Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75405.45101169492,
-      "required_pages_per_second_upper_estimate": 0.0052167240657408406
+      "required_token_positions_per_second": 76162.94955690732,
+      "required_pages_per_second_upper_estimate": 0.005269129572737414
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9425.681376461866,
-      "required_pages_per_second_upper_estimate": 0.003946335961715088
+      "required_token_positions_per_second": 9520.368694613415,
+      "required_pages_per_second_upper_estimate": 0.0039859795645290925
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18851.36275292373,
-      "required_pages_per_second_upper_estimate": 0.003926241717566742
+      "required_token_positions_per_second": 19040.73738922683,
+      "required_pages_per_second_upper_estimate": 0.003965683460163676
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28277.0441293856,
-      "required_pages_per_second_upper_estimate": 0.0016951656190102842
+      "required_token_positions_per_second": 28561.106083840245,
+      "required_pages_per_second_upper_estimate": 0.00171219469944235
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28277.0441293856,
-      "required_pages_per_second_upper_estimate": 0.002198415803487232
+      "required_token_positions_per_second": 28561.106083840245,
+      "required_pages_per_second_upper_estimate": 0.00222050037098959
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28277.0441293856,
-      "required_pages_per_second_upper_estimate": 0.3627075605672785
+      "required_token_positions_per_second": 28561.106083840245,
+      "required_pages_per_second_upper_estimate": 0.3663512023170591
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.05208874725001067,
+    "producer_batch_seconds": 0.08082158564843667,
     "producer_p95_seconds": 0.1727614250003171,
-    "gpu_update_seconds": 0.977754247349641,
-    "producer_batches_per_second": 19.1980044211909,
-    "gpu_batches_per_second": 1.0227518854667823,
-    "producer_headroom": 18.770930363456586,
+    "gpu_update_seconds": 0.9680297366229498,
+    "producer_batches_per_second": 12.372932205882092,
+    "gpu_batches_per_second": 1.0330261170370458,
+    "producer_headroom": 11.977366304513655,
     "sustainable": true,
-    "buffer_seconds": 60.620763335677744,
+    "buffer_seconds": 60.01784367062289,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 302608.8738357323,
-    "upload_bytes_per_second": 4743967.783598639,
+    "generation_bytes_per_second": 302429.45187419973,
+    "upload_bytes_per_second": 4764561.588323765,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.93281995109535,
+    "estimated_full_shards_per_hour": 64.89432017487997,
     "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": [
@@ -298,6 +298,12 @@
       "bytes": 11076576,
       "pages": 34,
       "shards": 4
+    },
+    {
+      "seconds": 1.1856196190001356,
+      "bytes": 6180989,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -63,3 +63,4 @@
 {"step": 726300, "loss": 0.5189372301101685, "lr": 5e-05, "grad_norm": 0.2189112038765758, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106310085132, "tokens_per_second": 151529.8049592878, "data_wait_s": 8.615000297140796e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 726400, "loss": 0.5483208298683167, "lr": 5e-05, "grad_norm": 0.2387842533931959, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106324813331, "tokens_per_second": 151435.70127050293, "data_wait_s": 1.1844000255223364e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 726500, "loss": 0.5062057971954346, "lr": 5e-05, "grad_norm": 0.22825353047214517, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106339248228, "tokens_per_second": 149008.91150842526, "data_wait_s": 9.424999916518573e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 726600, "loss": 0.5636150240898132, "lr": 5e-05, "grad_norm": 0.2252468873031572, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106353808082, "tokens_per_second": 150484.9628767221, "data_wait_s": 1.3204000424593687e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:21b2076fd79bf8db6f5d07027d287e1a401f0a45381edbaf543834e96ca7818e
-size 138618
+oid sha256:745642a076088d7051fdf0a86d7be0e9bf4a8fa02dfef30a2ca5df368c17b187
+size 140736