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": 74983.14774476313,
-      "required_pages_per_second_upper_estimate": 0.006499349203465567
+      "required_token_positions_per_second": 75131.19669556635,
+      "required_pages_per_second_upper_estimate": 0.006512181711294555
     },
     {
       "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": 9372.893468095392,
-      "required_pages_per_second_upper_estimate": 0.0037389873416688174
+      "required_token_positions_per_second": 9391.399586945794,
+      "required_pages_per_second_upper_estimate": 0.0037463697091693767
     },
     {
       "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": 18745.786936190783,
-      "required_pages_per_second_upper_estimate": 0.004336507893105643
+      "required_token_positions_per_second": 18782.799173891588,
+      "required_pages_per_second_upper_estimate": 0.004345070022904588
     },
     {
       "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": 28118.680404286177,
-      "required_pages_per_second_upper_estimate": 0.0017122801093064585
+      "required_token_positions_per_second": 28174.198760837382,
+      "required_pages_per_second_upper_estimate": 0.0017156608859381216
     },
     {
       "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": 28118.680404286177,
-      "required_pages_per_second_upper_estimate": 0.00216465385501113
+      "required_token_positions_per_second": 28174.198760837382,
+      "required_pages_per_second_upper_estimate": 0.002168927811783089
     },
     {
       "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": 28118.680404286177,
-      "required_pages_per_second_upper_estimate": 0.34913990345165796
+      "required_token_positions_per_second": 28174.198760837382,
+      "required_pages_per_second_upper_estimate": 0.3498292556320372
     }
   ],
   "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": 0.10037802072668,
-    "producer_p95_seconds": 0.4020711079938337,
-    "gpu_update_seconds": 0.9832609355233317,
-    "producer_batches_per_second": 9.96234028884577,
-    "gpu_batches_per_second": 1.017024030826323,
-    "producer_headroom": 9.795580032412271,
+    "producer_batch_seconds": 0.12023116313247328,
+    "producer_p95_seconds": 0.40723651299776975,
+    "gpu_update_seconds": 0.981323381534143,
+    "producer_batches_per_second": 8.317311202405806,
+    "gpu_batches_per_second": 1.0190320732363056,
+    "producer_headroom": 8.161971954416675,
     "sustainable": true,
-    "buffer_seconds": 60.962178002446564,
+    "buffer_seconds": 60.84204965511687,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 303094.16884719016,
-    "upload_bytes_per_second": 6075374.058751848,
+    "generation_bytes_per_second": 302662.1220396342,
+    "upload_bytes_per_second": 6080067.595438789,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.03695296346453,
+    "estimated_full_shards_per_hour": 64.94424577609796,
     "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": 0.9876477170037106,
-      "bytes": 6238360,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.2104173959960463,
       "bytes": 11923116,
@@ -562,6 +556,12 @@
       "bytes": 11694230,
       "pages": 35,
       "shards": 4
+    },
+    {
+      "seconds": 0.9577424080052879,
+      "bytes": 6389346,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -344,3 +344,4 @@
 {"step": 119400, "loss": 0.6104825809597969, "lr": 0.0003, "grad_norm": 0.16794767334880456, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17472747252, "tokens_per_second": 149110.36216746346, "data_wait_s": 1.1146999895572662e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 119500, "loss": 0.5576545484364033, "lr": 0.0003, "grad_norm": 0.12738941614743302, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17487473060, "tokens_per_second": 149518.90413774241, "data_wait_s": 1.3624994608107954e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 119600, "loss": 0.6462280862033367, "lr": 0.0003, "grad_norm": 0.15123167555158018, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17502101396, "tokens_per_second": 148654.39952012605, "data_wait_s": 1.2291005987208337e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 119700, "loss": 0.6458818539977074, "lr": 0.0003, "grad_norm": 0.13656010856277226, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17516653535, "tokens_per_second": 146939.63549080683, "data_wait_s": 1.2688004062511027e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a7213e199ae9849550e44048f54a4cbe8e56d01dd26716e81918b706c624f4c9
-size 737691
+oid sha256:6bd1dc013d91673d69606a83eea7d851ea1528b22b08ae912f312a703ec04362
+size 739809