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": 75217.95158068182,
-      "required_pages_per_second_upper_estimate": 0.006519701404858118
+      "required_token_positions_per_second": 75328.29453607876,
+      "required_pages_per_second_upper_estimate": 0.00652926565257026
     },
     {
       "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": 9402.243947585228,
-      "required_pages_per_second_upper_estimate": 0.003750695686766087
+      "required_token_positions_per_second": 9416.036817009845,
+      "required_pages_per_second_upper_estimate": 0.003756197868601342
     },
     {
       "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": 18804.487895170456,
-      "required_pages_per_second_upper_estimate": 0.004350087326863997
+      "required_token_positions_per_second": 18832.07363401969,
+      "required_pages_per_second_upper_estimate": 0.0043564688020543444
     },
     {
       "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": 28206.731842755682,
-      "required_pages_per_second_upper_estimate": 0.0017176419799390566
+      "required_token_positions_per_second": 28248.110451029534,
+      "required_pages_per_second_upper_estimate": 0.0017201617200861548
     },
     {
       "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": 28206.731842755682,
-      "required_pages_per_second_upper_estimate": 0.0021714322984864943
+      "required_token_positions_per_second": 28248.110451029534,
+      "required_pages_per_second_upper_estimate": 0.002174617738294734
     },
     {
       "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": 28206.731842755682,
-      "required_pages_per_second_upper_estimate": 0.35023320762824145
+      "required_token_positions_per_second": 28248.110451029534,
+      "required_pages_per_second_upper_estimate": 0.3507469914577093
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.07465409509376286,
-    "producer_p95_seconds": 0.19084727699919313,
-    "gpu_update_seconds": 0.9801915427185804,
-    "producer_batches_per_second": 13.395112468298436,
-    "gpu_batches_per_second": 1.0202087616737443,
-    "producer_headroom": 13.129775955190336,
+    "producer_batch_seconds": 0.10442788073436304,
+    "producer_p95_seconds": 0.21792881500005024,
+    "gpu_update_seconds": 0.978755731217142,
+    "producer_batches_per_second": 9.575986728522587,
+    "gpu_batches_per_second": 1.0217053837901307,
+    "producer_headroom": 9.372551892600772,
     "sustainable": true,
-    "buffer_seconds": 60.77187564855199,
+    "buffer_seconds": 60.682855335462804,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 316990.6801276196,
+    "generation_bytes_per_second": 305665.3268107316,
     "upload_bytes_per_second": 5737993.573265884,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.01882079001847,
+    "estimated_full_shards_per_hour": 65.58866360894643,
     "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": [

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -109,3 +109,4 @@
 {"step": 365900, "loss": 0.5644729752093554, "lr": 0.0003, "grad_norm": 0.1614524000850837, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53557545575, "tokens_per_second": 151592.21904008932, "data_wait_s": 1.0149999070563354e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 366000, "loss": 0.6050304397940636, "lr": 0.0003, "grad_norm": 0.15895203433216606, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53572241822, "tokens_per_second": 150119.314526114, "data_wait_s": 1.3007000234210864e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 366100, "loss": 0.6828296761959791, "lr": 0.0003, "grad_norm": 0.2705137944203427, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53586887872, "tokens_per_second": 149607.94873409357, "data_wait_s": 1.0300000212737359e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 366200, "loss": 0.5273890234529972, "lr": 0.0003, "grad_norm": 0.13614282773534725, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53601520104, "tokens_per_second": 148820.06353189057, "data_wait_s": 1.6324000171152875e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ef0da326b2607ed0d9bbe1fe827eade3c9d2c5d022be70bdd0233a96f2634f35
-size 236829
+oid sha256:a92c67ce88d9503275439424a4c87f067875bf26c35231b45e8d6d9c25148c34
+size 238947