Module qwen
Expand description
Caller-buffered Qwen vector and Hyper-Connection arithmetic.
Modules§
- hyper_
connection - Qwen4Exp four-stream Hyper-Connection vector operations.
- numerics
- Small caller-buffered numerical kernels used by the Qwen4Exp decoder.
- ple
- Trained PLE n-gram row selection over caller-owned history and output. No table I/O, layer admission, residual arithmetic or full model execution. Kernel operations and the borrowed view allocate nothing. A custom projection trait implementation is external code and must be qualified separately.
Enums§
Functions§
- add_
assign - Add
updateintostatein place, avoiding a temporary residual vector. - add_
into - Copy a supplied vector into an output buffer and add a second vector. This is useful where the decoder must preserve a pre-norm residual separately.
- group_
rms_ norm_ in_ place - In-place variant for a transient stream buffer whose unnormalized values
are not needed by the caller after the operation. Same pre-folded gamma
semantics as
group_rms_norm_into. - group_
rms_ norm_ into - Normalize each Hyper-Connection stream independently with RMSNorm.
The GGUF converter pre-folds the zero-centered
(1 + weight)scale into the stored gamma, so the checkpoint bytes multiply directly — callers must not add a further bias (llama.cppbuild_hc_mix). - inject_
stream_ update - Add a block update into each Hyper stream.
inject_gateis the sigmoid output of the streamed HC inject projection; Qwen scales it by two. - mix_
streams_ into - Form the token-mixer or MoE input from the gated average of Hyper streams.
mix_gateis the sigmoid output of the streamed HC up-projection. - rms_
norm_ into - Apply Qwen RMSNorm with a separately stored scale vector.