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": 77248.31812813367,
-      "required_pages_per_second_upper_estimate": 0.005344217888896907
+      "required_token_positions_per_second": 76905.16716354029,
+      "required_pages_per_second_upper_estimate": 0.005320477908946401
     },
     {
       "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": 9656.03976601671,
-      "required_pages_per_second_upper_estimate": 0.004042782208991515
+      "required_token_positions_per_second": 9613.145895442536,
+      "required_pages_per_second_upper_estimate": 0.004024823441107982
     },
     {
       "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": 19312.07953203342,
-      "required_pages_per_second_upper_estimate": 0.004022196872736777
+      "required_token_positions_per_second": 19226.291790885072,
+      "required_pages_per_second_upper_estimate": 0.004004329548630449
     },
     {
       "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": 28968.11929805013,
-      "required_pages_per_second_upper_estimate": 0.001736594520161039
+      "required_token_positions_per_second": 28839.43768632761,
+      "required_pages_per_second_upper_estimate": 0.001728880254023718
     },
     {
       "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": 28968.11929805013,
-      "required_pages_per_second_upper_estimate": 0.00225214385813248
+      "required_token_positions_per_second": 28839.43768632761,
+      "required_pages_per_second_upper_estimate": 0.002242139428831646
     },
     {
       "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": 28968.11929805013,
-      "required_pages_per_second_upper_estimate": 0.37157193081220263
+      "required_token_positions_per_second": 28839.43768632761,
+      "required_pages_per_second_upper_estimate": 0.369921341264576
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09930000465627131,
-    "producer_p95_seconds": 0.2482575470021402,
-    "gpu_update_seconds": 0.9544285466216308,
-    "producer_batches_per_second": 10.070492981964275,
-    "gpu_batches_per_second": 1.0477473704445215,
-    "producer_headroom": 9.611565980539496,
+    "producer_batch_seconds": 0.0941097524844281,
+    "producer_p95_seconds": 0.24742205999791622,
+    "gpu_update_seconds": 0.9586872081457936,
+    "producer_batches_per_second": 10.625891298199571,
+    "gpu_batches_per_second": 1.0430930876131224,
+    "producer_headroom": 10.186906062731628,
     "sustainable": true,
-    "buffer_seconds": 60.12899843716274,
+    "buffer_seconds": 60.39729411318499,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 314940.4007743477,
+    "generation_bytes_per_second": 314717.03048790275,
     "upload_bytes_per_second": 5613198.370573173,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.57887856886695,
+    "estimated_full_shards_per_hour": 67.53094850519001,
     "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-v1/1791143623306220418-eb05a537/train.jsonl

@@ -228,3 +228,4 @@
 {"step": 707800, "loss": 0.5774579644203186, "lr": 8e-05, "grad_norm": 0.20599089814819257, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103601807208, "tokens_per_second": 152264.95125941405, "data_wait_s": 1.475400131312199e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 707900, "loss": 0.517055869102478, "lr": 8e-05, "grad_norm": 0.20318982154904014, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103616605217, "tokens_per_second": 154302.6934459139, "data_wait_s": 1.4085999282542616e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 708000, "loss": 0.585803747177124, "lr": 8e-05, "grad_norm": 0.19999829250549198, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103631406402, "tokens_per_second": 153242.46500055972, "data_wait_s": 1.530900044599548e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 708100, "loss": 0.586965799331665, "lr": 8e-05, "grad_norm": 0.2340161612950941, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103645940375, "tokens_per_second": 143789.36842867962, "data_wait_s": 1.3051998394075781e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1c6bd07353dabbdf6a16ac1a42e1aa1748006fb65787d3a1312d465771959f83
-size 490437
+oid sha256:443507def31fafb5b88a1d62a9d554d595f4a1bce27b83af45aea3b0ab8163f2
+size 492555