Skip to content

Fix PD Scorers Configurations #133

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
May 7, 2025
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
12 changes: 12 additions & 0 deletions README.md
Original file line number Diff line number Diff line change
Expand Up @@ -59,6 +59,12 @@ export PREFILL_ENABLE_LOAD_AWARE_SCORER=true
export PREFILL_LOAD_AWARE_SCORER_WEIGHT=1.0
```

To enable and configure the prefix aware scorer for prefill, the following environment variables must be configured:
```
export PREFILL_ENABLE_PREFIX_AWARE_SCORER=true
export PREFILL_PREFIX_AWARE_SCORER_WEIGHT=1.0
```

Decode configuration:

To enable and configure the kv cache scorer for decode, the following environment variables must be configured:
Expand All @@ -72,6 +78,12 @@ To enable and configure the load aware scorer for decode, the following environm
export DECODE_ENABLE_LOAD_AWARE_SCORER=true
export DECODE_LOAD_AWARE_SCORER_WEIGHT=1.0
```

To enable and configure the prefix aware scorer for decode, the following environment variables must be configured:
```
export DECODE_ENABLE_PREFIX_AWARE_SCORER=true
export DECODE_PREFIX_AWARE_SCORER_WEIGHT=1.0
```
---
[Inference Gateways]:#concepts-and-definitions

Expand Down
5 changes: 5 additions & 0 deletions pkg/epp/scheduling/config_utils.go
Original file line number Diff line number Diff line change
Expand Up @@ -29,13 +29,17 @@ import (
const (
prefillKvCacheScorerEnablementEnvVar = "PREFILL_ENABLE_KVCACHE_AWARE_SCORER"
prefillLoadAwareScorerEnablementEnvVar = "PREFILL_ENABLE_LOAD_AWARE_SCORER"
prefillPrefixScorerEnablementEnvVar = "PREFILL_ENABLE_PREFIX_AWARE_SCORER"
decodeKvCacheScorerEnablementEnvVar = "DECODE_ENABLE_KVCACHE_AWARE_SCORER"
decodeLoadAwareScorerEnablementEnvVar = "DECODE_ENABLE_LOAD_AWARE_SCORER"
decodePrefixScorerEnablementEnvVar = "DECODE_ENABLE_PREFIX_AWARE_SCORER"

prefillKvCacheScorerWeightEnvVar = "PREFILL_KVCACHE_AWARE_SCORER_WEIGHT"
prefillLoadAwareScorerWeightEnvVar = "PREFILL_LOAD_AWARE_SCORER_WEIGHT"
prefillPrefixScorerWeightEnvVar = "PREFILL_PREFIX_AWARE_SCORER_WEIGHT"
decodeKvCacheScorerWeightEnvVar = "DECODE_KVCACHE_AWARE_SCORER_WEIGHT"
decodeLoadAwareScorerWeightEnvVar = "DECODE_LOAD_AWARE_SCORER_WEIGHT"
decodePrefixScorerWeightEnvVar = "DECODE_PREFIX_AWARE_SCORER_WEIGHT"

pdEnabledEnvKey = "PD_ENABLED"

Expand All @@ -46,6 +50,7 @@ const (
const (
loadAwareScorerName = "LoadAwareScorer"
kvCacheAwareScorerName = "KVCacheAwareScorer"
prefixAwareScorerName = "PrefixAwareScorer"
)

func addScorerByEnvironment(ctx context.Context, config *SchedulerConfig, scorerName string, scorerEnabledEnvKey string, weightEnvKey string, logger logr.Logger) {
Expand Down
16 changes: 12 additions & 4 deletions pkg/epp/scheduling/pd_config.go
Original file line number Diff line number Diff line change
Expand Up @@ -66,12 +66,20 @@ func init() {

func loadPrefillConfiguration(ctx context.Context, logger logr.Logger) {
// add scorers
addScorerByEnvironment(ctx, prefillConfig, kvCacheAwareScorerName, kvCacheScorerEnablementEnvVar, kvCacheScorerWeightEnvVar, logger)
addScorerByEnvironment(ctx, prefillConfig, loadAwareScorerName, loadAwareScorerEnablementEnvVar, loadAwareScorerWeightEnvVar, logger)
addScorerByEnvironment(ctx, prefillConfig, kvCacheAwareScorerName, prefillKvCacheScorerEnablementEnvVar,
prefillKvCacheScorerWeightEnvVar, logger)
addScorerByEnvironment(ctx, prefillConfig, loadAwareScorerName, prefillLoadAwareScorerEnablementEnvVar,
prefillLoadAwareScorerWeightEnvVar, logger)
addScorerByEnvironment(ctx, prefillConfig, prefixAwareScorerName, prefillPrefixScorerEnablementEnvVar,
prefillPrefixScorerWeightEnvVar, logger)
}

func loadDecodeConfiguration(ctx context.Context, logger logr.Logger) {
// add scorers
addScorerByEnvironment(ctx, decodeConfig, kvCacheAwareScorerName, kvCacheScorerEnablementEnvVar, kvCacheScorerWeightEnvVar, logger)
addScorerByEnvironment(ctx, decodeConfig, loadAwareScorerName, loadAwareScorerEnablementEnvVar, loadAwareScorerWeightEnvVar, logger)
addScorerByEnvironment(ctx, decodeConfig, kvCacheAwareScorerName, decodeKvCacheScorerEnablementEnvVar,
decodeKvCacheScorerWeightEnvVar, logger)
addScorerByEnvironment(ctx, decodeConfig, loadAwareScorerName, decodeLoadAwareScorerEnablementEnvVar,
decodeLoadAwareScorerWeightEnvVar, logger)
addScorerByEnvironment(ctx, decodeConfig, prefixAwareScorerName, decodePrefixScorerEnablementEnvVar,
decodePrefixScorerWeightEnvVar, logger)
}