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": 75308.61800825993,
-      "required_pages_per_second_upper_estimate": 0.006527560141008628
+      "required_token_positions_per_second": 75901.1839546183,
+      "required_pages_per_second_upper_estimate": 0.006578922255394312
     },
     {
       "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": 9413.577251032491,
-      "required_pages_per_second_upper_estimate": 0.0037552167109591874
+      "required_token_positions_per_second": 9487.647994327288,
+      "required_pages_per_second_upper_estimate": 0.0037847646379157844
     },
     {
       "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": 18827.154502064983,
-      "required_pages_per_second_upper_estimate": 0.0043553308474503895
+      "required_token_positions_per_second": 18975.295988654576,
+      "required_pages_per_second_upper_estimate": 0.0043896007731717744
     },
     {
       "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": 28240.731753097476,
-      "required_pages_per_second_upper_estimate": 0.0017197123960951285
+      "required_token_positions_per_second": 28462.943982981862,
+      "required_pages_per_second_upper_estimate": 0.0017332439550376197
     },
     {
       "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": 28240.731753097476,
-      "required_pages_per_second_upper_estimate": 0.002174049705702381
+      "required_token_positions_per_second": 28462.943982981862,
+      "required_pages_per_second_upper_estimate": 0.002191156218281703
     },
     {
       "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": 28240.731753097476,
-      "required_pages_per_second_upper_estimate": 0.3506553727243065
+      "required_token_positions_per_second": 28462.943982981862,
+      "required_pages_per_second_upper_estimate": 0.35341450492297777
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.11680671857820357,
-    "producer_p95_seconds": 0.28171148800174706,
-    "gpu_update_seconds": 0.9790114591123347,
-    "producer_batches_per_second": 8.561151380436112,
-    "gpu_batches_per_second": 1.0214385038012685,
-    "producer_headroom": 8.381465304642337,
+    "producer_batch_seconds": 0.23423475675781447,
+    "producer_p95_seconds": 0.2861575970018748,
+    "gpu_update_seconds": 0.9713682469575486,
+    "producer_batches_per_second": 4.269221245564097,
+    "gpu_batches_per_second": 1.0294756938289158,
+    "producer_headroom": 4.146985957177519,
     "sustainable": true,
-    "buffer_seconds": 60.69871046496475,
+    "buffer_seconds": 60.224831311368014,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 358129.6508733222,
-    "upload_bytes_per_second": 5770466.541366329,
+    "generation_bytes_per_second": 352402.46950510534,
+    "upload_bytes_per_second": 5751181.566466037,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.84628624582052,
+    "estimated_full_shards_per_hour": 75.61736644616003,
     "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": 1.1229757969995262,
-      "bytes": 5828242,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 1.539654905001953,
       "bytes": 12116214,
@@ -562,6 +556,12 @@
       "bytes": 11789498,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 0.8314693719985371,
+      "bytes": 2720719,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -232,3 +232,4 @@
 {"step": 23200, "loss": 0.6290844772011042, "lr": 0.0003, "grad_norm": 0.1338471451423039, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3394519508, "tokens_per_second": 148713.17656576616, "data_wait_s": 1.089999932446517e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 23300, "loss": 0.6249117664992809, "lr": 0.0003, "grad_norm": 0.13694267293745865, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3409082308, "tokens_per_second": 131761.32919313002, "data_wait_s": 1.4543002180289477e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 23400, "loss": 0.6614399738609791, "lr": 0.0003, "grad_norm": 0.13861435202380004, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3423522740, "tokens_per_second": 147256.87673126796, "data_wait_s": 1.0941999789793044e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 23500, "loss": 0.6564783193171024, "lr": 0.0003, "grad_norm": 0.1383765923556441, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3438227822, "tokens_per_second": 150066.6958219596, "data_wait_s": 1.1198000720469281e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e343242ed4df91583fbf689f1e25b903da716fdab86f8679a084da28b8e9e712
-size 493135
+oid sha256:22b48980d64073c0d4059cae5667d3e6c3a7d236e8c2d1e4b6ecadd8b750b0c2
+size 495253