sgl-project/sglang
Documented errors, page 9 of 32. Back to sgl-project/sglang
| Code / Message | Type | Severity | Tags |
|---|---|---|---|
| Packed decode kernel only supports NK=1 | exception | error | kernel-limit, triton-kernel, packed-decode |
| prefix-valid commit is unsupported for MXFP8 KV cache (it… | exception | error | mxfp8, kv-cache, prefix-caching, not-implemented, sglang |
| SGLANG_DISAGG_STAGING_BUFFER is designed for non-MLA models… | validation | error | disaggregation, staging-buffer, mla, env-var, unsupported-feature |
| --sidecar requires importable module | exception | critical | sidecar, module-import, startup, sglang, cli |
| --swa-full-tokens-ratio should be in range (0, 1.0]. | validation | error | sglang, server-args, swa, validation, config |
| Tag mismatch: expected CMD_LAYERWISE, got | exception | critical | flexkv, pipeline-parallel, protocol-mismatch, distributed |
| tasks must contain canonical task names, got | validation | error | minimax-h3, task-names, canonicalization, model-index |
| `A_log` and `dt_bias` must have | exception | error | fla, fused-recurrent, mamba-params, tensor-parallel |
| Cosmos3 rollout does not support action/sound modalities. | validation | error | cosmos3, rollout, action-latents, sound-latents, not-implemented |
| Cosmos3TokenizationStage requires a tokenizer; expected the… | validation | error | cosmos3, tokenizer, qwen2, checkpoint, init-validation |
| requires cache_head_start when cache heads ( ) differ from… | validation | error | kv-cache, gqa, head-slicing, shape-mismatch |
| experimental_sgl_marlin LoRA requires --lora-backend triton | validation | error | lora, backend, triton, marlin, experimental, sglang |
| input image is empty | validation | error | image-processing, mask, segmentation, mesh3d, validation |
| Kimi expert-pack range mismatch at index | validation | critical | kimi, moe, expert-pack, range-validation |
| kv-canary: expected input tensors are required when… | validation | error | kv-canary, argument-validation, assert-mode |
| no staged embedding for /send req_id= | http | error | encoder, send, request-lifecycle, race-condition |
| num_heads ( ) must be divisible by tp_size ( ). | validation | critical | tensor-parallel, model-config, startup, tp-sharding |
| num_kv_tokens must fit in the provided source pages, got… | validation | error | disaggregation, dcp, capacity-check, kv-transfer |
| prompt has image placeholder token(s) but image(s) were… | validation | error | multimodal, tokenization, placeholder-mismatch, input-validation |
| qkv weight has incompatible output dim for grouped… | exception | error | qkv, gqa, checkpoint-loading, weight-reorder, shape-mismatch |
| Subclasses of BaseScheduler must define | exception | error | scheduler, abstract-base, subclass-contract, diffusion |
| T5 attention bias with bucketed positions is not yet tested | validation | error | phi4, t5, attention-bias, not-implemented |
| Unsupported Comfy NVFP4 companion format(s): + "… | validation | error | minimax-h3, nvfp4, quantization, unsupported-format |
| Unsupported device type | validation | error | platform-plugin, device, configuration |
| unsupported profile stage | exception | error | profiling, forward-mode, scheduler, internal-error, sglang |
| Video file is corrupted or cannot be decoded | http | error | video, decode, http-exception, corrupt-file |
| world_size ( ) is not equal to tensor_model_parallel_size (… | exception | error | parallelism, tensor-parallel, pipeline-parallel, world-size, config-validation, sglang |
| Deterministic inference with absorbed-MLA models on the fa4… | validation | error | sglang, fa4, cuda-arch, deterministic-inference, blackwell |
| Failed to read JSON file | validation | error | config, file-permissions, mooncake, io |
| MiniMax H3 text encode failed on rank | error_code | error | minimax-h3, data-parallel, error-propagation, text-encoding |
| --mm-feature-transport=cuda_ipc only supports a single node. | validation | error | sglang, cuda-ipc, multi-node, multimodal, distributed |
| Not support norm_type | exception | error | kimi-k3, vision, config-validation, not-implemented |
| raw_action_dim must be in | validation | error | cosmos3, validation, range-check, action-dim |
| raw q_proj.weight has shape | validation | error | shape-mismatch, fused-gate, packaging |
| safetensors metadata | validation | error | lora, safetensors, metadata, peft |
| Unknown type | validation | error | type-error, interpreter, dsl, validation |
| Unsupported layout | validation | error | sglang, kv-cache, layout, host-pool, initialization |
| Unsupported native SD cross-attention arguments | validation | error | cross-attention, unsupported-argument, stable-diffusion |
| Z-Image text embeddings must have shape [seq, dim] or… | validation | error | z-image, multimodal, tensor-shape, text-embeddings, validation |
| Expected BasicTransformerBlock, got | validation | error | initialization, diffusers, type-mismatch |
| experimental_sgl_marlin EP requires trivial expert… | validation | error | moe, expert-parallelism, eplb, marlin, experimental, sglang |
| Failed to load LoRA adapter | validation | error | lora, pinned, capacity, config, sglang |
| I2I mode is not supported yet via external SGLang encoder… | validation | error | glm-image, external-encoder, image-to-image, notimplemented |
| image payload requires b64_json | validation | error | image-payload, base64, input-validation, action-endpoint |
| Invalid head config inferred from mixed_qkv: H= | exception | error | kda, helion, head-config, shape-validation |
| item_first is not supported when embeddings are supplied | validation | error | sglang, scoring, embedding-overrides, argument-validation |
| transition actions must be a list | validation | error | control-events, schema-validation, realtime |
| LoRA targets the DSA indexer | validation | error | lora, dsa, indexer, fusion, env-var, sglang |
| LTX2DurationHead requires at least one of video_tokens /… | validation | error | multimodal, duration-head, argument-validation, ltx-2 |
| ModelOpt quantization config | exception | error | modelopt, version-mismatch, attributeerror, quantization, sglang |
| SANA-WM first-frame conditioning failed; refusing to… | exception | critical | sglang, sana-wm, first-frame-conditioning, fail-fast, runtime-wrapper |
| SANA-WM height/width must be divisible by the LTX-2 spatial… | validation | error | sana-wm, vae-stride, resolution-validation, video-generation |
| The MLX tensor bridge requires MLX >= 0.32.0 | exception | error | sglang, mlx, dependency, import-error, macos |
| tokenspeed_mla backend can only be used with MLA models. | validation | error | attention-backend, tokenspeed-mla, mla, model-arch-mismatch, sglang |
| TP size must be positive. | validation | critical | config, tensor-parallel, minimax-h3, validation, init |
| action output dimensions must be non-zero, got | validation | error | action-inference, shape-validation, numpy |
| allowed media domains must be strings | validation | error | validation, media, security, ssrf |
| Cosmos3 action requests require domain_name or domain_id | validation | error | cosmos3, domain-config, missing-required-field |
| cuda.bindings.driver is required for CUDA VMM operations | exception | critical | cuda, vmm, import-error, environment, dependencies |
| Expected len(mlp_layer_types) == num_hidden_layers, got | exception | critical | mellum, config-validation, moe, layer-config |
| External ngram corpus exceeds the configured token limit | validation | error | sglang, ngram, token-limit, budget-exceeded |
| generate_action requires an ACTION pipeline, got | validation | error | input-validation, action-pipeline, data-type, diffusion |
| Invalid device_uuid= | exception | error | cuda, device-uuid, gpu, torch-patch, sglang |
| Invalid ltx2_two_stage_device_mode= | validation | error | ltx2, device-mode, env-var, invalid-value, sglang |
| Invalid packed Q size | exception | error | kda, helion, gqa, head-config |
| invalid predicate : ; allowed names are . | validation | error | predicate, dsl, name-error, whitelist |
| kill_process_tree: process(es) not reaped within s after… | exception | error | process-management, sigkill, subprocess, cuda, zombie-process |
| Kimi-K3 GGUF ssm_a must contain only -exp(A_log) values | validation | error | kimi-k3, gguf, ssm, a-log |
| --load-publish-endpoint needs an active --kv-events-config… | validation | error | sglang, kv-events, config-validation, load-publish |
| Qwen-Image-Layered generated latent shapes must match, got | validation | error | qwen-image, layered-generation, shape-mismatch, latent-shapes |
| Request mixes standalone audio and video-with-audio; EPD… | exception | error | audio, video, not-implemented, epd, multimodal |
| Required `vision_config.model_type` is not found in… | validation | error | multimodal, config, model-loading, missing-field |
| sparse_attn_v4_paged_decode expects fp16/bf16 q, got | validation | error | attention, dtype, triton, deepseek, gpu |
| Stochastic rounding for the Mamba SSM cache requires… | validation | error | sglang, mamba, stochastic-rounding, dtype, server-args |
| Unsupported activation scheme | validation | error | quantization, fp8, config-validation, checkpoint-config |
| Unsupported dtype | validation | error | numpy, msgpack, serialization, dtype-validation |
| Unsupported ModelSlim MoE schemes for layer | validation | error | modelslim, quantization, moe, unsupported-scheme, version-mismatch |
| Cannot split GLM-Image AR output for sequential inference… | exception | error | glm-image, sequential-inference, batch-split, shape-mismatch, runtimeerror |
| --enable-dsa-cache-layer-split is only supported on PD… | validation | error | dsa, pd-disaggregation, config-validation |
| Encoded prompt has tokens, expected at least | validation | error | sana-video, prompt-window, sequence-length, validation, sglang |
| Feature attrs ( ) not found in | validation | error | multimodal, schema, preprocessor, validation |
| Invalid style | validation | error | conversation, chat-template, parser |
| match text found multiple times | exception | error | source-patching, text-match, ambiguity, sglang |
| server_args is required to resolve Ideogram4 NVFP4 paths | validation | error | ideogram, nvfp4, missing-argument, lazy-init, sglang |
| Weight source is neither a local path nor an owner/repo… | validation | error | weights, source-parsing, huggingface |
| --enable-linear-replayssm requires Triton, or Helion for… | validation | error | sglang, replayssm, linear-attention, backend-validation |
| Failed to connect to metadata server | exception | critical | hf3fs, metadata-server, connection-refused, retries-exhausted |
| Incomplete Diffusers H3 fused parameters | exception | error | checkpoint-loading, weight-mapping, diffusers, minimax-h3, state-dict |
| KDA `a` must be a contiguous 2D or 3D tensor. | exception | error | kda, replayssm, contiguity, tensor-ndim |
| Kimi image placeholders must map one-to-one to image data… | validation | error | kimi, k25, multimodal, loader, count-mismatch |
| MiniMax H3 Qwen3-VL encoders smaller than 32B require… | exception | critical | minimax-h3, conditioning-projection, missing-component, server-args |
| --model-variant requires ' ' in the checkpoint, but does… | validation | error | ltx2, model-variant, partial-download, checkpoint, sglang |
| NIXL KVReceiver Exception | exception | critical | sglang, nixl, kv-receive, disaggregation, distributed-inference |
| Packed pixel_values token count does not match… | validation | error | siglip2, multimodal, preprocessing, shape-mismatch |
| PD KV layout mismatch on the whole-envelope path: prefill… | exception | error | pd-disagg, unified-memory, kv-layout, page-size |
| prefix-valid commit is unsupported under the page-major… | exception | error | kv-cache, prefix-cache, page-major, not-implemented |
| select/choices is not supported for chat models. Please try… | exception | error | frontend, openai, choices, chat-model, not-supported, sglang |
| Shared memory not found | exception | critical | sglang, shared-memory, startup, race-condition, file-not-found |
| The number of intermediate state indices is expected to be… | validation | error | pytorch, state-management, gdn, validation |
| DeepEP v2 does not forward deterministic=True to… | validation | error | deepep, determinism, moe, server-args, elastic-buffer |