Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75648.81231440976,
-      "required_pages_per_second_upper_estimate": 0.00655704732125111
+      "required_token_positions_per_second": 75092.46868100748,
+      "required_pages_per_second_upper_estimate": 0.006508824865147847
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9456.10153930122,
-      "required_pages_per_second_upper_estimate": 0.0037721802853443514
+      "required_token_positions_per_second": 9386.558585125935,
+      "required_pages_per_second_upper_estimate": 0.0037444385611640077
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18912.20307860244,
-      "required_pages_per_second_upper_estimate": 0.004375005338828505
+      "required_token_positions_per_second": 18773.11717025187,
+      "required_pages_per_second_upper_estimate": 0.004342830261759984
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28368.30461790366,
-      "required_pages_per_second_upper_estimate": 0.0017274809142387316
+      "required_token_positions_per_second": 28159.675755377804,
+      "required_pages_per_second_upper_estimate": 0.001714776511101954
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28368.30461790366,
-      "required_pages_per_second_upper_estimate": 0.0021838706179794533
+      "required_token_positions_per_second": 28159.675755377804,
+      "required_pages_per_second_upper_estimate": 0.0021678097906205583
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28368.30461790366,
-      "required_pages_per_second_upper_estimate": 0.3522394007462863
+      "required_token_positions_per_second": 28159.675755377804,
+      "required_pages_per_second_upper_estimate": 0.34964892850960183
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.08149615258595588,
-    "producer_p95_seconds": 0.20987265300027502,
-    "gpu_update_seconds": 0.9746088239108562,
-    "producer_batches_per_second": 12.27051791120172,
-    "gpu_batches_per_second": 1.0260526843859832,
-    "producer_headroom": 11.958955030213405,
+    "producer_batch_seconds": 0.41927316270309234,
+    "producer_p95_seconds": 0.2694568269998854,
+    "gpu_update_seconds": 0.9818294869648815,
+    "producer_batches_per_second": 2.3850799167609695,
+    "gpu_batches_per_second": 1.0185067909207828,
+    "producer_headroom": 2.341741791043665,
     "sustainable": true,
-    "buffer_seconds": 60.42574708247309,
+    "buffer_seconds": 60.873428191822654,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 229028.18321057403,
-    "upload_bytes_per_second": 6476735.6464460725,
+    "generation_bytes_per_second": 223015.6725227285,
+    "upload_bytes_per_second": 6456586.368092027,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 49.14411661374965,
+    "estimated_full_shards_per_hour": 47.853971784223475,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
-    {
-      "seconds": 1.0383053309997194,
-      "bytes": 7595630,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.335573416000443,
       "bytes": 5731224,
@@ -562,6 +556,12 @@
       "bytes": 2351431,
       "pages": 32,
       "shards": 2
+    },
+    {
+      "seconds": 1.204088584998317,
+      "bytes": 7306502,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -129,3 +129,4 @@
 {"step": 12900, "loss": 0.8119467496871948, "lr": 0.0003, "grad_norm": 0.2041011837749955, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1886921849, "tokens_per_second": 150302.32083414425, "data_wait_s": 1.843299833126366e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 13000, "loss": 0.7229449599981308, "lr": 0.0003, "grad_norm": 0.13446527130833577, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1901787377, "tokens_per_second": 151888.65565807078, "data_wait_s": 1.1818001439678483e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 13100, "loss": 0.7219682410359383, "lr": 0.0003, "grad_norm": 0.15079662284683937, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1916585449, "tokens_per_second": 151079.43025943745, "data_wait_s": 1.0460000339662656e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 13200, "loss": 0.7260879389941692, "lr": 0.0003, "grad_norm": 0.154266199946037, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1931266750, "tokens_per_second": 148320.34945739104, "data_wait_s": 1.2554999557323754e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:399aa530c7735679d909ab61abe99a55509a80224581d1a0f41103cee5002b50
-size 274545
+oid sha256:412c39ff50493ff04225c6b201256434a44a1efba131c4ad5845834133397c6d
+size 276628