sgl-project/sglang
Documented errors, page 8 of 32. Back to sgl-project/sglang
| Code / Message | Type | Severity | Tags |
|---|---|---|---|
| Unknown disagg_role | validation | error | cli-args, dispatch, enum-validation, disaggregated-serving |
| Cannot combine tool calls with constrained decoding… | validation | error | responses-api, tool-calling, constrained-decoding, structured-output, sglang |
| Comfy W4A8 layer has an incompatible correction tensor | validation | error | quantization, shape-mismatch, w4a8, correction-tensor |
| Could not connect to remote scheduler at | exception | critical | connection, scheduler, remote-mode, startup |
| DSPARK requires explicit layer_ids for aux hidden capture. | exception | error | dspark, speculative-decoding, argument-validation, kimi |
| --enable-dsa-cache-layer-split requires --enable-prefill-cp… | validation | error | dsa, context-parallel, config-validation |
| GenerativeModel | error_code | error | import-error, optional-dependency, vertexai, environment |
| Inconsistent receive_count for req_id= | http | error | validation, refcount, dp, tensor-parallel |
| `initial_state` must be contiguous in the last dim. | validation | error | kda, helion, contiguity, tensor-stride |
| `max_possible_layers` must be provided alongside… | validation | error | siglip2, api-misuse, layer-selection |
| must be a torch.Tensor | validation | error | diffusion, scheduler, type-validation, timestep |
| Not support merge_type | exception | error | kimi-k3, vision, config-validation, not-implemented |
| PD state transfer failed: state component count mismatch | exception | error | disaggregation, hybrid-model, state-transfer, version-mismatch |
| QwenImageEditPlus expects either one shared condition image… | validation | error | qwen-image, image-editing, batch-validation, multimodal |
| Cosmos3 action_horizon must be divisible by 4 so num_frames… | validation | error | cosmos3, temporal-vae, divisibility, action-horizon |
| cuda-graph mode cu_seqlens should be a list | exception | error | sglang, vision-transformer, cuda-graph, type-validation, cu-seqlens |
| DSV4 HiSparse direct PD transfer currently requires the… | validation | error | disaggregation, hisparse, dsv4, mooncake, unsupported-backend |
| Duplicate tensor names detected across safetensors files… | validation | critical | safetensors, duplicate-keys, checkpoint, weights |
| engine_info_bootstrap_port | exception | error | sglang, port-conflict, multi-instance, startup, bootstrap |
| experimental_sgl_marlin configuration is unsupported: + "… | validation | error | gpu, compute-capability, marlin, hardware, experimental, sglang |
| InklingMultimodalProcessor v1 requires pre-rendered… | validation | error | multimodal, inkling, pre-tokenized-input, missing-tokenizer |
| Kimi-K3 GGUF ssm_a must contain finite floating values | validation | error | kimi-k3, gguf, kda, mamba, nan |
| PD DCP source/destination KV geometry differs: src= | exception | critical | disaggregation, dcp, kv-cache, geometry-mismatch |
| --prefill-only-disable-kv-cache expected… | exception | error | prefill-only, kv-cache, pool-family, boot-config |
| The 'checksums' format is deprecated. Please regenerate… | console | info | model-verification, manifest, deprecation, checksums |
| Unsupported type | validation | error | cutedsl, dtype, type-mismatch, kernel |
| width and height must be non-zero, got width= | validation | error | cosmos3, resolution, aspect-ratio, validation |
| Cannot msgpack decode object of type | exception | error | serialization, msgpack, version-mismatch |
| Found static activation scheme for checkpoint that was not… | validation | error | quantization, fp8, moe, activation-scheme, checkpoint-config |
| In-place vision RoPE requires complex64 frequencies, got | validation | error | vision-rope, dtype, complex-tensor |
| KernelSpec.target must be 'module:attr', got | validation | error | kernels, spec-validation, configuration, sglang |
| ltx2_two_stage_device_mode=resident conflicts with explicit… | validation | error | ltx2, config-conflict, device-placement |
| num_attention_heads must be positive. | validation | critical | config, model-architecture, validation, minimax-h3 |
| Prefill out of memory. Try to lower your batch size.\nTry… | exception | critical | sglang, memory, paged-kv, prefill |
| prompt has tokens, exceeds max_text_tokens= | validation | error | ideogram, prompt-length, token-limit, validation |
| speculative_ngram_external_sam_budget must be less than or… | validation | error | speculative-decoding, ngram, external-corpus, draft-tokens, server-args, validation |
| Unsupported grid type for kimi image tokens | validation | error | kimi, multimodal, type-validation, grid-metadata |
| auxiliary output does not support pipeline-parallel… | exception | error | pipeline-parallel, sampling, type-validation, distributed |
| Comfy INT8 layer has incompatible weight/scale shapes: and | validation | error | quantization, int8, rowwise, shape-mismatch, checkpoint-validation |
| Failed to load customized | error_code | critical | model-loading, fallback, quantization, wrapper-exception |
| must be , got . | validation | error | speculative-decoding, dflash, config-validation, range-check |
| FP8 weight_block_size must contain two positive integers… | validation | error | quantization, fp8, config-validation, humming |
| Multi-node weight cache daemons (nnodes > 1) require… | exception | error | sglang, multi-node, distributed, weight-cache, startup |
| Multiple serve backends matched this request | exception | error | sglang, cli, auto-detection, ambiguity, model-type |
| .kind unsupported for ref2va | validation | error | validation, enum, multimodal, minimax-h3 |
| SGLang currently only supports inferring LoRA target… | validation | error | lora, target-modules, type-error, peft-config, sglang |
| Weight output_size_per_partition = | validation | error | marlin, tensor-parallel, pack-factor, shape-validation |
| Conditioning (prompt_embeds) must be provided | validation | error | hunyuan3d, pipeline-order, missing-state, conditioning |
| CPU offloading is unsupported under the page-major layout… | exception | error | kv-cache, cpu-offload, page-major, not-implemented |
| Either `prompt` or `prompt_embeds` must be provided | validation | error | prompt-validation, task-type, input-validation |
| GGUF tensor is not aligned to -element super blocks | validation | error | gguf, quantization, super-block, alignment |
| Hunyuan3D only supports a single image input. | validation | error | hunyuan3d, input-validation, single-image-constraint |
| kv-canary: forward_batch token count= | validation | error | kv-canary, capacity, chunked-prefill, max-prefill-tokens, sglang |
| paired audio memory slot requires audio_latent | validation | error | joy-echo, audio, memory-slot, missing-argument |
| Pi05 v1 expects one prompt per action request | validation | error | pi05, vla, prompt-validation, batch-size |
| prepared reference videos payload must carry a non-empty… | validation | error | minimax-h3, ref2va, reference-video, payload-validation |
| Unhandled known MessagePack extension code | exception | error | serialization, msgpack, ipc, version-mismatch |
| chunk plan does not cover the incoming frames | validation | error | sana-wm, streaming, chunk-plan, frame-count, valueerror |
| denoising_step_list must end with 0, got | validation | error | sana-wm, self-forcing, sigma-schedule, denoising-steps, valueerror |
| [FlexKV] Failed to connect to eventfd socket | exception | critical | flexkv, unix-socket, worker-startup, connection-refused |
| Found different quantization schemes for | validation | error | quantization, compressed-tensors, fused-layers, config-mismatch |
| Host cache is not supported yet | exception | error | radix-cache, cpp-backend, hicache, not-implemented, host-offload |
| Invalid allowed media domain | validation | error | validation, media, security, ssrf, url-parsing |
| Invalid stacked fused KV projection shape: got | validation | error | shape-validation, speculative-decoding, fused-kernel |
| kv_cache_quant_config must be QVGKVQuantArgs or a dict | validation | error | configuration, type-validation, quantization, kv-cache |
| MiniMax H3 AdaLN cache has invalid timestep plans | validation | error | minimax-h3, adaln-cache, corrupt-cache, validation |
| MiniMax H3 audio encode failed on rank 0 | exception | critical | minimax-h3, audio-encoding, distributed, rank0 |
| is deprecated; use instead. | console | warning | deprecation, sglang, dcp, distributed |
| Staging is enabled but kv_manager._staging_ctx.allocator is… | exception | error | disaggregation, staging-buffer, env-var, allocator, initialization |
| Subclasses of BaseDiT must define | validation | error | sglang, dit, post-init, attribute-validation |
| TensorRT-LLM MLA backend only supports kv-cache-dtype of… | validation | error | sglang, trtllm, mla, kv-cache-dtype, config-validation |
| thinking.budget_tokens is not allowed when thinking.type is… | validation | error | anthropic, thinking, validation, request-validation, pydantic |
| aiter is required when SGLANG_USE_AITER is set to True | validation | critical | moe, rocm, aiter, import-error, env-var, amd |
| already declares checkpoint quantization; drop the explicit… | validation | error | quantization, online-quantization, conflicting-config, text-encoder |
| error: unrecognized arguments | console | error | cli, argparse, argument-validation, startup |
| expected recurrent state layout [pool, H, V=128, K=128] | validation | error | kda, mtp, recurrent-state, dtype-stride |
| expert cache and staging budgets, and read splits, must be… | exception | error | moe, expert-pack, configuration, vram-budget, argument-validation |
| must be a tensor, list of tensors, list of sequence-length… | validation | error | type-validation, conditioning, multimodal |
| FlashAttention did not return the softmax LSE required by… | exception | critical | flash-attention, ring-attention, lse, contract-violation |
| invalid reasoning effort | validation | error | openai-api, reasoning, validation, sglang |
| {label}: {err} | exception | error | cuda, driver-api, vmm, error-code |
| min_new_tokens must be in [0, max_new_tokens], got | validation | error | sampling-params, min-new-tokens, validation, sglang |
| missing captured KV on | exception | error | sana-wm, streaming-refiner, kv-capture, runtimeerror, state-lifecycle |
| Not a canonical UMMA_MN Layout: Expected profile failure. | validation | error | cutlass, sm100, layout, shared-memory |
| at is not available in seconds. | exception | error | network, port, timeout, ci |
| Rank namespace ' ' not initialized. Please call /… | http | error | hf3fs, metadata-server, http-404, initialization-order |
| recycle_interval must be positive | validation | error | multimodal, memory-pool, argument-validation |
| SGLANG_USE_MLX requires an available MLX Metal device | exception | error | mlx, metal, apple-silicon, device-unavailable, sglang |
| Unknown image_mode ' '. Supported | exception | error | multimodal, ocr, invalid-argument, config-validation |
| Unsupported dtype . Supported: float16, bfloat16, float32 | validation | error | dtype-validation, mamba, causal-conv1d, jit-kernel |
| Cosmos3 action input accepts one image field; use a list or… | validation | error | cosmos3, observation, image-input, validation |
| Failed to load MiniMax H3 Qwen3-VL weight | exception | critical | minimax-h3, load-weights, weight-loading, tensor-parallel |
| get_split_heads_page_buffer_meta requires… | validation | error | sglang, hicache, not-implemented, kv-cache |
| Got LoRA adapter that has never been loaded | validation | error | lora, eviction, state-desync, cache |
| Invalid IPv6 address inside brackets | validation | error | network, ipv6, address-parsing, sglang |
| kv-canary: RealKvSource.page_size must be >= 1, got | validation | error | kv-cache, config-validation, page-size |
| latent_window_size must be positive, got | validation | error | joyecho, memory, window-size, validation |
| MLA model is not supported without global metadata server… | validation | error | hf3fs, mla, missing-config, env-var |
| MultiPlatformOp is deprecated; subclass… | console | warning | deprecation, sglang, operators, subclassing |
| num_inference_steps is required for dual-transformer mode… | validation | error | cache-dit, dual-transformer, missing-config, num-inference-steps |