Pad value vectors to the query-head width

Read the equation

The FlashAttention path appends zero-valued components to V when the value width is smaller than the query width. With the displayed 192 and 128 widths, it appends 64 zeros. The operation occurs before cache update in the embedded implementation, so the padded width affects that cache allocation. The output is later sliced back to the original value width.

Terms & definitions

The recovered passage does not supply a separate symbol table. Read the source context for the notation used in this version; familiar letters may carry different meanings in other builds.

How it works

Read the definition from its right-hand expression to the quantity on the left. Its surrounding source records the intended input, state, or parameter context. Where the source leaves an operation unspecified, this page preserves that gap.

Assumptions & limits

Source recovery establishes that the expression appears in this record. Its mathematical validity, implementation behavior, and interpretation must be assessed separately. See the linked source and research notes for the checks available for this version.

Original context

key_states[:, :, :, self.qk_nope_head_dim:] = k_pe

        if self.q_head_dim != self.v_head_dim:
            value_states = F.pad(value_states,
                                 [0, self.q_head_dim - self.v_head_dim])

        if past_key_value is not None:
            cache_kwargs = {"sin": sin, "cos": cos}  # Specific to RoPE models

value_states = F.pad(value_states,
                                 [0, self.q_head_dim - self.v_head_dim])

Related builds