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": 76749.45336819721,
-      "required_pages_per_second_upper_estimate": 0.005309705267278783
+      "required_token_positions_per_second": 76870.04996486272,
+      "required_pages_per_second_upper_estimate": 0.00531804841965876
     },
     {
       "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": 9593.681671024651,
-      "required_pages_per_second_upper_estimate": 0.0040166741768034395
+      "required_token_positions_per_second": 9608.75624560784,
+      "required_pages_per_second_upper_estimate": 0.004022985586388507
     },
     {
       "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": 19187.363342049302,
-      "required_pages_per_second_upper_estimate": 0.003996221779350191
+      "required_token_positions_per_second": 19217.51249121568,
+      "required_pages_per_second_upper_estimate": 0.004002501052035016
     },
     {
       "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": 28781.045013073955,
-      "required_pages_per_second_upper_estimate": 0.0017253797024226112
+      "required_token_positions_per_second": 28826.26873682352,
+      "required_pages_per_second_upper_estimate": 0.0017280907956087695
     },
     {
       "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": 28781.045013073955,
-      "required_pages_per_second_upper_estimate": 0.0022375996553284003
+      "required_token_positions_per_second": 28826.26873682352,
+      "required_pages_per_second_upper_estimate": 0.0022411156009317906
     },
     {
       "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": 28781.045013073955,
-      "required_pages_per_second_upper_estimate": 0.3691723427492459
+      "required_token_positions_per_second": 28826.26873682352,
+      "required_pages_per_second_upper_estimate": 0.3697524241200539
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.18640661994531627,
-    "producer_p95_seconds": 0.22745046799900592,
-    "gpu_update_seconds": 0.960632249018086,
-    "producer_batches_per_second": 5.364616344062015,
-    "gpu_batches_per_second": 1.040981084095557,
-    "producer_headroom": 5.153423463715475,
+    "producer_batch_seconds": 0.1035155791796285,
+    "producer_p95_seconds": 0.2496087700001226,
+    "gpu_update_seconds": 0.9591251733763807,
+    "producer_batches_per_second": 9.66038163458198,
+    "gpu_batches_per_second": 1.0426167801223785,
+    "producer_headroom": 9.265515210150445,
     "sustainable": true,
-    "buffer_seconds": 60.51983168813942,
+    "buffer_seconds": 60.42488592271199,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 428839.1840307309,
-    "upload_bytes_per_second": 5672089.532033722,
+    "generation_bytes_per_second": 480339.02166338806,
+    "upload_bytes_per_second": 5737058.287820015,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.0189060277123,
+    "estimated_full_shards_per_hour": 103.06957232881767,
     "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": 1.0935950399998546,
-      "bytes": 4381259,
-      "pages": 34,
-      "shards": 2
-    },
     {
       "seconds": 1.3099932099994476,
       "bytes": 13324819,
@@ -562,6 +556,12 @@
       "bytes": 5880079,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.5804328610010998,
+      "bytes": 12088466,
+      "pages": 42,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -159,3 +159,4 @@
 {"step": 413000, "loss": 0.5962061882019043, "lr": 0.0003, "grad_norm": 0.1740489759002041, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60452450337, "tokens_per_second": 151370.53684027633, "data_wait_s": 1.420499938831199e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 413100, "loss": 0.581549882888794, "lr": 0.0003, "grad_norm": 0.16567601139019777, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60467029185, "tokens_per_second": 150324.64960564225, "data_wait_s": 1.538399919809308e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 413200, "loss": 0.5733972191810608, "lr": 0.0003, "grad_norm": 0.1556347186809304, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60481558612, "tokens_per_second": 150772.12100413707, "data_wait_s": 1.2974998753634281e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 413300, "loss": 0.5443356037139893, "lr": 0.0003, "grad_norm": 0.13792302804523293, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60496215030, "tokens_per_second": 151755.15753279935, "data_wait_s": 1.3458000466926023e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:80a31d0b862ca36bbf9ec1d595e3ff792c3759ab4b12a37fb63803d1646da517
-size 343512
+oid sha256:d5360a0c0999d03b0d7c6f5629b2307908d5d2f4dd4235e4c1b76b84239a036d
+size 345630