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": 74802.17719747382,
-      "required_pages_per_second_upper_estimate": 0.006483663135092189
+      "required_token_positions_per_second": 74684.86353529182,
+      "required_pages_per_second_upper_estimate": 0.006473494684183011
     },
     {
       "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": 9350.272149684228,
-      "required_pages_per_second_upper_estimate": 0.003729963359535754
+      "required_token_positions_per_second": 9335.607941911478,
+      "required_pages_per_second_upper_estimate": 0.003724113587805759
     },
     {
       "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": 18700.544299368456,
-      "required_pages_per_second_upper_estimate": 0.00432604180532001
+      "required_token_positions_per_second": 18671.215883822955,
+      "required_pages_per_second_upper_estimate": 0.004319257192546043
     },
     {
       "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": 28050.816449052683,
-      "required_pages_per_second_upper_estimate": 0.0017081475504873945
+      "required_token_positions_per_second": 28006.82382573443,
+      "required_pages_per_second_upper_estimate": 0.0017054686305387693
     },
     {
       "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": 28050.816449052683,
-      "required_pages_per_second_upper_estimate": 0.002159429499877804
+      "required_token_positions_per_second": 28006.82382573443,
+      "required_pages_per_second_upper_estimate": 0.0021560428259553895
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28050.816449052683,
-      "required_pages_per_second_upper_estimate": 0.34829726025370555
+      "required_token_positions_per_second": 28006.82382573443,
+      "required_pages_per_second_upper_estimate": 0.3477510191059318
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.11051778796087319,
-    "producer_p95_seconds": 0.3691559370054165,
-    "gpu_update_seconds": 0.9856397602620837,
-    "producer_batches_per_second": 9.048317184506368,
-    "gpu_batches_per_second": 1.0145694606862226,
-    "producer_headroom": 8.91838118051215,
+    "producer_batch_seconds": 0.1380724586716724,
+    "producer_p95_seconds": 0.3534850310024922,
+    "gpu_update_seconds": 0.987187985757788,
+    "producer_batches_per_second": 7.242574005131154,
+    "gpu_batches_per_second": 1.012978292308103,
+    "producer_headroom": 7.1497820438271384,
     "sustainable": true,
-    "buffer_seconds": 60.1240253759871,
+    "buffer_seconds": 60.21846713122507,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 298776.5082164618,
-    "upload_bytes_per_second": 5988553.391733698,
+    "generation_bytes_per_second": 295457.2600267157,
+    "upload_bytes_per_second": 6042550.601170147,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.11048350210562,
+    "estimated_full_shards_per_hour": 63.39825010872939,
     "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": 0.8513431389983452,
-      "bytes": 631745,
-      "pages": 18,
-      "shards": 1
-    },
     {
       "seconds": 1.1167246349978086,
       "bytes": 13461458,
@@ -562,6 +556,12 @@
       "bytes": 5766570,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.1578597999978228,
+      "bytes": 6291400,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -277,3 +277,4 @@
 {"step": 112700, "loss": 0.6256785094738007, "lr": 0.0003, "grad_norm": 0.14013686517648433, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16491584949, "tokens_per_second": 147834.45150944236, "data_wait_s": 1.581400283612311e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 112800, "loss": 0.5562449060380459, "lr": 0.0003, "grad_norm": 0.13376561918239996, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16506040126, "tokens_per_second": 146979.2012667166, "data_wait_s": 1.4928002201486379e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 112900, "loss": 0.672998858615756, "lr": 0.0003, "grad_norm": 0.15856373016345113, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16520758339, "tokens_per_second": 149601.41419452222, "data_wait_s": 9.977004083339125e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 113000, "loss": 0.6347733959555626, "lr": 0.0003, "grad_norm": 0.14740347821185226, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16535376761, "tokens_per_second": 149178.66304026908, "data_wait_s": 1.59679984790273e-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 sha256:6a7803a445545d3ac6dd51125d00230ff2d9517c1d03bd2dff444cbbfcadffd4
-size 595002
+oid sha256:9712b772a87bda70cfae9dbd9bece19abea3ffa65e82d19c8bcb8c020ce6721d
+size 597120