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": 77100.53261989559,
-      "required_pages_per_second_upper_estimate": 0.005333993744527373
+      "required_token_positions_per_second": 77716.70250245683,
+      "required_pages_per_second_upper_estimate": 0.0053766218067140634
     },
     {
       "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": 9637.566577486949,
-      "required_pages_per_second_upper_estimate": 0.004035047870718147
+      "required_token_positions_per_second": 9714.587812807104,
+      "required_pages_per_second_upper_estimate": 0.004067295053560407
     },
     {
       "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": 19275.133154973897,
-      "required_pages_per_second_upper_estimate": 0.004014501916736764
+      "required_token_positions_per_second": 19429.175625614207,
+      "required_pages_per_second_upper_estimate": 0.004046584900998007
     },
     {
       "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": 28912.699732460846,
-      "required_pages_per_second_upper_estimate": 0.0017332722018247382
+      "required_token_positions_per_second": 29143.76343842131,
+      "required_pages_per_second_upper_estimate": 0.0017471241181832176
     },
     {
       "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": 28912.699732460846,
-      "required_pages_per_second_upper_estimate": 0.0022478352306728204
+      "required_token_positions_per_second": 29143.76343842131,
+      "required_pages_per_second_upper_estimate": 0.002265799417469414
     },
     {
       "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": 28912.699732460846,
-      "required_pages_per_second_upper_estimate": 0.3708610681297167
+      "required_token_positions_per_second": 29143.76343842131,
+      "required_pages_per_second_upper_estimate": 0.37382490525289963
     }
   ],
   "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": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.07612961767182469,
+    "producer_batch_seconds": 0.05581874067188153,
     "producer_p95_seconds": 0.17961665599978005,
-    "gpu_update_seconds": 0.9562579854471029,
-    "producer_batches_per_second": 13.135492211595547,
-    "gpu_batches_per_second": 1.0457429012030108,
-    "producer_headroom": 12.560919320116469,
+    "gpu_update_seconds": 0.9486763800570317,
+    "producer_batches_per_second": 17.915130079309478,
+    "gpu_batches_per_second": 1.0541002401049375,
+    "producer_headroom": 16.995660751890156,
     "sustainable": true,
-    "buffer_seconds": 60.244253083167486,
+    "buffer_seconds": 60.71528832365003,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 265197.07556358655,
-    "upload_bytes_per_second": 4582484.716336945,
+    "generation_bytes_per_second": 265135.29319237365,
+    "upload_bytes_per_second": 4544197.60484799,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 56.90511894398401,
+    "estimated_full_shards_per_hour": 56.89186188534171,
     "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": [
@@ -550,6 +550,12 @@
       "bytes": 7612762,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 0.9550784620005288,
+      "bytes": 883731,
+      "pages": 26,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -92,3 +92,4 @@
 {"step": 534200, "loss": 0.545046329498291, "lr": 0.0003, "grad_norm": 0.14561254477901242, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78190043077, "tokens_per_second": 154572.469506171, "data_wait_s": 1.1749998520826921e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 534300, "loss": 0.5961855053901672, "lr": 0.0003, "grad_norm": 0.15559007899240454, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78204758922, "tokens_per_second": 153818.65808628243, "data_wait_s": 1.0103998647537082e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 534400, "loss": 0.5679291486740112, "lr": 0.0003, "grad_norm": 0.1727571190424735, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78219399659, "tokens_per_second": 153467.69600136526, "data_wait_s": 7.64700052968692e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 534500, "loss": 0.5572027564048767, "lr": 0.0003, "grad_norm": 0.1427465521688393, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78233896276, "tokens_per_second": 142063.26615893075, "data_wait_s": 9.800998668652028e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:da68ab853cec09d18afe2c919e25a1f5362df1374d4fdd8914b8c4c0596a6032
-size 200040
+oid sha256:0d49a1281873d0d63de679b046873c8fa080abcc4f486c5761c11b256569664b
+size 202158