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": 75145.33582710207,
-      "required_pages_per_second_upper_estimate": 0.00651340725538077
+      "required_token_positions_per_second": 75690.19662237438,
+      "required_pages_per_second_upper_estimate": 0.006560634408177913
     },
     {
       "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": 9393.166978387759,
-      "required_pages_per_second_upper_estimate": 0.0037470747480404334
+      "required_token_positions_per_second": 9461.274577796798,
+      "required_pages_per_second_upper_estimate": 0.0037742438877440554
     },
     {
       "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": 18786.333956775517,
-      "required_pages_per_second_upper_estimate": 0.0043458877327147314
+      "required_token_positions_per_second": 18922.549155593595,
+      "required_pages_per_second_upper_estimate": 0.004377398721655673
     },
     {
       "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": 28179.500935163276,
-      "required_pages_per_second_upper_estimate": 0.0017159837605362097
+      "required_token_positions_per_second": 28383.823733390393,
+      "required_pages_per_second_upper_estimate": 0.0017284259469493672
     },
     {
       "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": 28179.500935163276,
-      "required_pages_per_second_upper_estimate": 0.002169335987840054
+      "required_token_positions_per_second": 28383.823733390393,
+      "required_pages_per_second_upper_estimate": 0.0021850653224492833
     },
     {
       "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": 28179.500935163276,
-      "required_pages_per_second_upper_estimate": 0.3498950908919289
+      "required_token_positions_per_second": 28383.823733390393,
+      "required_pages_per_second_upper_estimate": 0.35243209622149313
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.07498196811720703,
-    "producer_p95_seconds": 0.18529686700003367,
-    "gpu_update_seconds": 0.9811387385324467,
-    "producer_batches_per_second": 13.336539772293836,
-    "gpu_batches_per_second": 1.0192238474813107,
-    "producer_headroom": 13.08499580857618,
+    "producer_batch_seconds": 0.2548729064609745,
+    "producer_p95_seconds": 0.18913675200019497,
+    "gpu_update_seconds": 0.9740759476136128,
+    "producer_batches_per_second": 3.923524135560158,
+    "gpu_batches_per_second": 1.0266139949866318,
+    "producer_headroom": 3.821810490330642,
     "sustainable": true,
-    "buffer_seconds": 60.830601789011695,
+    "buffer_seconds": 60.392708752043994,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 298084.3582800597,
-    "upload_bytes_per_second": 4437226.450920933,
+    "generation_bytes_per_second": 271801.23127415625,
+    "upload_bytes_per_second": 4434634.924709495,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.96196423817962,
+    "estimated_full_shards_per_hour": 58.32221702259555,
     "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": [
@@ -472,6 +472,12 @@
       "bytes": 7595630,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.335573416000443,
+      "bytes": 5731224,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -54,3 +54,4 @@
 {"step": 5400, "loss": 0.7493402753025293, "lr": 0.0003, "grad_norm": 0.17096302890595227, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 789905642, "tokens_per_second": 146897.6928870207, "data_wait_s": 9.242000487574842e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 5500, "loss": 0.7404255270957947, "lr": 0.0003, "grad_norm": 0.162354875326267, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 804478038, "tokens_per_second": 148527.42950388734, "data_wait_s": 1.1267999980191234e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 5600, "loss": 0.7228264361619949, "lr": 0.0003, "grad_norm": 0.1703213232573938, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 818964145, "tokens_per_second": 147434.22191567856, "data_wait_s": 1.6070999663497787e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 5700, "loss": 0.6885922737419605, "lr": 0.0003, "grad_norm": 0.18319819000259852, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 833519878, "tokens_per_second": 148250.98255790243, "data_wait_s": 1.0204000318481121e-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:b0f5472a6d1b2263949b3fcd2a256bc847b3b5639c9049782132e0a1fe4f24cc
-size 117547
+oid sha256:8114a3f4a00767db823baa17998aeef54dbd689d8d30da8f5f05a6af1e89fcb8
+size 119630