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": 77264.19359887656,
-      "required_pages_per_second_upper_estimate": 0.005345316190799067
+      "required_token_positions_per_second": 76893.9247977905,
+      "required_pages_per_second_upper_estimate": 0.005319700135997948
     },
     {
       "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": 9658.02419985957,
-      "required_pages_per_second_upper_estimate": 0.00404361305000183
+      "required_token_positions_per_second": 9611.740599723813,
+      "required_pages_per_second_upper_estimate": 0.004024235073136465
     },
     {
       "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": 19316.04839971914,
-      "required_pages_per_second_upper_estimate": 0.004023023483209634
+      "required_token_positions_per_second": 19223.481199447626,
+      "required_pages_per_second_upper_estimate": 0.004003744176554309
     },
     {
       "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": 28974.07259957871,
-      "required_pages_per_second_upper_estimate": 0.0017369514114975123
+      "required_token_positions_per_second": 28835.221799171435,
+      "required_pages_per_second_upper_estimate": 0.0017286275180259925
     },
     {
       "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": 28974.07259957871,
-      "required_pages_per_second_upper_estimate": 0.002252606701140521
+      "required_token_positions_per_second": 28835.221799171435,
+      "required_pages_per_second_upper_estimate": 0.0022418116621489813
     },
     {
       "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": 28974.07259957871,
-      "required_pages_per_second_upper_estimate": 0.37164829337205413
+      "required_token_positions_per_second": 28835.221799171435,
+      "required_pages_per_second_upper_estimate": 0.36986726439080353
     }
   ],
   "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": 0.09408797603893504,
-    "producer_p95_seconds": 0.2630074289991171,
-    "gpu_update_seconds": 0.9542324402266462,
-    "producer_batches_per_second": 10.628350636283054,
-    "gpu_batches_per_second": 1.0479626952972625,
-    "producer_headroom": 10.141916963244807,
+    "producer_batch_seconds": 0.093044692616985,
+    "producer_p95_seconds": 0.2682080099984887,
+    "gpu_update_seconds": 0.958827374124601,
+    "producer_batches_per_second": 10.747523280198932,
+    "gpu_batches_per_second": 1.0429406032686428,
+    "producer_headroom": 10.30501952509616,
     "sustainable": true,
-    "buffer_seconds": 60.11664373427871,
+    "buffer_seconds": 60.40612456984986,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 286602.7173367766,
-    "upload_bytes_per_second": 3913889.079196053,
+    "generation_bytes_per_second": 327307.6379312962,
+    "upload_bytes_per_second": 3902646.1585181947,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.49827137067293,
+    "estimated_full_shards_per_hour": 70.23259976820148,
     "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": 1.2274471419987094,
-      "bytes": 6988556,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.103131982999912,
       "bytes": 7144691,
@@ -562,6 +556,12 @@
       "bytes": 6995458,
       "pages": 46,
       "shards": 3
+    },
+    {
+      "seconds": 1.261231005002628,
+      "bytes": 5881548,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -294,3 +294,4 @@
 {"step": 554400, "loss": 0.5891408920288086, "lr": 0.0003, "grad_norm": 0.15826079270071136, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81146437215, "tokens_per_second": 153105.11426150508, "data_wait_s": 1.3337001291802153e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 554500, "loss": 0.6127526760101318, "lr": 0.0003, "grad_norm": 0.16399397993627016, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81161096845, "tokens_per_second": 153161.7267546149, "data_wait_s": 1.1164000170538202e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 554600, "loss": 0.5827692747116089, "lr": 0.0003, "grad_norm": 0.14074798741976163, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81175624209, "tokens_per_second": 151229.45465750332, "data_wait_s": 1.5508001524722204e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 554700, "loss": 0.6151465177536011, "lr": 0.0003, "grad_norm": 0.17316984258980894, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81190191635, "tokens_per_second": 144846.02677313812, "data_wait_s": 7.919999916339293e-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:d910fc7e2ddd5e71a5d970f641a2a227e360922369a9c1491ff4f4a9d3a9f3b7
-size 631008
+oid sha256:229cf0433aeef8457ffaf2792b0d95f829bdea1bfabb537ff51f587bf291942b
+size 633126