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": 77075.2615037061,
-      "required_pages_per_second_upper_estimate": 0.0053322454300723165
+      "required_token_positions_per_second": 77263.06664650404,
+      "required_pages_per_second_upper_estimate": 0.0053452382256190464
     },
     {
       "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": 9634.407687963263,
-      "required_pages_per_second_upper_estimate": 0.004033725309639694
+      "required_token_positions_per_second": 9657.883330813005,
+      "required_pages_per_second_upper_estimate": 0.004043554071073713
     },
     {
       "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": 19268.815375926526,
-      "required_pages_per_second_upper_estimate": 0.004013186089972234
+      "required_token_positions_per_second": 19315.76666162601,
+      "required_pages_per_second_upper_estimate": 0.004022964804594768
     },
     {
       "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": 28903.223063889793,
-      "required_pages_per_second_upper_estimate": 0.0017327040900139389
+      "required_token_positions_per_second": 28973.649992439015,
+      "required_pages_per_second_upper_estimate": 0.0017369260768447714
     },
     {
       "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": 28903.223063889793,
-      "required_pages_per_second_upper_estimate": 0.0022470984613748805
+      "required_token_positions_per_second": 28973.649992439015,
+      "required_pages_per_second_upper_estimate": 0.002252573845294262
     },
     {
       "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": 28903.223063889793,
-      "required_pages_per_second_upper_estimate": 0.3707395116005412
+      "required_token_positions_per_second": 28973.649992439015,
+      "required_pages_per_second_upper_estimate": 0.3716428726214263
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.10116056198449996,
-    "producer_p95_seconds": 0.2966433399997186,
-    "gpu_update_seconds": 0.9565715193383397,
-    "producer_batches_per_second": 9.885275253346478,
-    "gpu_batches_per_second": 1.0454001397529582,
-    "producer_headroom": 9.455972768171332,
+    "producer_batch_seconds": 0.7236898014377289,
+    "producer_p95_seconds": 0.3052730050039827,
+    "gpu_update_seconds": 0.9542463585780543,
+    "producer_batches_per_second": 1.3818075064942679,
+    "gpu_batches_per_second": 1.0479474100274528,
+    "producer_headroom": 1.3185847813279763,
     "sustainable": true,
-    "buffer_seconds": 60.2640057183154,
+    "buffer_seconds": 60.11752059041742,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 38981.311455425144,
-    "upload_bytes_per_second": 5314254.954739658,
+    "generation_bytes_per_second": 41073.75797558385,
+    "upload_bytes_per_second": 5376825.621018861,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 8.364481999846133,
+    "estimated_full_shards_per_hour": 8.813472313410156,
     "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": 0.9430077399993024,
-      "bytes": 1135014,
-      "pages": 33,
-      "shards": 1
-    },
     {
       "seconds": 1.1626808530018025,
       "bytes": 5979744,
@@ -562,6 +556,12 @@
       "bytes": 7666603,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 0.9383765300008235,
+      "bytes": 5995522,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -339,3 +339,4 @@
 {"step": 718900, "loss": 0.577467679977417, "lr": 8e-05, "grad_norm": 0.22120885357620004, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105227313897, "tokens_per_second": 154028.4385952602, "data_wait_s": 1.5370002074632794e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 719000, "loss": 0.5348767638206482, "lr": 8e-05, "grad_norm": 0.21006133985819403, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105242001386, "tokens_per_second": 152235.51998967447, "data_wait_s": 9.997995221056044e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 719100, "loss": 0.5993387699127197, "lr": 8e-05, "grad_norm": 0.22667573341498234, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105256470148, "tokens_per_second": 150256.08197712986, "data_wait_s": 1.6924997908063233e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 719200, "loss": 0.5463426113128662, "lr": 8e-05, "grad_norm": 0.23426403055278208, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105271034141, "tokens_per_second": 151377.8714554051, "data_wait_s": 1.4236000424716622e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dd977e9a79e936699d48254c098126f3831ed3749de53844cb169a1f37b76d14
-size 727101
+oid sha256:744e1ed775701bb849d42eaef96ac18eb06952d218847175519b504bd2ed26f1
+size 729219