ErrLookup › hiyouga/LlamaFactory
hiyouga/LlamaFactory
Unified Efficient Fine-Tuning of 100+ LLMs & VLMs (ACL 2024) · Python · 259 source files
Analyzed at f28afaf635 on 2026-08-14. 401 documented errors.
| Code / Message | Type | Severity | Tags |
|---|---|---|---|
| Qwen2VL requires 3D position ids for mrope. | exception | error | dependency, qwen3-5, packing, flash-linear-attention, installation |
| Please upgrade `transformers` to 4.34.0 | exception | error | model, internvl, multimodal, checkpoint-format, config |
| Unable to process key {key} | exception | error | model, llava, multimodal, checkpoint-format, config |
| All 'dcp_path', 'hf_path', and 'config_path' are required. | exception | error | version, transformers, internlm3, dependency |
| Both 'hf_path' and 'dcp_path' are required. | exception | error | version, transformers, lfm2-vl, dependency |
| `num_layers` {num_layers} should be divisible by `num_expand | exception | error | version, transformers, dpo, dependency |
| bf16 and fp16 cannot be both True. | exception | error | dpo, loss-function, reference-model, config |
| The model does not have a submodule named '{submodule_name}' | exception | error | hyper-parallel, context-parallel, dataset, streaming |
| Unsupported model type: {getattr(config, 'model_type')}. | exception | error | dataset, config, hyper-parallel, training |
| Stage does not supported: {stage}. | exception | error | dependency, hyper-parallel, installation |
| Device not supported: {device_name}. | exception | error | version, transformers, kto, dependency |
| Stage does not supported: {stage}. | exception | error | kto, dataset, data-quality, collator |
| Pipeline parallel size should be smaller than the number of | exception | error | dependency, megatron, mcore-adapter, installation |
| Invalid API key. | http | error | megatron, model-support, version, config |
| Not allowed | http | error | megatron-bridge, chat-template, jinja, dataset |
| Invalid length | http | error | megatron-bridge, model-support, multimodal, config |
| Only supports u/a/u/a/u... | http | error | dependency, megatron-bridge, installation, nemo |
| Invalid role | http | error | ppo, rlhf, eval-dataset, config, not-implemented |
| Invalid input type {input_item.type}. | http | error | api, multimodal, validation, http-400, chat |
| Invalid tools | http | error | api, tools, function-calling, validation, http-400 |
| Cannot stream function calls. | http | error | api, streaming, tools, http-400, function-calling |
| Cannot stream multiple responses. | http | error | api, streaming, sampling, http-400 |
| Invalid request | http | error | api, validation, scoring, http-400 |
| Local file access is disabled. | http | error | security, lfi, api, http-403, multimodal |
| File access is restricted to the safe media directory. | http | error | security, lfi, path-traversal, http-403, multimodal |
| Invalid or inaccessible file path. | http | error | security, filesystem, permissions, http-400, lfi |
| Only HTTP/HTTPS URLs are allowed. | http | error | security, ssrf, url-validation, http-400 |
| Invalid URL hostname. | http | error | security, ssrf, url-validation, http-400 |
| Access to private or reserved IP addresses is not allowed. | http | error | security, ssrf, network, http-403 |
| Could not resolve hostname: {parsed_url.hostname} | http | error | security, ssrf, dns, network, http-400 |
| Invalid URL: {e} | http | error | security, ssrf, url-validation, http-400 |
| vLLM not install, you may need to run `pip install vllm` or | exception | error | installation, vllm, backend, import-error, inference |
| SGLang not install, you may need to run `pip install sglang[ | exception | error | installation, sglang, backend, import-error, inference |
| Unknown backend: {model_args.infer_backend} | exception | error | configuration, backend, enum, validation |
| The current model does not support `chat`. | exception | error | model-loading, hf-engine, inference, model-type |
| The current model does not support `stream_chat`. | exception | error | model-loading, hf-engine, streaming, model-type |
| Cannot get scores using an auto-regressive model. | exception | error | model-loading, hf-engine, scoring, reward-model |
| SGLang server initialization failed: {str(e)}. | exception | error | sglang, server, subprocess, inference, startup |
| SGLang only supports n=1. | exception | error | sglang, sampling, not-implemented, inference |
| SGLang server error: {response.status_code}, {response.text} | exception | error | sglang, http, inference, api-error |
| SGLang engine does not support `get_scores`. | exception | error | sglang, not-implemented, reward-model, inference |
| vLLM engine does not support `get_scores`. | exception | error | vllm, not-implemented, reward-model, inference |
| Template is required for MultiModalDataCollator. | exception | error | collator, multimodal, config, training |
| Merged position_ids shape mismatch: got {features['position_ | exception | error | collator, mrope, multimodal, tensor-shape, training |
| Omni models are not supported for packed sequences for now. | exception | error | packing, omni, collator, training, audio |
| {self.model.config.model_type} requires 3D position ids for | exception | error | mrope, collator, qwen-vl, multimodal, training |
| Neat packing is not supported for gemma4, gpt_oss models for | exception | error | neat-packing, flash-attention, gemma, gpt-oss, training |
| Dataset converter {name} already exists. | exception | error | registry, plugin, dataset, converter |
| Dataset converter {name} not found. | exception | error | dataset, converter, registry, config |
| Unknown mixing strategy: {data_args.mix_strategy}. | exception | error | dataset, mixing, config, validation |
| Cannot specify `val_size` if `eval_dataset` is not None. | exception | error | dataset, config, validation, training |
| Unsupported protocol in path: {path}. Use 's3://' or 'gs://' | exception | error | cloud-storage, fsspec, dataset, config |
| Path does not exist: {path}. | exception | error | cloud-storage, fsspec, dataset, permissions |
| No JSON/JSONL files found in the specified path: {cloud_path | exception | error | cloud-storage, dataset, file-format, fsspec |
| Empty formatter should not contain any placeholder. | exception | error | formatter, template, config, custom-template |
| A placeholder is required in the string formatter. | exception | error | formatter, template, config, custom-template |
| Expected a string, got {value} | exception | error | formatter, dataset, data-quality, template |
| Input must be string, set[str] or dict[str, str], got {type( | exception | error | templates, data-format, custom-template |
| Invalid JSON format in function message: {str([content])}. | exception | error | json, tool-calls, data-format, sharegpt |
| Invalid JSON format in tool description: {str([content])}. | exception | error | json, tools, data-format, sharegpt |
| File {local_path} not found. | exception | error | dataset, file-not-found, path, config |
| Allowed file types: {}. | exception | error | dataset, file-format, config |
| File types should be identical. | exception | error | dataset, file-format, directory, config |
| Unknown load type: {dataset_attr.load_from}. | exception | error | dataset, dataset-info, config |
| The dataset is not applicable in the current training stage. | exception | error | dataset, training-stage, ranking, config |
| Cannot find sufficient samples, consider increasing dataset | exception | error | dataset, pretrain, empty-dataset, preprocessing |
| Cannot find valid samples, check `data/README.md` for the da | exception | error | dataset, data-format, empty-dataset, preprocessing |
| Turn off `streaming` when saving dataset to disk. | exception | error | streaming, tokenized-path, config, dataset |
| This model does not support image input. Please check whethe | exception | error | multimodal, image, template, config |
| This model does not support video input. Please check whethe | exception | error | multimodal, video, template, config |
| This model does not support audio input. Please check whethe | exception | error | multimodal, audio, template, config |
| Processor was not found, please check and update your model | exception | error | multimodal, processor, model-files, config |
| Image processor was not found, please check and update your | exception | error | multimodal, image-processor, model-files, transformers |
| Video processor was not found, please check and update your | exception | error | multimodal, video-processor, model-files, transformers |
| Audio feature extractor was not found, please check and upda | exception | error | multimodal, audio, feature-extractor, model-files |
| The number of images does not match the number of {IMAGE_PLA | exception | error | multimodal, image, data-format, placeholders |
| The number of videos does not match the number of {VIDEO_PLA | exception | error | multimodal, video, dataset-validation, data-preprocessing |
| Expect input is a list of images, but got {type(image)}. | exception | error | multimodal, image, type-validation, data-preprocessing |
| Invalid image found in video frames. | exception | error | multimodal, video, frames, path-resolution |
| MOSS-VL encountered nested video token blocks after tokeniza | exception | error | multimodal, moss-vl, video, truncation, cutoff-len |
| MOSS-VL encountered a video end token without a matching sta | exception | error | multimodal, moss-vl, video, truncation, cutoff-len |
| MOSS-VL encountered an incomplete video token block after to | exception | error | multimodal, moss-vl, video, truncation, cutoff-len, video-maxlen |
| MOSS-VL media tokens do not match the provided media after t | exception | error | multimodal, moss-vl, truncation, cutoff-len, media-count |
| MOSS-VL video frame tokens do not match the processed video | exception | error | multimodal, moss-vl, video, frames, truncation |
| MOSS-VL does not support audio inputs. | exception | error | multimodal, moss-vl, audio, unsupported-feature |
| MOSS-VL batch metadata must have one entry per sample. | exception | error | multimodal, moss-vl, batching, internal-invariant |
| MOSS-VL media lengths do not consume all provided inputs. | exception | error | multimodal, moss-vl, batching, truncation, media-count |
| MiniCPM-V model does not support input images and videos at | exception | error | multimodal, minicpm-v, mixed-media, unsupported-feature |
| Number of videos ({len(videos)}) must match number of audios | exception | error | multimodal, qwen2-omni, audio, video, dataset-validation |
| Each {VIDEO_PLACEHOLDER} must be followed by an {AUDIO_PLACE | exception | error | multimodal, qwen2-omni, audio, video, placeholder-order |
| Multimodal plugin {name} already exists. | exception | error | plugin-registry, multimodal, extension-api, name-collision |
| Multimodal plugin `{name}` not found. | exception | error | plugin-registry, multimodal, config-typo, version-compat |
| Cannot open {config_path} due to {str(err)}. | exception | error | dataset-config, json, file-io, hf-hub, config-path |
| Undefined dataset {name} in {DATA_CONFIG}. | exception | error | dataset-config, config-typo, json, data-preparation |
| The length of packed example should be identical to the cuto | exception | critical | data, packing, sft, config |
| Input must be string, set[str] or dict[str, str], got {type( | exception | error | template, custom-template, data |
| Unexpected role: {} | exception | error | data, roles, sharegpt, template |
| Stop words are required to replace the EOS token. | exception | error | template, tokenizer, custom-template |
| Dict is not supported. | exception | error | template, jinja, export |
| Template {name} already exists. | exception | error | template, custom-template, idempotency |
| Template {data_args.template} does not exist. | exception | error | template, config, typo |
| Current template does not support `train_on_prompt`. | exception | error | template, config, train-on-prompt |
| GLM-4 does not support parallel functions. | exception | error | tools, glm4, data |
| Unknown identifier: {node.id} | exception | error | tools, json, ast, glm4-moe |
| Tool utils `{name}` not found. | exception | error | tools, config, typo |
| Unknown logging level: {env_level_str}. | exception | error | logging, environment, config |
| Cannot specify `val_size` if `dataset` is None. | exception | error | config, data-args, validation-split |
| Cannot specify `val_size` if `eval_dataset` is not None. | exception | error | config, data-args, validation-split |
| `interleave_probs` is only valid for interleaved mixing. | exception | error | config, data-args, mixing |
| The length of eval dataset and interleave probs should be id | validation | error | config, data-args, mixing |
| Streaming mode should have an integer val size. | validation | error | config, streaming, data, validation-split |
| `max_samples` is incompatible with `streaming`. | validation | error | config, streaming, data, sampling |
| `mask_history` is incompatible with `train_on_prompt`. | validation | error | config, multi-turn, loss-masking, data |
| `save_dir` already exists, use another one. | validation | error | config, filesystem, evaluation, idempotency |
| `reward_model` is necessary for PPO training. | validation | critical | config, ppo, rlhf, reward-model |
| `reward_model_type` cannot be lora for Freeze/Full PPO train | validation | error | config, ppo, lora, reward-model |
| `reward_model_type` cannot be oft for Freeze/Full PPO traini | validation | error | config, ppo, oft, reward-model |
| `dpo_label_smoothing` is only valid for sigmoid loss functio | validation | error | config, dpo, preference-learning, hyperparameters |
| `use_llama_pro` is only valid for Freeze or LoRA training. | validation | error | config, llama-pro, finetuning-type |
| Cannot use LoRA with GaLore, APOLLO or BAdam together. | validation | error | config, lora, galore, apollo, badam, optimizer |
| Cannot use GaLore, APOLLO or BAdam together. | validation | error | config, galore, apollo, badam, optimizer |
| Cannot use PiSSA for current training stage. | validation | error | config, pissa, lora, ppo, kto, dpo |
| `loraplus_lr_ratio` is only valid for LoRA training. | validation | error | config, lora, loraplus, optimizer |
| `use_rslora` is only valid for LoRA training. | validation | error | config, lora, rslora |
| `use_dora` is only valid for LoRA training. | validation | error | config, lora, dora |
| `pissa_init` is only valid for LoRA training. | validation | error | config, lora, pissa |
| `tensor_model_parallel_size` must be >= 1. | validation | error | config, megatron, parallelism, distributed |
| `pipeline_model_parallel_size` must be >= 1. | validation | error | config, megatron, parallelism, distributed |
| `expert_model_parallel_size` must be >= 1. | validation | error | config, megatron, moe, parallelism |
| `context_parallel_size` must be >= 1. | validation | error | config, megatron, context-parallelism, long-context |
| `virtual_pipeline_model_parallel_size` must be >= 1 when set | validation | error | megatron, parallelism, config-validation, hparams |
| `sequence_parallel` requires `tensor_model_parallel_size` > | validation | error | megatron, sequence-parallel, config-validation, hparams |
| `recompute_granularity` must be 'full' or 'selective'. | validation | error | megatron, activation-checkpointing, config-validation |
| `recompute_method` must be 'uniform' or 'block'. | validation | error | megatron, activation-checkpointing, config-validation |
| `recompute_num_layers` must be >= 1 when set. | validation | error | megatron, activation-checkpointing, config-validation |
| `moe_token_dispatcher_type` must be 'allgather', 'alltoall', | validation | error | megatron, moe, config-validation |
| `extra_config` file not found: {self.extra_config} | validation | error | megatron, file-not-found, config-validation |
| Please provide `model_name_or_path`. | validation | critical | hparams, required-field, config-validation, quickstart |
| YAML config must be a dictionary mapping tokens to descripti | validation | error | hparams, tokenizer, special-tokens, yaml, config-validation |
| `image_max_pixels` cannot be smaller than `image_min_pixels` | validation | error | hparams, multimodal, vision, config-validation |
| `video_max_pixels` cannot be smaller than `video_min_pixels` | validation | error | hparams, multimodal, video, config-validation |
| Quantization dataset is necessary for exporting. | validation | error | export, quantization, hparams, required-field |
| `kt_cpu_activation` must be `retain` or `recompute`. | validation | error | ktransformers, cpu-offload, config-validation |
| `kt_cpu_activation` is only valid when `use_kt: true`. | validation | error | ktransformers, config-validation, dependent-fields |
| `kt_cpu_activation: recompute` requires GPU gradient checkpo | validation | error | ktransformers, gradient-checkpointing, memory |
| LLaMA-Factory `kt_config` must be a flat mapping. | validation | error | ktransformers, yaml, type-error, config-validation |
| These `kt_config` values are derived from LLaMA-Factory argu | validation | error | ktransformers, config-conflict, derived-fields |
| Put KTransformers settings in the LLaMA-Factory training YAM | validation | error | ktransformers, accelerate, config-conflict |
| LLaMA-Factory YAML and Accelerate config cannot define diffe | validation | error | ktransformers, accelerate, config-conflict |
| KTransformers cannot be combined with Unsloth checkpoint wra | validation | error | ktransformers, unsloth, incompatible-backends |
| KTransformers uses LLaMA-Factory's `disable_gradient_checkpo | validation | error | ktransformers, gradient-checkpointing, config, llamafactory |
| KTransformers supplies its checkpoint context; remove `gradi | validation | error | ktransformers, gradient-checkpointing, config, llamafactory |
| Disable FSDP activation checkpointing when using KTransforme | validation | error | ktransformers, fsdp, config, llamafactory |
| KTransformers thin integration currently supports LoRA finet | validation | error | ktransformers, lora, config, llamafactory |
| `kt_model_max_length` must be a positive integer. | validation | error | ktransformers, config, validation, llamafactory |
| KTransformers accepts a single `adapter_name_or_path`. | validation | error | ktransformers, lora, adapter, llamafactory |
| `adapter_folder` must stay inside the KT adapter directory. | validation | error | ktransformers, adapter, path-traversal, config |
| KTransformers {operation} requires a local adapter directory | validation | error | ktransformers, adapter, filesystem, llamafactory |
| The installed Transformers-KT does not provide `TrainingArgu | exception | critical | ktransformers, dependencies, version-mismatch, llamafactory |
| `kt_config` requires `use_kt: true`. | validation | error | ktransformers, config, llamafactory |
| KTransformers inference requires `infer_backend: huggingface | validation | error | ktransformers, vllm, sglang, inference, llamafactory |
| The installed Transformers-KT does not provide `configure_kt | exception | critical | ktransformers, dependencies, import-error, llamafactory |
| Some specified arguments are not used by the HfArgumentParse | validation | error | cli, arguments, typo, llamafactory |
| `--project` must be specified when using Trackio. | validation | error | trackio, reporting, config, llamafactory |
| Adapter is only valid for the LoRA method. | validation | error | lora, adapter, config, llamafactory |
| Quantization is only compatible with the LoRA or OFT method. | validation | error | quantization, qlora, config, llamafactory |
| Please use scripts/pissa_init.py to initialize PiSSA for a q | validation | error | pissa, quantization, config, llamafactory |
| Cannot resize embedding layers of a quantized model. | validation | error | quantization, tokenizer, vocabulary, llamafactory |
| Cannot create new adapter upon a quantized model. | validation | error | config, qlora, adapter, validation, llamafactory |
| Quantized model only accepts a single adapter. Merge them fi | validation | error | config, qlora, adapter, merging, validation |
| Total Megatron Bridge parallel size ({parallel_size}) exceed | validation | error | megatron, distributed, parallelism, config, validation |
| Total Megatron Bridge parallel size ({parallel_size}) must d | validation | error | megatron, distributed, parallelism, config, validation |
| megatron-bridge is required when USE_MEGATRON_BRIDGE=1. Plea | exception | error | megatron, dependency, environment, import-error |
| `predict_with_generate` cannot be set as True except SFT. | validation | error | config, validation, evaluation, stage, sft |
| `neat_packing` cannot be set as True except SFT. | validation | error | config, validation, packing, stage, sft |
| `train_on_prompt` or `mask_history` cannot be set as True ex | validation | error | config, validation, data-mask, stage, sft |
| Please enable `predict_with_generate` to save model predicti | validation | error | config, validation, prediction, sft, generation |
| Megatron Bridge cannot be used together with MCA or HyperPar | validation | error | megatron, distributed, config, backend-conflict, validation |
| Megatron Bridge only supports the `pt` and `sft` stages. | validation | error | megatron, config, stage, validation |
| Megatron Bridge only supports `full` and `lora` finetuning. | validation | error | megatron, config, finetuning-type, validation |
| Quantized models are not supported with Megatron Bridge. | validation | error | megatron, quantization, config, validation |
| Megatron Bridge is incompatible with DeepSpeed. | validation | error | megatron, deepspeed, distributed, config, backend-conflict |
| Megatron Bridge arguments are missing. Please set USE_MEGATR | validation | error | megatron, config, environment, validation |
| RM and PPO stages do not support `load_best_model_at_end`. | validation | error | config, validation, reward-model, ppo, checkpoints |
| PPO training does not support evaluation, use the SFT stage | validation | error | config, ppo, validation, evaluation |
| PPO training is incompatible with S^2-Attn. | validation | error | config, ppo, attention, long-context, validation |
| KTransformers does not support lora reward model. | validation | error | ppo, ktransformers, reward-model, lora, config |
| Unsloth does not support lora reward model. | validation | error | ppo, unsloth, reward-model, lora, config |
| PPO only accepts wandb, tensorboard, or trackio logger. | validation | error | ppo, rlhf, logging, wandb, tensorboard, config-validation |
| Please launch distributed training with `llamafactory-cli` o | validation | error | launcher, distributed, torchrun, config-validation |
| Please use `FORCE_TORCHRUN=1` to launch DeepSpeed training. | validation | error | deepspeed, distributed, torchrun, config-validation |
| Please specify `max_steps` in streaming mode. | validation | error | streaming, dataset, max-steps, config-validation |
| Please specify dataset for training. | validation | error | dataset, training, config-validation, yaml |
| Please make sure eval_dataset be provided or val_size >1e-6 | validation | warning | evaluation, dataset, val-size, config-validation |
| `predict_with_generate` is incompatible with DeepSpeed ZeRO- | validation | error | deepspeed, zero-3, predict-with-generate, evaluation, incompatible-flags |
| Cannot use `predict_with_generate` and `compute_accuracy` to | validation | error | evaluation, mutually-exclusive, compute-accuracy, predict-with-generate, config-validation |
| Cannot use device map for quantized models in training. | validation | error | quantization, device-map, qlora, training, config-validation |
| Please use scripts/pissa_init.py to initialize PiSSA in Deep | validation | error | pissa, lora, deepspeed, zero-3, incompatible-flags |
| This device does not support `pure_bf16`. | validation | error | pure-bf16, hardware, gpu-compat, precision, config-validation |
| `pure_bf16` is incompatible with DeepSpeed ZeRO-3. | validation | error | pure-bf16, deepspeed, zero-3, incompatible-flags |
| Distributed training does not support layer-wise GaLore. | validation | error | galore, optimizer, distributed, layerwise, incompatible-flags |
| Distributed training does not support layer-wise APOLLO. | validation | error | apollo, optimizer, distributed, layerwise, incompatible-flags |
| Radio-based BAdam does not yet support distributed training, | validation | error | badam, optimizer, distributed, layerwise, incompatible-flags |
| Layer-wise BAdam only supports DeepSpeed ZeRO-3 training. | validation | error | badam, optimizer, deepspeed, zero-3, distributed, incompatible-flags |
| GaLore and APOLLO are incompatible with DeepSpeed yet. | validation | error | galore, apollo, deepspeed, optimizer, incompatible-flags |
| FP8 training is not compatible with quantization. Please dis | validation | error | fp8, quantization, precision, incompatible-flags, config-validation |
| vLLM/SGLang backend is only available for API, CLI and Web. | validation | error | vllm, sglang, infer-backend, training, config-validation |
| Unsloth is incompatible with DeepSpeed ZeRO-3. | validation | error | unsloth, deepspeed, zero-3, lora, incompatible-flags |
| KTransformers is incompatible with DeepSpeed ZeRO-3. | validation | error | ktransformers, deepspeed, zero3, training-config |
| Output directory already exists and is not empty. Please set | validation | error | output-dir, checkpoint, resume, training-config |
| vLLM engine only supports auto-regressive models. | validation | error | vllm, infer-backend, stage, reward-model |
| vLLM engine does not support bnb quantization (GPTQ and AWQ | validation | error | vllm, bnb, quantization, gptq, awq |
| vLLM engine does not support RoPE scaling. | validation | error | vllm, rope-scaling, long-context, infer-config |
| vLLM only accepts a single adapter. Merge them first. | validation | error | vllm, lora, adapter-merge, export |
| mcore_adapter is required when USE_MCA=1. Please install `mc | exception | critical | megatron, mca, environment, import-error, dependency |
| Evaluation will be deprecated in the future. | exception | warning | deprecated, eval, launcher, not-implemented |
| Current model does not support freeze tuning. | validation | error | freeze-tuning, model-config, architecture, layer-count |
| `num_layers` {num_layers} should be divisible by `num_layer_ | validation | error | llama-pro, freeze-tuning, layer-count, divisibility |
| Module {} is not found, please choose from {} | validation | error | freeze-tuning, module-names, architecture, config |
| DoRA is not compatible with PTQ-quantized models. | validation | error | dora, lora, gptq, awq, quantization |
| KTransformers only supports LoRA finetuning. | validation | error | ktransformers, lora, finetuning-type |
| Unsloth is currently not supported for OFT. | validation | error | unsloth, oft, finetuning-type |
| Quantized models can only be used for the LoRA or OFT tuning | validation | error | qlora, quantization, finetuning-type, full-tuning |
| Cannot initialize PiSSA adapter on quantized models. | validation | error | pissa, qlora, lora-init, quantization |
| Unknown finetuning type: {finetuning_args.finetuning_type}. | exception | error | finetuning-type, typo, config, not-implemented |
| Failed to load tokenizer. | exception | critical | tokenizer, huggingface-hub, network, auth, offline |
| Unsupported torch version detected: torch 2.9.x with Conv3D. | exception | error | torch, compatibility, conv3d, multimodal, version-check |
| Unknown attention type: {model_args.flash_attn} | exception | error | flash-attention, config, enum, attention |
| The installed kt-kernel does not provide the activation chec | exception | error | kt-kernel, dependency, gradient-checkpointing, environment |
| {self.__class__.__name__} does not support gradient checkpoi | exception | error | gradient-checkpointing, vram, custom-model, training-config |
| Cannot resize embedding layers of a quantized model. | exception | error | quantization, tokenizer, embedding, resize |
| Current model does not support resizing embedding layers. | exception | error | embedding, tokenizer, lm-head, resize |
| Model was not supported. | exception | error | lora, num-layer-trainable, config, custom-model |
| `num_layers` {num_layers} should be divisible by `num_layer_ | exception | error | lora, num-layer-trainable, config-validation |
| Current model is not supported by mixture-of-depth. | exception | error | mixture-of-depths, architecture, model-support |
| Cannot find satisfying example, considering decrease `export | exception | error | gptq, export, quantization, calibration, dataset |
| DeepSpeed ZeRO-3 or FSDP is incompatible with PTQ-quantized | exception | error | quantization, gptq, deepspeed, fsdp, zero3, training-config |
| `quantization_bit` cannot be combined with KT weight caches. | exception | error | fp8, kt-kernel, quantization, config-conflict |
| AutoGPTQ only accepts 2/3/4/8-bit quantization. | exception | error | gptq, export, quantization, config-validation |
| ChatGLM model is not supported yet. | exception | error | gptq, export, chatglm, model-support |
| Bitsandbytes only accepts 4-bit or 8-bit quantization. | exception | error | bitsandbytes, quantization, config-validation |
| Only 4-bit quantized model can use fsdp+qlora or auto device | exception | error | bitsandbytes, qlora, deepspeed, fsdp, zero3, device-map |
| HQQ only accepts 1/2/3/4/5/6/8-bit quantization. | exception | error | hqq, quantization, config-validation |
| HQQ quantization is incompatible with DeepSpeed ZeRO-3 or FS | exception | error | hqq, quantization, deepspeed, fsdp, zero3 |
| EETQ only accepts 8-bit quantization. | exception | error | eetq, quantization, config-validation |
| EETQ quantization is incompatible with DeepSpeed ZeRO-3 or F | exception | error | eetq, quantization, deepspeed, fsdp, zero3 |
| Qwen3.5 packing-seq forwarding requires `flash-linear-attent | exception | error | |
| Please download the internvl models in a Hugging Face–compat | exception | error | |
| Please download llava models with hf-compatible format: http | exception | error | |
| InternLM3 model requires transformers>=4.47.1, please upgrad | exception | error | |
| LFM2.5-VL model requires transformers>=4.58.0 or install fro | exception | error | |
| Please update `transformers`. | exception | error | |
| Unknown loss type: {self.loss_type}. | exception | error | |
| HyperParallel trainer requires Accelerate FSDP2 mode to be e | exception | error | |
| HyperParallel CP training requires a map-style dataset becau | exception | error | |
| Trainer: training requires a train_dataset. | exception | error | |
| hyper_parallel is not installed. Please install it with `pip | exception | error | |
| Please update `transformers`. | exception | error | |
| Mismatched shape of inputs and labels. | exception | error | |
| mcore_adapter is not installed. Please install it with `pip | exception | error | |
| Model {model_type} is not supported by mcore_adapter.You can | exception | error | |
| Cannot inject {% generation %} into chat template: no `{% el | exception | error | |
| Failed to inject {% generation %} into chat template. | exception | error | |
| Model type `{model_type}` is not supported by the Megatron B | exception | error | |
| megatron-bridge is not installed. Please install it with `pi | exception | error | |
| PPOTrainer does not support eval dataset yet. | exception | error | |
| `resume_from_checkpoint` will be supported in the future ver | exception | error | ppo, reinforcement-learning, checkpoint, resume |
| Cannot process the logits. | exception | error | sft, evaluation, logits, metrics |
| `predict_with_generate` is not supported in KTransformers SF | exception | error | ktransformers, sft, evaluation, generation |
| `compute_accuracy` is not supported in KTransformers SFT yet | exception | error | ktransformers, sft, evaluation, accuracy |
| Unknown optim: {training_args.optim}. | exception | error | galore, optimizer, sft, config |
| Per-layer GaLore does not support gradient accumulation. | exception | error | galore, layerwise, gradient-accumulation, config |
| Per-layer APOLLO does not support gradient accumulation. | exception | error | apollo, layerwise, gradient-accumulation, config |
| Logits (batchsize x seqlen) and labels must have the same sh | exception | critical | dpo, logits, shape-mismatch, preference-training |
| hyper_parallel is not installed. Please install it with `pip | exception | error | hyper-parallel, fsdp, optional-dependency, installation |
| megatron-bridge is not installed. Please install it with `pi | exception | error | megatron, optional-dependency, installation, distributed |
| Megatron Bridge arguments are missing. Please set USE_MEGATR | exception | error | megatron, config, environment-variables |
| mcore_adapter is not installed. Please install it with `pip | exception | error | mca, megatron-core, optional-dependency, installation |
| Unknown task: {finetuning_args.stage}. | exception | error | config, stage, dispatch, typo |
| Please specify `export_dir` to save model. | exception | error | export, config, merge |
| Please merge adapters before quantizing the model. | exception | error | export, quantization, lora, merge |
| Cannot merge adapters to a quantized model. | exception | error | export, quantization, lora, merge |
| The model is not a `PreTrainedModel`, export aborted. | exception | error | export, ktransformers, model-type |
| Failed to export model: weight conversion reversal is not su | exception | error | export, transformers-5, version-incompatibility, mistral |
| The number of devices in the Ray cluster ({total_devices}) s | exception | error | ray, distributed, gpu, resource-planning |
| The `master_addr` ({master_addr}) is not in Ray cluster or n | exception | error | ray, distributed-training, configuration, network |
| torch.accelerator is not available, please upgrade torch to | exception | error | torch, version-mismatch, v1, environment |
| world_size ({helper.get_world_size()}) must be divisible by | exception | error | distributed-training, v1, configuration, device-mesh |
| mp_replicate_size * mp_shard_size must equal to world_size, | exception | error | distributed-training, v1, configuration, device-mesh |
| world_size ({helper.get_world_size()}) must be divisible by | exception | error | distributed-training, v1, context-parallelism, configuration |
| dp_size * cp_size must equal to world_size, got {self.dp_siz | exception | error | distributed-training, v1, configuration, context-parallelism |
| dim must be specified. | exception | error | v1, api-misuse, distributed-training, device-mesh |
| Some specified arguments are not used by the HfArgumentParse | exception | error | v1, configuration, cli, argument-parsing |
| Plugin configuration must have a 'name' field. | exception | error | v1, plugins, configuration, validation |
| Unsupported `flash_attn`: {self.flash_attn}. Supported value | exception | error | v1, configuration, attention, enum-validation |
| `dynamic_batching` requires `max_steps` because it is step-d | exception | error | v1, configuration, batching, training-args |
| `save_epochs` is not supported with `dynamic_batching`; use | exception | error | v1, configuration, checkpointing, batching |
| Unknown sample backend: {args.sample_backend} | exception | error | v1, sampling, backend, configuration |
| Context parallelism currently requires `dist_config.name: fs | exception | error | v1, context-parallelism, deepspeed, fsdp2, configuration |
| Sequence parallel is not supported for qwen3.5 model due to | exception | error | v1, context-parallelism, qwen3, model-support |
| Sequence parallelism requires flash attention. Please set `f | exception | error | v1, context-parallelism, flash-attention, configuration |
| `padding_free` requires `flash_attn: flash_attention_2`. | exception | error | v1, flash-attention, batching, configuration |
| All datasets must be streaming or non-streaming. | validation | error | v1, datasets, streaming, configuration |
| Streaming dataset does not support index access. | validation | error | v1, datasets, streaming, api-misuse |
| Currently lora stage does not support loading model by meta. | validation | error | lora, peft, model-loading, meta-device, v1 |
| special-token escape failed: the tokenizer normalized away t | validation | error | tokenizer, special-tokens, data-cleaning, prompt-injection, rendering |
| tool_call value is not valid JSON: {content['value']!r} | validation | error | tool-calls, json, multimodal, data-format, rendering |
| tool_call must be a JSON object with 'name' and 'arguments' | validation | error | tool-calls, json, schema, multimodal, rendering |
| {kind} placeholder count ({seen}) != number of {kind} blocks | validation | error | multimodal, placeholders, data-format, rendering, vision |
| tools is not valid JSON: {tools!r} | validation | error | tools, json, function-calling, rendering |
| training render expects the last message to be the supervise | validation | error | training-data, chat-template, supervision, rendering |
| prompt is not a token-prefix of the full sequence; the chat | validation | error | chat-template, tokenization, prefix-stability, training-data, labels |
| Unsupported dummy media modality: {modality!r} (expected ima | validation | error | multimodal, api-misuse, dummy-data, rendering |
| Cannot build a dummy media fragment for a text-only processo | exception | error | multimodal, text-only, api-misuse, processor |
| Processor did not emit {modality} placeholder tokens for the | exception | error | multimodal, processor, dummy-data, compatibility, transformers-version |
| No valid messages or chosen_messages/rejected_messages found | validation | error | data-format, dataset, preference-data, rendering |
| Global batch size must be divisible by DP size and micro bat | validation | error | batching, distributed, config, training |
| Drop last must be True. | validation | error | batching, config, dataloader, training |
| Iterable dataset is not supported yet. | exception | error | dataset, iterable-dataset, streaming, not-implemented, v1 |
| batching_strategy={self.batching_strategy.value!r} does not | exception | error | batching, multimodal, packing, not-implemented, training |
| Unexpected missing keys when loading checkpoint model weight | exception | critical | checkpoint, resume, state-dict, model-mismatch |
| Checkpoint directory does not exist: {ckpt_dir} | validation | error | checkpoint, resume, filesystem, config |
| Batch infer is not implemented. | exception | error | inference, not-implemented, v1-api |
| Environment information is not implemented yet. | exception | warning | cli, not-implemented, v1-api, environment |
| Version information is not implemented yet. | exception | warning | cli, not-implemented, v1-api, version |
| More {segment} tags than provided media files. | validation | error | data, multimodal, dataset-conversion, validation |
| Fewer {tag} tags than provided media files ({unused} unused) | validation | error | data, multimodal, dataset-conversion, validation |
| Unknown dataset filetype: {filetype}. | validation | error | data, file-format, dataset-loading, validation |
| Can not load dataset from {filepath}. | validation | error | data, path, dataset-loading, config |
| Invalid index type {type(index)}. | validation | error | data, indexing, api-misuse, v1-api |
| DeepSpeed only supports bf16 mixed precision for now, fp16 i | validation | error | deepspeed, precision, config, distributed |
| DeepSpeed config_file is required in dist_config | validation | error | deepspeed, config, distributed, zero3 |
| DeepSpeed ZeRO-3 model-loading bootstrap failed: transformer | exception | critical | deepspeed, environment, distributed, zero3, version-conflict |
| HFModel instance is required for {cls.__name__}. | validation | error | kernels, model-loading, validation |
| kernel_config.name must be a string. | exception | error | kernels, config, type-error, validation |
| kernel_config.name must contain at least one kernel name. | validation | error | kernels, config, validation |
| LigerKernel requires CUDA or NPU, current accelerator is {cu | exception | error | kernels, device, liger, hardware |
| Liger kernel is not installed. | exception | error | kernels, liger, dependencies, install |
| Unknown Liger op(s) {sorted(ops)} for model_type={model_type | validation | error | kernels, liger, config, validation |
| cross_entropy and fused_linear_cross_entropy cannot both be | validation | error | kernels, liger, config, conflict |
| FlashLinearAttentionKernel requires CUDA or NPU, current acc | exception | error | kernels, device, fla, hardware |
| Flash Linear Attention and FSDPTurbo are required for this k | exception | error | kernels, fla, fsdp-turbo, dependencies, install |
| kernel_config.include_kernels must be 'auto' or a comma-sepa | validation | error | configuration, type-error, kernel-plugin, linear-attention |
| kernel_config.include_kernels must select at least one FLA k | validation | error | configuration, validation, kernel-plugin |
| Unsupported Flash Linear Attention kernels: {sorted(unsuppor | validation | error | configuration, validation, version-skew, kernel-plugin |
| chunk_size must be one of {SUPPORTED_CHUNK_SIZES}, got {chun | validation | error | configuration, validation, kernel-plugin |
| FLA operator `{op_name}` did not match any model module attr | exception | error | runtime, model-mismatch, kernel-plugin, version-skew |
| CudaFusedMoEKernel requires CUDA, current accelerator is {cu | exception | error | device-check, cuda, moe, kernel-plugin |
| cuda_fused_moe requires Triton. | exception | error | dependencies, triton, cuda, moe |
| NpuFusedMoEKernel requires NPU, current accelerator is {curr | exception | error | device-check, npu, moe, kernel-plugin |
| NpuFusedMoEKernel requires torch_npu. | exception | error | dependencies, npu, torch-npu, moe |
| NpuSwiGluKernel requires NPU, current accelerator is {curren | exception | error | device-check, npu, swiglu, kernel-plugin |
| NpuSwiGluKernel requires torch_npu. | exception | error | dependencies, npu, torch-npu, swiglu |
| {self.__class__.__name__} has no RMSNorm weight for NPU RMSN | exception | error | runtime, npu, rmsnorm, model-state |
| {self.__class__.__name__} requires a gate tensor for NPU Gat | validation | error | runtime, npu, rmsnorm, api-contract |
| NpuRMSNormKernel requires NPU, current accelerator is {curre | exception | error | device-check, npu, rmsnorm, kernel-plugin |
| NpuRMSNormKernel requires torch_npu. | exception | error | dependencies, npu, torch-npu, rmsnorm |
| NpuRoPEKernel requires NPU, current accelerator is {current} | exception | error | device-check, npu, rope, kernel-plugin |
| NpuRoPEKernel requires torch_npu. | exception | error | dependencies, npu, torch-npu, rope |
| Other sequence parallel modes are to be implemented. | exception | error | not-implemented, sequence-parallel, attention, distributed |
| When `adapter_name_or_path` is provided for training, only a | validation | error | configuration, lora, peft, training |
| Current model does not support freeze tuning. | validation | error | peft, freeze-tuning, model-config, configuration |
| Module {module_name} not found in hidden modules: {hidden_mo | validation | error | peft, freeze-tuning, module-names, configuration |
| Module {module_name} not found in non-hidden modules: {non_h | validation | error | peft, freeze-tuning, module-names, configuration |
| Please specify peft_config to merge and export model. | validation | error | peft, export, merge, configuration |
| Currently merge and export model function is only supported | validation | error | peft, export, lora, configuration |
| Please specify export_dir. | validation | error | peft, export, configuration |
| Please set adapter_name_or_path to merge adapters into base | validation | error | peft, export, lora, configuration |
| compute_dtype={self.compute_dtype!r} is not a torch dtype na | validation | error | quantization, bitsandbytes, dtype, configuration |
| compute_dtype must be str or torch.dtype, got {type(self.com | validation | error | quantization, bitsandbytes, dtype, type-error |
| Unsupported quantization bit: {quant_config.quantization_bit | validation | error | quantization, bitsandbytes, configuration |
| Bitsandbytes only accepts 4-bit or 8-bit quantization. | validation | error | quantization, bitsandbytes, configuration |
| Only 4-bit quantized model can use fsdp+qlora or auto device | validation | error | quantization, bitsandbytes, qlora, export, bug |
| DeepSpeed config_file is required in dist_config | validation | error | deepspeed, distributed, configuration |
| No checkpoint files found in {hf_model_path} | validation | error | fsdp2, model-loading, checkpoint, filesystem |
| hf_model_path='{hf_model_path}' does not exist locally and h | validation | error | fsdp2, model-loading, huggingface-hub, environment |
| ep_size must be positive, got {ep_size}. | validation | error | fsdpturbo, expert-parallel, distributed, configuration |
| dp_size must be divisible by ep_size, got {dp_size} % {ep_si | validation | error | fsdpturbo, expert-parallel, distributed, topology |
| FSDPTurbo parallel state is already initialized with {curren | exception | error | fsdpturbo, distributed, singleton, lifecycle |
| FSDPTurbo expert parallelism requires an initialized distrib | exception | error | fsdpturbo, expert-parallel, device-mesh, initialization-order |
| FSDPTurbo parallel state must be initialized before clipping | exception | error | fsdpturbo, gradient-clipping, distributed, initialization-order |
| No FSDPTurbo EP spec is registered for model_type={_get_mode | validation | error | distributed, fsdp, expert-parallelism, moe, config |
| FSDPTurbo EP mesh is not initialized. | exception | critical | distributed, fsdp, expert-parallelism, device-mesh |
| FSDPTurbo EFSDP mesh is not initialized. | exception | critical | distributed, fsdp, expert-parallelism, device-mesh, config |
| ep_size must be positive, got {self.ep_size}. | validation | error | config, validation, distributed |
| DeepSpeed config_file is required. | validation | error | config, deepspeed, distributed, validation |
| DPO training dataset is empty: {dataset_path} | validation | error | data, dataset, dpo, validation |
| DPO training requires pair-format samples containing chosen/ | validation | error | data, dataset, dpo, preference-data, validation |
| DPO trainer currently only supports cp_size == 1. | validation | error | dpo, distributed, context-parallelism, not-implemented |
| Unknown pref_loss: {self.pref_loss} | validation | error | dpo, config, loss-function, validation |
| RM training dataset is empty: {dataset_path} | validation | error | data, dataset, reward-model, validation |
| RM training requires pair-format samples containing chosen/r | validation | error | data, dataset, reward-model, preference-data, validation |
| RM trainer currently only supports cp_size == 1. | validation | error | reward-model, distributed, context-parallelism, not-implemented |
| RM training requires pair data with token_type_ids. Ensure t | validation | error | reward-model, data, batch, token-type-ids |
| No valid RM pairs found in this micro-batch. This is usually | validation | error | reward-model, truncation, cutoff-len, data |
| Unexpected precision: {precision} | validation | error | dtype, precision, validation, config |
| Unknown logging level: {env_level_str}. | validation | error | logging, environment, config, startup |
| Buffer size exceeds max buffer size {self._max_buffer_size}. | exception | error | buffer, memory, generation, validation |
| {cls.__name__} config must be a mapping or {params_cls.__nam | validation | error | plugin, config, type-error, validation |
| Unknown params for {cls.__name__}.{params_cls.__name__}: {so | validation | error | config, validation, llamafactory, v1-plugins, yaml |
| Plugin {self.name!r} is not registered under {cls.__name__}. | validation | error | plugin-registry, llamafactory, v1-plugins, import, config |