Fix variable names

Signed-off-by: Azam Ikram <azamikram@google.com>

Author: azamikram

pkg/epp/framework/plugins/scheduling/scorer/prefix/README.md

@@ -14,24 +14,22 @@ The scorer computes a score between `0.0` and `1.0` for each endpoint using a we
 ### Scoring Formula
 
 ```text
-score = 0
-score += matchLengthWeight * matchLengthRatio
-score += (1.0 - matchLengthWeight) * matchRatio
+score = matchLengthWeight*matchLengthScore + (1.0-matchLengthWeight)*matchRatioScore
 ```
 
 Where:
-*   **`matchRatio`** = `matchBlocks` / `totalBlocks`
-*   **`matchLengthRatio`** = `min(1.0, matchedTokens / matchLengthScaleTokens) ^ 2`
+*   **`matchRatioScore`** = `matchBlocks` / `totalBlocks`
+*   **`matchLengthScore`** = `min(1.0, matchedTokens / matchLengthScaleTokens) ^ 2`
     *   `matchedTokens` = `matchBlocks` * `blockSize`
     *   **`matchLengthScaleTokens`** is the scaling factor.
 
-If `matchLengthWeight` is `0.0` (default), the score simplifies to just the `matchRatio`.
+If `matchLengthWeight` is `0.0` (default), the score simplifies to just the `matchRatioScore`.
 
 If the attribute is missing, has the wrong type, or `totalBlocks` is zero, the endpoint receives score `0`.
 
 ### Why the Quadratic Term?
 
-The squaring of `matchLengthRatio` introduces a non-linearity motivated by the fact that attention computation grows quadratically as a function of prompt length. This ensures that the scoring function assigns a higher priority to longer requests, where the computational and latency savings of KV-cache reuse are significantly more critical.
+The squaring of `matchLengthScore` introduces a non-linearity motivated by the fact that attention computation grows quadratically as a function of prompt length. This ensures that the scoring function assigns a higher priority to longer requests, where the computational and latency savings of KV-cache reuse are significantly more critical.
 
 ## Inputs Consumed
 
@@ -44,7 +42,7 @@ The attribute is typically produced by the approximate prefix cache data produce
 | Parameter | Type | Description | Default |
 | :--- | :--- | :--- | :--- |
 | `matchLengthWeight` | float | Weight of the absolute match length in the score. Must be between `0.0` and `1.0`. | `0.0` |
-| `matchLengthScaleTokens` | integer | The number of tokens used to normalize `matchLengthRatio`. | `8192` |
+| `matchLengthScaleTokens` | integer | The number of tokens used to normalize `matchLengthScore`. | `8192` |
 
 ### Example
 

pkg/epp/framework/plugins/scheduling/scorer/prefix/plugin.go

@@ -154,18 +154,16 @@ func (p *Plugin) Score(ctx context.Context, _ *fwksched.InferenceRequest, endpoi
 			continue
 		}
 
-		matchLengthRatio := 0.0
-		matchRatio := float64(matchBlocks) / float64(totalBlocks)
+		matchRatioScore := float64(matchBlocks) / float64(totalBlocks)
 		blockSize := prefixMatchInfo.BlockSizeTokens()
-		if blockSize > 0 && p.matchLengthScaleTokens > 0 {
+		matchLengthScore := 0.0
+		// Calculate matchLengthScore when match length is considered
+		if p.matchLengthWeight > 0.0 && blockSize > 0 {
 			// (matchBlocks * blockSize / matchLengthScaleTokens) ^ 2
-			// Capped at 1.0 as the normalized score term cannot be greater than 1.
-			ratio := float64(matchBlocks) * float64(blockSize) / float64(p.matchLengthScaleTokens)
-			matchLengthRatio = math.Min(1.0, ratio)
-			matchLengthRatio *= matchLengthRatio
+			normalizedMatchLength := math.Min(1.0, float64(matchBlocks)*float64(blockSize)/float64(p.matchLengthScaleTokens))
+			matchLengthScore = normalizedMatchLength * normalizedMatchLength
 		}
-		scores[endpoint] += p.matchLengthWeight * matchLengthRatio
-		scores[endpoint] += (1.0 - p.matchLengthWeight) * matchRatio
+		scores[endpoint] += p.matchLengthWeight*matchLengthScore + (1.0-p.matchLengthWeight)*matchRatioScore
 	}
 	return scores
 }