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": 75493.39317404832,
-      "required_pages_per_second_upper_estimate": 0.00654357598407079
+      "required_token_positions_per_second": 75961.7254763804,
+      "required_pages_per_second_upper_estimate": 0.006584169841059566
     },
     {
       "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": 9436.67414675604,
-      "required_pages_per_second_upper_estimate": 0.003764430407992676
+      "required_token_positions_per_second": 9495.21568454755,
+      "required_pages_per_second_upper_estimate": 0.0037877835026916985
     },
     {
       "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": 18873.34829351208,
-      "required_pages_per_second_upper_estimate": 0.004366016968118712
+      "required_token_positions_per_second": 18990.4313690951,
+      "required_pages_per_second_upper_estimate": 0.0043931020770630914
     },
     {
       "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": 28310.022440268123,
-      "required_pages_per_second_upper_estimate": 0.0017239318354036813
+      "required_token_positions_per_second": 28485.64705364265,
+      "required_pages_per_second_upper_estimate": 0.0017346264529270554
     },
     {
       "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": 28310.022440268123,
-      "required_pages_per_second_upper_estimate": 0.002179383894609676
+      "required_token_positions_per_second": 28485.64705364265,
+      "required_pages_per_second_upper_estimate": 0.0021929039635072926
     },
     {
       "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": 28310.022440268123,
-      "required_pages_per_second_upper_estimate": 0.35151573115795376
+      "required_token_positions_per_second": 28485.64705364265,
+      "required_pages_per_second_upper_estimate": 0.3536964010782951
     }
   ],
   "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": 0.20320066571107986,
-    "producer_p95_seconds": 0.2783711450010742,
-    "gpu_update_seconds": 0.9766152626100902,
-    "producer_batches_per_second": 4.921243719850044,
-    "gpu_batches_per_second": 1.0239446773823828,
-    "producer_headroom": 4.8061617278296085,
+    "producer_batch_seconds": 0.2329691403359817,
+    "producer_p95_seconds": 0.25063624900030845,
+    "gpu_update_seconds": 0.9705940661251178,
+    "producer_batches_per_second": 4.2924140019481865,
+    "gpu_batches_per_second": 1.0302968407712185,
+    "producer_headroom": 4.16619155964328,
     "sustainable": true,
-    "buffer_seconds": 60.5501462818256,
+    "buffer_seconds": 60.176832099757306,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 472513.4550720909,
-    "upload_bytes_per_second": 5344257.70017976,
+    "generation_bytes_per_second": 449563.1473585996,
+    "upload_bytes_per_second": 5318211.38612779,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 101.39038790819211,
+    "estimated_full_shards_per_hour": 96.4657861286973,
     "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": 1.2902597590000369,
-      "bytes": 12650540,
-      "pages": 36,
-      "shards": 4
-    },
     {
       "seconds": 1.0202606470002138,
       "bytes": 1124000,
@@ -562,6 +556,12 @@
       "bytes": 2528980,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 1.481450629999017,
+      "bytes": 11596666,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -131,3 +131,4 @@
 {"step": 168100, "loss": 0.5387906767427921, "lr": 0.0003, "grad_norm": 0.1564837904446995, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24602123606, "tokens_per_second": 150961.6990159981, "data_wait_s": 1.2391001291689463e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 168200, "loss": 0.5773624591529369, "lr": 0.0003, "grad_norm": 0.14285495224951889, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24616734383, "tokens_per_second": 149160.76455226296, "data_wait_s": 1.1449999874457717e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 168300, "loss": 0.610111890360713, "lr": 0.0003, "grad_norm": 0.14470366306935767, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24631421174, "tokens_per_second": 150422.06956699205, "data_wait_s": 1.1874999472638592e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 168400, "loss": 0.6451338455080986, "lr": 0.0003, "grad_norm": 0.13721324623497358, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24646131979, "tokens_per_second": 151058.34318716076, "data_wait_s": 1.389499993820209e-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:acbf3daca9cf47b5d703734593184fe5e9c336b8654b92d86fd21af0355313b9
-size 283425
+oid sha256:7112fc1bfb7d9b1df5b04e64c2160b8c6a4dca2cdb0e5fa9cacd20cb6443a2a6
+size 285543