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": 75823.16868503424,
-      "required_pages_per_second_upper_estimate": 0.005245622743620991
+      "required_token_positions_per_second": 76510.13681539625,
+      "required_pages_per_second_upper_estimate": 0.005293148792865106
     },
     {
       "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": 9477.89608562928,
-      "required_pages_per_second_upper_estimate": 0.003968197170076367
+      "required_token_positions_per_second": 9563.767101924532,
+      "required_pages_per_second_upper_estimate": 0.004004149571408458
     },
     {
       "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": 18955.79217125856,
-      "required_pages_per_second_upper_estimate": 0.003947991611416927
+      "required_token_positions_per_second": 19127.534203849063,
+      "required_pages_per_second_upper_estimate": 0.003983760947663558
     },
     {
       "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": 28433.68825688784,
-      "required_pages_per_second_upper_estimate": 0.00170455619527231
+      "required_token_positions_per_second": 28691.301305773595,
+      "required_pages_per_second_upper_estimate": 0.0017199997042006612
     },
     {
       "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": 28433.68825688784,
-      "required_pages_per_second_upper_estimate": 0.0022105941953958308
+      "required_token_positions_per_second": 28691.301305773595,
+      "required_pages_per_second_upper_estimate": 0.0022306224768266483
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28433.68825688784,
-      "required_pages_per_second_upper_estimate": 0.3647168232435171
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28691.301305773595,
+      "required_pages_per_second_upper_estimate": 0.36802120683128225
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.11613496818745261,
-    "producer_p95_seconds": 0.22942908799814177,
-    "gpu_update_seconds": 0.9723676981406901,
-    "producer_batches_per_second": 8.610670977116103,
-    "gpu_batches_per_second": 1.0284175440135936,
-    "producer_headroom": 8.372738317465231,
+    "producer_batch_seconds": 0.11548192772650623,
+    "producer_p95_seconds": 0.23647541599711985,
+    "gpu_update_seconds": 0.9636370168555705,
+    "producer_batches_per_second": 8.659363587766581,
+    "gpu_batches_per_second": 1.0377351456081305,
+    "producer_headroom": 8.344483295583139,
     "sustainable": true,
-    "buffer_seconds": 60.28679728472279,
+    "buffer_seconds": 60.70913206190094,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 443978.77285473206,
+    "generation_bytes_per_second": 470448.6907151498,
     "upload_bytes_per_second": 1882427.3294040847,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 95.26750935775253,
+    "estimated_full_shards_per_hour": 100.9473375424468,
     "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-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -166,3 +166,4 @@
 {"step": 661600, "loss": 0.5818714499473572, "lr": 0.0001, "grad_norm": 0.18780524546508917, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96841124320, "tokens_per_second": 153551.00608325418, "data_wait_s": 1.268800042453222e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 661700, "loss": 0.5681452751159668, "lr": 0.0001, "grad_norm": 0.17807320642351002, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96855865914, "tokens_per_second": 153883.15016461324, "data_wait_s": 1.0070998541777954e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 661800, "loss": 0.5450122356414795, "lr": 0.0001, "grad_norm": 0.18417547114940816, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96870638696, "tokens_per_second": 154010.79874109037, "data_wait_s": 7.2549992182757705e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 661900, "loss": 0.533353328704834, "lr": 0.0001, "grad_norm": 0.18042469287622923, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96885118580, "tokens_per_second": 151442.23393889933, "data_wait_s": 1.3823999324813485e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6bea7ff1eaaacbfddfc9977d94d29b388d56a090f5f9159615ca425234e70877
-size 358338
+oid sha256:4d38782f527f8f213893392da22beabc9f6a8f8f03831eae009ce40fb2ce61fb
+size 360456