Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77312.46282452886,
-      "required_pages_per_second_upper_estimate": 0.005348655567829728
+      "required_token_positions_per_second": 76452.5634140618,
+      "required_pages_per_second_upper_estimate": 0.005289165731372093
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9664.057853066108,
-      "required_pages_per_second_upper_estimate": 0.004046139214602401
+      "required_token_positions_per_second": 9556.570426757726,
+      "required_pages_per_second_upper_estimate": 0.004001136473799783
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19328.115706132216,
-      "required_pages_per_second_upper_estimate": 0.004025536784899207
+      "required_token_positions_per_second": 19113.14085351545,
+      "required_pages_per_second_upper_estimate": 0.00398076319236724
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28992.173559198323,
-      "required_pages_per_second_upper_estimate": 0.001738036536388143
+      "required_token_positions_per_second": 28669.711280273175,
+      "required_pages_per_second_upper_estimate": 0.0017187054151379712
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28992.173559198323,
-      "required_pages_per_second_upper_estimate": 0.002254013971133239
+      "required_token_positions_per_second": 28669.711280273175,
+      "required_pages_per_second_upper_estimate": 0.0022289439473084766
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28992.173559198323,
-      "required_pages_per_second_upper_estimate": 0.3718804730467583
+      "required_token_positions_per_second": 28669.711280273175,
+      "required_pages_per_second_upper_estimate": 0.3677442731657261
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.06540388839081857,
-    "producer_p95_seconds": 0.2294216049995157,
-    "gpu_update_seconds": 0.9536366752063727,
-    "producer_batches_per_second": 15.289610825957872,
-    "gpu_batches_per_second": 1.0486173885705412,
-    "producer_headroom": 14.58073363326583,
+    "producer_batch_seconds": 0.47997545155476473,
+    "producer_p95_seconds": 0.26831112500076415,
+    "gpu_update_seconds": 0.9643626937751485,
+    "producer_batches_per_second": 2.0834398858540393,
+    "gpu_batches_per_second": 1.036954256375621,
+    "producer_headroom": 2.0091917006407893,
     "sustainable": true,
-    "buffer_seconds": 60.07911053800148,
+    "buffer_seconds": 60.75484970783435,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 385672.25190008257,
-    "upload_bytes_per_second": 5547138.020430888,
+    "generation_bytes_per_second": 337773.2865665607,
+    "upload_bytes_per_second": 5524723.961741462,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.75628726722582,
+    "estimated_full_shards_per_hour": 72.47828433749785,
     "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": 1.1795985310000106,
-      "bytes": 8204299,
-      "pages": 31,
-      "shards": 3
-    },
     {
       "seconds": 0.9598508790004416,
       "bytes": 6041026,
@@ -562,6 +556,12 @@
       "bytes": 6127707,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.116981245999341,
+      "bytes": 6115742,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -175,3 +175,4 @@
 {"step": 502500, "loss": 0.5735471248626709, "lr": 0.0003, "grad_norm": 0.15704754788065672, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73549106769, "tokens_per_second": 150982.9314653788, "data_wait_s": 1.1591000657062978e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 502600, "loss": 0.5724793076515198, "lr": 0.0003, "grad_norm": 0.1434232357324958, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73563901469, "tokens_per_second": 154233.70855758447, "data_wait_s": 1.748100112308748e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 502700, "loss": 0.537597119808197, "lr": 0.0003, "grad_norm": 0.14868456586983175, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73578574033, "tokens_per_second": 152649.4251514213, "data_wait_s": 1.2917000276502222e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 502800, "loss": 0.5820568203926086, "lr": 0.0003, "grad_norm": 0.1621224558029401, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73592991407, "tokens_per_second": 149879.22073531325, "data_wait_s": 1.3794000551570207e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8d41a831ed7d1eb87083a818ff88004e20a9aea29b509ce5472a1e880b35b251
-size 377400
+oid sha256:5d36aac99fc6f151d26fc581bf11548d5be4dbc0721625928bfea6725d15bcfa
+size 379518