chore: refactorings for more code reorganizing
This commit is contained in:
@@ -7,7 +7,7 @@ mod tokenizer;
|
||||
mod validation;
|
||||
|
||||
#[cfg(target_os = "macos")]
|
||||
use crate::metrics::{KvLookup, Metrics};
|
||||
use crate::metrics::{KvLookup, Metrics, SsdStats};
|
||||
use crate::model::ModelChoice;
|
||||
#[cfg(target_os = "macos")]
|
||||
use crate::settings::TurnSettings;
|
||||
@@ -509,37 +509,7 @@ impl Generator {
|
||||
settings.steering.clone(),
|
||||
settings.diagnostics.expert_profile_path.as_deref(),
|
||||
)?;
|
||||
let stats = executor.execution_stats();
|
||||
metrics.speculative_stats(
|
||||
stats.speculative_mode,
|
||||
stats.speculative_cycles,
|
||||
stats.drafted_tokens,
|
||||
stats.accepted_draft_tokens,
|
||||
stats.verifier_passes,
|
||||
stats.verifier_ms,
|
||||
);
|
||||
metrics.ssd_stats(
|
||||
stats.ssd_enabled,
|
||||
stats.ssd_resident_bytes,
|
||||
stats.ssd_cache_bytes,
|
||||
stats.ssd_cache_experts,
|
||||
stats.ssd_cache_entries,
|
||||
stats.ssd_preloaded_experts,
|
||||
stats.ssd_cache_hits,
|
||||
stats.ssd_cache_misses,
|
||||
stats.ssd_cache_evictions,
|
||||
stats.ssd_cache_wraps,
|
||||
stats.ssd_buffer_allocs,
|
||||
stats.ssd_buffer_reuses,
|
||||
stats.ssd_pread_bytes,
|
||||
stats.ssd_pread_ms,
|
||||
stats.ssd_evict_advise_bytes,
|
||||
stats.ssd_willneed_advise_bytes,
|
||||
stats.ssd_selected_requests,
|
||||
stats.ssd_requested_bytes,
|
||||
stats.ssd_wait_ms,
|
||||
);
|
||||
Ok(Self {
|
||||
let generator = Self {
|
||||
executor,
|
||||
_simulated_memory: simulated_memory,
|
||||
checkpoint: None,
|
||||
@@ -553,7 +523,9 @@ impl Generator {
|
||||
.filter(|limit| *limit > 0)
|
||||
.unwrap_or(1)
|
||||
.saturating_sub(1),
|
||||
})
|
||||
};
|
||||
generator.publish_execution_stats();
|
||||
Ok(generator)
|
||||
}
|
||||
|
||||
pub(crate) fn summary(&self) -> ModelSummary {
|
||||
@@ -622,27 +594,27 @@ impl Generator {
|
||||
stats.verifier_passes,
|
||||
stats.verifier_ms,
|
||||
);
|
||||
self.metrics.ssd_stats(
|
||||
stats.ssd_enabled,
|
||||
stats.ssd_resident_bytes,
|
||||
stats.ssd_cache_bytes,
|
||||
stats.ssd_cache_experts,
|
||||
stats.ssd_cache_entries,
|
||||
stats.ssd_preloaded_experts,
|
||||
stats.ssd_cache_hits,
|
||||
stats.ssd_cache_misses,
|
||||
stats.ssd_cache_evictions,
|
||||
stats.ssd_cache_wraps,
|
||||
stats.ssd_buffer_allocs,
|
||||
stats.ssd_buffer_reuses,
|
||||
stats.ssd_pread_bytes,
|
||||
stats.ssd_pread_ms,
|
||||
stats.ssd_evict_advise_bytes,
|
||||
stats.ssd_willneed_advise_bytes,
|
||||
stats.ssd_selected_requests,
|
||||
stats.ssd_requested_bytes,
|
||||
stats.ssd_wait_ms,
|
||||
);
|
||||
self.metrics.ssd_stats(SsdStats {
|
||||
enabled: stats.ssd_enabled,
|
||||
resident_bytes: stats.ssd_resident_bytes,
|
||||
cache_bytes: stats.ssd_cache_bytes,
|
||||
cache_experts: stats.ssd_cache_experts,
|
||||
cache_entries: stats.ssd_cache_entries,
|
||||
preloaded_experts: stats.ssd_preloaded_experts,
|
||||
cache_hits: stats.ssd_cache_hits,
|
||||
cache_misses: stats.ssd_cache_misses,
|
||||
cache_evictions: stats.ssd_cache_evictions,
|
||||
cache_wraps: stats.ssd_cache_wraps,
|
||||
buffer_allocs: stats.ssd_buffer_allocs,
|
||||
buffer_reuses: stats.ssd_buffer_reuses,
|
||||
pread_bytes: stats.ssd_pread_bytes,
|
||||
pread_ms: stats.ssd_pread_ms,
|
||||
evict_advise_bytes: stats.ssd_evict_advise_bytes,
|
||||
willneed_advise_bytes: stats.ssd_willneed_advise_bytes,
|
||||
selected_requests: stats.ssd_selected_requests,
|
||||
requested_bytes: stats.ssd_requested_bytes,
|
||||
wait_ms: stats.ssd_wait_ms,
|
||||
});
|
||||
}
|
||||
|
||||
pub(crate) fn generate_transient(
|
||||
|
||||
Reference in New Issue
Block a user