quarot Configuration Reference¶
1. Configuration Overview¶
QuaRot (offline rotation) processor configuration.
| Item | Content |
|---|---|
| Config Class | QuaRotProcessorConfig |
| Source | quarot.py |
2. Parameter List¶
2.1 QuaRotProcessorConfig
| Field Path | Type | Required/Optional | Default | Value Range or Format | Description | Referenced Config |
|---|---|---|---|---|---|---|
type |
string |
Optional | quarot |
quarot |
Processor type, fixed to quarot. |
None |
online |
bool |
Optional | false |
— | Whether to use online rotation (offline by default). | None |
block_size |
int |
Optional | -1 |
— | Rotation block size; -1 means rotating the whole hidden_dim as one block. | None |
down_proj_online_layers |
list[int] |
Optional | [] |
— | Index list of down_proj layers that need online rotation. | None |
max_tp_size |
int |
Optional | 4 |
— | Maximum TP (Tensor Parallelism) degree; must be a power of 2. | None |
export_extra_info |
bool |
Optional | true |
— | Whether to export the optional.quarot.global_rotation rotation information for downstream deployment. |
None |
Configuration Constraints
- Validates max_tp_size: it must be at least 1 and a power of 2
- Validates block_size: the value must be -1 or a non-negative integer power of 2