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": 77016.43071815216,
-      "required_pages_per_second_upper_estimate": 0.005328175379821465
+      "required_token_positions_per_second": 76667.01460214957,
+      "required_pages_per_second_upper_estimate": 0.005304001962159317
     },
     {
       "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": 9627.05383976902,
-      "required_pages_per_second_upper_estimate": 0.004030646406966577
+      "required_token_positions_per_second": 9583.376825268697,
+      "required_pages_per_second_upper_estimate": 0.004012359753075071
     },
     {
       "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": 19254.10767953804,
-      "required_pages_per_second_upper_estimate": 0.004010122864682549
+      "required_token_positions_per_second": 19166.753650537394,
+      "required_pages_per_second_upper_estimate": 0.003991929324122376
     },
     {
       "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": 28881.16151930706,
-      "required_pages_per_second_upper_estimate": 0.0017313815341022322
+      "required_token_positions_per_second": 28750.13047580609,
+      "required_pages_per_second_upper_estimate": 0.0017235264231171678
     },
     {
       "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": 28881.16151930706,
-      "required_pages_per_second_upper_estimate": 0.0022453832733220526
+      "required_token_positions_per_second": 28750.13047580609,
+      "required_pages_per_second_upper_estimate": 0.0022351961860345025
     },
     {
       "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": 28881.16151930706,
-      "required_pages_per_second_upper_estimate": 0.37045652979447496
+      "required_token_positions_per_second": 28750.13047580609,
+      "required_pages_per_second_upper_estimate": 0.36877580425861767
     }
   ],
   "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": 0.26335356919528863,
-    "producer_p95_seconds": 0.27975313900242327,
-    "gpu_update_seconds": 0.9573022186631001,
-    "producer_batches_per_second": 3.797176560225218,
-    "gpu_batches_per_second": 1.0446021961554928,
-    "producer_headroom": 3.6350455457591195,
+    "producer_batch_seconds": 0.3406237537342065,
+    "producer_p95_seconds": 0.25486802700106637,
+    "gpu_update_seconds": 0.9616652009028773,
+    "producer_batches_per_second": 2.9357905578725845,
+    "gpu_batches_per_second": 1.0398629367696068,
+    "producer_headroom": 2.8232476166453093,
     "sustainable": true,
-    "buffer_seconds": 60.3100397757753,
+    "buffer_seconds": 60.58490765688127,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 474833.3444772473,
-    "upload_bytes_per_second": 3759097.5559630464,
+    "generation_bytes_per_second": 428357.4482073569,
+    "upload_bytes_per_second": 3812333.0122269276,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 101.88818217027725,
+    "estimated_full_shards_per_hour": 91.91553673425227,
     "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": 1.001895314999274,
-      "bytes": 6152237,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 0.8757686259996262,
       "bytes": 2112854,
@@ -562,6 +556,12 @@
       "bytes": 1201136,
       "pages": 33,
       "shards": 1
+    },
+    {
+      "seconds": 1.2566547139977047,
+      "bytes": 13157311,
+      "pages": 30,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -251,3 +251,4 @@
 {"step": 550100, "loss": 0.5845863819122314, "lr": 0.0003, "grad_norm": 0.1454932906480027, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80517419446, "tokens_per_second": 133977.610676426, "data_wait_s": 1.0357001883676276e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 550200, "loss": 0.5769320130348206, "lr": 0.0003, "grad_norm": 0.2435106766307711, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80531903113, "tokens_per_second": 150930.9935901468, "data_wait_s": 1.0010000551119447e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 550300, "loss": 0.5711182355880737, "lr": 0.0003, "grad_norm": 0.1465889956531403, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80546357747, "tokens_per_second": 150211.515778511, "data_wait_s": 1.018900002236478e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 550400, "loss": 0.6104316711425781, "lr": 0.0003, "grad_norm": 0.16719218383811282, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80561022373, "tokens_per_second": 152615.41820878926, "data_wait_s": 1.1588999768719077e-05, "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 sha256:7360dbaf3f5f92b6a7adf541d08127cda93a3d0f11508fccc76cab87034e0c96
-size 539934
+oid sha256:3168b908634b5804f811307b5519a4121345662fc1b5f551460ba44e64a558ff
+size 542052