires… | validation | error | kimi-k3, dspark, speculative-decoding, env-var, sglang |
| kv-canary: scatter_req_token_ids flat_in must be int64, got | validation | error | kv-cache, dtype-validation, torch |
| Length mismatch: lora_nicknames has | validation | error | lora, argument-validation, length-mismatch |
| LSE partial tensor must be Float32 | validation | error | flash-attention, dtype-mismatch, numerical, validation |
| MXFP8 KV cache does not support SGLANG_USE_HND_KVCACHE. | validation | error | kv-cache, mxfp8, hnd-layout, env-var |
| Only CUDA, MUSA and NPU support GGUF quantization currently. | console | warning | sglang, gguf, quantization, platform-support |