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": 76585.08832986189,
-      "required_pages_per_second_upper_estimate": 0.005298334112547313
+      "required_token_positions_per_second": 76609.40518205908,
+      "required_pages_per_second_upper_estimate": 0.005300016408805185
     },
     {
       "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": 9573.136041232736,
-      "required_pages_per_second_upper_estimate": 0.004008072150651103
+      "required_token_positions_per_second": 9576.175647757385,
+      "required_pages_per_second_upper_estimate": 0.004009344770428772
     },
     {
       "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": 19146.27208246547,
-      "required_pages_per_second_upper_estimate": 0.003987663553628266
+      "required_token_positions_per_second": 19152.35129551477,
+      "required_pages_per_second_upper_estimate": 0.003988929693386807
     },
     {
       "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": 28719.40812369821,
-      "required_pages_per_second_upper_estimate": 0.0017216846650186157
+      "required_token_positions_per_second": 28728.526943272158,
+      "required_pages_per_second_upper_estimate": 0.0017222313243284426
     },
     {
       "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": 28719.40812369821,
-      "required_pages_per_second_upper_estimate": 0.0022328076582914603
+      "required_token_positions_per_second": 28728.526943272158,
+      "required_pages_per_second_upper_estimate": 0.0022335166064038864
     },
     {
       "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": 28719.40812369821,
-      "required_pages_per_second_upper_estimate": 0.36838173091286935
+      "required_token_positions_per_second": 28728.526943272158,
+      "required_pages_per_second_upper_estimate": 0.3684986973393384
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09038238322631287,
-    "producer_p95_seconds": 0.31688419800047996,
-    "gpu_update_seconds": 0.9626939343915615,
-    "producer_batches_per_second": 11.064103028751202,
-    "gpu_batches_per_second": 1.0387517405851492,
-    "producer_headroom": 10.651344875262087,
+    "producer_batch_seconds": 0.10530479371908541,
+    "producer_p95_seconds": 0.3137233930028742,
+    "gpu_update_seconds": 0.9623883624313289,
+    "producer_batches_per_second": 9.496243852560344,
+    "gpu_batches_per_second": 1.0390815590014524,
+    "producer_headroom": 9.139074570514124,
     "sustainable": true,
-    "buffer_seconds": 60.649717866668375,
+    "buffer_seconds": 60.63046683317372,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 450464.52220969455,
-    "upload_bytes_per_second": 4897340.299530888,
+    "generation_bytes_per_second": 450271.2581740131,
+    "upload_bytes_per_second": 4771348.523071785,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.65920018880965,
+    "estimated_full_shards_per_hour": 96.61773022570891,
     "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": [
-    {
-      "seconds": 1.4461447649991896,
-      "bytes": 15647853,
-      "pages": 34,
-      "shards": 4
-    },
     {
       "seconds": 2.7048667290000594,
       "bytes": 2731260,
@@ -562,6 +556,12 @@
       "bytes": 10855257,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 1.1245584639982553,
+      "bytes": 2759850,
+      "pages": 40,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -338,3 +338,4 @@
 {"step": 793800, "loss": 0.5176961421966553, "lr": 3e-05, "grad_norm": 0.26464572361485117, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116189747580, "tokens_per_second": 150848.97817983935, "data_wait_s": 6.815003871452063e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 793900, "loss": 0.498049259185791, "lr": 3e-05, "grad_norm": 0.28876484965245436, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116204357103, "tokens_per_second": 151941.34767091012, "data_wait_s": 1.3454999134410173e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 794000, "loss": 0.4984740614891052, "lr": 3e-05, "grad_norm": 0.25053805706835874, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116219109402, "tokens_per_second": 153060.61836415264, "data_wait_s": 1.3880002370569855e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 794100, "loss": 0.5055521726608276, "lr": 3e-05, "grad_norm": 0.258596852103173, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116233781402, "tokens_per_second": 153481.87996062965, "data_wait_s": 9.018003765959293e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:62909570e65a543ebeaf665f179646ee629808cae192e02bf05092232bbd0e56
-size 724983
+oid sha256:59fa78fa2ebb5e4ac1e7c4d407b8a6a975dd3bf444e15cc372d42b7b3e9873d3
+size 727101