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": 77420.95173127238,
-      "required_pages_per_second_upper_estimate": 0.005356161082127189
+      "required_token_positions_per_second": 76745.39204928796,
+      "required_pages_per_second_upper_estimate": 0.0053094242958130825
     },
     {
       "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": 9677.618966409047,
-      "required_pages_per_second_upper_estimate": 0.00405181696957084
+      "required_token_positions_per_second": 9593.174006160994,
+      "required_pages_per_second_upper_estimate": 0.004016461628126275
     },
     {
       "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": 19355.237932818094,
-      "required_pages_per_second_upper_estimate": 0.004031185629456658
+      "required_token_positions_per_second": 19186.34801232199,
+      "required_pages_per_second_upper_estimate": 0.003996010312944037
     },
     {
       "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": 29032.856899227143,
-      "required_pages_per_second_upper_estimate": 0.0017404754405030073
+      "required_token_positions_per_second": 28779.522018482985,
+      "required_pages_per_second_upper_estimate": 0.0017252884012223604
     },
     {
       "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": 29032.856899227143,
-      "required_pages_per_second_upper_estimate": 0.0022571769218733785
+      "required_token_positions_per_second": 28779.522018482985,
+      "required_pages_per_second_upper_estimate": 0.0022374812491978975
     },
     {
       "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": 29032.856899227143,
-      "required_pages_per_second_upper_estimate": 0.37240231525028084
+      "required_token_positions_per_second": 28779.522018482985,
+      "required_pages_per_second_upper_estimate": 0.3691528074098971
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.048634109374990686,
-    "producer_p95_seconds": 0.14790451299995766,
-    "gpu_update_seconds": 0.9523003573491244,
-    "producer_batches_per_second": 20.561700683969633,
-    "gpu_batches_per_second": 1.0500888635426484,
-    "producer_headroom": 19.58091490905002,
+    "producer_batch_seconds": 0.2303537506875024,
+    "producer_p95_seconds": 0.15871822700000848,
+    "gpu_update_seconds": 0.9606830850854198,
+    "producer_batches_per_second": 4.34114919776843,
+    "gpu_batches_per_second": 1.0409259989323996,
+    "producer_headroom": 4.170468604128271,
     "sustainable": true,
-    "buffer_seconds": 60.947222870343964,
+    "buffer_seconds": 60.523034360381445,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 410161.1903236289,
-    "upload_bytes_per_second": 3942940.5998206283,
+    "generation_bytes_per_second": 396703.7814950977,
+    "upload_bytes_per_second": 4098194.445974557,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 88.01104337960864,
+    "estimated_full_shards_per_hour": 85.12339671745012,
     "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": [
@@ -328,6 +328,12 @@
       "bytes": 12316311,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 1.3365711859996736,
+      "bytes": 11197506,
+      "pages": 36,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -27,3 +27,4 @@
 {"step": 845200, "loss": 0.4966509938240051, "lr": 3e-05, "grad_norm": 0.3019810721601624, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123713823583, "tokens_per_second": 154940.3224176908, "data_wait_s": 1.1714999800460646e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 845300, "loss": 0.5320813059806824, "lr": 3e-05, "grad_norm": 0.29579668042968593, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123728522712, "tokens_per_second": 153232.44099634653, "data_wait_s": 1.904400005514617e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 845400, "loss": 0.5353712439537048, "lr": 3e-05, "grad_norm": 0.35950841787826937, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123743053716, "tokens_per_second": 150599.3986406735, "data_wait_s": 1.2543999673653161e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 845500, "loss": 0.5478113889694214, "lr": 3e-05, "grad_norm": 0.31219902425806406, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123757522136, "tokens_per_second": 150237.6673645429, "data_wait_s": 1.7101000139518874e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:90eeadd21a9b4288dc41d988b4e416e20a085bd69f5162982fa30e0060954cb6
-size 62370
+oid sha256:f0f5350796ce7696a17fe22f1d526ae972536103a228c8baf96ba6372c80429b
+size 64488