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": 77000.30291867639,
-      "required_pages_per_second_upper_estimate": 0.005327059621232081
+      "required_token_positions_per_second": 76768.05628092766,
+      "required_pages_per_second_upper_estimate": 0.0053109922599461805
     },
     {
       "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": 9625.037864834549,
-      "required_pages_per_second_upper_estimate": 0.004029802360359858
+      "required_token_positions_per_second": 9596.007035115957,
+      "required_pages_per_second_upper_estimate": 0.004017647758189346
     },
     {
       "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": 19250.075729669097,
-      "required_pages_per_second_upper_estimate": 0.004009283115854517
+      "required_token_positions_per_second": 19192.014070231915,
+      "required_pages_per_second_upper_estimate": 0.003997190403382679
     },
     {
       "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": 28875.113594503648,
-      "required_pages_per_second_upper_estimate": 0.001731018970244914
+      "required_token_positions_per_second": 28788.02110534787,
+      "required_pages_per_second_upper_estimate": 0.0017257979085025562
     },
     {
       "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": 28875.113594503648,
-      "required_pages_per_second_upper_estimate": 0.002244913073770597
+      "required_token_positions_per_second": 28788.02110534787,
+      "required_pages_per_second_upper_estimate": 0.00223814201581811
     },
     {
       "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": 28875.113594503648,
-      "required_pages_per_second_upper_estimate": 0.37037895350885247
+      "required_token_positions_per_second": 28788.02110534787,
+      "required_pages_per_second_upper_estimate": 0.3692618245705913
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.12191876294536996,
-    "producer_p95_seconds": 0.35151034699811134,
-    "gpu_update_seconds": 0.9575027266823557,
-    "producer_batches_per_second": 8.20218296053484,
-    "gpu_batches_per_second": 1.0443834488752766,
-    "producer_headroom": 7.853612549459666,
+    "producer_batch_seconds": 0.11866160643745616,
+    "producer_p95_seconds": 0.3584676699974807,
+    "gpu_update_seconds": 0.9603994626384342,
+    "producer_batches_per_second": 8.427325653366047,
+    "gpu_batches_per_second": 1.0412334022478253,
+    "producer_headroom": 8.093599028971843,
     "sustainable": true,
-    "buffer_seconds": 60.32267178098841,
+    "buffer_seconds": 60.505166146221356,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 359469.504044859,
-    "upload_bytes_per_second": 5065432.420462855,
+    "generation_bytes_per_second": 360671.5076130559,
+    "upload_bytes_per_second": 5192570.242931546,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.13378754624678,
+    "estimated_full_shards_per_hour": 77.39170953077085,
     "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": 4.102204398001049,
-      "bytes": 6262944,
-      "pages": 38,
-      "shards": 2
-    },
     {
       "seconds": 1.1364409210000304,
       "bytes": 7535674,
@@ -562,6 +556,12 @@
       "bytes": 18015559,
       "pages": 40,
       "shards": 5
+    },
+    {
+      "seconds": 0.9437920670025051,
+      "bytes": 911442,
+      "pages": 27,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -390,3 +390,4 @@
 {"step": 436100, "loss": 0.5766322016716003, "lr": 0.0003, "grad_norm": 0.15279914796720345, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63832285536, "tokens_per_second": 151045.89412842927, "data_wait_s": 2.7783004043158144e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 436200, "loss": 0.5508224964141846, "lr": 0.0003, "grad_norm": 0.15514427238020523, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63846978055, "tokens_per_second": 136587.17837558212, "data_wait_s": 1.2106000212952495e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 436300, "loss": 0.6405804753303528, "lr": 0.0003, "grad_norm": 0.1689496082436297, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63861577180, "tokens_per_second": 150784.97162904218, "data_wait_s": 8.994997187983245e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 436400, "loss": 0.6233225464820862, "lr": 0.0003, "grad_norm": 0.17652776534212125, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63876259565, "tokens_per_second": 151066.72195197208, "data_wait_s": 1.0071998985949904e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c220f194f2f2e08d357eacd0eb3a35b467c8eb0df5855eb17b167f65c8b3e053
-size 836685
+oid sha256:70d00f5c02e4d33304f6213fa2338003ce2111fa3cc2d6858e3f16a6715c7edd
+size 838803