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": 76041.30865448438,
-      "required_pages_per_second_upper_estimate": 0.005260714172862016
+      "required_token_positions_per_second": 75604.30661243413,
+      "required_pages_per_second_upper_estimate": 0.005230481357608542
     },
     {
       "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": 9505.163581810548,
-      "required_pages_per_second_upper_estimate": 0.003979613501317394
+      "required_token_positions_per_second": 9450.538326554266,
+      "required_pages_per_second_upper_estimate": 0.003956743047646633
     },
     {
       "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": 19010.327163621096,
-      "required_pages_per_second_upper_estimate": 0.0039593498121415814
+      "required_token_positions_per_second": 18901.076653108532,
+      "required_pages_per_second_upper_estimate": 0.003936595811931524
     },
     {
       "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": 28515.490745431645,
-      "required_pages_per_second_upper_estimate": 0.0017094601295553433
+      "required_token_positions_per_second": 28351.614979662798,
+      "required_pages_per_second_upper_estimate": 0.0016996360276213067
     },
     {
       "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": 28515.490745431645,
-      "required_pages_per_second_upper_estimate": 0.0022169539790689906
+      "required_token_positions_per_second": 28351.614979662798,
+      "required_pages_per_second_upper_estimate": 0.0022042133590937786
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28515.490745431645,
-      "required_pages_per_second_upper_estimate": 0.365766097733888
+      "required_token_positions_per_second": 28351.614979662798,
+      "required_pages_per_second_upper_estimate": 0.3636640753666936
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09493991454687034,
-    "producer_p95_seconds": 0.28007562800121377,
-    "gpu_update_seconds": 0.9695782635068056,
-    "producer_batches_per_second": 10.532977670906957,
-    "gpu_batches_per_second": 1.0313762567068736,
-    "producer_headroom": 10.212546199713927,
+    "producer_batch_seconds": 0.2996538372655664,
+    "producer_p95_seconds": 0.29529358700165176,
+    "gpu_update_seconds": 0.9751825432107654,
+    "producer_batches_per_second": 3.3371840291628105,
+    "gpu_batches_per_second": 1.0254490371695169,
+    "producer_headroom": 3.254363608721339,
     "sustainable": true,
-    "buffer_seconds": 60.11385233742195,
+    "buffer_seconds": 60.461317679067456,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 526629.8402404939,
+    "generation_bytes_per_second": 489622.771880808,
     "upload_bytes_per_second": 3255229.205933221,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 113.00250440035927,
+    "estimated_full_shards_per_hour": 105.06164901083164,
     "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-v1/1791300814123050567-72be911d/train.jsonl

@@ -254,3 +254,4 @@
 {"step": 825400, "loss": 0.5109246373176575, "lr": 3e-05, "grad_norm": 0.2816684029767808, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120815394808, "tokens_per_second": 150872.49485313363, "data_wait_s": 1.995899947360158e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 825500, "loss": 0.5523240566253662, "lr": 3e-05, "grad_norm": 0.2925730057574528, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120830104871, "tokens_per_second": 152308.17124629006, "data_wait_s": 1.6930996935116127e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 825600, "loss": 0.5206471681594849, "lr": 3e-05, "grad_norm": 0.269644409841144, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120844770070, "tokens_per_second": 151627.88770341396, "data_wait_s": 7.342998287640512e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 825700, "loss": 0.5310903191566467, "lr": 3e-05, "grad_norm": 0.28555193636918114, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120859220936, "tokens_per_second": 149203.39291495833, "data_wait_s": 1.4158002159092575e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:33800d644f98d7de250fd3202e23e528644e4f44b054b84bb1141099da1ea717
-size 546288
+oid sha256:a437d57967c8b3e2abaa60135dc7d1ad83c37f470b8eb2a16b1a34c2c49a8c84
+size 548406