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": 73687.11704012184,
-      "required_pages_per_second_upper_estimate": 0.0063870125467470505
+      "required_token_positions_per_second": 75655.25452668693,
+      "required_pages_per_second_upper_estimate": 0.006557605715883656
     },
     {
       "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": 9210.88963001523,
-      "required_pages_per_second_upper_estimate": 0.0036743615884854116
+      "required_token_positions_per_second": 9456.906815835866,
+      "required_pages_per_second_upper_estimate": 0.0037725015221939787
     },
     {
       "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": 18421.77926003046,
-      "required_pages_per_second_upper_estimate": 0.004261554419566293
+      "required_token_positions_per_second": 18913.813631671732,
+      "required_pages_per_second_upper_estimate": 0.00437537791193632
     },
     {
       "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": 27632.668890045687,
-      "required_pages_per_second_upper_estimate": 0.0016826845579945608
+      "required_token_positions_per_second": 28370.720447507596,
+      "required_pages_per_second_upper_estimate": 0.001727628025586733
     },
     {
       "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": 27632.668890045687,
-      "required_pages_per_second_upper_estimate": 0.0021272393432789183
+      "required_token_positions_per_second": 28370.720447507596,
+      "required_pages_per_second_upper_estimate": 0.0021840565952298088
     },
     {
       "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": 27632.668890045687,
-      "required_pages_per_second_upper_estimate": 0.34310526702069466
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28370.720447507596,
+      "required_pages_per_second_upper_estimate": 0.35226939726470563
     }
   ],
   "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": 60,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.5285637208592675,
-    "producer_p95_seconds": 0.2946540390003065,
-    "gpu_update_seconds": 1.0005548182846657,
-    "producer_batches_per_second": 1.8919194801609447,
-    "gpu_batches_per_second": 0.999445489367972,
-    "producer_headroom": 1.8929691516816531,
+    "producer_batch_seconds": 0.40433110567974495,
+    "producer_p95_seconds": 0.32220593800229835,
+    "gpu_update_seconds": 0.9745258338136038,
+    "producer_batches_per_second": 2.473220551060104,
+    "gpu_batches_per_second": 1.0261400624821904,
+    "producer_headroom": 2.4102173197267884,
     "sustainable": true,
-    "buffer_seconds": 60.033289097079944,
+    "buffer_seconds": 60.420601696443434,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 189105.6061419702,
-    "upload_bytes_per_second": 6183311.493122145,
+    "generation_bytes_per_second": 185953.15116242765,
+    "upload_bytes_per_second": 6246819.088888134,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 40.57766092485742,
+    "estimated_full_shards_per_hour": 39.901217471643655,
     "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": 1.0279631710000103,
-      "bytes": 7068347,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 0.8251235360003193,
       "bytes": 2046300,
@@ -562,6 +556,12 @@
       "bytes": 5928833,
       "pages": 24,
       "shards": 2
+    },
+    {
+      "seconds": 1.330314847000409,
+      "bytes": 13379246,
+      "pages": 41,
+      "shards": 5
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -169,3 +169,4 @@
 {"step": 16900, "loss": 0.6956988908350468, "lr": 0.0003, "grad_norm": 0.18371533973157148, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2473089397, "tokens_per_second": 150360.3606912104, "data_wait_s": 1.3584001862909645e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 17000, "loss": 0.671393072232604, "lr": 0.0003, "grad_norm": 0.1515170674137935, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2487627387, "tokens_per_second": 147841.38964624522, "data_wait_s": 9.729999874252826e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 17100, "loss": 0.6575552709400654, "lr": 0.0003, "grad_norm": 0.1446292805032547, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2502147483, "tokens_per_second": 136587.49712598158, "data_wait_s": 1.0157000360777602e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 17200, "loss": 0.6627637622877955, "lr": 0.0003, "grad_norm": 0.14520324430035478, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2516775760, "tokens_per_second": 149344.46905586615, "data_wait_s": 1.0747997293947265e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:70b934c271322cd0b7382c5fd88220ddef669b138aa186af699305336babd45f
-size 358918
+oid sha256:5a0f9540ab74e6f96713dbde8e8a18ab6e95240c41473a6dd9ef3d77f0719113
+size 361036