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": 76553.82654368595,
-      "required_pages_per_second_upper_estimate": 0.005296171349642324
+      "required_token_positions_per_second": 77078.6105657682,
+      "required_pages_per_second_upper_estimate": 0.005332477125956667
     },
     {
       "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": 9569.228317960744,
-      "required_pages_per_second_upper_estimate": 0.004006436068519661
+      "required_token_positions_per_second": 9634.826320721026,
+      "required_pages_per_second_upper_estimate": 0.004033900582433324
     },
     {
       "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": 19138.456635921488,
-      "required_pages_per_second_upper_estimate": 0.00398603580222034
+      "required_token_positions_per_second": 19269.65264144205,
+      "required_pages_per_second_upper_estimate": 0.004013360470298942
     },
     {
       "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": 28707.68495388223,
-      "required_pages_per_second_upper_estimate": 0.0017209818788883917
+      "required_token_positions_per_second": 28904.478962163077,
+      "required_pages_per_second_upper_estimate": 0.0017327793791977765
     },
     {
       "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": 28707.68495388223,
-      "required_pages_per_second_upper_estimate": 0.0022318962334030427
+      "required_token_positions_per_second": 28904.478962163077,
+      "required_pages_per_second_upper_estimate": 0.002247196101941516
     },
     {
       "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": 28707.68495388223,
-      "required_pages_per_second_upper_estimate": 0.3682313586778291
+      "required_token_positions_per_second": 28904.478962163077,
+      "required_pages_per_second_upper_estimate": 0.3707556209151124
     }
   ],
   "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": 0.07934976096871083,
-    "producer_p95_seconds": 0.20891952500096522,
-    "gpu_update_seconds": 0.9630870634262367,
-    "producer_batches_per_second": 12.60243241809285,
-    "gpu_batches_per_second": 1.038327725473171,
-    "producer_headroom": 12.13723962956865,
+    "producer_batch_seconds": 0.07814268122648116,
+    "producer_p95_seconds": 0.20523150699955295,
+    "gpu_update_seconds": 0.956529956350092,
+    "producer_batches_per_second": 12.797103763328739,
+    "gpu_batches_per_second": 1.0454455643143474,
+    "producer_headroom": 12.240813104144436,
     "sustainable": true,
-    "buffer_seconds": 60.674484995852914,
+    "buffer_seconds": 60.2613872500558,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 75,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 420081.26457692706,
-    "upload_bytes_per_second": 3417935.965116692,
+    "generation_bytes_per_second": 405273.616012472,
+    "upload_bytes_per_second": 3460226.9915787955,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.13966038685665,
+    "estimated_full_shards_per_hour": 86.96228370934125,
     "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": 4.260054792999654,
-      "bytes": 10427634,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 1.4442174710002291,
       "bytes": 7300139,
@@ -562,6 +556,12 @@
       "bytes": 8234938,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.4250404260001233,
+      "bytes": 6072497,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -125,3 +125,4 @@
 {"step": 937500, "loss": 0.5770094990730286, "lr": 3e-05, "grad_norm": 0.3401494716194435, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137223151794, "tokens_per_second": 147682.17443085133, "data_wait_s": 1.4368999472935684e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 937600, "loss": 0.5389680862426758, "lr": 3e-05, "grad_norm": 0.3437692404653344, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137237949541, "tokens_per_second": 154653.54816237182, "data_wait_s": 6.243999450816773e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 937700, "loss": 0.5223715901374817, "lr": 3e-05, "grad_norm": 0.33024077437116944, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137252705765, "tokens_per_second": 154431.41308355043, "data_wait_s": 1.3341999874683097e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 937800, "loss": 0.5168602466583252, "lr": 3e-05, "grad_norm": 0.33473344251680687, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137267246678, "tokens_per_second": 152357.613139161, "data_wait_s": 2.5258001187467016e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:900aeb4aa9bca15200a292f1e02a5113842fbbda39ccca03705146fc8b14f4dc
-size 270717
+oid sha256:6ce7c54ae02fe67350777e528f9ae0546a6694ee4b3b4876a182fcd6ec949ca0
+size 272835