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": 77339.97337202121,
-      "required_pages_per_second_upper_estimate": 0.005350558811338513
+      "required_token_positions_per_second": 76933.99300121887,
+      "required_pages_per_second_upper_estimate": 0.005322472147282164
     },
     {
       "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": 9667.496671502651,
-      "required_pages_per_second_upper_estimate": 0.004047578976071086
+      "required_token_positions_per_second": 9616.749125152359,
+      "required_pages_per_second_upper_estimate": 0.004026332038143492
     },
     {
       "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": 19334.993343005302,
-      "required_pages_per_second_upper_estimate": 0.00402696921528439
+      "required_token_positions_per_second": 19233.498250304718,
+      "required_pages_per_second_upper_estimate": 0.004005830464080446
     },
     {
       "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": 29002.490014507952,
-      "required_pages_per_second_upper_estimate": 0.0017386549921316401
+      "required_token_positions_per_second": 28850.24737545708,
+      "required_pages_per_second_upper_estimate": 0.00172952827838159
     },
     {
       "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": 29002.490014507952,
-      "required_pages_per_second_upper_estimate": 0.0022548160301562713
+      "required_token_positions_per_second": 28850.24737545708,
+      "required_pages_per_second_upper_estimate": 0.0022429798346146647
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 29002.490014507952,
-      "required_pages_per_second_upper_estimate": 0.3720128014585235
+      "required_token_positions_per_second": 28850.24737545708,
+      "required_pages_per_second_upper_estimate": 0.37005999635018894
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.07659373584360196,
-    "producer_p95_seconds": 0.26923336899926653,
-    "gpu_update_seconds": 0.9532974577758533,
-    "producer_batches_per_second": 13.055897965884792,
-    "gpu_batches_per_second": 1.0489905242515896,
-    "producer_headroom": 12.446154339858907,
+    "producer_batch_seconds": 0.13240694131272335,
+    "producer_p95_seconds": 0.28252544299903093,
+    "gpu_update_seconds": 0.9583280046160079,
+    "producer_batches_per_second": 7.552474138332106,
+    "gpu_batches_per_second": 1.0434840630590667,
+    "producer_headroom": 7.237747470901811,
     "sustainable": true,
-    "buffer_seconds": 60.05773983987876,
+    "buffer_seconds": 60.3746642908085,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 223182.65970923155,
+    "generation_bytes_per_second": 235619.20399195547,
     "upload_bytes_per_second": 4690420.650504971,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 47.88980334718428,
+    "estimated_full_shards_per_hour": 50.5583962423229,
     "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-v1/1791143623306220418-eb05a537/train.jsonl

@@ -281,3 +281,4 @@
 {"step": 713100, "loss": 0.5497522950172424, "lr": 8e-05, "grad_norm": 0.21310864575031135, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104377844279, "tokens_per_second": 144989.59008227877, "data_wait_s": 1.6006997611839324e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 713200, "loss": 0.5147819519042969, "lr": 8e-05, "grad_norm": 0.23424810404783758, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104392597617, "tokens_per_second": 153461.35665833578, "data_wait_s": 1.5017998521216214e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 713300, "loss": 0.5463555455207825, "lr": 8e-05, "grad_norm": 0.21938065296435114, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104407196969, "tokens_per_second": 151590.80574856538, "data_wait_s": 1.593500201124698e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 713400, "loss": 0.5498261451721191, "lr": 8e-05, "grad_norm": 0.2149240752611408, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104421857417, "tokens_per_second": 152653.58740029114, "data_wait_s": 1.7070000467356294e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9cd7cfad4fb19b122d888f89ee030e4b747e720a33fb2b97e48d8e24d87d9693
-size 603474
+oid sha256:7bb12aeb5b65eaef034a4ab143fbc9d1a9b49f76ea0260f9ee00a7143056d309
+size 605592