sgl-project/sglang
Documented errors, page 12 of 32. Back to sgl-project/sglang
| Code / Message | Type | Severity | Tags |
|---|---|---|---|
| Comfy tensorwise INT8 layer | validation | error | quantization, int8, tensorwise, rank-mismatch, checkpoint-validation |
| --enable-unified-memory supports monolithic (decode)… | validation | error | sglang, unified-memory, cuda-graph, prefill, piecewise |
| Expected image placeholder(s), found . | validation | error | multimodal, kimi-k3, cpu-fallback, string-split, validation |
| expert-pack data offset is invalid | exception | critical | moe, expert-pack, binary-format, offset, alignment |
| Failed to import sglang.private.private_model_loader | exception | error | load-format, private-module, import-error, sglang, internal-build |
| fl2va Qwen preparation requires one or two ordered images… | validation | error | minimax-h3, fl2va, keyframes, validation |
| gRPC mode requires the smg-grpc-servicer package. If not… | exception | critical | grpc, dependency, import-error, installation, sglang |
| grpcs:// is not supported; use grpc:// | validation | error | grpc, url-scheme, tls, network, configuration |
| MiniMax H3 text payload must contain positive.hidden_states… | validation | error | minimax-h3, text-encoding, tensor-shape, payload-validation |
| is not implemented for | validation | error | scheduler, diffusion, solver, unipc |
| sound generation was requested (sound_duration > 0) but the… | validation | error | cosmos3, audio, checkpoint-capability, validation |
| tensor_model_parallel_size | exception | error | parallelism, decode-context-parallel, tensor-parallel, divisibility, config-validation, sglang |
| Unsupported mlp_layer_types | exception | critical | mellum, config-validation, moe |
| video latent spatial/time dims must be divisible by… | validation | error | validation, tensor-shape, patchify, divisibility, minimax-h3 |
| Weight cache daemon (pid= ) exited prematurely with code | exception | critical | sglang, weight-cache, daemon-crash, startup |
| When enabling two batch overlap without an EP a2a backend… | validation | error | sglang, two-batch-overlap, dp-attention, moe, server-args |
| Activation function is not supported. | validation | error | activation, registry, lookup-failed, model-config |
| adapt_shape_v1 ratio must be within the inclusive range 1:4… | validation | error | minimax-h3, aspect-ratio, range-check, spatial |
| AITer backend does not have a metadata builder. | exception | error | aiter, attention-backend, rocm, not-implemented |
| Cannot collate mixed VLA noise presence | validation | error | vla, diffusion-noise, batching, validation |
| Component attention backend key must be a string | validation | error | config, type-error, attention-backend |
| checkpoint declares quantization metadata in (quant_method=… | validation | error | quantization, model-loading, fail-closed, state-dict |
| DSPARK layer capture is not available in encoder-only mode | exception | error | kimi-k3, encoder-only, dspark, attribute-error |
| For FP8 Fused MoE layer, we require either per tensor or… | validation | error | quantization, fp8, moe, static-scales, input-quantization |
| --gpu-ids contains a non-integer GPU id | validation | error | gpu, cli-arguments, validation |
| [Grafter] tags= matched BOTH grafter_b2t_filter and… | exception | error | distributed, grafter, filter-config |
| Hunyuan3D Paint does not use extra UNet conditioning. | validation | error | runtime, api-misuse, diffusion, unet |
| Kimi K3 tool property schemas must be JSON schemas | validation | error | kimi-k3, json-schema, tool-calling, validation |
| kv-canary: real_kv_sources | validation | error | kv-canary, shape-validation, tensor-layout |
| media URL timeout must be positive | validation | error | validation, media, timeout, network |
| MHATokenToKOnlyPool: use set_index_k_buffer on the parent… | exception | error | kv-cache, k-only-pool, sparse-attention, minimax, api-misuse, sglang |
| noise_aug must be in [0, 1], got | validation | error | minimax-h3, noise-aug, parameter-validation, range-check |
| only one of 'replacement', 'prepend', 'append' may be set… | validation | error | validation, edit-spec, mutually-exclusive, sglang |
| .aspect_ratio must be "auto" for task , got | validation | error | minimax-h3, aspect-ratio, validation |
| --prefill-only-disable-kv-cache is incompatible with… | validation | error | sglang, context-parallelism, kv-cache, distributed, server-args |
| q can only be None when only_qv=True | validation | error | sglang, flash-attention, invalid-argument, validation |
| Scheduler not initialized; call set_timesteps() first | exception | error | scheduler, initialization-order, flow-match, state-error |
| speculative miss_count must have shape [batch]. | validation | error | hisparse, shape-validation, miss-plan |
| --ssl-ca-certs has no effect without --ssl-certfile and… | validation | error | sglang, ssl, tls, argument-validation |
| task resolves outside partition | validation | error | minimax-h3, task-routing, partition |
| The quantization config must be a subclass of… | exception | error | quantization, type-validation, subclass, config |
| VMM handle export failed: FABRIC export failed on at least… | exception | critical | cuda, distributed, fabric, p2p, driver-support |
| consumer_count must be 1, the attention TP size, or the… | validation | error | cuda, vmm, validation, parallelism |
| Decode attention backend for Kimi-K3 DCP must be… | validation | error | kimi-k3, attention-backend, dcp, sglang |
| DeepEP v2 MoE currently supports only --moe-runner-backend… | validation | error | moe, deepep, runner-backend, deep-gemm, server-args |
| DSpARK KDA MTP requires a fixed 1 + num_spec dense tokens… | validation | error | kda, mtp, cu-seqlens, uniform-batch |
| --enable-unified-memory only supports hybrid Mamba and… | validation | error | unified-memory, model-architecture, kv-cache, boot-config |
| Encoder request was released | http | error | lifecycle, race-condition, encoder, send |
| Expected a PIL image, got | validation | error | dots3, type-error, pil, multimodal, input-validation |
| Failed to implicitly load LoRA adapter | validation | error | lora, adapter-load, oom, incompatible-weights |
| Kimi expert object is not contiguous at index | validation | critical | kimi, moe, expert-pack, layout |
| Kimi-K3 expects one vision grid per MultimodalDataItem… | exception | error | kimi-k3, multimodal, grid-thws, data-shape |
| --kv-cache-dtype=nvfp4 requires Blackwell SM100 or SM120… | validation | error | sglang, nvfp4, kv-cache-dtype, blackwell, gpu-architecture |
| kv-canary: at most RealKvSource entries supported by the… | validation | error | kv-canary, cuda-abi, limit-exceeded |
| layer_types has entries but num_hidden_layers is | validation | error | config-validation, layer-types, muse-glimmer |
| mask_candidates is required for STA_searching mode | validation | error | sta, missing-argument, kwargs-validation |
| mooncake encoder_transfer_backend requires HTTP encoders… | validation | error | mooncake, rdma, grpc, disaggregation, configuration, epd |
| Multiple distributions register serve backend | exception | error | sglang, cli, entry-points, duplicate-registration, plugin-conflict |
| Pi05 v1 expects one state vector per request | validation | error | pi05, vla, state-input, batch-size, torch |
| --radix-cache-backend= | validation | error | radix-cache, backend-registry, plugin, server-args, value-error |
| Speculative algorithm | validation | error | speculative-decoding, plugin-api, duplicate-registration |
| This browser cannot encode H.264 MP4 | error_code | error | ngram, config, parsing, speculative-decoding |
| This layer doesn't support feature dim >= 64KB. | validation | error | triton, l2norm, feature-dim, kda, attention |
| Warning: system prompt is not supported in VertexAI. | console | warning | vertexai, system-prompt, chat-roles, warning |
| Action normalization stats not found at | exception | error | cosmos3, file-not-found, normalization-stats, checkpoint |
| DSV4 target and draft pools must share the SWA index mapping | validation | error | disaggregation, deepseek-v4, swa, index-mapping |
| experimental_sgl_marlin EP requires --moe-a2a-backend none | validation | error | moe, expert-parallelism, a2a, marlin, experimental, sglang |
| forward_batch with seq_lens is required for TopK retrieval | validation | error | sparse-attention, retrieval, forward-batch, seq-lens, value-error |
| Invalid allowed media domain | validation | error | validation, media, security, idna, dns |
| Invalid projection layers | exception | critical | mimo-audio, config-validation, audio-encoder, version-mismatch |
| Invalid VLA prefix cache layer | exception | error | vla, prefix-cache, kv-cache, index-error |
| num_inference_steps is required for transformer-only mode… | validation | error | cache-dit, missing-config, inference-steps, transformer-only |
| num_target_layers must be positive, got | validation | error | speculative-decoding, dflash, config-validation |
| --speculative-ngram-external-corpus-max-tokens must be… | validation | error | speculative-decoding, ngram, external-corpus, server-args, validation |
| SWA Radix tree sanity check failed, ping @hanming-lu | exception | critical | swa, radix-cache, internal-bug, sanity-check |
| The quantization method moe_wna16 + awq is not supported… | validation | error | quantization, awq, moe, gpu-capability, hardware |
| actions must be a list[list[str]] | validation | error | sglang, lingbot-world, type-validation, actions, embodied-ai |
| AITER Sage attention is not available, please update AITER… | exception | critical | aiter, sage-attention, rocm, dependency-version, import-error |
| Comfy W4A8 layer needs I8 weights and FP8 group scales, got… | validation | error | quantization, dtype-mismatch, w4a8, safetensors |
| CUDA VMM proxy has no shareable handle | exception | error | cuda, vmm, ipc, handle, multiprocessing |
| --enable-linear-replayssm-spec with DSPARK/DFLASH requires… | validation | error | speculative-decoding, kda, kimi-linear, dspark, dflash, boot-config |
| Failed to find config.json for | validation | error | model-repo, diffusers, config-not-found |
| Failed to register buffer to SiMM | exception | critical | rdma, memory-registration, hierarchical-cache, sim |
| Feature size mismatch | exception | error | mimo-audio, tensor-shape, batching, audio |
| Gemma4AssistantForCausalLM draft requires… | validation | error | sglang, speculative-decoding, eagle3, gemma4, model-incompatibility |
| Invalid . Expected tp_size >= 1. | exception | error | parallelism, tensor-parallel, init-order, distributed, ltx2 |
| k_cache can only be None when only_qv=True | validation | error | sglang, flash-attention, kv-cache, invalid-argument |
| Kimi K3 required parameter | validation | error | kimi-k3, json-schema, required-fields, validation |
| Model config does not contain a _class_name attribute. Only… | exception | error | model-loading, diffusers, adapter, config |
| native MXFP8 MoE only supports gated swiglu-oai, got | exception | error | moe, mxfp8, rocm, not-implemented, activation |
| No browser tool call found | validation | error | tool-calling, browser, validation, sglang |
| Quanto layer is missing tensors | validation | error | quantization, quanto, missing-tensor, checkpoint |
| Quanto layers collide after parameter mapping at | validation | error | quantization, quanto, name-mapping, collision |
| SGLANG_ROLE_NAMESPACES= | validation | error | environment-variable, config-validation, typo, sglang |
| Tag mismatch: expected CMD_PUT_META, got | exception | critical | flexkv, pipeline-parallel, protocol-mismatch, distributed |
| The installed Mooncake version does not support tenant_id… | exception | error | mooncake, version-mismatch, tenant, dependency |
| Unknown dtype | validation | error | frontend, openai, dtype, invalid-argument, sglang |
| Weight output_size_per_partition = | validation | error | marlin, tensor-parallel, shape-validation, gptq, awq |
| handle_types must be 'auto', an integer, or None | validation | error | cuda, vmm, validation, handle, type-error |
| Invalid image | validation | error | image, multimodal, validation, input-validation |