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": 74713.63632596082,
-      "required_pages_per_second_upper_estimate": 0.006475988636754229
+      "required_token_positions_per_second": 75098.17578157528,
+      "required_pages_per_second_upper_estimate": 0.00650931954216055
     },
     {
       "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": 9339.204540745102,
-      "required_pages_per_second_upper_estimate": 0.0037255483248544367
+      "required_token_positions_per_second": 9387.27197269691,
+      "required_pages_per_second_upper_estimate": 0.0037447231421321643
     },
     {
       "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": 18678.409081490205,
-      "required_pages_per_second_upper_estimate": 0.004320921212230406
+      "required_token_positions_per_second": 18774.54394539382,
+      "required_pages_per_second_upper_estimate": 0.004343160321078689
     },
     {
       "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": 28017.61362223531,
-      "required_pages_per_second_upper_estimate": 0.0017061256725359856
+      "required_token_positions_per_second": 28161.81591809073,
+      "required_pages_per_second_upper_estimate": 0.0017149068357826054
     },
     {
       "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": 28017.61362223531,
-      "required_pages_per_second_upper_estimate": 0.002156873454358095
+      "required_token_positions_per_second": 28161.81591809073,
+      "required_pages_per_second_upper_estimate": 0.0021679745462705492
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28017.61362223531,
-      "required_pages_per_second_upper_estimate": 0.3478849922673468
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28161.81591809073,
+      "required_pages_per_second_upper_estimate": 0.34967550216783255
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.13740962660151013,
-    "producer_p95_seconds": 0.3317472179987817,
-    "gpu_update_seconds": 0.9868078121420742,
-    "producer_batches_per_second": 7.277510497135796,
-    "gpu_batches_per_second": 1.0133685482579322,
-    "producer_headroom": 7.181504211519553,
+    "producer_batch_seconds": 0.06878276410935769,
+    "producer_p95_seconds": 0.2637965119974979,
+    "gpu_update_seconds": 0.9817548726408419,
+    "producer_batches_per_second": 14.538525936673619,
+    "gpu_batches_per_second": 1.018584198426314,
+    "producer_headroom": 14.273268679344586,
     "sustainable": true,
-    "buffer_seconds": 60.195276540666526,
+    "buffer_seconds": 60.8688021037322,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 266684.90651030256,
-    "upload_bytes_per_second": 5363050.686726327,
+    "generation_bytes_per_second": 287338.1547453034,
+    "upload_bytes_per_second": 5233258.676304946,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.22437283021743,
+    "estimated_full_shards_per_hour": 61.65607911843611,
     "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": 1.056229023000924,
-      "bytes": 11737562,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 0.9255671579994669,
       "bytes": 1181038,
@@ -562,6 +556,12 @@
       "bytes": 7736202,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.0140705839985458,
+      "bytes": 1147707,
+      "pages": 33,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -192,3 +192,4 @@
 {"step": 174200, "loss": 0.6223274301737547, "lr": 0.0003, "grad_norm": 0.1410694148798424, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25495207625, "tokens_per_second": 149775.84034037788, "data_wait_s": 1.4260996977100149e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 174300, "loss": 0.606553552672267, "lr": 0.0003, "grad_norm": 0.14896050249629975, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25509914704, "tokens_per_second": 149257.32553951207, "data_wait_s": 9.665000106906518e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 174400, "loss": 0.5848895721137524, "lr": 0.0003, "grad_norm": 0.14330444334816816, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25524521088, "tokens_per_second": 149186.86521065643, "data_wait_s": 9.627998224459589e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 174500, "loss": 0.6052178870886564, "lr": 0.0003, "grad_norm": 0.14196068976940465, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25539019678, "tokens_per_second": 148059.00687164412, "data_wait_s": 1.4252000255510211e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:68199aa604dea26116df4392238865731ac33b033d28db8b2fc049086b6bb57d
-size 413406
+oid sha256:5b79a5bb9f01c96c0cdfd95916c2b9fdcbc394b47d51d72c9ffa38d83aa9c1c2
+size 415524