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": 74362.39532803131,
-      "required_pages_per_second_upper_estimate": 0.00644554395726587
+      "required_token_positions_per_second": 74738.85865214404,
+      "required_pages_per_second_upper_estimate": 0.006478174844062389
     },
     {
       "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": 9295.299416003914,
-      "required_pages_per_second_upper_estimate": 0.003708033914155064
+      "required_token_positions_per_second": 9342.357331518006,
+      "required_pages_per_second_upper_estimate": 0.0037268060202321706
     },
     {
       "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": 18590.59883200783,
-      "required_pages_per_second_upper_estimate": 0.004300607856420267
+      "required_token_positions_per_second": 18684.71466303601,
+      "required_pages_per_second_upper_estimate": 0.004322379897546583
     },
     {
       "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": 27885.898248011745,
-      "required_pages_per_second_upper_estimate": 0.0016981048973029314
+      "required_token_positions_per_second": 28027.07199455402,
+      "required_pages_per_second_upper_estimate": 0.0017067016377859507
     },
     {
       "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": 27885.898248011745,
-      "required_pages_per_second_upper_estimate": 0.0021467336402388733
+      "required_token_positions_per_second": 28027.07199455402,
+      "required_pages_per_second_upper_estimate": 0.00215760158604164
     },
     {
       "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": 27885.898248011745,
-      "required_pages_per_second_upper_estimate": 0.3462495281424903
+      "required_token_positions_per_second": 28027.07199455402,
+      "required_pages_per_second_upper_estimate": 0.3480024335964093
     }
   ],
   "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": 0.08007204931251977,
-    "producer_p95_seconds": 0.2921649759991851,
-    "gpu_update_seconds": 0.9914688691073917,
-    "producer_batches_per_second": 12.488752424669663,
-    "gpu_batches_per_second": 1.0086045373268135,
-    "producer_headroom": 12.382209243049427,
+    "producer_batch_seconds": 0.32235900739064505,
+    "producer_p95_seconds": 0.33590944000025047,
+    "gpu_update_seconds": 0.9864747914221051,
+    "producer_batches_per_second": 3.1021314034143543,
+    "gpu_batches_per_second": 1.0137106479511726,
+    "producer_headroom": 3.0601744291471378,
     "sustainable": true,
-    "buffer_seconds": 60.4796010155509,
+    "buffer_seconds": 60.17496227674842,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 493545.12244336243,
-    "upload_bytes_per_second": 3974204.3136536246,
+    "generation_bytes_per_second": 452770.20355092007,
+    "upload_bytes_per_second": 3944657.9583660485,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 105.90329413390783,
+    "estimated_full_shards_per_hour": 97.15394573112204,
     "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": 1.3172718099995109,
-      "bytes": 5670029,
-      "pages": 25,
-      "shards": 2
-    },
     {
       "seconds": 1.2858697970004869,
       "bytes": 11849581,
@@ -562,6 +556,12 @@
       "bytes": 7681898,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 0.8942368999996688,
+      "bytes": 850966,
+      "pages": 25,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -220,3 +220,4 @@
 {"step": 137000, "loss": 0.5834247525781393, "lr": 0.0003, "grad_norm": 0.15340312132563966, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20049573944, "tokens_per_second": 150249.84502143835, "data_wait_s": 1.5495999832637608e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 137100, "loss": 0.633556991815567, "lr": 0.0003, "grad_norm": 0.13615735735379328, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20064270632, "tokens_per_second": 150067.10070603818, "data_wait_s": 1.2589000107254833e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 137200, "loss": 0.5906205177307129, "lr": 0.0003, "grad_norm": 0.1406812911077866, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20078968164, "tokens_per_second": 149795.1198935231, "data_wait_s": 1.862299905042164e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 137300, "loss": 0.5933730937540531, "lr": 0.0003, "grad_norm": 0.1486073915231052, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20093579691, "tokens_per_second": 148801.1000828846, "data_wait_s": 1.5335001080529764e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:23931eef58a123c86a3add224598961fb74a4d79aab30edbe2ef03d582b5cdb7
-size 473493
+oid sha256:2b8070464aaea078bb2d93f2c1d9e549c06d3254f033b95cebd064d47a3c3938
+size 475611