ruvnet/ruflo · error · Error

FlashAttention: Keys and values must have same count. Got ${

Error message

FlashAttention: Keys and values must have same count. Got ${keys.length} keys, ${values.length} values

What it means

Attention scores every query against every key and mixes the corresponding value, so keys and values must be parallel arrays of the same length; validateInputs() enforces keys.length === values.length and reports both counts when they differ. A mismatch means K and V were built from different sequences — an off-by-one, a bad slice, or caching one side from a previous step.

Source

Thrown at v3/@claude-flow/neural/src/flash-attention.ts:775

    }

    return vectors;
  }

  /**
   * Validate input arrays
   */
  private validateInputs(
    queries: Float32Array[],
    keys: Float32Array[],
    values: Float32Array[],
  ): void {
    if (!queries.length || !keys.length || !values.length) {
      throw new Error('FlashAttention: Empty input arrays');
    }

    if (keys.length !== values.length) {
      throw new Error(
        `FlashAttention: Keys and values must have same count. Got ${keys.length} keys, ${values.length} values`,
      );
    }

    const qDim = queries[0]?.length ?? 0;
    const kDim = keys[0]?.length ?? 0;
    const vDim = values[0]?.length ?? 0;

    if (qDim !== kDim) {
      throw new Error(
        `FlashAttention: Query and key dimensions must match. Got Q=${qDim}, K=${kDim}`,
      );
    }

    if (kDim !== vDim) {
      throw new Error(
        `FlashAttention: Key and value dimensions must match. Got K=${kDim}, V=${vDim}`,
      );

View on GitHub (pinned to fa13ee4ad6)

Solutions

Derive keys and values from the same source array so lengths match by construction
Assert keys.length === values.length before calling computeAttention
Re-check slice bounds: both must use the same [i, i+n) window

Example fix

// before
const out = computeAttention(q, cachedKeys, freshValues);

// after
if (keys.length !== values.length) {
  throw new Error(`K/V count mismatch: ${keys.length} vs ${values.length} - rebuild both from the same window`);
}
const out = computeAttention(q, keys, values);

Defensive patterns

Strategy: validation

Validate before calling

if (keys.length !== values.length) {
  throw new Error(
    `K/V count mismatch: ${keys.length} keys vs ${values.length} values - rebuild both from the same window`,
  );
}
const out = computeAttention(queries, keys, values);

Prevention

Derive keys and values from the same array so counts match by construction
Keep KV-cache extensions symmetric: append to K and V together
Use identical slice windows for K and V in sliding-window code

When it happens

Trigger: Building keys from one window and values from another (e.g. keys.slice(0, n) vs values.slice(1, n+1)); concatenating past-sequence keys for KV-cache without extending values the same way; example code adapted with different dummy lengths.

Common situations: Sliding-window/KV-cache attention implementations; inconsistent preprocessing of the same token stream; partial refactors that update one array's construction.

Understand the failure class

Background: Tensor shape mismatch errors ("must have shape", "expected shape ... got ..."): when tensor dimensions disagree with what an op or layer was told to expect — this error's family across 6 libraries.

Related errors

AI-assisted analysis of ruvnet/ruflo@fa13ee4ad6 (2026-08-18). Data as JSON: /api/errors/fe368dcba780ce5c. Report an issue: GitHub.