ErrLookup › deepseek-ai/DeepSeek-V3
deepseek-ai/DeepSeek-V3
· Python · 5 source files
Analyzed at 9b4e9788e4 on 2026-08-14. 16 documented errors.
| Code / Message | Type | Severity | Tags |
|---|---|---|---|
| Vocabulary size must be divisible by world size | exception | critical | distributed, tensor-parallelism, embedding, model-construction, deepseek |
| Prompt length exceeds model maximum sequence length | validation | error | generation, context-length, kv-cache, validation, deepseek |
| Output features must be divisible by world size | exception | critical | distributed, tensor-parallelism, linear-layer, model-construction, deepseek |
| Key not found in mapping | validation | error | checkpoint-conversion, safetensors, huggingface, key-mapping, deepseek |
| Warning: Missing scale_inv tensor for | console | warning | fp8, dequantization, checkpoint-conversion, safetensors, silent-corruption, deepseek |
| Input features must be divisible by world size | exception | critical | distributed, tensor-parallelism, linear-layer, model-construction, deepseek |
| Dimension must be divisible by | validation | error | checkpoint-conversion, model-parallelism, sharding, validation, deepseek |
| Input tensor must be contiguous | validation | error | triton, fp8, quantization, tensor-layout, deepseek |
| Either input-file or interactive mode must be specified | validation | error | cli, argument-validation, usage-error, deepseek |
| Number of prompts exceeds maximum batch size | validation | error | batching, generation, cli, validation, deepseek |
| Scaling factor tensors must be contiguous | validation | error | triton, fp8, gemm, tensor-layout, deepseek |
| Number of experts must be divisible by world size | exception | critical | distributed, moe, experts, model-construction, deepseek |
| Input tensors must be contiguous | validation | error | triton, fp8, dequantization, tensor-layout, deepseek |
| Last dimension size must be divisible by block_size | validation | error | triton, fp8, quantization, shape-validation, deepseek |
| Number of experts must be divisible by model parallelism | validation | error | checkpoint-conversion, cli, moe, model-parallelism, deepseek |
| Input tensors must have 2 dimensions | validation | error | triton, fp8, dequantization, shape-validation, deepseek |